In recent years there has been great interest in reducing fossil fuel reliance. This comes in an attempt for countries to deliver on pledges outlined in the Paris Agreement of 2016, which was compiled to tackle climate change by lowering greenhouse gas emissions (GHG) (1, 2). Individual nationally determined contributions (NDCs) are central to this agreement, detailing the collective efforts required to achieve longer-term global aims. Within these NDCs, several countries made specific references to the increased use of renewable energy resources, given their widely recognised environmental advantages over more traditional fossil fuel equivalents. Now, atmospheric CO₂ levels are higher than any recorded in the previous 800,000 years, with a rise from 300 ppm to 400 ppm being recorded over the last 70 years (1950–2020) (3). These continually rising levels are commonly attributed to the increased consumption of fossil fuels, which are conveniently employed to satisfy ever-changing energy demands. Yet the consequences are serious: during the 20th century a 1°C increase in average global temperature accompanied the growth of CO₂ emissions, causing drastic changes in weather patterns and rising sea-levels (4, 5).

Fossil fuels are extremely widely used because of convenience, availability and the economic advantage of a lower initial capital expenditure. Developing a ‘greener’ system with a comparable energy density and transport efficiency is a challenge (6). Despite several years of research into alternatives, there is still a significant way to go before achieving the goals proposed in the Paris Agreement by 2050. In 2016 just 19.5% of the global energy demand was fulfilled by renewable sources (7). Germany, however, has pledged that by 2050, 80% of its energy will be produced from renewable sources, whereas other countries have implemented strategies to increase collaborative research to achieve net-zero emissions (7, 8). As an example, Portugal and The Netherlands signed a memorandum of understanding to develop a strategic export-import value chain to ensure production and transport of green hydrogen from Portugal to The Netherlands and its hinterland via the ports of Sines and Rotterdam (9).

Hydrogen is a well-studied alternative to fossil fuels. With its combustion producing only water as a byproduct, the environmental advantages of employing hydrogen as an energy carrier are obvious. However, at present, around 96% of the annual global hydrogen production is generated from fossil fuels (grey or brown hydrogen) (Table I). This comprises steam reforming of methane (48%), reforming of oil/naphtha (30%) and coal gasification (18%) (11). Thus, in order to completely remove reliance on the non-renewable, finite, fossil fuel resources, an alternative method of hydrogen production is required. This can be achieved with the electrolysis of water. Provided the energy for this process is obtained from renewable sources, sustainable, greenhouse gas emission-free hydrogen production is possible. Hence, hydrogen produced in this manner is termed green hydrogen.

The European Union (EU) has pledged billions of Euros to develop the so-called European Green Deal in the coming years. Here, hydrogen is considered ‘a key’ to fulfilling the ambitious target of halving carbon emissions by 2030 (8). To achieve this, the European Green Deal acknowledges the need to increase green hydrogen production capacity, setting targets for around 10 million tonnes of hydrogen to be produced annually by 2030 (8). Currently, the hydrogen supplied from electrolysers stands at just 4% of demand (8).

A third colour-coded category of hydrogen refers to blue hydrogen (Table I). Blue hydrogen is often considered a vital tool in transitioning from grey to green hydrogen production and, like grey hydrogen, uses a fossil fuel feedstock. However, carbon capture and storage (CCS) technologies are also employed to capture CO₂, reducing the greenhouse gas emissions (12). Given the increase in carbon-tax expected over the coming decades, blue hydrogen is an important improvement upon grey hydrogen, despite the currently higher initial capital expenditure (12–14). Brown, grey, blue and green hydrogen (Table I) are the most discussed colour-coded categories of hydrogen within the energy industry. Nevertheless, countless other hydrogen colours (i.e. the hydrogen colour spectrum), such as yellow, pink and turquoise hydrogen, also exist with each colour code describing the different type of source or process used to produce hydrogen (15, 16). For instance: pink hydrogen is generated by electrolysis of water using electricity from a nuclear power plant; yellow hydrogen is produced via electrolysis using solar power; turquoise hydrogen is generated via methane pyrolysis through direct splitting of methane into hydrogen and solid carbon (15, 16).

Some countries are well-positioned to the transition to blue hydrogen production, with a potential access to large, offshore CCS facilities (17). In contrast, many landlocked countries in Europe have a greater focus on the transition to green, rather than blue, hydrogen due to their limited access to offshore facilities (18). In addition, with blue hydrogen still requiring a finite resource, the demand of one country may eventually be pushed onto another, where resources are more available (12, 19). For example, an exhaustion of natural gas supplies in one location would require a country to source this elsewhere, resulting in the production of hydrogen and its utilisation at different locations. This dependence (expected to be governed by maintaining amicable international relations) is another factor causing several countries to seriously consider the favourability of blue over green hydrogen, or vice versa (12).

Producing hydrogen from renewable sources and development of technologies for this purpose have huge environmental benefits and supports the implementation of the Paris Agreement and the United Nations Sustainable Development Goals (SDG) (20). It contributes to multiple SDGs, such as SDG 7 (Affordable and Clean Energy), SDG 9 (Industry, Innovation and Infrastructure), SDG 11 (Sustainable Cities and Communities) and SDG 13 (Climate Action) (20). According to SDG 7: Affordable and Clean Energy, a substantial increase in the share of renewable energy in the global energy mix is required by 2030 (i.e. Target 7.2 of SDG 7) (21). However, the energy output from renewable sources is intermittent and dependent on geographic, seasonal and temporal factors. Furthermore, sites of highest energy potential for renewable hydrogen production (such as a desert, offshore wind and tidal farms) are rarely located in areas of highest energy demand, such as densely populated cities in central and southern Europe (1, 22). When moving towards a global low carbon hydrogen economy with the aim of meeting net-zero climate goals, a reliable storage and transportation of hydrogen at scale is a challenge which needs to be tackled to achieve a widespread usage of renewable hydrogen. The possibility to store and transport hydrogen is essential for the integration of high shares of renewable energy source with positive effects on SDGs (23).

Numerous technologies and options are currently being explored for effective hydrogen storage and transportation to facilitate a smooth transition to the hydrogen economy (24). LOHC is one such technology which has gained considerable attention in recent years (1, 6, 7, 25–27). In our work, which consists of the present paper and two accompanying papers (28, 29), we provide an overview and new perspectives on the LOHC technology among different hydrogen storage and transportation technologies. We analyse the advantages and disadvantages of the LOHC technology and future considerations for its optimisation which might accelerate its commercial deployment. Furthermore, in our following second paper we describe the potential deployment and integration of LOHCs within different industries: the transportation sector (automobiles, ships, trains); steel and cement industries; the use of stored hydrogen to produce fuels and chemicals from flue gases; and system integration of fuel cells and LOHCs for energy storage (28). Due to numerous possibilities for the commercial deployment and integration of LOHCs within different industries, the use of different LOHC systems might be considered to accommodate specific requirements. A review of the most prominent LOHC systems, focusing on properties of LOHCs and catalytic materials used for hydrogenation and dehydrogenation of LOHCs, is presented in our third paper dedicated to the analysis of LOHC systems (29).

Despite the attractively high gravimetric energy density of hydrogen (120 MJ kg^–1), the low volumetric energy density at ambient conditions necessitates the use of pressurised or liquified hydrogen to ensure economic viability (30). On a large-scale, the use of stored, highly pressurised hydrogen in transport systems presents serious safety concerns, such as an explosion. The technology is also costly, much like that of liquified hydrogen storage, which requires low temperatures (–252°C). In addition, such liquified hydrogen technologies can result in losses of between 0.3% and 3% of the hydrogen due to boil-off from the storage system (1).

Compressed hydrogen is typically transported through pipelines. As a result, long-distance transport is significantly more challenging when compared to the infrastructure, such as ships or railroads, currently available to transport liquid fossil fuels. Although attempts have been made to inject hydrogen into the natural gas grid and numerous countries are exploring blending hydrogen with natural gas, it can be argued that the intercontinental transport of hydrogen would require innovative solutions (31–33). These solutions are predicted to resemble current methods, relying on tanker vessels, trucks and railroads.

To overcome the challenges raised with the transportation and storage of hydrogen as an energy vector, more sophisticated concepts have been developed (Figure 1) (24). One such concept involves the use of low- and high-temperature metal hydride systems, where reversible adsorption and desorption of hydrogen is permitted through formation and breakage of chemical bonds with the storage material (1, 11, 34). Here, the terms low- and high-temperature systems refer to the dehydrogenation reaction temperature. These methods of solid-state hydrogen storage, therefore, do not require the demanding pressures associated with storing compressed hydrogen.

Fig. 1.

Low temperature metal hydrides are capable of releasing hydrogen at ambient temperatures and pressures but are typically limited to a maximum hydrogen capacity of 2.15 wt% (for example, LaNi₅H₆) (35). On the contrary, high temperature hydrides are attractive for hydrogen storage because of their high hydrogen storage capacities (6.5 hydrogen atoms per cm³ (7.6 wt%) for the metal hydride MgH₂, compared to a hydrogen storage density of 0.99 hydrogen atoms per cm³ for hydrogen gas and 4.2 hydrogen atoms per cm³ for liquified hydrogen). However, disadvantages of high temperature metal hydride systems include slow reaction kinetics of the adsorption-desorption process. Moreover, the high desorption enthalpies of these systems require high temperatures (300°C for MgH₂) for hydrogen release at atmospheric pressure, raising questions around the economic viability of the technology (34–36). The reversibility of the process is also limited by the decomposition of the metal hydride, making a regular replacement of this storage material necessary.

A different concept for solid-state hydrogen storage is the use of metal-organic frameworks (MOFs). Here, hydrogen is stored in the pores of the MOF framework, where the rate of hydrogen adsorption is dependent on its diffusivity in the particular MOF chosen (37). MOFs can be tailored to specific applications, including supercapacitors, fuel storage and batteries and have been noted for their short refuelling times (37). However, this technology typically requires low temperatures (ca. –196°C) and high pressures (100 bar) to achieve a reasonable energy density (7.2 MJ l^–1, 4.5 wt%). To illustrate this, the MOF displaying the highest hydrogen storage capacity to date (MOF‐210: total hydrogen gravimetric uptake of 17.6 wt%) must be operated at 80 bar and –196°C (38). It is therefore possible that a low system efficiency in terms of energy consumption, and thus cost, will be observed. It has been predicted that the hydrogen storage capacity of MOFs can be improved by increasing MOF surface area, but there are experimental limitations in synthesising such structures (38). Moreover, as with using compressed hydrogen, the requirement of a higher-pressure system implicates explosion risks (37).

Borohydrides have also been considered for hydrogen storage applications, given their attractively high theoretical storage capacities. For example, LiBH₄ has a gravimetric hydrogen storage capacity of up to 18.5 wt% (24). However, in practice this is rarely achievable as stable hydrides, such as LiH, can form during the hydrogen unloading process, for which high temperatures (typically greater than 300°C) are also required. Direct employment of borohydrides for hydrogen storage applications is therefore unfeasible. Yet, a technology combining both borohydrides and metal hydrides is now being considered. Such a system reduces the endothermicity of the dehydrogenation process, facilitating hydrogen release at lower temperatures, but the kinetics of both the hydrogenation and dehydrogenation reactions are slow and require suitable catalytic additives (24). This increases the complexity of the technology.

Chemical hydrides, both organic and inorganic, are another method available to store hydrogen, and rely on chemical reactions. Chemical hydrogen storage is used to describe storage technologies in which hydrogen is both released and restored through a chemical reaction. Like metal hydrides, chemical hydrides also form chemical bonds between the storage material and hydrogen. Yet, these two hydride classes have very different properties. Arguably the most significant of these is that chemical hydrides (such as methanol) are generally liquids at ambient conditions. This simplifies transportation and storage issues associated with either gaseous hydrogen or solid-state hydrides. Provided ammonia is in its liquid form, which can easily be achieved by applying a pressure of 10 bar at ambient temperature, ammonia can also be considered a chemical hydride, with a very attractive storage density (17.7 wt%) (24). In addition, given both methanol and ammonia are already synthesised on a large scale, there is the possibility of using existing infrastructure and production plants (24). Both chemicals have also been reported to have potential as hydrogen-alternative energy vectors (i.e. direct use as a fuel) rather than as hydrogen storage materials (24).

Ammonia is currently of industrial interest for hydrogen storage and transportation, given its high gravimetric hydrogen density (17.7 wt%) (24, 39–42). Additionally, the high ignition temperature of ammonia increases the safety of the technology and is considered advantageous (41, 42). When green hydrogen is used for the production of ammonia, such ammonia is classified as green ammonia (43–46). Green ammonia, when liquified, facilitates the transport and storage of hydrogen, allowing existing infrastructure to be used (42). With the potential for worldwide transport of the green ammonia, for example via ships, the distribution of green hydrogen to growing zero-emission markets will be facilitated (42). Once transported to the site of use, ammonia can be reconverted into carbon-free hydrogen which can be used at hydrogen refuelling stations, for example (42, 47).

Within the framework of the NEOM project it was recently announced that 650 tons of carbon-free hydrogen per day will be produced using 4 GW of renewable power from solar and wind in Saudi Arabia (48). To facilitate hydrogen storage and transportation, the hydrogen will be converted into 3500 tons of green ammonia per day (or 1.2 million tons per year) which will be transported around the world and then converted back into carbon-free hydrogen at hydrogen refuelling stations (48). By supplying the fuel cells currently used within the transport sector (specifically in buses and trucks) with this hydrogen, it is predicted that over 3 million tons per year of CO₂ emission can be prevented: equivalent to all emissions from 700,000 cars (48).

Although ammonia is less flammable than hydrogen, concerns around its toxicity to both humans and the environment have been raised (49). A spillage of this feedstock (for example during transport) could therefore have serious consequences, and questions around its suitability have been raised. However, it is important to remember that suitable controls have been employed to mitigate the risks associated with fossil fuels (in the forms of gasoline and diesel), which are also very harmful substances. Moreover, as ammonia is already synthesised on an industrial scale (Haber-Bosch process) for example, in the fertiliser industry, it can be argued that the safety concerns are known and can be managed effectively. If the ammonia were not to be dehydrogenated at its destination, but instead used as a fuel itself, the ammonia would be classified as a renewable fuel (50). As an example, using ammonia as a transportation marine fuel is currently being explored within the shipping sector to cut fossil fuel use in ocean-going vessels (51–55).

Liquid hydrocarbons and formic acid can also fall into the category of hydrogen carriers. Provided such carriers are produced using green hydrogen and atmospheric CO₂, or CO₂ from waste streams, the cycle can be labelled as carbon-neutral (11). However, as the carriers are used as liquid fuels, regeneration of the carrier material is not possible and thus new material must be purchased for every cycle, much like employing fossil fuels as energy vectors (11).

As previously discussed, renewable methanol can also be considered as a chemical hydride. Although the gravimetric energy density of methanol is lower than that of ammonia (12.5 wt% and 17.7 wt%, respectively), it is significantly higher than a typical metal hydride, such as MgH₂ (7.6 wt%, Table II) (24, 37). Most commonly, methanol is synthesised via the hydrogenation of CO₂ and carbon monoxide, whereas the release of hydrogen from methanol is done in the methanol steam reforming process, which involves the reaction of methanol with water (24). This process is generally preferred over methanol decomposition, permitting the release of three moles of hydrogen in comparison to the two moles from methanol decomposition, where the extra mole of hydrogen is provided by the water (24). Yet for the steam reforming reaction, temperatures of between 220°C and 330°C are often required to meet the thermodynamic demands of the endothermic reaction (24).

Although renewable methanol (produced via the hydrogenation of CO₂ waste streams) is produced on a much smaller scale than its non-renewable equivalent (in which natural gas is used to produce a mixture of carbon monoxide, CO₂ and hydrogen), similarities in the two processes exist (24). With an overlap in the technology, progress in renewable methanol production is better facilitated than other hydrogen storage technologies and as a result the first renewable methanol production plant was constructed in Iceland in 2011 by Carbon Recycling International (24). Interestingly, it has been reported that the separation of methanol and water (via distillation) is not required when using methanol as a hydrogen storage medium: the hydrogen can simply be released from the mixture in a steam-reforming reaction (24). This simplifies the process and eliminates costs associated with the energy intensive distillation step. However, CO₂ is also stored within the methanol-water mixture and hence is also released upon steam reforming. If a pure hydrogen output stream is required, it has been predicted that CO₂ could be separated from the hydrogen relatively easily, but this process would require additional separation technologies (24). Alternatively, the gaseous hydrogen and CO₂ mixture could be used directly within proton exchange membrane (PEM) fuel cells (56).

