TiO₂ nanoparticles were successfully synthesised by a simple sol-gel method (37) using Ti[OCH(CH₃)₂]_4, generally referred to as titanium tetra-isopropoxide (TTIP), as a precursor purchased from Sigma-Aldrich Company (USA) with purity of 97%. Titanium(IV) isopropoxide was dropped slowly into the mixed solution of distilled water and ethanol in the ratios of 1:4:1 (TTIP: water: ethanol). The solution was stirred continuously for 1 h at room temperature to obtain a white slurry. HNO₃ was used to adjust pH value in the range 2–3. The white slurry mixture was dried at 120°C for 3 h on a hot plate; the dried powder was sintered at 450°C for 3 h. Finally, we obtained the required TiO₂ nanoparticles. The flow chart of synthesis of TiO₂ nanoparticles is given in Figure 1.

Fig. 1

The synthesised sample of TiO₂ nanoparticles were analysed with XRD pattern using a SmartLab^® X-ray diffractometer (Rigaku Corporation, Japan) (with λ = 1.5406 Å CuK_α radiation) operating at 40 kV, 30 mA and at room temperature. The XRD patterns were used to determine the crystallite size, lattice parameter and phase identification. The structural and morphological analysis of TiO₂ nanoparticles were done by HR-TEM and the selected area electron diffraction (SAED) pattern using the model Tecnai^TM G² F30 field emission gun transmission electron microscope (FEI Company, USA) operating at 200 kV accelerating voltage with resolution point:0.17 Angstrom line:1.24 Å and magnification 1500 LM to 520 kx. Tescan MAIA3 field emission scanning electron microscope (Tescan, Czech Republic) operating at 12.0 kV and magnification 21.4 Kx was used for SEM-EDX analysis of the morphology and average particle size of the TiO₂ nanoparticles. The UV-vis absorption spectrum was recorded using Shimadzu UV-2330 spectrometer (Shimadzu Corporation, Japan) in the range 200–700 nm. The UV-vis spectrum was used to determine direct energy band gap of the TiO₂ nanoparticles.

TiO₂-ethylene glycol nanofluids were prepared at different concentrations, 0.2 wt%, 0.5 wt% and 1.0 wt% of TiO₂ nanoparticles. When TiO₂ nanoparticles are added to the ethylene glycol base fluid, the nanoparticles produce a sediment within a few minutes because they remain in clusters without being dispersed. For the uniform dispersion of nanoparticles in the base fluid, we used an ultrasonic homogeniser VC 505 (Sonics & Materials Inc, USA) working at 20–40 kHz, 500 W.

The crystal phases of the synthesised TiO₂ nanoparticles were determined by XRD patterns as shown in Figure 2. The obtained peaks in the diffraction pattern are identified with the JCPDS Card No. 88-1175. The interplanar spacing has been calculated using Equation (i):

(i)

where λ represents the wavelength of CuK_α (1.5406 Å) radiation, θ is the angle between incident beam and the reflection lattice planes and n = 1 is the order of the XRD spectra. The highest peak is observed at 2θ = 25.4° which was indicated to plane (101) and d spacing corresponding to this peak is 3.12 Å. The other peaks in XRD pattern are observed at 2θ = 27.6°, 37.9°, 48.2°, 54.1°, 55.1°, 62.8°, 69°, 70.4°, 75.1° and 82.8° correspond to the (110), (004), (200), (105), (211), (002), (116), (112), (215) and (312) planes of TiO₂ nanoparticles and d spacing are calculated as 2.83 Å, 2.43 Å, 1.88 Å, 1.69 Å, 1.66 Å, 1.47 Å, 1.35 Å, 1.33 Å, 1.26 Å and 1.16 Å, respectively. The intensity of the obtained peaks indicates the well-formed crystalline nature of the sample. The average crystallite size has been computed with Scherrer’s equation (Equation (ii)) (38):

(ii)

where β_hkl represents the full width at half maxima (FWHM) and K is the Scherrer constant. From this formula, the calculated average crystallite size of the given sample is approximately ~23 nm.

Fig. 2

The TEM image of a crystalline sample is shown in Figure 3(a). The average particle size of the TiO₂ nanoparticles ranged from 20–26 nm as shown in the histogram (Figure 3(b)). The SAED pattern in Figure 3(c) shows principally 10 rings which are ascribed to (101), (110), (103), (004), (111), (200), (105), (211), (002) and (116) planes, respectively. These planes are consistent with the XRD results. The d spacings are in agreement with the tetragonal structure of TiO₂ nanoparticles (JCPDS Card No. 88-1175). For the structural analysis TiO₂ nanoparticles were also examined by HR-TEM as shown in Figure 3(d). The crystalline nature of the nanoparticles is visible in the HR-TEM micrograph. The lattice spacing 0.31 nm and 0.28 nm corresponds to (101) and (110) planes respectively. The size and morphology of the TiO₂ nanoparticles were also determined using SEM. Figure 4 shows typical SEM images of TiO₂ nanoparticles. The SEM image shows random distribution of TiO₂ nanoparticles having sizes in the range 18–26 nm. In Figure 4, there is a soft agglomeration of the nanoparticles: isolated particles are connected to each other by attractive physical interactions like Van der Waals force. The agglomeration of nanoparticles in the base fluid probably affects the thermal conductivity performance of the nanofluids. Agglomeration of nanoparticles affects the Brownian motion of the nanoparticles resulting in a decrease in thermal performance of the nanofluids. To remove agglomerations of nanoparticles in the base fluid, a sonication process has been used to break the intermolecular interactions. The EDX spectrum (Figure 5) of the TiO₂ nanoparticles provides information about the constituent components of our sample, which contains titanium and oxygen. The high intensity peaks for titanium and oxygen justifies that the sample contains mainly TiO₂.