LOHCs, also categorised under chemical hydrides, are another option for the storage and transport of hydrogen. The first studies into this technology were completed in the 1980s by Japanese researchers, studying a benzene/cyclohexane system (1). The LOHC process comprises a two-step process, which is based on the loading of hydrogen onto the chosen LOHC in a catalytic hydrogenation reaction, followed by the unloading of hydrogen in a catalytic dehydrogenation reaction. This second step thus produces a stream of gaseous hydrogen alongside the unloaded form of the LOHC, which can then be reused in subsequent cycles (Figure 2). Between these two steps, the hydrogen-rich form of the LOHC can be easily stored and transported at ambient pressures, given its liquid state. The LOHC technology eliminates the expense associated with repeatedly purchasing a feedstock (i.e. the LOHC) and hence may be considered advantageous. Yet one must also consider the expense of returning the unloaded LOHC to the hydrogenation plant. In some cases, finding an alternative use for the unloaded LOHC may be most cost-effective.

Fig. 2.

Aromatic molecules are typically used as LOHCs due to their high hydrogen loading capacities (57). In addition, the cyclic compounds hydrogenated form of aromatic compounds have relatively good thermodynamic properties for the more challenging, endothermic dehydrogenation reaction. This can be explained by the stability gained on formation of the aromatic system (58). In contrast, dehydrogenation of bonds outside of such an aromatic system is difficult (even if the double bond formed can become conjugated with the system). For instance, dehydrogenation of ethylcyclohexane would form ethylbenzene not styrene as a result of thermodynamic limitations (58).

Importantly, the high hydrogen loading capacities of LOHCs enable high hydrogen storage and transport efficiencies, increasing the economic viability of the technology. This is highlighted by previous research that identified several potential carriers which meet the objectives for storage capacity and volumetric energy density, set by the United States Department of Energy (US DoE), as 6.5 wt% and 1.7 kWh l^–1, respectively (7). However, it is important to consider that the respective hydrogenation and dehydrogenation reactions may not always go to 100% completion.

As noted above, catalysts are required to facilitate hydrogen loading and unloading via hydrogenation and dehydrogenation reactions. Despite some reports detailing advantages of homogeneous catalysts (including lower operating temperatures and improved dehydrogenation product specificity), heterogeneous catalysts are considered preferable for both reactions in large-scale applications (26, 59).

A great majority of current studies examine the hydrogenation and dehydrogenation steps individually, often employing different catalysts to complete the two transformations. Most commonly, platinum group metal (pgm) catalysts, namely platinum, palladium or ruthenium-based, are employed (7, 30, 60, 61).

The LOHC technology is also an attractive option for stationary on-site energy storage (on-grid and off-grid), enabling long-term storage of large amounts of energy, such as seasonal storage and energy production buffering (26). In this case, a catalyst which allows both hydrogenation and dehydrogenation to be carried out in the same reactor through altering process conditions (such as temperature and pressure) can be deployed. The development of such a bifunctional catalyst would be an attractive research objective and facilitate a lower capital expenditure since a set-up in which only a single reactor and its associated pipework is necessary. From a process safety perspective, the ambient hydrogen storage pressures facilitated with the use of LOHC technology are an improvement on using a pressurised hydrogen storage tank.

Since the catalytic hydrogenation and dehydrogenation processes are exothermic and endothermic reactions, respectively, the hydrogenation step is typically performed at lower temperatures (100–240°C) and higher pressures (10–50 bar) than the dehydrogenation step (150–400°C, atmospheric pressure) (62). Thus, a higher level of heat is needed for the dehydrogenation reaction, which is often provided by an external heating source (62). To improve efficiency, system integration and intensification can be achieved, with the heat produced in the hydrogenation step being used to drive the dehydrogenation reaction when using LOHCs for on-site hydrogen storage (63). This would not be possible when the hydrogen is produced in a different location to where it is needed.

As discussed, a timely migration away from fossil fuel reliance is of the utmost environmental importance, and is achievable with a combination of renewable energy, production of green hydrogen via electrolysis and hydrogen storage and transportation technologies. However, implementation of a LOHC hydrogen storage and transportation system also has advantages over simply using compressed hydrogen, or direct employment of renewable electricity. On the contrary to storage and transportation of compressed hydrogen, the small quantity of gas present in a LOHC system minimises the risk of explosion and facilitates the safe handling of hydrogen. Importantly, this allows for the long-term storage of large amounts of loaded LOHCs at ambient conditions, without the loss of hydrogen or negative impact upon carrier storage density, which has been reported with the use of compressed hydrogen and alternative storage technologies such as metal hydrides (64).

The elimination of pressure-related hazards also makes LOHCs suitable for long-distance transport. The liquid state of the carriers enables existing infrastructure, originally built for crude-oil transportation such as pipeline networks, to be used (1). However, unlike liquified or compressed hydrogen, the term ‘infrastructure’ also includes the use of ships and trucks, which enables worldwide transport of hydrogen using LOHCs. This too is often considered as one of the main advantages of the LOHC technology, meaning the dehydrogenation plant required to release hydrogen from LOHCs does not need to be located within proximity of the hydrogen production site. Thus, the LOHC technology allows green hydrogen release (and indirect renewable energy use) in locations which are not best suited for renewable energy production. Moreover, as LOHCs are in the liquid state, they resemble current fossil fuel-based energy vectors, such as diesel and gasoline. Given societal familiarity with these systems, it is predicted that public acceptance of the LOHC technology will be increased in comparison to concepts such as metal hydrides, where public understanding is limited (7). This potential for greater acceptance of LOHC systems is considered advantageous (7).

In theory, the reversibility of the hydrogen loading and unloading processes allows LOHC to be used continually without replacement. Practically, however, this is unlikely due to LOHC material losses from side-reactions and incomplete unloading reactions (1). The choice of LOHC is also critical: for maximum efficiency, both the dehydrogenated and hydrogenated form must remain in their liquid state throughout the cycle. For instance, if the dehydrogenated form of the carrier material is a solid at ambient conditions, transport complications arise as it cannot be pumped through a pipe or into a truck or ship. To maintain the liquid state, incomplete hydrogen unloading or dilution of the LOHC would be required, reducing storage and transport efficiencies (7).

Moreover, the toxicity of the LOHCs themselves must be considered and evaluated in terms of projected applications; some LOHCs, such as benzene and toluene, have a toxicity so great that their use in practical applications is unfeasible, despite attractively high hydrogen storage capacities (65).

The choice of carrier can also influence the overall cost of the technology in other ways. Aside from the obvious cost associated with purchasing the LOHC feedstock, the released hydrogen from some carriers, like 1,2-dihydro-1,2-diazaborine, requires further hydrogen purification steps, while others (for example, dibenzyltoluene) have higher dehydrogenation heating demands (1). Both factors increase energy consumption. Moreover, the cost of transporting the unloaded LOHC via ships back to the site of hydrogenation (i.e. for hydrogen loading) should be taken into account. Mainly long-distance transportation of renewable hydrogen in LOHCs would be more economically viable than transportation of compressed hydrogen, which is more suited to transport over short distances using the existing pipeline infrastructure (25). Methanol has also been reported to be economical for long-distance transport, with some studies suggesting that methanol can be a more cost-effective option than LOHCs (66). The same applies to ammonia (41). The Committee on Climate Change has reported that using ammonia as a hydrogen carrier for long-distance transport could be economically viable (25, 67). One of the key disadvantages of the LOHC technology is additional costs required for transporting of the LOHC loaded with hydrogen to end users followed by a transport of the dehydrogenated LOHC back to a chemical plant for loading with hydrogen.

To improve economic viability of the LOHC technology, additional developments into optimising the LOHC technology might be needed. Reducing energy intensity during loading and particularly unloading of the LOHCs with hydrogen and efficient system integration can contribute to the cost reduction of the LOHC technology. In our following work, we discuss efficient system integration and potential deployment of the LOHC technology within different industries (28) and provide a detailed analysis of the most promising LOHC candidates, catalysts used for hydrogenation as well as dehydrogenation of LOHCs along with operating conditions (29). Efficient system integration of the LOHC technology as well as selection of the most suitable LOHC system, optimisation of reaction conditions and catalysts might improve economic viability and facilitate widespread commercial deployment of the LOHC technology.

Reactor configuration can also contribute significantly to the overall efficiency of the LOHC technology. The particular challenge for the LOHC technology is the hydrogen release from hydrogen-rich LOHC systems during the endothermic dehydrogenation process which is combined with the reaction mixture volume expansion. As an example, 1 ml of fully hydrogenated dibenzyltoluene can release more than 650 ml of hydrogen (68). A range of reactor types have been proposed for dehydrogenation of LOHCs, such as fixed-bed, continuous stirred tank reactor batch-type, tubular, spray-pulsed, pressure-swing and three-dimensional (3D) structured monolith reactors, among others (Figure 3) (68).

Fig. 3.

Tubular-type reactors are often discussed for the LOHC technology. To optimise the process, different orientations of the tubular reactors have been evaluated (69). Tubular reactors can have either a horizontal or vertical orientation and have two possible operational modes: (a) LOHC flowing through the central tube and heat transfer fluid in the annulus; (b) heat transfer fluid in the central tube and the LOHC in the annulus (70). The heat transfer fluid enables external heating of the LOHC (69). In comparison to their horizontal equivalents, vertical tubular reactors have the advantage of a more even heat distribution, as a result of rising bubbles mixing the LOHC (70). Moreover, several vertical tubular reactors are suitable for both the hydrogenation and dehydrogenation reaction. Often, this is not the case for horizontal tubular reactors, where the interfacial region between the gaseous hydrogen and LOHC is often too small to achieve the mass transfer rates required for a hydrogenation process (69). Vertical dehydrogenation tubular reactors can also be constructed to produce either a co-flow or counter-flow between the LOHC and released hydrogen, offering more flexibility in a system design. In a study using N -ethylcarbazole as the carrier, different orientations of the reactor (horizontal/vertical) were studied, revealing that the maximum power densities are similar in both orientations, but radial heat transfer is improved compared to that in a horizontal orientation (69). Despite this, several examples of horizontal tubular reactors can be found, with advantages including convenient removal of hydrogen from the top of the reactor and prevention of a multi-phase flow (71).

Helical reactors are more complex and comprise the LOHC flow in the central tube, with the heat transfer fluid in the surrounding shell space (70). A more effective heat transfer can be achieved with a helical reactor than a tubular reactor, as a result of a ‘shear’ formation, while the rate of reaction (hydrogen release) is also increased. A double helical reactor is also discussed for the LOHC technology (70). While this enables a significant saving in terms of space, the construction and assembly is typically complex and expensive. The deployment of a hot pressure-swing reactor was demonstrated for stationary hydrogen storage in which hydrogenation and dehydrogenation were performed within the same reactor using the same catalyst with dibenzyltoluene as the LOHC (72). This demonstrates the benefit of potential capital expenditure and operational expenditure savings.

Structured reactors, such as monolithic, 3D printed or foam reactors, are an attractive research area within the LOHC technology as they provide high heat conductivity, allowing for a good heat input for the endothermic dehydrogenation reaction (73, 74). Furthermore, the high porosity of such reactor systems lowers pressure drop, on the contrary to fixed-bed reactors where catalyst pellets are typically used, and facilitates an efficient hydrogen removal. This is important for dehydrogenation of hydrogen-rich LOHCs, during which significant reaction mixture volume expansions take place upon hydrogen release (for example, 1 ml of fully hydrogenated dibenzyltoluene can release more than 650 ml of hydrogen) (68, 73).

Depending on the intended application, the hydrogen purity released from LOHCs during dehydrogenation should be considered. Given that high temperatures are required for dehydrogenation of LOHCs, side products might be formed during hydrogen release, contaminating the hydrogen stream. In such cases, separation systems are required to purify the hydrogen, which increases the cost of the LOHC technology. The use of membrane reactors, which combine the benefit of shifting the equilibrium of dehydrogenation to the product side with integrated purification of the released hydrogen, have been reported in the literature for the LOHC technology (71, 75–80). Byun et al . performed the technoeconomic assessment of methylcyclohexane dehydrogenation in both a packed-bed reactor and a membrane reactor (78). This study demonstrates the cost effectiveness of an membrane reactor: the unit hydrogen production cost of a packed-bed reactor is US$11.76, US$9.50, US$8.50 and US$8.08 and that of an membrane reactor is US$9.37, US$7.43, US$6.58 and US$6.23 for hydrogen production capacities of 30 m³ h^–1, 100 m³ h^–1, 300 m³ h^–1 and 700 m³ h^–1, respectively (78).

A lower temperature of operation for the dehydrogenation reaction, deploying a reactive distillation column under reduced pressure, could also allow for an increase in the efficiency of the LOHC system (81). This was demonstrated for perhydrobenzyl toluene dehydrogenation. A lowering of reaction temperature reduces the energy intensity of the overall process and facilitates a simpler and more effective integration of heat with waste heat sources or from subsequent hydrogen utilisation steps (81). Another recent study demonstrated that an electrochemical hydrogen compression (EHC) unit, which is connected to the LOHC dehydrogenation unit, reduces hydrogen pressure and shifts the thermodynamic equilibrium towards dehydrogenation (82). This accelerates the hydrogen release for the perhydrodibenzyltoluene LOHC and lowers dehydrogenation temperatures to 240°C. In addition, the EHC unit produces high value compressed hydrogen and purifies hydrogen contaminated with impurities such as traces of methane (82).

Several approaches to effective hydrogen storage technologies must be explored in parallel to facilitate a smooth transition to the hydrogen economy. LOHCs for hydrogen storage and transportation are an attractive option for storing and transporting green hydrogen. Key advantages of the LOHC technology are: high storage capacity compared to alternative hydrogen storage technologies such as MOFs; no hydrogen losses during extended storage periods; ambient storage pressures and compatibility with existing fossil fuel infrastructure (pipes, ships, trucks). LOHCs ideally have high, reversible hydrogen-loading capacities, enabling large quantities of hydrogen to be stored in the liquid carrier. Importantly, LOHCs facilitate the storage and intercontinental transport of hydrogen and can capitalise on infrastructure originally constructed for fossil fuels. Thus, over long distances, the transport of hydrogen using the LOHC technology has the potential to be economically viable in comparison to other hydrogen storage and transportation technologies. Though costs required for transporting of the LOHC loaded with hydrogen to end users followed by a transport of the dehydrogenated LOHC back to a chemical plant for reloading with hydrogen should be considered. Furthermore, additional developments and research into optimising the technology might be needed to improve economic viability of the LOHC technology. Reducing energy intensity during loading and particularly unloading of the LOHCs and efficient system integration can contribute to cost reduction of the LOHC technology.

Ammonia is a vital commodity chemical incorporated into the fertilisers that are needed to feed the growing global population. The conventional industrial process to produce ammonia involves the conversion of hydrocarbons into hydrogen through purification, steam reforming, water-gas shift and separation. Nitrogen is incorporated from the air during secondary steam reforming. Ammonia is made in the Haber-Bosch process; a synthesis loop that operates at high pressure (150–350 bar) to favour the gaseous reaction of nitrogen and hydrogen and high temperature (400–450°C) to promote the reaction kinetics (1). The reaction is catalysed by metallic iron. The process is summarised in Figure 1.

Fig. 1.

Equation (i) shows the equilibrium reaction of nitrogen, hydrogen and ammonia:

(i)

The Haber-Bosch process has provided ammonia-based fertiliser to feed our increasing population for a century. Haber-Bosch supports nearly half of global food production and demand for ammonia is expected to increase as the world’s population grows (2, 3). The use of ammonia as a carbon-free fuel, or as a hydrogen vector, would also increase demand. Fuel ammonia is a nascent concept but the well-established supply line, stability as a liquid under relatively mild conditions and high hydrogen density make it an attractive zero-carbon fuel (4). Indeed, the use of ammonia as a green fuel could cut carbon emissions from shipping by 50% by 2050 (5–7). As a hydrogen vector, ammonia can be converted to hydrogen (and nitrogen) by catalytic cracking. The resulting hydrogen could be used in fuel cells to generate electricity emitting water as a byproduct (8).

Haber-Bosch and the associated processes require considerable hydrocarbon input and ammonia synthesis is the source of an estimated 1% of global CO₂ emissions (9). There is an evident need for the development of ammonia synthesis routes with reduced CO₂ emissions if greenhouse gas targets are to be met.