Fig. 3

Fig. 4

Fig. 5

The UV-vis absorption spectrum at room temperature of TiO₂ nanoparticles has been recorded in the wavelength range 200–700 nm and is shown in Figure 6(a). It is obvious from the UV-vis absorption spectrum that the peak observed at 315 nm represents a blue shift compared with its bulk counterpart. This indicates that the particle size of the TiO₂ nanoparticles has been reduced (38–40). The optical absorption of the TiO₂ nanoparticles is analysed by Equation (iii):

(iii)

where E_g represents the optical band gap of nanoparticles, B is a constant, α is the optical absorption coefficient of the nanoparticles. The exponent m depends on the nature of the transition, m = 1/2, 2, 3/2, 3 for allowed direct, allowed indirect, forbidden direct and forbidden indirect transitions respectively. Figure 6(b) shows the Tauc plot of TiO₂ nanoparticles, a satisfactory fit is obtained for (αhν)² vs. hν indicating the presence of a direct band gap. The optical energy gap of the TiO₂ nanoparticles has been determined as 3.28 eV by extrapolating the linear portion of this plot at (αhν)² = 0.

Fig. 6

The thermal conductivity of the nanofluids was measured by using a Hot Disk TPS-500 S thermal constant analyser. The Hot Disk TPS-500 S is the newest transient plane source (TPS) thermal constants analyser. The TPS technique has been used to determine the thermal conductivity of a nanofluid. The temperature dependent thermal conductivity of the TiO₂-ethylene glycol nanofluids is plotted in Figure 7(a) at 0.2 wt%, 0.5 wt% and 1.0 wt%. The results show that the thermal conductivity of TiO₂-ethylene glycol nanofluids increases with concentration of TiO₂ nanoparticles. The thermal conductivity exhibits a slow increase for 0.2 wt% nanofluids while it shows relatively fast increase for 0.5 wt% and 1.0 wt% nanofluid in the temperature range 20–80°C. At 20°C, the value of thermal conductivity of pure ethylene glycol is 0.285 W mK⁻¹ and it has been increased to 0.314 W mK⁻¹ for 1.0 wt% concentration of TiO₂ nanoparticles in ethylene glycol base fluid. The expression of TCE is given by Equation (iv) (41):

(iv)

where TC_nf and TC_bf are the thermal conductivity of nanofluid and base fluid respectively.

Fig. 7

A number of investigators have developed models for determining the thermal conductivity of nanofluids containing spherical particles. They only consider the effect of volume fraction of the particles. However the thermal conductivity of nanofluids depends on various factors such as size, shape, volume fraction of the suspended particles as well as temperature of suspensions. A few models also propose that the TCE is due to the ordered layering of liquid molecules near the solid particles (42, 43).

In addition to describing the TCE in nanofluids, we consider the effect of three possible mechanisms for heat transfer in nanofluids: (a) translational Brownian motion, (b) the existence of an interparticle potential and (c) convection in the liquid due to the Brownian movement. In the low temperature region, the mean free path due to the collision of nanoparticles increases and leads to TCE due to Brownian particle (K_Brownian) as given as Equation (v):

(v)

where ϕ is the volume fraction, C_N is the heat capacity per unit volume of the nanoparticles, l is the mean free path and V_N is root mean square velocity of the particles.

This model explains the individual effect of temperature on the TCE in nanofluids with the help of Brownian motion but does not consider the effect of surface functionality and particle loading of nanoparticles. To overcome the shortcomings of this model, Prasher et al. (44) presents an order-of-magnitude justification to show a local convection effect caused by the Brownian movement of the nanoparticles. Based on the Brownian motion induced convection effect from multiple nanoparticles, the model of Prasher et al. for the TCE ratio of a nanofluid is given in Equation (vi):

(vi)

where K is the thermal conductivity of nanofluids, K_f is the thermal conductivity of fluids, A and m are the best fit constants, and should be same for different experimental data for a particular fluid. R_e and P_r are Reynolds and Prandtl numbers respectively (Equation (vii)):

(vii)

where d_N is the particle diameter, R_b is the interfacial resistance (Equation (viii)):

(viii)

The Reynolds number (R_e) is based on the root-mean-square velocity (ν_N) of a Brownian particle defined as Equation (ix) (45, 46):

(ix)

where ρ_N is the density of the particles, k_b is the Boltzmann constant and T is the temperature in Kelvin scale.

Figure 7(b) shows the variation of the TCE with temperature ranging from 20–80°C. It is clear from Figure 7(b) that the enhancement in thermal conductivity of TiO₂-ethylene glycol nanofluids is achieved with increasing concentrations of nanoparticles and temperatures. At 20°C, we observed 3.3% to 6.5% and 11.2% TCE on 0.2 wt%, 0.5 wt% and 1.0 wt%, and at 80°C, the TCE becomes 9.9%, 18.3% and 23.8% for 0.2 wt%, 0.5 wt% and 1.0 wt% respectively for TiO₂-ethylene glycol nanofluids. This enhancement is due to better uniformity and stability of suspensions. It has been found that ultrasonication increases the stability and uniformity of the nanofluids. The achieved values of TCE are higher than any of the results reported previously for TiO₂-ethylene glycol or TiO₂-water based nanofluids (13, 14, 30, 31) at such small concentrations. In preparation of the nanofluids, we used very small amounts of nanoparticles, so the fluidic properties of the liquid are almost unaffected, allowing for the easy flow of liquids and better transfer of heat. As the temperature increases, the TCE in the TiO₂-ethylene glycol nanofluids may be attributed to Brownian motion of nanoparticles. It is obvious from Equation (ix) that the root-mean-square velocity (ν_N) of a Brownian particle depends upon particle diameter. If the particle diameter is small, root-mean-square velocity of a Brownian particle is large. Since the synthesised nanoparticles are small in diameter (approximately 22 nm) this results in the increase of Brownian motion, causing convection which in turn increases the thermal conductivity of the nanofluids. The high TCEs are probably due to the small size of nanoparticles because as the particle size decreases, the surface-to-volume ratio of particles increases, which can lead to enhanced thermal conductivity of nanofluids.

The ultrasonic velocity in nanofluids was measured using an ultrasonic interferometer (model nanofluid-10X, Mittal Enterprises, India) at 3 MHz frequency in temperature range 20–80°C. The measured ultrasonic velocity in ethylene glycol matrix and three nanofluids samples containing 0.2 wt%, 0.5 wt% and 1.0 wt% of TiO₂ in temperature range 20–80°C are shown in Figure 8. It is obvious from Figure 8 that the ultrasonic velocity in the nanofluids increases with the temperature. The plot also indicates that the ultrasonic velocity in the nanofluids is larger than that of pure ethylene glycol matrix (1410 m s⁻¹) at 20°C and the velocity increases with the particle concentration (1430 m s⁻¹) for 1.0 wt% loading at the same temperature of 20°C.

Fig. 8

If we consider (ρ_m,ρ_s) and (k_m,k_s) are the density and the compressibility of fluid and suspended particles respectively, B and ϕ are the bulk modulus and the particle volume fraction; then the effective density (ρ_eff) and compressibility (k_eff) of the suspension becomes as Equation (x) (47–49):

(x)

The ultrasonic velocity (V) in a medium is given by Equation (xi):

(xi)

where B, ρ and k represent the bulk modulus, density and compressibility of the medium respectively. λ and μ are the material dependent quantities known as Lamé moduli or Lamé coefficients. The compressibility and density of a fluid medium are changed by the dispersion of nanoparticles and are the function of the particle volume fraction. From Equation (x), it is clear that the evaluation of the effective bulk modulus and compressibility of the suspension is performed with calculation of effective Lamé moduli, which depends on particle volume fraction of suspended particles.

It is obvious from Equations (x) and (xi) that the bulk modulus and change in density of the nanoparticles suspension as a function of volume fraction causes an enhancement in the ultrasonic velocity. An increase in the wave velocity with increase in the particle concentration of given nanofluids indicates that there is positive change in the bulk modulus and density of the nanofluids. It may be predicted that the comparative change in the density with respect to bulk modulus is small. As the particle concentration in nanofluids increases, the compressibility of the given matrix decreases. A strong cohesive interaction occurs among the molecules after dispersion of TiO₂ nanoparticles in the ethylene glycol matrix. Thus for the TiO₂ nanofluids, the ultrasonic velocities are larger in comparison to the ethylene glycol matrix and increase with the nanoparticle concentration.

In the low frequency region, the velocity in nanofluids is independent of particle size (49, 50). Here all the nanofluids have been prepared with nanoparticles fabricated at low evaporation rate and velocity of the ultrasonic wave is measured at different temperatures and low frequency (3 MHz). Thus it was concluded that the temperature dependent velocity at low frequency in the nanofluids depends only on the particle concentration. At low frequency, the ultrasonic velocity in a nanofluid is a quadratic function of temperature (Equation (xii) (51):

(xii)

where V₀ is the ultrasonic velocity at 0°C, V₁ and V₂ are the absolute temperature coefficients of velocity and T is the temperature difference between experimental and initial temperature (0°C). The first and second terms in Equation (xii) are in good agreement for a simple liquid system, but the third nonlinear term is caused by non-linear change in bulk modulus and density of the nanofluid system with temperature.