There are several approaches that can be used to decarbonise ammonia. One route would be to capture and store the CO₂ emissions associated with conventional synthesis gas (syngas) production, known as blue ammonia. Carbon emissions could be eliminated entirely by supplying green hydrogen to the Haber-Bosch process. Green hydrogen can be produced from the electrolysis of water, which is powered by renewable electricity. There are several recent industrial examples that implement this green hydrogen concept and rapid growth is expected as demand for lower carbon ammonia intensifies (10).

Finally, a more economical route to green ammonia would be to eliminate the Haber-Bosch process entirely and use renewable electrical energy (electrochemical) or sunlight (photochemical) to reduce nitrogen from air to ammonia in the presence of water under ambient conditions. Direct synthesis of green ammonia in this manner is non-trivial owing to the chemical inertness of the nitrogen molecule. It is unlikely to be achievable at the necessary scale and cost to compete with conventional Haber-Bosch or green hydrogen Haber-Bosch for decades. However, green ammonia synthesis could be the most intrinsically economical process because it would not require the combination of electrolysers, air separation units and the high-pressure Haber-Bosch plant (11).

The introduction of these technologies can be understood further by considering them as three generations as presented by Professor Douglas MacFarlane at Monash University, Australia, Figure 2. Present day (2020s) ammonia production is Generation 1 with the option of capturing CO₂ to make blue ammonia. Generation 2 is electrolysed hydrogen into the conventional process which is predicted to be widespread by 2030. Generation 3 breaks the paradigm of the Haber-Bosch process with direct synthesis of ammonia under mild conditions with a target for commercialisation by 2050 (12).

Fig. 2.

This review highlights key trends within green ammonia synthesis and identifies opportunities for decarbonisation at the peaks of Generation 2 and Generation 3 (2030 and 2050 respectively). Green hydrogen from the electrolysis of water coupled with an electrically powered Haber-Bosch process (Generation 2) receives considerable attention in the industry already. Several major ammonia producers have announced plans for green ammonia plants though total capacity of the green ammonia projects remains a fraction of the global ammonia production with challenges of high capital investment and access to sufficient renewable energy. With the drive to zero carbon emissions and increasing prominence of ammonia as a hydrogen vector, direct ammonia production (Generation 3) could become widespread.

2.1 Overview of Green Hydrogen for Ammonia Synthesis

2.2 Alternative Catalysts for the Haber-Bosch Process

2.3 Separation of Ammonia from a Low-Pressure Haber-Bosch Process

Electrochemical ammonia synthesis harnesses electrical energy, which could be renewably sourced, to directly convert the hydrogen of water and nitrogen in air to ammonia at ambient temperature and pressure. If electrochemical ammonia synthesis could be achieved at high efficiency at potentials close to that of the reaction (i.e. low overpotential), it could compete with conventional Haber-Bosch synthesis in terms of overall cost. However, the current level of development for electrocatalytic ammonia synthesis systems is not far advanced beyond laboratory scale.

The recurring challenge for ammonia synthesis is the inertness of the nitrogen molecule. Electrocatalytic reduction of nitrogen requires significant energy input and a catalyst site with strong binding for nitrogen but a weak interaction with ammonia so it is readily released. An additional challenge for electrochemical synthesis is the competing reaction for hydrogen evolution from water rather than hydrogen incorporation into ammonia. An ideal electrocatalyst maximises conversion to ammonia (measured from current density or turnover frequency), has long life, minimises overpotential (electrochemical potential above the thermodynamic potential that is required to drive the reaction) and has a high Faradaic efficiency (FE) (efficiency with which the electric charge is transferred to the electrochemical reaction) (11). The US Department of Energy (US DOE) has set a target rate for viable electrochemical ammonia production of 10^–4 mol h^–1 cm^–1 and FE of 50% but current systems suffer from insufficient production rates less than 10^–6 mol h^–1 cm^–1 and FE less than 30% indicating how much technology advancement is required. The US DOE estimates that it will take until 2050 for electrochemical ammonia production to compete with Haber-Bosch (30).

If or when electrochemical systems reach the targets, the cost per unit of electrochemical ammonia would be cheaper than that of ammonia from electrolysed hydrogen followed by electrically powered Haber-Bosch. Estimates from Professor Gal Hochman and Professor Alan Goldman from Rutgers University, USA are that if renewable electricity cost is US$50 MWh^–1, the cost for electrochemical ammonia is ~US$500 tonne^–1 compared to ~US$630 tonne^–1 for ammonia from electrolysed green hydrogen feeding into 2000 tonne day^–1 Haber-Bosch. The estimate for ammonia from conventional 2000 tonne day^–1 Haber-Bosch is US$159 tonne^–1 based on cost of US$2.62 per one thousand British thermal units (MBtu) natural gas (11).

Technoeconomic analysis from Jamie R. Gomez, University of New Mexico, USA, confirms the conclusion that direct electrochemical synthesis of ammonia is intrinsically lower cost than the Haber-Bosch process fed with green hydrogen though the calculated costs differ from those of Hochman and Goldman. Gomez assumes that direct electrochemical synthesis of ammonia will require the same infrastructure as a Haber-Bosch plant: renewably powered hydrogen generation, cryogenic nitrogen separation prior to ammonia synthesis coupled with ammonia liquefication and separation post synthesis. Using the US DOE targets for electrochemical ammonia production of 10^–4 mol ^–1 cm^–1 and efficiency of 50% and electrochemical reactor operating at 200°C, ambient pressure, the energy requirement is 17 MWh per tonne of ammonia. The cost of a tonne of electrochemically derived ammonia from this study is US$951 whereas the equivalent process with Haber-Bosch ammonia synthesis was calculated to be US$975 (29).

3.1 Electrochemical Ammonia Synthesis Mechanism

The typical electrochemical ammonia synthesis reaction is described by Equation (ii), the nitrogen reduction reaction (NRR):

(ii)

Reduction of nitrogen occurs at the cathode with six protons and six electrons required to form ammonia (cathode reaction, Equation (iii)). Oxidation of water to hydrogen and oxygen occurs at the anode (anode reaction, Equation (iv)) (11).

(iii)

(iv)

The relatively high number of requisite protons and electrons for nitrogen reduction suggests considerable optimisation is required to improve the reaction kinetics to deliver the charge carriers (30). The mechanism for the reduction of the nitrogen molecule by the six protons and electrons incorporates many intermediates with multiple proton-electron transfer steps. It is likely that electrochemical ammonia synthesis follows an associative mechanism with full cleavage of the nitrogen triple bond after proton-electron transfer, Figure 3. The Haber-Bosch process over iron catalysts is known to follow a dissociative mechanism where the first step is nitrogen triple bond breakage. The difference in the mechanisms means that electrochemical ammonia synthesis via the NRR is intrinsically lower energy than the Haber-Bosch reaction. The reactions that constitute the NRR together with their potentials are summarised in Table III (31). The most negative and therefore most energetically demanding step is the formation of N₂^– (–4.16 V vs. normal hydrogen electrode (NHE)). The formation of N₂H is also negative (–3.2 V vs. NHE). The negative potentials suggest that these steps are likely to be rate limiting in the electrochemical ammonia synthesis process.

Fig. 3.

Possible routes to ammonia via the NRR associative mechanism. The dissociative mechanism that occurs over the iron catalyst of the Haber-Bosch process is provided for comparison

Table III

Summary of the Reactions Constituting the Nitrogen Reduction Reaction and Associated Electrode Potentials

Reaction	E0, Va
H2O → 0.5O2 + 2H+ + 2e–	0.81 vs. NHE at pH 7
2H+ + 2e– → H2	–0.42 vs. NHE at pH 7
N2 + e– → N2–	–4.16 vs. NHE at pH 0
N2 + H+ + e– → N2H	–3.2 vs. NHE at pH 0
N2 + 2H+ + 2e– → N2H2	–1.10 vs. RHE
N2 + 4H+ + 4e– → N2H4	–0.36 vs. NHE at pH 0
N2 + 5H+ + 4e– → N2H4+	–0.23 vs. NHE at pH 0
N2 + 6H+ + 6e– → 2NH3	0.55 vs. NHE at pH 0
N2 + 8H+ + 8e– → 2NH4+	0.27 vs. NHE at pH 0

Theoretical studies are often used to explore the potential mechanisms for electrochemical reactions. Schematic representation of the associative and dissociative mechanisms is presented in Figure 3.

The electrochemical system has a significant impact on the efficiency of the process. Factors such as electrode potential, solvent and pH values of electrolytes need to be optimised to achieve the best yields. Electrochemical cell potential is fundamentally dependent on temperature (see box). For the NRR to ammonia from water and nitrogen, the higher the temperature, the lower the potential required to drive the system forward. Electrochemical potential is also critical; different reaction systems tend to have different potentials that will suit ammonia synthesis (33). Furthermore, application of appropriate potential for the NRR can reduce propensity for the hydrogen evolution reaction.

Electrochemical Cell Potential

The Nernst equation (Equation (v)) summarises the relationship between reduction potential of an electrochemical reaction to the standard electrode potential, temperature and activities of the chemical species involved.

(v)

where F = Faraday’s constant (eN_A, e = charge of an electron; N_A = Avogadro constant); v = stoichiometric coefficient of electrons in the electrochemical reaction; Q = reaction quotient, product activity/reactant activity; R = molar gas constant; T = temperature in Kelvin (32).

The electrolyte also plays a role. For example, a low pH solution and the ready availability of protons may favour hydrogen evolution over NRR so higher pH might suit some electrochemical ammonia synthesis processes. However, there is no definitive conclusion on the optimal NRR pH range that universally applies to every system.

Catalyst design is also pivotal for achieving the required rates and yields from electrochemical ammonia synthesis. As mentioned, electrocatalysts should have appropriate active sites to bind nitrogen, easily release ammonia and limit the hydrogen evolution reaction. There is intense academic research into catalysts for electrochemical synthesis of ammonia with many reviews summarising progress. The recent paper from Hui Xu of Giner Inc, USA, and Professor Gang Wu of the University at Buffalo, USA (30) and the review by Professor Liang-Xin Ding and Professor Haihui Wang of the South China University of Technology in Guangzhou, China (33) were highly informative. Review papers from Muhammad Aziz from The University of Tokyo, Japan (34) and Sarb Giddey of CSIRO Energy Technology, Australia (35) provide valuable summaries of electrochemical ammonia production. In many cases, advances in nanomaterials have supported recent developments in electrocatalysis. Table IV highlights a few recent examples with commentary in the sections below.

Table IV

Examples of Electrocatalytic Systems for Generating Ammonia

Catalyst	Electrolyte	Faraday efficiency, %	Ammonia production rate, μg h–1 mgcat–1	Eθ vs. RHE at 25°C	Comment	Reference
Pd/C	0.1 M phosphate buffered saline (PBS)	8.24	4.5	–0.2	Neutral electrolyte (pH 7) improved FE compared to FE of less than 0.1% in NaOH pH 12.9 and H2SO4 pH 1.2	(36)
Au-CeO2/reduced graphene oxide	0.1 M HCl	10.1	8.3	–0.2	Amorphous gold nanoparticles and structural distortion from ceria provides active sites for NRR	(37)
Au1C3N4	0.005 H2SO4	11.5	1305	–0.1	Gold single atom carbon nitride catalyst achieved 22 times more ammonia than equivalent system prepared with gold nanoparticles	(38)
Ru@ZrO2/NC	0.1 M HCl	21	183	–0.2	Ruthenium single-atom supported on nitrogen-doped porous carbon. ZrO2 supresses hydrogen evolution. Oxygen vacancy sites on ZrO2 promote catalytic activity of ruthenium for ammonia synthesis	(39)
1T-phase MoS2 nanodots on g-C3N4		20.5	30	–0.3	1T MoS2 nano dots possesses high surface area with many active edge sites. Graphitic carbon nitride (g-C3N4) electronic effect makes catalyst highly selective for NRR	(40)
Mo2C	0.1 M HCl	10	95	–0.2	Durable catalyst, 58 h operation. Greater FE than other molybdenum catalysts: MoS2 1.17%, MoO3 1.9% MoN 1.15% and Mo2N 4.5%. 25 mA cm–2 current efficiency	(41)
FeTPPCl	0.1 M Na2SO4 PBS (phosphate buffered saline)	17	18	–0.3	Tetraphenylporphyrin iron chloride. FeN4 site displays strong interaction with nitrogen. Activity retained for 36 h of catalytic testing	(42)
p-Fe2O3/CC	0.1 M Na2SO4	8	14	–0.4	Porous Fe2O3 nanorods grown on carbon cloth. Porosity provides facile access to active sites	(43)
Li+/Li	Li+ in THF	37	28 ppm	–1 vs. Li+/Li	Li+ deposited on metal electrode as lithium which reacts with nitrogen in presence of H+ to form ammonia. –3 V required to deposit lithium which makes process unstable. Cycling between potentials for Li+ in solution and deposited lithium has a stabilising effect, 125 h of testing	(44)
LiCl–KCl	LiCl–KCl–LiH	4.2	2.8 × 10–8 mol cm–2 s–1	1 V vs. Li+/Li	High rate for electrochemical ammonia synthesis. Molten salt electrolyte with LiH to provide H–. Isotope study to prove 15N2 incorporated into 15NH3	(45)

3.2 Precious Metal Electrocatalysts

3.3 Transition Metal Electrocatalysts

3.4 Molten Salt Electrolytes

3.5 Electrochemical Lithium Metal Cycling

3.6 Electrocatalysis Summary

Photochemical reactions are driven by light and photocatalysed ammonia synthesis is regarded as a potential route to green ammonia. The benefit of photocatalysis is that energy for the reaction would be provided directly from sunlight with water and air to provide hydrogen and nitrogen respectively. Unlike the electrochemical process, there would be no need to supply electricity, making photochemical synthesis a potential candidate for decentralised off-grid ammonia production. An evident drawback of photochemical processes is that they only operate when the sun is shining.

The concept would likely feature the catalyst dispersed in a panel to optimise light exposure, potentially suspended in water or as a coated catalyst with water and air bubbled over the surface. Ammonia would need to be separated from the mixture before application as a fertiliser. Photocatalytic activities of ~500 mmol_NH3 g_cat^–1 h^–1 (51) have been reported from the current state of the art systems which is 1000 times less than the electrochemical systems. As Table V indicates, the area of a coated 500 mmol_NH3 g_cat^–1 h^–1 photocatalyst required to match a 2000 tonne per day Haber-Bosch plant would be 1265 km², a considerable area equivalent to the North York Moors National Park in the UK. A 100-fold improvement in catalyst activity would reduce the area required to 13 km² which would be a more feasible area to cover. For comparison, the world’s largest solar park at the time of writing is 57 km² at Bhadla, India (52).

As mentioned, photocatalysis might suit demands of off-grid distributed ammonia production. In this case ammonia demand would be significantly less than a 2000-tonne-per day plant. Agricultural fertiliser demands vary according to crop, soil type and geography but if nitrogen requirement of 200 kg nitrogen hectare^–1 year^–1 (53) is used as a conservative average and 400 hectares the size of a large arable farm (54), the quantity of ammonia required is ~100 tonnes year^–1. With a 500 mmol_NH3 g_cat^–1 h^–1 coated photocatalyst, the area required is 0.2 km² or 20 hectares, 5% of the size of the farm.

A complicating factor is that molecular ammonia is rarely applied as a fertiliser. Its high volatility and solubility mean it would rapidly evaporate or leach away. Ammonia is converted to a range of compounds such as urea or ammonium salts (for example, (NH₄)₂SO₄ or NH₄NO₃) to be applied as a fertiliser. Distributed ammonia production would need additional technology for conversion of ammonia to fertiliser compounds. Furthermore, different fertiliser compounds have varying CO₂ emissions associated with their production and use. The CO₂ emissions from urea are 8% higher than from ammonium nitrate (55). However, ammonium nitrate can be highly explosive if not manufactured, stored and handled properly according to recognised standards and decentralised production poses significant safety and security risks. Together with advances in ammonia production, decentralisation also requires developments in fertiliser compounds, their application and stability and ease of production from ammonia.

4.1 Mechanism of Photocatalytic Ammonia Synthesis

4.2 Photochemical Ammonia Synthesis Catalysts

4.2.1 Defect Incorporation

4.2.2 Metal Doping

4.2.3 Cocatalyst Incorporation

4.3 Photocatalysis Summary

Green routes to ammonia are receiving considerable attention from academia, governments and industry to mitigate the high carbon footprint of the conventional Haber-Bosch process which is estimated to contribute 1% of global CO₂ emissions. Many key players of the ammonia industry have already announced plans to incorporate green hydrogen from electrolysis into their existing plants. It is likely that more will follow.

The development of lower pressure ammonia synthesis systems (less than 20 bar) is also an area receiving attention though the current industrial trend is towards higher pressure systems owing to the challenges of separating ammonia at low pressure and lower conversion. However, there may be instances where smaller, lower pressure plants make sense. These plants would require catalyst development to operate at the lower pressure and novel separation techniques to isolate the product ammonia.

Although direct ammonia synthesis via electro- or photocatalysis is a distant prospect, the gains to be made are significant considering that the direct route from water and air or (hydrogen and nitrogen) is inherently lower cost than electrolysis and Haber-Bosch. Efforts on direct ammonia synthesis would be a long-term undertaking as the technology is not anticipated to be economically viable for another 30 years. However, if ammonia demand for fertilisers and fuel increases as expected, production at lower cost with zero carbon emissions presents an attractive opportunity. Furthermore, the pursuit of intrinsically lower cost routes to ammonia synthesis such as electrochemical or photochemical will drive innovation in these fields which may accelerate breakthroughs. If ammonia is to be produced from water and air, separation techniques to isolate ammonia will be critical here too.

DoE and multivariate statistical approaches have been identified before as a clear way of saving time and resources (1). They are systematic and structured approaches to product development and process improvement. The methodology is based on introducing variability into the system by changing a limited number of variables at controlled levels simultaneously, but systematically, in order to study the parameter space. The aim of a DoE is to maximise the knowledge obtained while minimising experimentation. It can also help to ‘fail quickly’; if, for example, the outcome of the study is that the variability cannot be explained by changes on any of the variables, further variables need to be considered. This leads again to saving time and resources.

Statistical modelling of designed or undesigned data can provide a predictive model. This model can be used to predict the output from a combination of the inputs that has not been tried before experimentally, providing that the combination is within the experimental space. Therefore, the predictive aspect of the model can potentially save unnecessary experimentation in the future.

Although it is difficult to quantify and compare like-for-like, some studies in the pharmaceutical space stated that projects involving multivariate experimentation resulted in a requirement for 50–70% fewer batches than traditional experimentation, such as a ‘one factor at a time’ (OFAT) approach. Therefore, the total number of product development weeks were reduced by at least 43% (1), illustrating that time can be saved using this approach.

Historically, multivariate experimentation has not been very accessible for non-statisticians, but now it is possible thanks to user-friendly statistical software packages like JMP^® (from SAS Institute, USA), which offers extensive DoE capabilities to design and analyse all types of DoE, and visualisation tools which allow the user to understand which experiments are being carried out and to visualise and communicate the results effectively. Despite potentially remarkable savings, the implementation cost should be relatively low, since it only requires making a software package like JMP^® available to scientists and having a good network of support and coaching internally within organisations to share good practices and new methods.

Although cost reduction is possibly the strongest advantage of using statistical tools, it is not the only one. The structured approach embedded in statistical experimentation allows a better project planning process with known schedules.

When planning an experimental programme, the use of DoE methodology provides several advantages over a traditional approach. The scientists involved in the work must first identify all the variables, thinking about the entire system, which helps to ensure that the project scope is properly assessed and clearly defined at the outset. The variables should be split into those which can be changed (factors or inputs) and those which are to be affected because of these changes and measured (responses or outputs). Are the factors continuous or categorical in data type? Can the factors be controlled during the experiments? If they cannot be controlled then should they be measured? Which of the factors will be fixed as part of the scope of the experimental programme and which will be varied? What are the ranges of each factor? How many levels for each factor will be used? These questions are best answered by a team of scientists with pre-existing knowledge and skills. The planning stages of the DoE are crucial to its outcome and should not be overlooked.

The DoE methodology of variable identification facilitates project definition and considers the experimental design space in its entirety before focusing on the parts of interest. Experimental design space is the total space defined by the factor ranges. This must be carefully chosen by the scientist to ensure that the aims of the experiment can be achieved. An illustration of the design space for a three-factor experimental design is shown in Figure 1, where the design space is the three-dimensional area within the cube, and experiments can take the form of any combination of the three factors within this design space. This can often be neglected during the planning of non-DoE type experimental work.

Fig. 1.

The experimental matrix generated by the DoE is also important in project scheduling. Access to the full set of planned experiments at the start of the project helps when assigning resources and provides a good estimate to management about exactly how long the programme will take. It also prevents interim interpretation of the data because the full set of results is necessary for analysis. This is in direct contrast to typical ‘reactive’ laboratory practice whereby each experiment is analysed immediately afterwards and used to inform the next experiment. In this traditional way, the end of the programme is not clear as the total number of experiments has not been defined and is therefore likely to take longer. DoE is a more proactive approach with a clear timeline for project management purposes and ensures that the full dataset is available before analysis, decreasing the chances of drawing incorrect conclusions or subjectively changing the parameters of the project based on the results of the latest experiment.

An example of this proactive approach coupled with demand for a tight project schedule was demonstrated within Johnson Matthey. An online analyser was loaned from an instrument manufacturer to investigate whether it could be used to monitor a chemical reaction in real-time on plant. The analyser was only available for two weeks and it was therefore important to study the effectiveness of the analyser as efficiently as possible, by collecting spectral data accounting for a range of reaction product mixtures. The aim was to provide enough variation to ensure that robust calibrations for each component in the product mixture could be established within the range of expected online process conditions. A screening DoE was generated to assess the influence of six factors on the spectral response for each reaction product mixture. Two centre points, set in the middle of the factor ranges, were included to determine whether non-linear relationships between the factors and the spectral response could be present, and to establish whether the spectral response was repeatable. The DoE generated 17 experiments, which were run in a randomised order (Table I). These 17 experiments were combinations of high- and low-level settings for each of the six factors, ensuring that there were no correlations between each of the two factors and that effects on the response can be independently quantified (Figure 2).

Fig. 2.

Excellent calibrations for all components of the reaction product mixture were obtained, and the instrument manufacturer commented on how well the design space had been explored in the time available using the DoE. Following successful demonstration that this online analyser could be used to monitor the reaction in all expected conditions, proposals were submitted recommending its purchase and operation on a customer plant. The advantage offered by statistical tools to draw trustworthy conclusions is an aspect which deserves proper consideration.

Trustworthy conclusions obtained from a study and its data are necessary to make the right decisions. The conclusions obtained from the data are as good as the data itself. Therefore, quality of the data is a key aspect. For example, if the data is biased or unbalanced, there is a possibility of obtaining inaccurate conclusions which could lead to unsuccessful or suboptimal decisions for the system or process. The use of statistical tools to plan the study should ensure good quality data and therefore increase the probability of drawing reliable conclusions.

‘Universal versus local optimum’ is an issue which can occur if the data to analyse is not a good representation of the experimental space being studied. In that case, the data analysis can lead to a local optimum of conditions to maximise the output, while the universal optimum is still to be discovered (Figure 3). Following traditional experimentation only data in the red path was obtained, leading the scientists to reach a local optimum. However, within the experimental space defined, a better outcome is possible, but has not been found. This is what it is referred to here as the universal optimum for the experimental space.

Fig. 3.

DoE leads to obtaining the right data since the experiments are designed to study the effect of the selected variables and understand the system or process in the most efficient way. It ensures the data is balanced and distributed within the experimental space, allowing unbiased and relevant conclusions about the system to be extracted. JMP^® software is a leader in statistical DoE, making multiple state-of-the-art designs available for scientists to choose from depending on the specific case.

When dealing with historical data, which could be biased, for example rich in certain areas of the experimental space and sparse in others, the risk to find a local optimum instead of the universal optimum is significant. Working with historical data can not only lead to suboptimal decisions but can also be time consuming, so the use of statistical tools within JMP^® can significantly support this process.

The use of advanced data analytics may be applied effectively to existing datasets. There are many instances in research and development (R&D) and manufacturing where large datasets have been generated from previous work programmes which could prove useful as a starting point for the current project of interest. Rather than starting completely from scratch, it may be possible to identify trends and relationships between variables from this existing data. This has the advantage of utilising historical data, much of which was probably expensive and resource-intensive to generate. The use of exploratory data analysis tools within JMP^® facilitates this process.

An example of analysing historical data with an exploratory approach has been demonstrated within Johnson Matthey at a catalyst manufacturing site. Two separate plants were involved successively in the production of a single catalyst product, and the multivariate tools within JMP^® were used to determine which of the process inputs most affected the properties of the intermediate material (output of Plant 1), and then which of these as inputs affected the properties of the finished catalyst product (output of Plant 2). The process data used in this analysis was taken from several years of production on both plants. An example of part of the exploratory data analysis used for this example is shown in Figure 4, where the distribution and graph builder platforms of JMP^® were used to visualise relationships between variables. Based on the results of the analysis and the predictive models created, process settings were changed to optimise catalyst product properties and both plants now have higher rates of meeting target specifications.

Fig. 4.

Limitations may exist in the historical data, and probably will be present if the data was collected using a traditional OFAT approach rather than from a designed set of experiments. In this case, it is important to identify where multicollinearity exists and how this affects the analysis of the dataset and the conclusions drawn. The multivariate and exploratory tools within JMP^® allow these limitations to be visualised and understood, enabling the scientist to make informed decisions about what the data is showing while being mindful of the underlying assumptions. It also provides an opportunity for sequential experimentation, whereby the existing data, although limited, can be used as a starting point for a subsequent DoE which can deconvolute the limitations in the historical data, resolving the correlated effects and suggesting the best combination of experiments in parts of the design space with fewer existing data points. Alternatively, the understanding gained from mining the historical dataset may be used to focus on fewer significant effects for a new experimental design with additional factors. Related to predictive models, numerous advantages can be drawn for the prediction capabilities, such as gaining control and adaptability.

An important advantage of the predictive capacity of a model is the control over the system or process that it offers. It allows the scientist to respond to the outputs and modify the inputs in a system or process to adapt to a new situation, keeping the system or process on target. For example, if the value of one of the input variables changes for external reasons out of our control, the model will point out what the value of the other inputs should be which can be controlled to keep the output on target, compensating for the changes in the input without any experimentation needed. This brings control back to the users and offers tremendous flexibility and adaptability; very important qualities in the fast-moving world. This is often used within Johnson Matthey in different businesses, for example, formulations for certain products to ensure the quality of the final or intermediate product, by proactively adapting to changes in the raw materials.

This task is performed easily in JMP^® using the interactive ‘prediction profiler’ (Figure 5). The profiler also allows the scientist to find a new optimum combination of input values if the output target changes (for example, a new customer specification), or when an input needs to be fixed at a certain value (for example, a new requirement or limitation). The profiler will find the optimal combination of the remaining input variables to stay on target.

Fig. 5.

Control over systems and processes is not the only advantage of data modelling. Another very important aspect is related to knowledge storage.

In a scientific process, data is generated to obtain answers to technical questions, prove and contradict hypotheses and corroborate assumptions in the process of discovery or optimisation. Therefore, the data itself is a vehicle to get knowledge. Knowledge is the final aim, but that knowledge ideally needs to be recordable, communicable and transferable to maximise its use.

Statistical modelling allows knowledge to be extracted from a study or from data in the shape of a model that helps to communicate and visualise the effects of the different inputs on the output. The model itself contains this knowledge and allows the rest of the world to utilise that knowledge.

Within JMP^®, statistical modelling is accessible to everyone with multiple modelling techniques available and the ability to compare them easily. In addition, the software offers the prediction profiler tool (Figure 5), which not only enables scientists to visualise and communicate their findings (contained in a model) dynamically and interactively, but also to transfer and share the learnings with colleagues in the same team and between different teams and functions. Utilising these tools can ensure that the knowledge obtained from experimentation stays in the company in a reusable format despite employees leaving or retiring.

Another aspect to facilitate knowledge sharing comes from the understanding of a chemical problem or question. Sometimes this can be very subjective and variable depending on the scientist’s background, expertise and interests. JMP^® tools offer enhanced visualisations for different stages in the process to ensure good communication and visualisation of problems and results.

Visualisation tools are used in different steps of data analysis and are key to helping understand and communicate the chemical problem studied. In the first instance, they are used to explore the data set initially. This process is very important as it helps the scientist to get to know the data. On one hand, it helps to understand the experimental space and identify possible gaps, outliers and errors. As mentioned before, this stage is particularly important when looking at historical data as this data tends to be limited. It can also help the scientist to identify correlations between the inputs and the outputs before embarking into model building. The ‘graph builder’ and ‘distribution’ platforms available in JMP^® are excellent tools to use at this stage (Figure 4). They are also great tools to present a point or argument in a meeting since they are interactive and easy to understand. All these visualisations can also be shared using dashboards that can be produced in JMP^® very easily, and the interactivity is retained (Figure 6). Dashboards, in the same way as other visualisations, can be converted into HTML so they can be explored without the need to have JMP^®. Dashboards allow scientists to present key findings and can support stakeholders with decision making.

Fig. 6.

Once the model is built, the effect of the inputs to the outputs can be visualised using the prediction profiler (Figure 5) which is one of the most powerful tools available in JMP^®. As already mentioned, this allows the scientist to explore the effect of the factors and better understand the chemical problem. It is also a great tool to communicate the process and the effect of the factors. JMP^® allows these visualisations to be saved in an interactive format which can be shared across different functions. An example of utilising these tools to generate value has been demonstrated within Johnson Matthey. When the commercial team received enquiries regarding the use of a product under certain conditions, they had to contact the development team to access the information. The research team has now built a model as a result of a response surface DoE. The model has been shared with the commercial team using the interactive prediction profiler. With this, the commercial team can predict the performance depending on the conditions suggested by the customers. This tool has provided the commercial team with more autonomy and a quicker response to the customer and has saved time for the development team. Statistical tools can not only help us to visualise data but also to make objective decisions.

The use of statistics in disciplines such as physics, biology, medicine and finance is common (2, 3). However, in our experience, its use in chemistry has been sparse despite it being a useful tool, and some would say, indispensable.

The aim of experimentation is typically exploratory, to gain understanding, or to optimise a process. Although the objective might be different, a tool that helps to differentiate between the experimental variability and the effect of a particular input is needed. This is where statistics can help to make more informed decisions. Statistical tests are carried out to understand if results are statistically significant or not. When talking about statistically significant results, we refer to those results obtained by testing or experimentation that are not likely to occur randomly or by chance, instead they are due to a specific cause. Often p-values are used to describe this. Although the inappropriate use of p-values in some cases has brought controversy (4–7), they can be very useful. It is important to remember that the conclusions drawn from statistical tests should be interpreted within the context of the study (sample size, reliability and validity of the instruments used to measure the outputs).

An example of this within Johnson Matthey has been a comparison study between several analysers (Figure 7). The statistical tool facilitated the visualisation and helped to establish the significance of the differences found between the measurements obtained in the analysers when dealing with the same samples. These types of studies are crucial to ensure the reproducibility of results.

As seen so far, the toolbox is quite extensive and sometimes that can be slightly overwhelming. For example, when generating a DoE, it is possible to be intimidated by the choice of design types available. However, JMP^® has features to help when evaluating and comparing designs.

Fig. 7.

The choice of design depends upon the aims of the project (screening or optimisation) and the resolution required (main effects, higher order terms). Classical DoEs (full factorial and fractional factorial designs) are no longer used as often as increasingly popular modern designs (definitive screening designs and bespoke custom designs) (8, 9). The design choice must then be carefully balanced against the resources available (timeframe, cost of running experiments) to decide upon the experimental matrix to be used. More experiments will provide more information about the system, but often this is not possible because of practical or financial constraints. It therefore becomes extremely important to compare multiple designs and understand the relative advantages and disadvantages of each.

This is made possible with the ‘evaluate design’ and ‘compare designs’ tools in JMP^®. Potential designs can be opened side-by-side and comparisons made. Power analysis helps to estimate the ability of the design to detect effects of importance by reporting the probability of detecting effects of a given size. Higher powers for model terms result in a greater chance of detecting their effect. Prediction variance profiling displays the uncertainty across the experimental space and can be altered depending on the focus of the design. For example, an optimisation design would try to minimise prediction variance at the centre of the experimental space. Colour maps of correlations show the absolute value of the correlation between any two effects that appear in the prediction model, represented visually with a sequential colour scheme (Figure 8). This helps to identify where factors and higher order terms in the models may be partially or fully confounded, and where one design might have the advantage over another.

Fig. 8.