The acoustic particle sizer APS-100 (Matec Applied Sciences, USA) was used to examine the PSD in the nanofluids. The APS-100 works on Epstein and Carhart theory (52) and is mainly based on the ultrasonic spectroscopic method. The APS-100 computes the sound attenuation (dB) per unit length (cm) over the 1–100 MHz frequency range in particle-liquid suspensions with high precision. This attenuation spectrum can be converted to PSD data. According to Epstein and Carhart theory (52), the attenuation of the ultrasonic wave in a nanofluid can be understood with the understanding of the thermal wave length (; K_S, ρ_S and C_S: thermal conductivity, density and specific heat of the dispersed particle: ω ; frequency of the wave) and the viscous wave length (; η: viscosity of the matrix). When the viscous wave length is comparable to particle radius (r), the viscous loss is a prominent cause behind the ultrasonic attenuation; while the viscous drag, scattering and thermal losses are effective when the thermal wave length λ_T ≈ r. The expressions for the ordinary viscous dissipation (α_V), the viscous drag loss (α_VD) (47, 50) of the sound waves are given as Equation (xiii) and Equation (xiv):

(xiii)

(xiv)

where η_d and η_V represent the dynamic and the volume viscosities of the nanofluid, k̅ is the wave number, . Biwa (53) calculated the change in the ultrasonic attenuation with respect to volume fraction caused by scattering at microscale in low frequency limit. The expression to compute the ultrasonic attenuation is given as Equation (xv) (53):

(xv)

where γ^sca represents the scattering cross-section which depends on the frequency of the ultrasonic wave, particle size, bulk modulus and density of the base/carrier fluid and suspended particles. The thermal attenuation is caused by temperature variation produced by propagation of the sound waves in different components of suspension. The thermal loss mainly depends on the frequency and particle size. The particle size has been obtained by APS-100 in range of 19 nm to 24 nm as visualised in Figure 9. It has been confirmed from Figures 3(a) and 9 that the PSD obtained by APS-100 is in good agreement with that obtained by the TEM micrograph. Ultrasonic spectroscopy is sensitive to particles with radius between about 10 nm to 1000 mm. The maximum particle concentration which can be analysed varies between about 1 wt% to 50 wt% depending on the nature of the system. On the other hand, the technique is unsuitable for analysing dilute suspensions i.e., particle concentrations below about 1 wt%. In the present study (Figure 9) the weight percentage of TiO₂ is 1 wt%. Systems with different weight percentages of TiO₂ show the same PSD because nanoparticles are dispersed in base fluid with the same technique and the same ultrasonication time.

Fig. 9

Taking into account that more than 99.4% of the diesel vehicles currently in China are compliant with Euro III–V, three typical LDDTs compliant with Euro III, Euro IV and Euro V, respectively, were selected from the market and their specifications are provided in Table I. These trucks have similar dimensions and powers, and their biggest difference is their aftertreatment technology. To eliminate the impact of fuel quality, all the diesel fuel used in the study was from a specified filling station, conforming to the China VI standard.

The test route was designed to simulate the real driving conditions of most diesel trucks in Zhengzhou, Henan province. The total length of the test route was approximately 68 km, including 14 km of urban roads, 18 km of connection roads and 36 km of highway. VOCs were sampled only when the trucks travelled on the urban and highway roads and cold start emissions of VOCs were not included during the whole test. Table II shows the driving condition parameters during each road type. The average speeds on urban and highway roads were 18.1–20.8 km h⁻¹ and 72.8–76.5 km h⁻¹, respectively. Driving conditions on urban roads are more aggressive than those on highway roads. The average accelerations on urban and highway roads were 0.22–0.26 m s⁻² and 0.10–0.13 m s⁻², respectively. During the measurement, the trucks were not in service and the load of each truck was approximately 500 kg during the experiment, containing the PEMS equipment, four batteries, two testers and one driver.

Under real driving conditions, some gaseous emissions may transform to secondary fine particles when the exhaust is cooled or diluted with the ambient atmosphere. Thus, VOC emissions might be overestimated if sampled directly from the vehicle exhaust because the temperature is very high. Therefore, a combined PEMS (Sensors Inc, USA) was employed to sample the exhaust VOC emissions. The schematic diagram of the emission testing and sampling system is shown in Figure 1.

Fig. 1

The microproportional sample system (MPS), a partial flow dilution system, was used to dilute and cool the exhaust from tailpipe. After the MPS, the gas temperature decreased from about 120ºC to about 40ºC. Two 3.2 L SUMMA^® canisters (Entech Instruments Inc, USA) were used to sample the VOCs during each test trip, one for the urban roads and the other for the highway roads. VOC emissions during the connection roads section were not sampled because the actual running speed could not meet the requirement due to unexpected road repairing. The sampling flow rate was controlled by a passive restrict valve at 0.1 L min⁻¹. Teflon^TM tubes were used to connect the canister and PEMS system to minimise the adsorption of VOCs. A laptop was used to control the system and collect data from the test module. It should be noted that these trucks were driven by their owners throughout the test to ensure these trucks were running under ordinary working conditions and each vehicle was tested twice to enhance the reliability of the results.

Analysis of the VOCs was carried out following the United States Environmental Protection Agency (US EPA) TO-15 method by a gas chromatography-mass selective detector (GC-MSD) (22). Samples collected in the SUMMA^® canister were preconcentrated using an 8900DS preconcentrator (Nutech Instruments Inc, USA) with three cold traps and a canister autosampler (Nutech Instruments Inc, USA, mode 3600DS). The moisture, CO₂ and methane would be removed through the traps. Then the concentration of the individual VOCs in the samples was determined by a GC-MSD system (7890A GC with a 5975 MSD, Agilent Technologies Inc, USA). Separation of the VOCs was achieved through a capillary column (60 mm × 0.25 mm internal diameter, 1.4 μm film thickness, DB-624 column, Agilent Technologies Inc). During sampling and analysis, strict quality assurance and quality control procedures were conducted to assure the data quality (22). The detection limits of the target non-methane hydrocarbons ranged from 7 parts per trillion by volume (pptv) to 141 pptv and the accuracy of the measurements was about 1–10%. Detailed description of the analysis procedures can be found in our previous study (23).

A total of 102 VOC species were identified and quantified, including 29 alkanes, 35 halocarbons, 17 aromatics, nine alkenes, five carbonyls and seven other compounds, which are presented in Table III. Due to the detection limitation of GC-MSD used in this study, some species (ethane, ethylene, propylene, acetylene, formaldehyde) were not detected and included.

EF per kilometre of a certain pollutant was calculated with the corresponding concentration, total exhaust volume and running distance during the test process. Prior to calculation, the results of the VOC measurements were time-aggregated. The total exhaust volumes in various driving conditions were the integration of the instantaneous exhaust flow rates, and the same for the total running distance. The EF of compound i was calculated as Equations (i)–(iii):

(i)

(ii)

(iii)

where V (m³) is the total exhaust volume of the sampling process; V_ins (m³ s⁻¹) is the instantaneous exhaust flow rate; DR_ins is the instantaneous dilution ratio of MPS; S (km) is the distance that the test vehicle travelled during the sample period; S_j is the travel distance at j second, which is equal to the value of instantaneous speed at time j recorded by the global positioning system (m s⁻¹); EF_i (mg km⁻¹) is the EF of compound i; C_i (parts per billion by volume) is the concentration of compound i; M_i (g mol⁻¹) is the molar mass of compound i; and V_m (l mol⁻¹) is the molar volume of compound i. The volumes and concentration data were all normalised to the standard ambient temperature and pressure condition (273.15 K, 101.33 kPa). The total EFs of the VOCs in a certain driving mode were summed by the individual VOC EFs in the driving mode.

The OFP refers to the amount of ozone generated by VOCs per unit mass (mg O₃ mg⁻¹ VOCs), which can reflect the ozone formation capacity of VOC species. In most cases, ratios of VOCs to NOx from the diluted exhaust were much higher than 20 in this study, which illustrated VOCs had the greater effect on the ozone formation (24). Therefore, the MIR scenarios developed by Cater (25) was applicable to evaluate the OFP of VOC species here. The OFP of a certain VOC is calculated according to Equation (iv) (26, 27):

(iv)

where OFP_i (mg O₃ km⁻¹) is the ozone formation of compound i; and MIR_i (mg O₃ mg⁻¹ VOCs) is the maximum incremental reactive of compound i obtained from Cater (25, 28). The total OFPs of a certain driving mode were summed by the individual VOC OFPs of the driving mode.

Figure 2 presents the EFs of regulated gaseous pollutants of three LDDTs compliant with different standards. Obviously, NOx, CO and HC emissions from LDDT-3 (Euro V) were the lowest and those from LDDT-1 (Euro III) were the highest, except for CO. In general, updated emission standards had a great effect on the reduction of regulated gaseous emissions. This is mainly because the three trucks adopted different aftertreatment technologies to meet different emission standards (29). For example, both selective catalytic reduction (SCR) and diesel oxidation catalyst (DOC) were utilised by LDDT-3 to be compliant with Euro V standards. SCR was often used to purify the NOx emissions and the DOC device could oxidise the CO and HC emissions efficiently (30–32). It is not difficult to understand why LDDT-1 produced the worst emissions because there is no aftertreatment requirement for Euro III trucks in most of China.

Fig. 2

As shown in Figure 2, NOx, CO and HC emissions under urban conditions were significantly higher than those under highway conditions. To be specific, NOx, CO and HC emissions under urban conditions were 1.3–1.8 times, 1.4–2.2 times and 2.5–4.1 times those under highway conditions. This phenomenon could be explained by the fact that the combustion quality in the engine was associated with the operation speed and frequent acceleration and deceleration (18, 33, 34). During this experiment, no traffic signals were encountered on the highway and the average speed was up to 73.8 km h⁻¹. However, there were 26 traffic signals on the urban roads and the average speed was only 19.4 km h⁻¹. In this operating condition, the combustion was insufficient and the temperature of aftertreatments might not be high enough for proper function, which caused the emissions to deteriorate.

Average weight percentage of individual VOC species of the entire trip was calculated based on the test trucks. On the whole, alkanes were the dominant group, accounting for 65.5 ± 10.3% of the total VOCs, followed by aromatics, carbonyls and alkenes, taking up 19.6 ± 5.0%, 5.4 ± 1.9% and 4.4 ± 1.8%, respectively. Additionally, though 35 halocarbons were quantified, they only took up 3.6 ± 1.5% of the VOCs. Thus, the following discussions on the VOCs are mainly focused on alkanes, aromatics, alkenes and carbonyls.