The eventual design choice will be unique to the scenario, but evaluation and comparison of multiple designs allows the requirements of the project to be considered against the real-world implications. Running more experiments will provide additional understanding of the system but resource may only be available for a predefined number of experiments. These tools allow the best choice to be made to carry out experimentation in the most efficient manner to maximise the information that can be gained while also identifying the limitations of the design. The efficiency of statistical design has already been mentioned several times, and this characteristic is due to the systematic and structured nature of this approach.

The traditional approach to experimentation, which is still taught in most universities, consists of changing one input while keeping the others constant. This provides the certainty, or so it is believed, that the variance observed in the output is due to this change. However, this approach has many pitfalls: there is no way of studying the interaction between two inputs, experimental error is not accounted for and the experimental space is not fully covered. DoE corrects all these pitfalls: it allows the study of interactions between inputs, the experimental error is accounted for and all the experimental space is fully covered. All this provides more control than traditional experimentation.

The process of carrying out statistically designed experimentation follows a structured approach. Initially, the experimental space is decided by the scientist based on experience or prior knowledge. If working in a new area, a pilot trial can be used to help the scientist. Once the first set of experiments has been completed and analysed, further experiments can be planned based on the results obtained, the aim of the experimentation and the number of experiments that can be performed. Also, experiments to validate the model should be carried out. The scientist has control over the experimental plan and the statistical tools are only there to facilitate the work. All this is made very easy by JMP^® as it provides different platforms to generate the different designs and augmentations. As already commented, tools to evaluate the designs can also be found in these platforms so the scientist can take an informed decision when selecting the design.

As emphasised extensively in this article, there are multiple benefits from utilising statistical tools for product development and process optimisation. However, their implementation has not been so widely applied, especially in the chemical industry. It is worth highlighting some of the challenges and how to overcome them.

Commonly, data logging from scientific equipment and data analysis from experiments is done within Microsoft Excel (Microsoft Corporation, USA). Scientists are familiar with this program, having probably used it daily during the entirety of their career. There is a reluctance to move away from something to which we are so accustomed, in some cases to the point where we can no longer see the limitations. Microsoft Excel is an excellent spreadsheet program with simple user interface, ensuring it is used universally. However, it was never designed with the intention of handling and interpreting large volumes of data. A recent example of its misuse resulted in Public Health England failing to report nearly 16,000 coronavirus cases in 2020 (10). Add-ins are available to perform simple statistical functions, but specialist software like JMP^® is required to thoroughly interrogate data and deliver greater understanding.

As well as the tools available within JMP^® to provide greater insight, it has been purposely designed to manipulate and visualise large datasets. This is exemplified by features such as ‘graph builder’ and ‘dynamic linking’, as previously shown in Figure 4. The click-and-drag interface when building graphs in JMP^® is a much simpler workflow to visualise data than creating graphs within Excel. There is also a JMP^® add-in available for use in Excel which allows the user to transfer data between the two programmes in a single step and quickly access some of the common analysis platforms of JMP^®. From experience within Johnson Matthey, we have found that the key to persuading people away from Excel and into specialist software is to show a direct comparison of a typical workflow with real-life data used in that part of the company. The improved visualisation and data analysis are immediately obvious, as are the time savings, which liberates more time for scientists to develop new technologies and products in the laboratory rather than handling and formatting data. However, there is also a barrier to overcome when learning to use new software.

Johnson Matthey has recognised the benefits of promoting and instilling a culture of advanced data analytics. However, a common barrier to overcome when transitioning to new ways of working is the initial investment of time required to get to grips with new software. For research professionals whose time is a precious commodity, the investment of time needed upfront to learn new techniques and navigate around the software can be deterring. This is especially true as this part of the learning curve does not provide any immediate, tangible output. Furthermore, the wealth and variety of training resources available to new software users can make the learning process seem initially overwhelming. From experience within Johnson Matthey, we have found that setting aside time at regular intervals to progress through a predetermined training plan helps to make the process as simple as possible for new users. The training plan can be developed alongside a more experienced software user and will be bespoke to the requirements of the individual, concentrating on the functionality of the software with which the user will primarily be working. The training plan typically includes different resources, such as individual learning (online webinars, e-learning subscriptions, ‘Statistical Thinking for Industrial Problem Solving Modules’ – a free online statistics course provided by JMP^®) and group learning (Johnson Matthey specific software introduction courses developed and run by experienced users). There is also an active JMP^® user community within Johnson Matthey which was created to support new users, provide an informal environment for sharing knowledge and as an open forum to ask questions about specific problems.

Demonstrations of Johnson Matthey projects to senior management where the software has been used to improve process understanding have been critical to increase awareness of the benefits the software can bring. This has resulted in management encouraging staff to dedicate time towards software training. The impact of coronavirus has also accelerated this process, as developing new software skills is a task that can be carried out while working from home, either during forced periods of self-isolation or by minimising regular operations on site. But obviously it is not all about learning to use new software, it is also about learning statistics.

As already mentioned, the use of statistics is more common in other fields such as biology or medicine than in chemistry. Traditionally statistics is not a featured component in chemistry undergraduate degrees. If some statistical content was taught in the first years, the learning was not normally reinforced with practical activities further on in the course. This can make chemists uncomfortable around statistics.

Within Johnson Matthey we believe that a practical understanding of statistics can be achieved to complement chemistry expertise by our scientists. The use of practical statistics is now totally accessible by using software packages like JMP^®. The learning curve of statistics goes hand-in-hand with learning new software from a practical point of view. It allows scientists to practise and learn with their own data, which has been proved to be the best way to learn, always supported by the most experienced users within the company and learning from each other’s cases. This process is not easy because it requires a total change of culture.

While DoE methodology has been applied experimentally for decades, it is only relatively recently that its usage has gathered momentum across many scientific disciplines. This is due to a combination of advances in the algorithms used to tailor designs to the experiments and an increasing industrial need for rapid experimentation and decision making. For example, addressing design space constraints (11), handling mixture-type factors (12), comparing the effectiveness of different designs (13), introducing uncertainty in the factors and optimising using a variability simulator (14). However, for traditionally trained scientists who are used to changing OFAT in accordance with the scientific method, the transition to DoE methodology can be met with trepidation. There can be a concern that the scientist’s skills are not being fully utilised and that the recommended experiments in the matrix will not be enough to understand the system. Overcoming these anxieties is a significant challenge, particularly within established R&D departments. At Johnson Matthey, the way this has been approached is to demonstrate the power of DoE on small projects across a range of technology areas, and actively promote these results to the rest of the company, increasing the visibility and viability of the DoE methodology. This generates additional interest and establishes confidence in the methods so that scientists have more faith in using DoE for larger, more complicated projects. The functionality of software such as JMP^® to create designs, analyse the results and present the conclusions is essential in facilitating this cultural shift.

At Johnson Matthey, the key principle when driving this transition to an advanced experimentation and statistical approach to data is to empower our scientists to do it themselves. The scientists are the technical experts in their respective areas, and by giving them the understanding, tools and training to create and analyse DoE it is believed that this will result in better outcomes for both the current project and future work programmes.

Indeed, recognising the technical expertise in their respective areas when deploying a statistical software like JMP^® is a very important step to overcome a very important concern; the fear of being substituted by a computer or a machine (15).

The media can be overwhelming in this respect: listening and reading continuously about artificial intelligence, robotics, automation and machine learning. However, technical expertise will always be necessary, and the human being has proved to be indispensable in many fields. Statistical techniques like DoE are not designed to substitute the chemical expertise of a human scientist but to work in conjunction with them to get the most out of experimentation, and to make them more efficient. Statistical tools are exactly that: tools to be used, not to substitute scientists.

Indeed, the first step in a statistical design is the planning. For this step, chemical expertise plays a crucial role. The DoE is not going to tell the scientist which factors or responses should be studied. It is the scientist who should feed all this valuable information into the design.

In the same way, as a result of a DoE it could be found that a variable has or does not have an effect on the output, but it will not say why. It is up to the scientist to interpret the result, try to understand why and continue designing more experiments to test and prove that hypothesis.

When teaching these techniques within Johnson Matthey we are very careful to emphasise these tools are to help scientists, not to replace them. It is crucial to motivate scientists to believe in the process to overcome other major challenges, such as the timings.

Another challenge that scientists experience when using DoE is the lack of immediate visibility of the factors’ effects. In traditional experimentation, the scientists can see the effect that changing an input has on the output once the experiment has finished and then, based on this result, decide the next experiment. However, there is no visibility of progress while carrying out experiments from a DoE as analysis of the results only makes sense once all the experiments of the design have been carried out. This requires some patience and trust from the scientists to see it through. Therefore, the first time that someone uses such tools they will struggle but once they see the results, they understand that the wait was worthwhile. For this reason, it is recommended to start with smaller sets of experiments instead of embarking on a large, complex DoE. Also, to start with a relatively simple DoE such as a full factorial design with only a few factors, to overcome another important challenge, the fear of the ‘black box’.

Another big challenge that pushes scientists away from using DoE is that it is seen as a ‘black box’. A lack of understanding of the technique together with a limited understanding of statistics creates uncertainty and the scientist can feel a loss of control. It is an understandable response and can only be helped by providing the information needed to understand the technique and its benefits.

Work is being done at Johnson Matthey to make sure that scientists are provided with the tools and support necessary, so they can understand the techniques and use them with confidence. Different approaches are taken for this: one-to-one training, in-house and external group training. Also, the use of software such as JMP^® has been critical to empower the scientists at Johnson Matthey to use such tools. The program is easy to use and there are lots of free learning materials available from JMP^®. For new users who do not feel very adventurous, starting with a simple and more intuitive design such as a small full factorial is recommended. Despite starting with something simple, the results will sometimes be unsatisfactory for the experimenters, and might reveal some difficult truths.

The use of statistics and DoE during experimentation might uncover some weaknesses in the way the experiments are carried out. Sometimes, inconclusive results will be obtained from DoE due to irreproducibility issues on the experimentation. It might be tempting to point at the DoE as the problem; however, DoE has only helped to uncover an issue that already existed even when performing traditional experimentation. Instead of seeing this as an issue it should be thought of as an opportunity to improve the way experimentation is carried out and to reduce the experimental error.

The variability observed could be due to many reasons, such as uncontrolled factors that affect the response or its measurement. These could be included in a subsequent DoE to be studied further and help to provide better understanding of the system (Figure 9). To be included in the study, the scientist needs to be able to measure and control the different inputs. Understanding the origin of the variability of an experiment can be used to improve the process. For example, if a more precise measurement of the output can be obtained, the scientist would be able to observe smaller effects of inputs which, combined, could drive larger improvements in the output.

Fig 9.

DoE and statistical tools allow experimenters to obtain reliable data in order to extract objective conclusions and take decisions, even if those conclusions are that the experimentation needs to be redesigned or the measurement system improved.

Pelleting processes are used in the manufacture of a range of consumer and industrial products, including tableted pharmaceuticals, health products, consumer goods, food products and catalysts. In the manufacture of these pelleted products, it is important to produce pellets that are consistent in size, shape, composition, density and strength to ensure that they are fit for purpose. Usually, several of the aforementioned variables will be listed in the product specification along with appropriate bounds that need to be met. Poorly controlled pelleting processes can result in production of out-of-specification material, which is associated with negative economic and environmental impacts, and with potential safety implications in the case of pharmaceuticals for example. Conversely, operating pelleting processes with tight control over the CQAs can allow manufacturers to operate closer to the specification limits and to improve the product, the process economics and sustainability.

To establish tight control over a manufacturing process, it is important to control all the variables in the manufacturing process that have the potential to impact the product CQAs. Before that can be achieved, it is necessary to understand the relationships between the manufacturing variables and the product CQAs and to identify those that are most influential, as well as those that can be manipulated to control the product CQAs (1). Henceforth, knowledge of the interactions between process variables and the response of the process is critical. Typically, models are deployed to help us gain understanding of process behaviour. In the literature, there are three main approaches that have been applied to modelling pelleting process behaviour, these are mechanistic, data-driven and hybrid modelling approaches.

Traditionally, mechanistic models such as the Drucker-Prager Cap (DPC) model and finite element models have been applied to better understand the behaviour of materials undergoing compaction and their resultant physical properties. Wu et al. (2) used finite element methods (FEM) to gain understanding of the behaviour of pharmaceutical powders during compaction. The DPC model was used as the yield surface of the medium, representing failure and yield behaviours. Experiments were carried out using a compaction simulator with an instrumented die to calibrate the DPC model and to investigate the relationship between relative density of the powder bed and the applied pressure during compaction. The DPC model generated realistic powder properties that were fed into finite element analysis (FEA), which was able to accurately model the relationship between relative powder bed density and the compaction force. FEA also allowed close examination of the evolution of stress distribution during relaxation, which revealed narrow bands of localised intensive shear stresses where potential failure mechanisms can initiate. Several other works have utilised purely mechanistic approaches to model pelleting behaviour (3–7).

The drawback to mechanistic approaches is that they take significant human resource to develop and they are not easily adapted to new processes or scenarios. In contrast, purely data-driven approaches are very fast to develop and deploy and are easily adapted to new contexts. Several works have focused on the application of data-driven models to pelleting processes (8–15). Haware et al. (9) applied multivariate analysis to quantify the relationships between material properties of α-lactose monohydrate grades, PPs and the tablet tensile strength. The materials were tableted on a compaction simulator and the collected data were analysed with PCA and PLS regression. PCA provided insights into relationships between different powder and compression properties of the studied materials. PLS was successfully used to predict tablet tensile strength from the compression parameters, punch velocity and the lubricant fraction.

Li et al. (10) used multivariate analysis to evaluate the fundamental and functional properties of natural plant product (NPP) powders and their suitability for direct compaction. NPP powders were prepared by three different methods and data were produced in a single-punch compaction simulator. Results from a one-way analysis of variance, cluster analysis and PCA showed that the physical properties of the NPP powders were mainly determined by their particle structure, which derived from the preparation method. Stepwise regression analysis indicated that the compaction properties of the NPP powders could be improved by controlling physical properties, such as density, particle size, morphology and texture. Overall, the work provided guidance on the development of NPP powders for compaction.

Matji et al. (16) conducted a multivariate analysis on data from the production of ibuprofen tablets. In their study, regression methods were used to predict the CQAs of the tablets, such as disintegration time, dissolution, hardness, porosity and tensile strength, from the pressure applied in roller compaction (dry granulation) and tabletting. Tabletting compaction pressure was found to be positively correlated to disintegration time, tensile strength and hardness, and negatively correlated to the porosity and percentage of drug dissolved. Roller compaction pressure during dry granulation was observed to have those same correlations with the CQAs inversed.

More recent works have also focused on the development of hybrid models for pelleting processes, which combine mechanistic models with data-driven models to leverage benefits from both approaches (17, 18). For example, Benvenuti et al. (17) trained an artificial neural network (ANN) to model the relationship between macroscopic experimental results and microscopic parameters of discrete element method (DEM) simulations. The work showed that the ANNs could be used to generically identify DEM material parameters for any given non-cohesive granular material. Hybrid modelling approaches can offer great benefits where existing mechanistic models are available because they leverage the accuracy and interpretability of the mechanistic model, while offering increased flexibility.

In this work, a data-driven approach is used to model pelleting performance in order to gain a deeper understanding of the process behaviour and to understand the potential of such models for use in process monitoring and control. The work expands upon previous data-driven modelling of pelleting processes by exploring the impact of two different feeder mechanisms: a force feeder and a vibration feeder, operated at different speeds. Furthermore, the product studied is an inorganic catalyst material. Analysis of such materials in pelleting processes has not been widely reported in the literature. The pelleted catalyst product studied is manufactured by Johnson Matthey.

2.1 Equipment

2.2 Experimental Work

2.3 Principal Component Analysis

2.4 Partial Least Squares Regression

In order to compare the effect of the powder feeder mechanism and speed on pellet density and hardness, the distributions of pellet density and hardness were plotted for each experiment and pairwise t-tests were used to check for statistically significant differences in average pellet density and hardness. Figures 2(a) and 2(b) show the distributions of pellet density and hardness, respectively.

Figure 2 shows that the feeder configuration greatly influenced the level of variability in pellet density and hardness. In particular, experiments ‘Left40’ and ‘Right70’ produced a large amount of variability, while the vibration feeder resulted in much less variability at all speeds tested. The pairwise t-tests revealed statistically significant shifts in average pellet density and pellet hardness between the six experiments. For example, with the vibration feeder and the force feeder in ‘left’ orientation, increasing the speed of the feeder resulted in statistically significant increases in pellet density. Importantly, the distributions indicate that the vibration feeder results in more consistent pellet density and hardness than the force feeder, although the force feeder may be used in the ‘left’ orientation at high speed to minimise variability.

Fig. 2.

PCA was used to assess the correlations between the variables in the compaction simulator summary dataset and their contributions to the overall process variability. The PCA model presented here was built on summary data from the experiments using the force feeder, namely ‘Left40’, ‘Left70’ and ‘Right70’. Figure 3 shows the variance explained by each PC in an eight PC model. The first PC explains 49.3% of the variation in the data, while PCs 2 and 3 explain 14.1% and 7.2%, respectively. Collectively, the first three PCs capture 70.6% of the variance in the data, while PCs 4 and beyond explain less than 5% of the variance each. Due to the exploratory nature of this analysis, it is preferable to consider all the PCs that are likely to be informative; however, determination of the number of PCs to include in the model is not critically important. In this work, the first three PCs were analysed because PC 4 and beyond capture very little variance and likely feature a low signal to noise ratio.

Fig. 3.

The scores of the PCA model for PCs 1 to 3 are displayed in Figure 4 and the most important variables – those with loadings of the highest magnitude for each PC – are displayed in Figure 5. PC 1 captures variation in the data that is present on a pellet-to-pellet basis within each of the three experimental runs and the scores overlap, i.e. there is no separation of the scores by experiment on PC 1. Figure 5(a) shows that the 49% of variation captured in PC 1 is attributed to the forces and energies involved in the pre-compaction, main-compaction and ejection events. The model also indicates that the forces and energies listed in Figure 5(a) are all positively correlated with one another.

Fig. 4.

Fig. 5.

In contrast to PC 1, PC 2 captures variation that separates the different experiments. Figure 5(b) shows this is largely attributed to differences in die filling height, ejection force and some derived parameters which are determined from die filling height, such as the compression and relaxation times. The large spread of the red markers in the vertical plane, corresponding to the ‘Right70’ experiment, indicates that this set up resulted in more variability in the die filling height compared to ‘Left40’ and ‘Left70’. The compaction simulator was calibrated to produce pellets of the same weight for each experiment; therefore, it is likely that the different feeder setups result in a different bulk density of the powder when it is initially filled into the die, explaining the differences in filling height.

Figure 4(b) shows the within group variability that is captured by PCs 1 and 3. All of the experiments overlap on PCs 1 and 3, while ‘Right70’ spreads across the largest area, indicating that this experiment has the largest variability on these PCs. Observing the distributions of the response variables, it is clear that ‘Right70’ produces pellets with the largest variation in pellet density and hardness. As shown in Figure 5(c), the main variables represented by PC 3 are the energies involved in the pellet ejection and the elastic energy calculated for the main-compaction event.

While the ejection plastic energy and compression energy correlated with the pre-compaction and main-compaction forces, the ejection energy and the ejection force appeared as key variables in PCs 2 and 3, indicating that that there is variance in the ejection force that is uncorrelated to pre-compaction and main-compaction force. PC 2 shows that ejection force is positively correlated to die filling height. An additional factor that may be influencing the ejection force that is not monitored, so not captured by the dataset, is the amount of lubricant present in the material.

PLS regression models were developed to predict pellet density and pellet hardness, using the methodology outlined in Section 2. Table II shows the performance metrics obtained from cross-validation and testing of the two models for pellet hardness and density.

The performance metrics shown in Table II are for the models obtained after the variable selection procedure and tuning of the number of latent variables included in the model, as described in Section 2.4. The models for both pellet hardness and density performed well in cross-validation and testing. Pellet hardness however proved to be the more difficult to model and predict accurately, as the performance metrics demonstrate. The pellet density PLS model explained approximately 86% of the variance in both 10-fold cross-validation and independent testing. The MAE for the pellet density model was 0.17 and 0.18 in 10-fold cross-validation and independent testing, respectively. The cross-validation metrics indicate that the pellet density PLS model should have excellent predictive performance and the independent test on the held-out data supports this. The fit of the density PLS model to the training data and the testing data is shown in Figure 6.

Fig. 6.

Figure 6 shows that the density PLS model fits the data well in both training and testing, with the exception of a few outliers, which are likely to result from a mismatch between the X and Y data that occurred during the experimental process.

The pellet hardness PLS model explained approximately 67% and 63% of the variance in pellet hardness in 10-fold cross-validation and testing, respectively. The MAE for this model was 0.44 and 0.53 in cross-validation and testing, respectively. While this performance is not as good as the pellet density model, it indicates that the model has good predictive capability. Observation of the measured versus fitted values and the residuals in Figure 7 reveals that the model is good at fitting and predicting the low-hardness pellets but is far less accurate for the high-hardness pellets. The residuals in Figure 7(b) and 7(d) ‘fan-out’ and become much larger from –1 and above on the x-axis (scaled pellet hardness). The increasing residuals with increasing pellet hardness could be related to missing factors that are not captured in the compaction simulator data. It could however equally be the result of changes in the sensitivity of the measurement device at different hardness levels. Unfortunately, it is difficult to gain a good understanding of the reliability of the hardness measurement due to the test being destructive. Given that the model offers accurate prediction of low hardness pellets, the model could still be valuable in a process monitoring or supervisory process control applications, where identification of low hardness pellets could allow operators to intervene early and adjust PPs accordingly.

Fig. 7.

5.1 Interpretation of the Model Coefficients

The workflow for this study began with exploratory data analysis to observe and compare the distributions of pellet density and hardness and to understand the variance and correlation in the compaction simulator data. The distribution plots clearly showed that feeder configuration impacted both the average pellet density and hardness and the level of variability in those properties. Pairwise t-tests revealed that the shifts in mean density and hardness were significant in many cases. To minimise overall variability, the vibration feeder mechanism should be favoured over the force feeder mechanism, however, the force feeder mechanism can be optimised for consistency by running in the ‘left’ orientation at higher speeds.

PCA showed that the most important component of variance in the dataset was attributed to the forces and energies involved in the pre-compaction, main-compaction and ejection events. The main differences between the force feeder experiments that were highlighted were due to the die filling height and associated parameters, and some observable differences in overall variability. The high level of variability in pellet density and hardness for experiments ‘Right70’ and ‘Left40’ was also reflected in the energies and forces recorded in the compaction simulator summary data through the PC scores.

The PLS models that were developed for pellet density and pellet hardness performed well in cross-validation and testing. The key predictor variable in the models for pellet density and hardness was the lower punch force, which correlated positively with both. The density PLS model explained around 86% of the variance in the response, while the hardness PLS model explained around 65%. Both models offer predictive capability, however, the hardness model underperforms at predicting the medium to high hardness pellets. Unfortunately, it is difficult to gain a good understanding of the reliability of the hardness measurement due to the test being destructive. The questions raised in this study highlight the need to further validate the pellet hardness measurement in future work. For now, the learning from this work indicates that the pellet density can be used more reliably as an indicator of product quality than pellet hardness. Pellet density should therefore be the preferred basis for process monitoring and control applications.

If a similar model for pellet density can be developed using the data available from production scale pelleting machines, then there would be the potential for such a model to be used for process monitoring and control. This could provide real-time process monitoring that greatly improves upon existing techniques for monitoring pelleting performance, which are based on random sampling and testing of the pellets. The information could be used by plant operators at a supervisory level or in an automated control system to help inform and guide decision making to keep the process on track and producing on-specification material.

Metal matrix composites (MMCs) are systematic combinations of two or more materials (one of the materials is a metal) engineered to achieve tailored properties (1). Thus, engineered MMCs have two or more chemically and physically distinct phases that are suitably distributed to provide properties not attainable with either of the individual phases (2). AMMCs exhibit better mechanical and physical properties than the aluminium-matrix alloy (3–5). The hardness and strength of AMMCs are significantly higher than that of the aluminium-matrix alloy, leading to improved wear resistance (6). AMMCs have found applications in aerospace, automotive, nuclear, telecommunications (7) and marine industries (6). The applications of AMMCs in the automotive and aerospace industry sectors can reduce fuel usage by replacing steel and cast-iron parts with lighter AMMCs. Some of these applications include pistons, piston rings, cylinder liners, connecting rods (1), cylinder blocks, driveshafts and brake drums (6). The tribological behaviour of particle reinforced AMMCs has been regularly reported. However, most of the studies have analysed the tribological behaviour of composites reinforced with SiC and Al₂O₃ particles. Besides these conventional reinforcement particles, aluminium alloys also can be reinforced with h-BN (8), TiC (9), TiO₂ (10), ZrO₂ (10) and B₄C (11) to impart wear resistance. Studies on AMMCs reinforced with B₄C particles have been limited mainly due to the higher cost of B₄C particles than SiC and Al₂O₃ particles (12). Al-B₄C composites are commonly used in automotive, sports (7) and neutron shielding applications (13).

B₄C possesses excellent properties such as high hardness, low density, high melting point, chemical inertness and wear resistance, making it suitable for many high-performance applications (14). The hardness of B₄C (Vickers Hardness under the load of 0.981 N (HV0.1) = 3200) is far superior to the hardness of conventional reinforcement particles, SiC (HV0.1 = 2500) and Al₂O₃ particles (HV0.1 = 1900) (15). The density of B₄C (2.52 g cm^–3) (16) is less than the density of solid aluminium (2.70 g cm^–3) (17), which significantly improves specific properties. The densities of SiC, Al₂O₃ and B₄C are 3.21 g cm^–3, 3.92 g cm^–3 and 2.52 g cm^–3, respectively. The density of molten aluminium is 2.38 g cm^–3 (17). Hence, it is evident that the difference in density between molten aluminium and B₄C is lower when compared to the difference in density between molten aluminium and conventional reinforcement phases (SiC and Al₂O₃). This phenomenon minimises the sedimentation of B₄C particles at the crucible bottom during stir casting (12). The abrasive resistance of B₄C (0.4–0.422 (expressed in arbitrary units)) is higher than that of SiC (0.314 (expressed in arbitrary units)) due to its high hardness and strength (18).

This overview aims to discuss the microstructural characteristics, mechanical properties and tribological behaviour of Al-B₄C composites. The different properties and microstructural characteristics of Al-B₄C, Al-SiC and Al-Al₂O₃ composites are compared. Furthermore, the statistical significance of physical parameters (applied load, sliding speed and sliding distance) on the tribological behaviour of the composites is analysed. However, the literature that compares the microstructural characteristics, mechanical properties and tribological behaviour of Al-B₄C, Al-SiC and Al-Al₂O₃ composites are insufficient. The literature on the statistical analysis of the tribological behaviour of Al-B₄C composites is also sparse. Despite these shortcomings, this overview discusses the mechanical and tribological properties of the composites mentioned above. Section 2 gives a brief insight into the fabrication of Al-B₄C composites through the stir casting technique. Section 3 compares the microstructural characteristics of Al-B₄C, Al-SiC and Al-Al₂O₃ composites. Section 4 analyses the tribological behaviour of Al-B₄C composites. The tribological properties of Al-B₄C and Al-SiC composites are also compared in Section 4.

Many methods are available to fabricate MMCs, and the commonly used two primary processes are: (a) solid-state processes; and (b) liquid state processes (6). Liquid state processes include infiltration techniques (pressure infiltration and squeeze casting) and dispersion techniques (stir casting and compocasting). The stir casting (vortex addition) technique has been the most studied method for producing AMMCs due to its simplicity, flexibility, commercial viability and ease of processing (19, 20). The core requirement of the stir casting of MMCs is close contact and bonding between the ceramic phase and the molten alloy. The wettability of the ceramic particles to molten melt is inherently weak. Thus, intimate contact and bonding between them are enhanced by artificially inducing wettability or using an external force to weaken the thermodynamic surface energy barrier. One of the commonly used methods to incorporate, wet and uniformly distribute the ceramic particles is to add the particles to a vigorously stirred molten melt. The stirring action (external force) enhances wetting and ensures homogenous dispersion of reinforcement particles through the matrix. Wettability is also induced artificially by modifying the chemical composition of the matrix alloy: small quantities of reactive elements, such as magnesium, calcium, lithium or sodium, are added (20). The addition of magnesium improves the wettability of Al₂O₃ and SiC particles to the alloy matrix, which increases the wear resistance of Al-Al₂O₃-SiC hybrid composites (21).

Lashgari et al. (22) reported that during stir casting, the addition of magnesium improved wettability between the matrix (A356) and reinforcement particles (B₄C). The reinforcement particles were preheated to enhance the wettability of the ceramic particles with the metal matrix. Details of the stir casting technique and the particle size of B₄C particles are shown in Table I. Mahesh et al. (23) preheated the reinforcement particles to remove impurities and to enhance the wetting characteristics. Canakci et al. (24) observed that the vortex formed due to stirring holds the reinforcement particles dispersed in the melt, which ensured their uniform distribution. After particle addition, the composite melt is poured into a permanent mould. Kalaiselvan et al. (25) fabricated AA6061-B₄C composites reinforced with 4 wt%, 6 wt%, 8 wt%, 10 wt% and 12 wt% B₄C particles through the stir casting process. Uniform distribution of reinforcement particles was observed at all weight percent additions. Furthermore, X-ray diffraction (XRD) analysis of the composites revealed that there is no reaction of the AA6061 matrix with the B₄C particles. This phenomenon shows the thermodynamic stability of B₄C particles at the temperature (920°C) used for the stir casting of AA6061-B₄C composites. The parameters used by Lashgari et al. (22), Mahesh et al. (23), Canakci et al. (24), Kalaiselvan et al. (25), Toptan et al. (26), Mazahery and Shabani (27), Toptan et al. (28) and Baradeswaran and Perumal (29) for the stir casting of Al-B₄C composites are listed in Table I.

Shorowordi et al. (30) studied the matrix-reinforcement interface of Al-20 vol% SiC (Figure 1(a)), Al-20 vol% Al₂O₃ (Figure 1(b)) and Al-13 vol% B₄C (Figure 1(c)) composites produced through the stir casting technique.

Fig. 1.

The microstructure and interfacial characteristics of Al-SiC, Al-Al₂O₃ and Al-B₄C composite are extensively reported in this study. The interfacial reaction product is not observed for the Al-B₄C composite, unlike the Al-SiC composite, which revealed an apparent interfacial reaction. Furthermore, it was observed from the fracture surfaces that Al-B₄C composite exhibited the strongest bonding at the matrix-reinforcement interface, and the bonding of Al-SiC composite is weak due to the low adherence of aluminium matrix to the SiC particles. In Al-Al₂O₃ composites, voids and microvoids are observed at the interface, indicating poor bonding. Moreover, particle distribution is found to be better for Al-B₄C composite when compared to Al-SiC and Al-Al₂O₃ composites.

The mechanical properties of spray-cast Al 6061-15 vol% B₄C and aluminium 6061-15 vol% SiC composites have been reported (31). The B₄C reinforced composite exhibited significantly greater strength, strain to failure in tension and strain hardening compared to the SiC reinforced ones, due to strong bonding at the Al-6061-B₄C interface (31). The strong bonding at the interface is ascribed to the chemical stability of B₄C particles, the absence of interfacial reaction products and the excellent wetting of the particles by the matrix alloy. The wetting characteristics of the Al-6061-SiC composite are weaker than that of the Al-6061-B₄C composite.

3.1 Influence of Boron Carbide Particles Addition on Hardness

An overview of the literature on the tribological properties of Al-B₄C composites is provided in the following subsections. The tribological properties are controlled by the physical parameters (applied load, sliding speed and sliding distance) and material parameters (the type of reinforcement and volume fraction) (35). Hence, the overview is focused on analysing the influence of physical and material parameters on the dry sliding tribological behaviour of the composites. The relevant details of sliding wear studies are shown in Table II.

4.1 Effect of Variation of Applied Load

4.2 Effect of Variation of Sliding Distance and Sliding Speed

4.3 Influence of Mechanically Mixed Layer

4.4 Beneficial Effects of Boron Carbide Particles Addition

4.5 Comparison of Tribological Properties of Aluminium-Boron Carbide and Aluminium-Silicon Carbide Composites

4.6 Inferences Obtained from the Statistical Analysis

The fabrication and tribological properties of Al-B₄C composites are discussed in this overview. The Al-B₄C composites exhibited better particle distribution than Al-SiC or Al-Al₂O₃ composites. The bonding at the matrix-reinforcement interface is also strong, and the interface is free of the interfacial reaction product, which is not the case with Al-SiC and Al-Al₂O₃ composites. The presence of a brittle phase at the matrix-reinforcement interface reduced the wear resistance of Al-SiC composites. The friction coefficient of Al-B₄C composites is lower than that of Al-SiC composites due to the presence of oxidised boron on the contact surfaces. The better tribological properties of Al-B₄C composites compared to those of pure aluminium are due to the abrasion resistance imparted by the B₄C particles. The wear mechanisms induced during wear studies of Al-B₄C composites are plastic deformation, adhesion, abrasion and delamination. Statistical analysis revealed that the influence of physical and material factors and their interaction on the tribological behaviour is statistically significant.