Weight percentages of the top 15 VOC species from the exhaust are presented in Table IV. These species accounted for approximately 83.4% of the total VOCs. Dodecane, n-undecane, naphthalene, n-decane and acetone were the major species, and their total weight percentages were over 80.1%. These results are partially consistent with the results obtained by Wang et al. (14), who indicated that n-decane, n-undecane and n-dodecane were the most abundant species. However, a study by Yao et al. (20) showed that carbonyls were the top group, which could account for 42.7–69.2% of the total VOCs. The difference was mainly attributed to the different VOC species quantified between the two studies. For example, Yao et al. (20) reported formaldehyde and acetaldehyde took up 47.9% and 21.0% of carbonyls, while these two species were not detected in this study.

The mean EFs and weight percentages for each VOC group for the entire trip of the three test trucks are plotted in Figure 3. The total VOC EFs of LDDT-1 (Euro III), LDDT-2 (Euro IV) and LDDT-3 (Euro V) were 186.9 ± 34.9 mg km⁻¹, 106.5 ± 26.2 mg km⁻¹ and 61.1 ± 16.9 mg km⁻¹, respectively. In other words, the VOC emissions decreased significantly as the standards tightened gradually from Euro III to Euro V. Most of the other species also showed a decreasing trend. Especially, dodecane and n-undecane presented the most significant decline, from 85.2 ± 3.7 mg km⁻¹ and 38.6 ± 11.7 mg km⁻¹ for Euro III to 16.7 ± 2.8 mg km⁻¹ and 9.7 ± 3.0 mg km⁻¹ for Euro V, respectively. The trend was partially consistent with that found by Zhang et al. (10), though the VOC EFs were a little higher than those in this work. This might be mainly attributed to the fact that Zhang et al. (10) employed tunnel measurement, which included evaporative emissions.

Fig. 3

Most VOC groups presented similar variation trends as the emission standards changed, especially for the dominant groups. For example, both alkane and aromatic emissions decreased noticeably as the standards varied from Euro III to Euro V. The progress in engine technology and application of aftertreatment devices played a major role in the subtraction of VOCs emissions. Additionally, there were no significant differences between emissions of carbonyls, alkenes and halocarbons from LDDT-1 (Euro III) and LDDT-2 (Euro IV), but they were much higher than those of LDDT-3 (Euro V). No coherent order was observed for other emissions among these diesel trucks, possibly because the absolute values of these species were too small to quantify accurately. On the whole, implementing stringent emissions standards could reduce most of the VOC species effectively in the freight transportation sector.

Figure 3 indicates that alkanes were the dominant group in tailpipe VOCs emissions from the test LDDTs, accounting for 57.2–80.0%, followed by aromatics (12.5–22.9%), carbonyls (3.1–7.7%) and alkenes (2.2–6.5%). This result was consistent with that observed by Wang et al. (14) (carbonyls < aromatics < alkanes) but inconsistent with that by Yao et al. (20) (alkenes < aromatics < alkanes < carbonyls). Discrepancy of the quantified VOC species was the main cause of the inconsistency. It can also be found that the proportion of alkanes decreased significantly, from 80.01% for LDDT-1 to 57.15% and 60.41% for LDDT-2 and LDDT-3, respectively. Additionally, LDDT-2 and LDDT-3 had similar VOC group distributions, while the aromatics weight percentage of LDDT-2 was significantly related to that of LDDT-3. This is probably due to the different aftertreatment used in LDDT-2 and LDDT-3 (as shown in Table III). Jung et al. (35) also observed that heavy-duty trucks equipped with DPF emitted higher quantities of aromatics compared with those with SCR.

Figure 4 shows the EFs of the top 15 VOC species from the exhaust of LDDTs. The EF of dodecane for LDDT-3 (Euro V) was 51.0% and for LDDT‐2 (Euro IV) it was only 19.6% relative to LDDT‐1 (Euro III). For several other species, LDDT-3 had the lowest EFs, while the EFs of LDDT-2 and LDDT-1 were comparable or even higher, such as naphthalene, acetone, 2-propenal. A hypothesis is that much higher temperatures and more oxidising conditions during the DPF regeneration process favour carbonyl formation (36). However, there is no direct evidence that DPF regeneration occurred. Additionally, there were several individual species whose emissions were not affected by the emission standards. Overall, most of the top 15 VOC species presented a decreasing trend as the emission standards tightened.

Fig. 4

Figure 5 shows several VOC group emissions from the exhaust of LDDTs under urban and highway driving conditions, respectively. It can be seen that VOC emissions on highway roads were much lower than those on urban roads. EFs of each VOC group decreased significantly, especially for alkanes and aromatics. Specifically, EFs of alkanes under highway conditions were only 20.4–46.2% of those under urban conditions, which was mainly attributed to the sharp decline of the most abundant alkane species, such as dodecane, n-undecane and n- ecane. For aromatics, the significant reduction of the EFs during highway driving could be attributed to the sharp reductions of naphthalene, 1,2,3-trimethylbenzene and 1,2,4-trimethylbenzene. Lower average speed and more acceleration and declaration were found during the urban road episodes, causing more incomplete combustion on non-highway road driving, resulting in higher VOCs emissions than on highways (37). Caplain et al. (38) also reported that tailpipe emissions in urban driving cycles were approximately four times those in motorway driving cycles. In addition, the reduction degrees of VOC EFs (urban vs. highway) for LDDT-3 were highest while those for LDDT-2 were lowest. This discrepancy is mainly due to the different aftertreatment devices used. For instance, optimum conditions of a DOC + SCR system used for LDDT-3 could be maintained under highway conditions because of the high exhaust temperature, resulting in more efficient reduction.