To summarise, Al-B₄C composites exhibit better microstructural characteristics than aluminium-matrix composites reinforced with SiC and Al₂O₃ particles. The tribological properties of Al-B₄C composites are better than those of aluminium and Al-SiC composites; thus, these composites may be considered as a potential candidate for different tribologically crucial applications.

In this research, desized, 100% cotton fabrics (warp/weft yarn density of 34/17 yarns per centimetre) were used. Mikrathermic^® P PCM capsules were provided by Devan Chemicals, Belgium. For the coating process, Mikracat B as a cross linker and L Mikrasoftener as a softener were supplied from Devan Chemicals. RUCO^®-COAT PU 1110 polyurethane coating material was used for coating process and supplied from Rudolf Duraner, Turkey. EDOLAN^® MR polyurethane binder was used for the impregnation method and provided by Tanatex, Switzerland to bond the microcapsules to the fabric. All other auxiliary chemicals used in the study were of laboratory-reagent grade.

The application of the capsules to the cotton fabrics was carried out by impregnation and coating methods. Fabrics were conditioned in accordance with ISO 139:2005 (43) at standard atmospheric conditions (20°C±2 and 65% RH±4) for 24 h. Capsule transfer prescriptions were made according to Tables I and II and in the same ratio to compare the application processes. Polyurethane was selected as binder and each experiment was repeated three times.

The capsules were impregnated in a solution bath containing capsules (125 g l⁻¹) and binding agent (30 g l⁻¹), and then squeezed between rollers to 90% wet pick-up. Achieving long lasting effect, the fabric was exposed to drying for 10 min at 80°C and fixation process for 3 min at 140°C in a laboratory stenter (Table I).

Viscosities of the coating pastes were measured using a DV-II+Pro viscometer (AMETEK Brookfield, USA) and the viscosity of the coating paste was determined to be 9000 cps. Cotton base fabrics were coated with the above mentioned coating pastes using a laboratory type blade coating machine, as two layers of coating. It was subjected to intermediate drying at 100°C for 2 min between each layer. Coated samples were cured at 140°C for 3 min.

SEM images were taken to obtain the existence of capsules on the textile surface from both coated and impregnated samples. Samples were gold-coated (15 mA, 2 min) to assure electrical conductivity. The measurements were taken at 2 kV accelerating voltage. The images were taken at 250× and 1000× magnification.

Thermal properties of the fabrics, such as melting and crystallising temperatures and enthalpies, were measured by DSC performed using a PYRIS^TM Diamond differential scanning calorimeter (PerkinElmer Inc, USA) to distinguish the capsules on the fabric with the help of characteristic endothermic or exothermic peaks. The samples were cooled down to −20°C and then heated up to 40°C at a constant rate of 10°C min⁻¹ under a nitrogen flow rate of 60 ml min⁻¹.

In order to examine the efficiency of the transferred capsules, the surface temperature of the raw fabric samples containing PCM was measured at a certain time interval by thermal camera as shown in Figure 1. Measurements in the system were made in an insulated box. Before measurement, the inner temperature of the box was heated to a constant temperature of 40°C and the test was carried out at this temperature. The inner temperature of the box was kept constant by means of a thermostat. Before measurement, the fabrics were conditioned for 12 h and placed in the box as quickly as possible. Once the fabric was placed in the box, the surface temperature was measured from a fixed point for 15 min. A thermal camera (Fluke Ti100 Thermal Imager, Fluke, USA; emission value 0.94) was used in the measurements and the temperature was recorded every 30 s.

Fig. 1.

When an interface exists between a liquid and a solid, the angle between the surface of the liquid and the outline of the contact surface is described as the contact angle θ (lower case theta). The contact angle (wetting angle) is a measure of the wettability of a solid by a liquid. In order to examine the hydrophilicity of the fabrics, the contact angle was examined. The measurements were carried out at 25°C using the Theta Lite T101 (Biolin Scientific, Sweden) model contact angle device. An image of approximately 5 μl of water droplet dropped onto the surface to be measured was recorded for 10 s by the device camera. Using the device software, an average of 200 data were recorded for 10 s for each sample and the arithmetic mean was taken.

Water vapour permeability is related to breathability of fabrics. Water vapour permeability of samples was determined by using M261 (SDL Atlas International, USA) model water vapour permeability tester according to BS 3424-34:1992-Method 37 (44). The amount of water vapour passed through the samples was determined after 24 h and permeability values were calculated. The test was repeated three times for each sample type.

SEM images of the Mikrathermic^® P PCM capsule are shown in Figure 2. Mikrathermic^® P was around 3 μm and had a spherical shape as expected. SEM images of the PCM capsules transferred to cotton fabrics by coating and impregnation methods, enlarged 250 times and 1000 times, are given in Table III.

Fig. 2.

Table III

SEM Photomicrographs of Fabrics Treated with PCM Capsules

Fabric	250×	1000×
Coated
Impregnated

When the images were examined morphologically, it was observed that the capsules transferred by the impregnation method preserved their spherical form. PCMs transferred by coating remain under the coating polymer and were homogeneously distributed over the entire surface. These images showed that capsule application was successful for both impregnation and coating methods. It was observed that capsules were covered with the binder and fixed onto the textile surface of the cotton fabrics.

The DSC diagrams of coated and impregnated fabrics are given in Figure 3. The heat storage capacity of the Mikrathermic^® P PCM microcapsule is 140 J g⁻¹ according to the literature (45–47). From the DSC curve given in Figure 3 and from Table IV, the amount of heat stored and emitted by the fabrics from the area under the endothermic and exothermic melting and solidification peaks and the temperatures at which heat storage and emission begins can be seen. According to the DSC analysis, similar values were obtained for coated and impregnated fabrics. The values are provided in Table IV in detail.

Fig. 3.

The melting process in fabrics coated with Mikrathermic^® P microcapsules occurred between 25.83°C–31.04°C and the amount of heat energy stored by the cotton fabric during the melting period was measured as 2.70 J g⁻¹. For the Mikrathermic^® P microcapsule, the crystallisation process occurred in the range of 25.70°C–23.45°C and the cotton fabric released −1.45 J g⁻¹ heat during crystallisation. Impregnated fabric absorbed 2.70 J g⁻¹ at 25.72°C during melting and released −1.39 J g⁻¹ at 25.61°C during crystallisation.

Thermal conductivity measures the capacity of temperature exchange between heat and cold passing through a material mass. Decreased thermal conductivity allows for a faster rate of heat transfer in a PCM, increasing the time required for the PCM to undergo a complete charge or discharge. The major shortcoming of PCM is its limited ability to exchange heat effectively due to low thermal conductivity. This suppresses the amount of heat that can be exchanged during melting processes and a lower thermal conductivity of solidification will occur at low temperatures. The effective thermal conductivity of PCM can be increased by many mechanisms such as inserting fins and adding a dispersion of high thermal conductivity nanoparticles (48, 49).

Although the process temperatures are very close to each other, coated fabrics changed state at higher temperatures compared to impregnated fabrics. The shifting of the process peaks to higher temperatures has been explained in the literature as the lower thermal conductivity of the fabric (50). This situation was interpreted as the lower thermal conductivity value of coated fabrics compared to impregnated fabrics resulting in melting and solidification at higher temperatures. However, considering that these data are very close to each other, it was thought that the capsules can be transferred to the fabrics by the coating method. Encapsulated PCMs which were transferred with coating and impregnation lead to lower thermal conductivity and increased heat capacity of a textile structure. They improve the thermal performance of textile material and therefore may save energy.

Depending on the change in ambient temperature, the fabric surface temperature change caused by PCM capsules was measured. For this purpose, a thermal camera was used to determine the heat regulation properties of fabrics that can store heat. Two measurements were taken from two different points in the fabric samples and their averages are shown in Figure 4.

Fig. 4.

The temperature-time curves are given in Figure 4. It can be seen from the graphs that the fabrics which were brought from a cold environment (4°C±2) to a warm environment (40°C±2) were warmed and the temperatures measured on their surfaces increased. On the other hand, it is observed that the heating time of the fabrics in a hot environment and the maximum temperatures reached were not equal. According to both measurement results, it can be seen that the raw fabric heats up the fastest. Similarly, the maximum surface temperature of the raw fabric was higher than the fabrics containing PCM. The raw fabric warmed to almost maximum temperature (about 42°C) in about 5 min. For fabrics containing PCM, the maximum temperature recorded was lower at the end of the measurement period. The maximum value recorded was 37°C for the fabric in which the PCM capsules were impregnated and 40–41°C for the fabric transferred with the coating. Thermal camera analysis was performed for 15 min. It was determined that the temperature of the fabrics remained at the last point which they reached for an extended period. During the measurement period, it was determined that the temperature measured on the surface of the fabric to which the PCM capsules were impregnated was 3°C to 5°C lower than the raw fabric surface temperature. It was determined that the surface temperature of the fabric to which the PCM capsules were transferred with the coating was 1–3.5°C lower than the raw fabric.

When the analysis results were evaluated, it was seen that the fabric with PCM transferred by the impregnation method has more effective temperature regulation. The impregnated fabric, which has the lowest temperature, absorbed more heat in the cold environment when the PCM structure was applied. It also appears that there was not a big difference between coating and impregnation methods in thermal camera analysis. The thermal camera method demonstrates the heat regulation ability of fabrics, but does not provide information about their performance in end-use areas. Therefore, for fabrics treated with coating and impregnation methods, performance evaluation according to the area of use will give the most accurate results. This shows that PCM capsules can also be transferred by the coating method, depending on the usage areas.

In order to evaluate the hydrophilicity properties of raw fabric and PCM-transferred fabrics with different methods, contact angle measurement was made as shown in Figure 5.

Fig. 5.

The angle between the surface of the liquid and the outline of the contact surface is described as the contact angle θ. The contact angle is a measure of the wettability of a solid by a liquid. In the case of complete wetting, the contact angle is 0°. Between 0° and 90°, the solid is wettable and above 90° it is not wettable. When the analysis results were examined, water was completely absorbed by raw fabric in 5 s and this indicates that the fabric is hydrophilic. When comparing the transfer methods of PCM capsules, contact angle of impregnated and coated fabric was obtained as 42° and 73°, respectively. In general, the coating paste has a more viscous structure and this structure causes a thick layer to form on the fabric. Due to this structure, the surface energy of the fabric decreases and it gains water repellency. In the impregnation method, since a viscous structure is not obtained and a layer is not formed on the fabric surface, the contact angle becomes lower causing the textile material to be more hydrophilic than the coated one. This result, as expected, was that the coated fabric was more hydrophobic than the impregnated fabric.

Water vapour permeability analysis was carried out to examine the comfort properties of the fabrics obtained. Water vapour permeability of samples are tabulated in Table V.

The highest water vapour permeability was obtained from raw fabric with 625 g m⁻² per 24 h permeability value. It was determined that the fabrics with PCM transferred by the impregnation method gave a similar result to the raw fabric. On the other hand, water vapour permeability of coated samples reduced to approximately 50% that of the raw base fabric in parallel with the contact angle results. This was due to the additional polyurethane coating layer which contributed mass transfer limitation through the fabric. Even the most breathable coating polymer applied to the samples would add a resistance to vapour flow by closing the pores and creating an additional layer (51). The water vapour permeability of a material plays an important part in evaluating the physiological wearing comfort of clothing systems or determining the performance characteristics of textile materials used in technical applications. Therefore, it is important to choose the transfer method of PCM capsules considering the area where the fabric will be used.

There are few mathematical breakthroughs that have had as dramatic impact on the scientific process as the Fourier transform. Defined in 1807 in a paper by Jean Baptiste Joseph Fourier (1) to solve a problem in heat conduction, the integral transform, Equation (i):

(i)

and its inverse, Equation (ii):

(ii)

provide the framework to determine the spectral make up of a time varying function g(t) using Equation (i). Conversely, if the frequency domain is understood G(ω), the time signal can be derived using Equation (ii). The same analysis can be applied to spatial functions to yield wave number spectra and is the basis for a significant portion of wave optics, and is used in techniques such as Fourier transform infrared (FTIR) spectroscopy (2).

The transform, which is part of a wider family of integral transforms (3), had a profound impact on the development of much of 19th and 20th century mathematical physics. Previously intractable problems in optics, electromagnetism and acoustics became soluble. The insights these breakthroughs yielded paved the way for quantum mechanics and much of modern science. The famous Heisenberg uncertainty principle is actually just a mathematical property of the Fourier transform in Schrödinger’s wave mechanics (4). Domínguez gives a good overview of this history and some of the mathematical properties of the transform that make it so useful (5).

A significant hurdle with the practical application of the Fourier transform in real-world problems is that it is mathematically challenging to calculate for even the simplest of functions. As a consequence the transform is not taught in the UK until undergraduate level and even then only in mathematically heavy courses such as mathematics, physics and engineering. To make progress in practical problems numerical methods are generally required, meaning the practical application of the Fourier transform can feel like an esoteric part of computer science, rather than the scientific core of the modern world.

Fortunately, the great leaps in understanding that quantum mechanics gave us in electronics has ultimately led to a situation where anyone who wants to, can with a few lines of Python (6) code use sophisticated algorithms that have been developed in the post-World War II period. As such, calculations of the Fourier transform are readily available to those that would like to make use of them.

Unfortunately, the education around how to do practical Fourier analysis has become something of a dark art, which is often picked up in an ad hoc manner in postgraduate studies. The advent of accessible artificial intelligence algorithms has further obscured the basic techniques of Fourier analysis and created a strange scenario where even basic spectral methods are being conducted with inefficient computationally heavy neural network approaches.

In this short article we outline some basic practical steps for successfully conducting Fourier analysis. We will also give a few example Python scripts so the interested reader may apply these techniques to their data.

The first challenge for any numerical method is the digitisation step during which the smooth curves of analytical functions must be turned into discrete numbers. There are two sources of data that are normally digitised:

Discussing these in turn, when an analytic function of time g(t) is evaluated, it is relatively trivial to generate the discretised function with N samples in the time window 0 < t ≤ T (Equations (iii) and (iv)):

(iii)

where

(iv)

The numerical value of δ is of crucial importance in numerical estimates of the Fourier transform. It places limits on what information is lost in the discretisation and plays a fundamental role in how experimental work should be designed. It is more usual to quote its reciprocal, which is the sampling frequency, f_s (Equation (v)):

(v)

It is this frequency that appears in one of the most important results associated with the Fourier transform: Nyquist’s theorem (7). This result states (Equation (vi)):

(vi)

where B is the highest frequency component in the signal in g(t).

Nyquist’s theorem is particularly important as we turn our discussion to sampling experimental data. In theoretical work one can choose, in principle, δ to be as small as is necessary. However, in experimental work this is not an available option; the cost of data loggers increases significantly with the sampling frequency and data storage problems quickly become limiting. Moreover, in nearly all applications where data is recorded by a computer, signals are voltages recorded by an analogue to digital converter (ADC). To conduct scientific work a 12 bit ADC is the standard level. This means that a voltage signal varying between a nominal full-scale deflection ±10 V is recorded to the nearest 5 mV as defined in Equation (vii):

(vii)

When numerical results are compared to experimental results this level of precision must always be borne in mind, as the limitations of the sampling frequency or the voltage level are both likely to be significantly more coarse grain in the experimental work. An example of the effects of this digitisation step is shown in Figure 1. A 5.01 Hz sine wave has been sampled for 1 s, with a sampling frequency of 200 Hz. The blue dots denote the locations of the sampled data and the red curve the analytic form of a sine curve with this frequency.

The popular data analytics tool Jupyter (8) was used to generate the graph shown in Figure 1, this is part of the open-source data analytics bundle Anaconda. The code used is shown in Figure 2. The majority of the code is presentational and associated with plotting the graph using the Python library matplotlib (9). However, the numerical analysis makes use of the versatile NumPy library (10). The key lines for our discussion are lines 17 and 18, which generate two vectors Vs and Vss. The vector Vss is the smooth underlying 5.01 Hz sinusoidal signal and Vs is the signal sampled with a sampling frequency of 200 Hz. It is these two vectors that are manipulated in the sections that follow.