Fig. 5

A breakdown of the C1–C12 VOCs for different driving conditions of the tested trucks is presented in Figure 6. There was no obviously consistent trend in the carbon number distribution of the VOC species between highway and urban road conditions except for C3 and C11, which showed a decreasing trend when driven on the highway compared to urban roads. This phenomenon illustrated that driving conditions had a weak correlation with carbon number distribution. On the whole, the carbon number of the VOCs was concentrated in C3–C4 and C10–C12, showing a distinct ‘double peak’ phenomenon. Lu et al. (39) summarised several previous studies and reached a similar conclusion. VOCs are expected to be a mixture of unburned and partially burned fuel species (40). Propane and acetone are the dominant species in C3–C4 group, and this portion of the VOCs is likely generated as a result of the high efficiency of the diesel engine. For the C10–C12 group, these species are considered to be components of diesel fuel. Durbin et al. (41) reported the C1–C3 species contributed most to non-methane organic gases (NMOG) and ethene, ethyne, acetaldehyde and formaldehyde made the largest contribution. The differences may be attributed to the VOCs species detected between these two studies.

Fig. 6

According to the EFs of each VOC species, OFPs based on the travelled distance were calculated and the results are plotted in Figure 7. As expected, the magnitude of OFP based on emission rate presented a decreasing trend. To be specific, LDDT‐1 and LDDT-2 had the higher OFPs, approximately 239.6 ± 57.3 mg O₃ km⁻¹ and 227.7 ± 69.2 mg O₃ km⁻¹, respectively, and that for LDDT-3 was 124.8 ± 47.6 mg O₃ km⁻¹. The OFP values in this work were lower but comparable to those for diesel trucks in some studies (20, 37, 42). The lower OFP values in this study were mainly because DOC and SCR dramatically reduced VOCs emission. Additionally, engine technologies, driving cycles and fuel quality were also important factors.

Fig. 7

The chemical structure of OFPs was different from the trend of VOCs emissions based on distance travelled, shown in Figure 3. Aromatics were the primary contributor to OFP, accounting for 49.3–57.6% of the OFPs. It was noteworthy that although alkenes accounted for only approximately 5.0% of the VOC emissions, the OFP contribution of alkenes (13.4–22.3%) was comparable with that of alkanes (13.7–27.9%), which was attributed to the higher MIR scales of alkenes related to alkanes. Similar conclusions have been reached in previous studies. Therefore, priority measures should be taken to reduce the VOCs with high MIR values, such as aromatics and alkenes, to control the formation of ozone originated from diesel exhaust.

The top 20 VOC species ranked by their OFP are given in Figure 8. The contribution of these substances accounted for approximately 90.0% of the total measured OFPs. Naphthalene, 1-butene, dodecane, 1,2,3-trimethylbenzene, 2-propenal, 1,2,4-trimethylbenzene and 3-ethyltoluene were the dominant species in the photochemical ozone formation process, and their OFP values were over 10 mg O₃ km⁻¹. Among the top 20 species, 11 belonged to the aromatic group and four were alkenes, which accounted for a lower mass percentage but higher MIR values. This indicates that substances present in small amounts but with high MIR values should not be ignored.

Fig. 8

Global energy demands are increasing and so too is the need for more renewable and sustainable sources of energy to help transition us to a post-fossil-fuel-powered world. The European Union (EU) has recently increased its renewable energy target to 32% for 2030 (1), with many countries planning to ban internal combustion engine powered cars by 2040 or sooner. However, the transportation industry is one of the most challenging sectors to adapt to using low-carbon fuels. Transportation modes such as aircraft, heavy-duty and marine vehicles demand high power and energy capacity that are currently unmet by renewable technologies. In the interim, we need clean, sustainable methods, continuous improvement and new innovations in renewable fuels to meet EU and other similar worldwide targets.

Johnson Matthey and bp have been collaborating for the past two decades (2, 3) to develop an efficient reactor system and catalyst for the FT process. This offers a cost-effective method of converting any carbon source into high-quality liquid hydrocarbon fuels.

Today the world consumes more than 55 million barrels (bbl) day^–1 of transportation fuels (4), the vast majority of which originates from crude oil. In addition to being a finite resource, each barrel of crude-oil-derived fuel typically contributes about 475 kg of CO₂ into the atmosphere over its life cycle (based on a 2010 European average) (5). At the same time, hundreds of millions of tonnes of municipal solid waste (MSW) are incinerated or sent to landfill each year, while similar quantities of woody biomass decompose to CO₂ and methane (a more potent greenhouse gas than CO₂) (6). Industrial processes release more than 8 billion tonnes of direct CO₂ emissions (7), and flaring of natural gas releases a further 275 million tonnes of CO₂ equivalent per year (8).