Fig. 1.

Fig. 2.

Having defined the digitised signal, discrete Fourier transform (DFT) can be defined as shown in Equations (viii) and (ix):

(viii)

where

(ix)

The DFT is simple enough to code from first principles that it is often used as an example numerical problem to teach students how to use loops in a given programming language, however it is rarely used in production code because it is computationally inefficient. As the number of samples increases, the number of calculations increases with the square of the number of samples (O(N²)). If this efficiency problem had not been solved in a paper by Cooley and Tukey (11), where they introduced what is known as the fast Fourier transform (FFT), a significant amount of the telecommunications sector would not have been possible. The algorithm they published was actually first discovered by Gauss in 1809 in an unpublished paper and uses a divide and conquer technique. The original time series is split into odd samples and even samples; and then a recursive approach used to construct the Fourier spectrum. This is the reason that many implementations of this algorithm impose the restriction that the number of samples should be a power of two, as this improves the operational efficiency. The efficiency of the FFT scales as O(N log N) opened up the possibility of using Fourier analysis in technical areas that previously would not have been possible.

It is not an exaggeration to say the FFT revolutionised electronic engineering and in turn computer science. Nearly all digital communications rely on the FFT in some form. A measure of how integral to the mathematical sciences the algorithm has become is that improvements to the algorithm continue to the modern day, for example a particularly fast and robust implementation of the FFT called the ‘fastest Fourier transform in the West’ (FFTW) was developed and maintained by academics at the Massachusetts Institute of Technology (MIT), USA (12), and remains an active project. Despite how readily available FFT algorithms have become it is still easy to make mistakes when using them in a real-world example. A raw power spectrum of the time series shown in Figure 1 is shown in Figure 3. The spectrum is shown on a log scale to highlight the detailed features that might otherwise be missed.

Fig. 3.

The first and most important point is that the spectrum plotted is actually a power spectrum. Theoretically this is defined as Equation (x):

(x)

where the G^*_k is the complex conjugate of each Fourier component. The process of finding the power spectrum is lossy, as all phase information in the signal is lost. Despite this, there are many situations where the power spectrum is a much more useful quantity than the raw time series. In this example the large peak at 5.01 Hz, which is seven orders of magnitude above the noise floor, easily identifies the main frequency present in original times series. The code snippet in Figure 4 illustrates how simple using the FFT is with a modern analytics package like Jupyter. Line 2 takes the sampled data Vs from Figure 2, calculates the FFT and converts it into a power spectrum (by taking the absolute value and squaring each component of the vector). Line 3 is simply the calculation of the frequency associated with each bin in the spectrum and is determined by the original sampling frequency fs of the signal. The remainder of the snippet is about presenting the spectrum on a graph.

Fig. 4.

The ideal nature of the original time series used to calculate the power spectrum shown in Figure 3 obfuscates some of the limitations of this naïve brute force use of the FFT. A typical experimental time series has underlying electrical noise and the time digitisation further distorts the signal. In the following sections we shall discuss the best practice that should be followed to get the best estimate of a power spectrum from an experimental signal. We first simulate what a noisy experimental signal might look like by adding Gaussian noise and then splitting the data into 20 different finite levels to simulate the effect of an analogue to digital converter. The three signals are shown in Figure 5. The digitised noisy signal is representative of many experimental signals met in practice.

Fig. 5.

The main challenge with any experimental setup is designing the experiment to give the best answers we can reasonably expect. The processing of a time series to give the most spectral insight is no different. In this section we will attempt to give some basic guidelines that a novice time series analyst should follow, where possible, when conducting spectral analysis.

4.1 Filter High Frequency Signals

Once a low pass filter has been applied to the signal it is sensible to resample the data at the lower frequency to enable more details of the spectrum to be resolved in the region of interest. This process is called downsampling (15) and should only be done if there are no higher frequency components that are likely to interfere with the results. Since we have applied a low pass filter there are no higher components in the time series data, hence downsampling can be applied safely. The reasons for doing this are perhaps not obvious at first sight, but as discussed in the next section, the computational impact of having an oversampled time series can be significant, particularly when fine frequency resolution is required in the power spectrum.

If one considers two notes of frequency f₁ and f₂ which are played at the same time, a third lower frequency can be heard. This is called a beat frequency, f_b (Equation (xi)):

(xi)

If the notes are nearly the same frequency, the beat frequency becomes very small, vanishing to zero when they are identical. Guitarists sometimes use this effect to tune their instruments. This point illustrates that in order to distinguish between two frequencies of slightly different tones, the frequency resolution is limited by the length of the time series recorded. To increase frequency resolution one must record longer time series. The impact of increasing the sampling time frame can be quite dramatic. The power spectrum shown in Figure 8 is what is obtained if 25.8 s of data are used at the 40 Hz sampling frequency (1024 data points). The fundamental peak at 5.01 Hz is much sharper allowing for a better resolution of the frequency.

Fig. 8.

If the downsampling step had not been performed to get the same resolution, the number of data points included in the FFT would need to be increased five-fold for no benefit. It is tempting to simply take very long time series and then calculate the power spectrum with a very large number of data points. However, this can be counter productive, not to say computationally inefficient. The spectrum shown in Figure 8 has 512 different frequency bins for 0 < f < 0. 5f_s, which gives a resolution of Equation (xii):

(xii)

If a frequency resolution finer than this is required then it is reasonable to use longer time series. However, fine resolution bins can lead to difficult-to-interpret noise floors. It is unlikely, for example, that there is a two order of magnitude difference in the power content of two adjacent bins outside of the main harmonics of any time series, yet that is what the blue spectrum shown in Figure 8 indicates. This wildly oscillating noise floor is an artefact of the discretisation, rather than a true reflection of the noise content of the signal.

4.4 Averaging Spectra and Window Functions

The application of the FFT to data is one of the most widespread numerical algorithms. It is integral to a huge amount of fundamental scientific research and engineering. In an industrial setting the power spectrum is used as a noise reduction method on many sensors, in the communication sector information is compressed using the FFT and in the laboratory many measurement techniques intrinsically make use of the FFT.

Many instruments report spectra directly, for example the output of an FTIR spectrometer, but it is always prudent to understand what analysis is being conducted on our behalf. As outlined here many analytical steps are happening and they may not be applicable to the analysis that we wish to conduct. Fortunately, many numerical packages are readily available that we as users can use to undertake our own Fourier analysis. All the graphs presented in this article have been generated from within a Jupyter notebook using the standard Python libraries bundled with Anaconda. These are readily available tools that we can all use if we have the inclination. Moreover, any time series can be analysed using Fourier analysis to reveal any possible underlying periodic behaviour. Atypical examples might be timesheets, holidays and production data.

The first stage of data analysis for nearly all time series data should be to understand the power spectra. The first step for a novice is to download the Anaconda bundle and start up the Jupyter executable, the second step is to search one of the many online tutorials (for example, (19)) in data analysis in Python and start experimenting. We are fortunate to live in an age when data analysis is an exceptionally easy thing to do. Let us all embrace this gift!

In recent years, whenever the subject of digitalisation or digital transformation is brought up for discussion, we normally observe two distinguishing reactions from the attendees: one group is excited and satisfied, the other, interested and worried. Of course, some have a good mixture of both. The former has been from companies, big or small, which have a clear digitalisation strategy in place from which obvious development and benefits have been achieved. For the latter, people are as keen as others on implementing solid steps to realise the long-waited benefit from business digitalisation. However, they are not quite sure where and what to start with, despite the continuously advancing technologies in the market. While still dealing with the COVID-19 pandemic, we were very curious about what the book “Digitalization” (1) would bring to help accelerate digital transformation for various organisations.

Professor Schallmo and Professor Tidd are the editors of “Digitalization” with a list of distinguished researchers on the editorial board. Professor Schallmo is a well-known key researcher focusing on business digitalisation at various stages, and the development and application of the methods to innovate business models. “Digitalization” continues his research focus following his previous book “Digital Transformation Now!” (2).

Besides his professorship of technology and innovation management at University of Sussex, UK, Professor Tidd has worked with numerous technology-based organisations globally on technology and innovation management projects. His view and experience of connecting innovation and digitalisation is always insightful. In conjunction with “Digitalization”, it is worth expanding the reader’s knowledge through his bestselling textbook on managing innovation (3).

The book “Digitalization” is a collection of 25 research-based studies which have been arranged in sections to emphasise five aspects of digitalisation: ‘Digital Drivers’, ‘Digital Maturity’, ‘Digital Strategy’, ‘Digital Transformation’ and ‘Digital Implementation’. This arrangement gives a clear statement of the focus of each part. Throughout the book, the literature review of all subjects is very rich which should give the audience a wide range of further reading if required.

The very early challenges that all organisations face in digital transformation are to discover the right opportunities and initiatives holistically. In the section ‘Digital Drivers’, four articles explore this subject from different angles. Disaster management and future‐led innovation framework, presented by Vettorello (Swinburne University of Technology, Australia) et al., and technology‐oriented future analysis by Urbano (Politecnico di Milano, Italy) et al., aim to provide guidance to organisations on innovation management with fast and accurate decision making within highly dynamic and complex environments. We feel these concepts may also have a place for individual business units within a large organisation where specific needs of that business unit can be addressed to capture local opportunity.

Chiaroni (Politecnico di Milano) et al. present a real example of how a circular business model has been applied in the building industry to realise business transformation from linear to circular by adopting digital technologies. Mutanov and Zhuparova (al-Farabi Kazakh National University, Kazakhstan) in the fourth article explain several fundamental reasons that commodity countries such as Kazakhstan and other post-Soviet countries are falling behind on digital transformation. These findings certainly show the great potential of digitalisation. Among the literature provided by the authors, two popular books written by Cross (4) and Tighe (5) are worthy of extra attention to expand ways of thinking and setting strategy.

‘Digital Maturity’ in Part 2 focuses on discovering digitalisation opportunities from a different angle, by assessing the current digital development status of an organisation and comparing with others within the same business sector or even wider to draw action plans for its own needs. First, a systematic literature review is conducted by Ochoa-Urrego and Peña‐Reyes (Universidad Nacional de Colombia) which includes 22 publications on formal maturity model applications.

The other two studies from Schallmo (Neu-Ulm University of Applied Sciences, Germany) et al. and Pierenkemper and Gausemeier (Heinz Nixdorf Institute, University of Paderborn, Germany) et al. emphasise a digital maturity models assessment of small and medium-sized enterprises (SMEs). It is recognised that the examined digital maturity models cannot provide a comprehensive digitalisation implementation plan for SMEs with an overarching vision like that typically seen at large corporations. Although Pierenkemper and Gausemeier list a few aspects of the presented model that may require further investigation, the study itself shows through examples how SMEs can produce a simple development plan for digitalisation using the model provided.

Once digitalisation objectives are determined, it is natural to move onto ‘Digital Strategy’ as presented in Part 3 on how we can capture the opportunities. The first paper in this part gives a deep dive on how disruptive innovation is used as business strategy or model for digital transformation among 80 companies in Germany. To expand the understanding of disruptive innovation, it is worth exploring relevant resources from the bestselling author (6). It is followed by Hartmann (HTW Berlin – University of Applied Science, Germany) et al. and Gernreich (Ruhr-Universität Bochum, Germany) et al. who separately address the importance of top management or an innovation manager who has the necessary knowledge in digitalisation and can drive to complete the plan for desired productivity and benefits.

Kruft and Gamber (Technische Universität Darmstadt, Germany) in the fourth paper present a critical component of digital transformation: continuous culture change, which often poses an even bigger challenge on the entire journey of digitalisation. All organisations need to recognise the significance of cultural renewal and work closely with their employees to bring them along with progress. It is one of the core strategies to empower people with the right tools, knowledge and communication via digital platforms in the era of ever-changing technology.

The focus in the paper from Koldewey (Heinz Nixdorf Institute, University of Paderborn) et al. falls in the mainstream of digitalisation, i.e., smart services interconnecting products with aftersales service. They demonstrate how they use a design research methodology to develop a smart service strategy through four comprehensive case studies. The last paper in Part 3, from Porté (Ecole Polytechnique Fédérale de Lausanne, Switzerland) et al., draws attention to the potential of using Systemic Enterprise Architecture Methodology (SEAM) to align business and IT perspectives on innovative projects. A project by the Society of Family Doctors (SFD) is used to showcase how we structure a problem based on who sees it and why, instead of the problem itself.

Part 4, ‘Digital Transformation’, expands on the first three parts of the book with papers from governments, universities and other parts of the public sector. Meier (University of Innsbruck, Austria) provides a systematic review of the literature on SME digitalisation. Her discovery agrees with a few other papers in the book on challenges that traditional SMEs face while adopting digitalisation: time, financial, human and technical resource constraints. For the public sector, Bjerke-Busch and Aspelund (Department of Industrial Economics & Technology Management, Norwegian University of Science and Technology) use Norwegian Court Administration (NCA) to explain the barriers for digital transformation in a typical public organisation.

The study from Haslam (Centre for IS Management, Department of Politics and Society, Aalborg University, Denmark) et al. identifies a few key elements of how digital transformation has been accelerated at a Danish university during the pandemic period. Staying connected with the Danish Government, Rosenstand (Aalborg University) shows early work on applying a digital ecosphere canvas for cultivating multiple digital ecosystems at Digital Hub Denmark, a private-public partnership organisation. Jütting (Fraunhofer IAO, Fraunhofer Institute for Industrial Engineering, Center for Responsible Research and Innovation (CeRRI), Germany) et al. introduce the pro-poor digitalisation canvas as a conceptual framework aiming to act as a practical tool to evaluate the potential of digital innovations. The particular interest is to practically turn the objectives of the United Nations Sustainable Development Goals (SDGs) 1 (‘no poverty’) and 10 (‘reduced inequality’) into actions to minimise the digitalisation gap between the advanced and developing world.

Digital implementation, the focus of Part 5, is the step to really make the transformation. Although it is impossible to cover all areas in the implementation stage, the authors have attempted in-depth discussion in several major subjects. Gfrerer (University of Innsbruck) et al. lead the discussion in the composition of digital leadership and gender diversity, particularly targeting female managers and how they envisage their roles and challenges to digitalisation and innovation. Reis and Hunt (Thinkergy Ltd, Hong Kong and Thailand) in the second paper also focus on the effectiveness of leadership in digitalisation. They conclude by highlighting the importance of creative leaders in the success of digitalisation and such leaders can be trained up through selective programmes combining effective methodology and pedagogy.

Schallmo and Williams (Neu-Ulm University of Applied Sciences) bring attention to an integrated theoretical approach to digital implementation which aims to realise digitalisation in four interactive dimensions and five procedural phases. The study presented in the fourth paper by Kruszelnicki (Creative Labs sp. zoo ul, Poland) and Breuer (UXBerlin Innovation Consulting and HMKW University of Applied Sciences for Media, Communication and Management, Germany) is particularly interesting. Three use cases are presented to show how Adobe Kickbox has effectively promoted ‘intrepreneurship’ to unlock innovation opportunities. Haag (TH Köln, Germany) et al. have sustainability at the centre of their research. Their main contribution is to provide the ‘design-to-sustainability matrix’ as a toolkit to address ecological challenges through the life cycle of both new and existing product development.

The last two studies in this part put weight on innovation management. Johnsson (Blekinge Institute of Technology, Sweden) et al. explore the key success factors in evaluating innovation teams. In the last paper Colucci and Forciniti (Evidentia srl, Italy) recount the story of how Ferrari has transformed its business through an innovation management programme which involves management at all levels and processes at different stages.

On completing the book, although the questions we had at the start of this review are not fully answered, we were delighted to see several useful case studies presented throughout the book. When it comes to real implementation, we understand that it is impossible to write down all details due to confidentiality and variations in organisational status and need. The richness of the literature resources in this book provided by all authors is hugely beneficial to the audience to gain a theoretical foundation. There is also wide discussion on how digitalisation is applied to various areas of focus, including SMEs, developing countries, gender diversity, SDGs, high-tech industry leaders and the public sector. Digitalisation practitioners such as management and innovation consultants and organisations would find it useful to navigate through the business models and frameworks presented by several authors at different stages of digitalisation. Readers who are very new to the digital transformation subject may find this book too profound and pre‐study is needed to bridge the knowledge gap. Finally, digital transformation is often bundled with innovation for many good reasons. We highly recommend readers continuously explore ways of innovation (7) to identify and truly drive ideas through to implementation.