These wastes and emissions are rich in carbon which can be extracted through gasification, reforming or capture of CO₂ (which can then be converted to synthesis gas (syngas) via reverse water gas shift). Converting this carbon to useful syngas rather than allowing it to be emitted to the atmosphere as CO₂ creates an opportunity to significantly reduce fuel life-cycle emissions and its impact on global warming. The carbon intensity of the resulting fuel can typically be reduced by more than 70% (relative to conventional fuel), with reductions of more than 100% possible as grid electricity becomes increasingly renewable and CO₂ sequestration is added to the technology mix.

Gasification of biomass is not new technology. However, it has typically been used to produce syngas for the generation of electricity, rather than chemical synthesis. Effective FT synthesis requires a specific ratio of hydrogen and carbon monoxide, and the catalyst used for this reaction is particularly susceptible to poisoning by impurities that may be present. Gasification of waste introduces the potential for a wide range of contaminants that need to be removed.

The range of potential poisons, and the low levels necessary to ensure continued high-level performance for the FT synthesis catalyst, make syngas purification a critical step of the process. Effective removal of these impurities also ensures that FT products are ultra-clean and high purity (9).

As a world leader in purification and pre- and post-treatment of syngas gas for downstream applications, Johnson Matthey has a range of solutions to condition the syngas (irrespective of its original source) ready for conversion into sustainable fuel.

The FT process was originally developed by Franz Fischer and Hans Tropsch in 1925. It is a way of converting any carbon source into liquid hydrocarbon via syngas, effectively creating synthetic fuel (see Equation (i)).

(i)

Syngas can be generated from various carbon sources, including coal, natural gas, MSW and biomass. The process mainly produces linear, long-chain paraffins that require further upgrading to produce liquid fuels, such as diesel and kerosene. The upgrading step comprises catalytic hydrocracking to both isomerise and crack the long-chain paraffins into smaller-chain paraffins with the correct properties for fuel applications. The various stages in the process are shown in Figure 1. In the quest for sustainable fuel solutions, FT-derived synthetic fuels provide a cleaner way to power cars, heavy-duty vehicles and aeroplanes. The latest developments in FT technology mean that the production of fuel from sustainable carbon sources is now closer to being commercially viable at all industrial scales.

Fig. 1

Catalysts are required for the FT process to increase the rate of reaction and make the process industrially viable. There are broadly two options for FT synthesis, using cobalt or iron catalysts (10, 11). While cobalt is more expensive than iron, it mainly produces normal paraffins. Iron catalysed FT synthesis also incorporates the water-gas shift reaction for CO₂ products and makes a mixture of olefins and paraffins.

The most commonly used catalyst is cobalt, due to its high activity, selectivity to liquid hydrocarbons and stability. Commercial synthesis of hydrocarbons occurs at moderate temperatures of 200–240ºC and pressures of 20–40 bara. During the process, hydrogen and carbon monoxide are converted into long-chain paraffins or waxes over the supported catalyst (see Figure 1). Pore diffusion and mass-transfer effects therefore play a key role in FT catalyst performance due to the need for hydrogen and carbon monoxide to move into and along the catalyst pores against the movement of product molecules going the other way.

The FT synthesis reactions are all highly exothermic, making efficient removal of heat essential for any reactor design. There are a number of benefits of using conventional fixed-bed tubular reactors (12, 13) which is why Johnson Matthey and bp have favoured this design. They are a proven technology with many manufacturers able to fabricate reactors at large scale. They work by holding the catalyst in place via a static bed, which has the advantage of preventing catalyst loss, which could lead to product contamination as can occur in slurry reactors. The reactors have a modular design, which makes increasing capacity as simple as adding tubes; however, conventional fixed-bed tubular reactors are limited by the need to balance tube diameter and catalyst pellet size to achieve effective temperature control without excessive pressure drop. These reactors generally contain tens of thousands of tubes of around 25 mm diameter, resulting in high construction costs with catalyst pellets in the range of 1–2 mm diameter, which reduces catalyst productivity and selectivity to hydrocarbon liquids.

An alternative synthesis route is through slurry reactors. This type of reactor is more efficient at heat removal and uses catalyst powder of the order of tens of microns diameter to minimise pore diffusion resistance. However, slurry reactors can suffer from catalyst attrition, which leads to catalyst loss and product purity issues, and are also less straightforward to scale up compared to fixed-bed alternatives.

Since 1996, Johnson Matthey and bp have been collaborating to bring FT synthesis to the industrial scale. The first major joint venture in 2002 was to build the Nikiski demonstration plant in Alaska, USA (Figure 2), based on the first generation (Gen1) FT catalyst contained within conventional tubular reactor technology (14). The Nikiski plant produced a nominal 300 bbl day^–1 of synthetic crude product from pipeline natural-gas feedstock; and by the time it was decommissioned in 2009, the plant had exceeded all its performance goals related to catalyst productivity, hydrocarbon selectivity, carbon monoxide conversion, methane selectivity and catalyst lifetime. A single charge of catalyst ran for just over 7000 h enabling Johnson Matthey to predict an expected three-year lifetime without any regeneration.

Fig. 2

The integrated plant combined three processes for testing FT technology: a novel compact reformer for syngas generation; a fixed-bed FT reactor; and mild hydrocracking of FT waxes to produce synthetic crude. The original fixed-bed tubular reactor technology was developed as a method of monetising stranded natural gas in remote locations. However, it was only competitive at large scale, above 30,000 bbl day^–1 (~3850 metric tonnes per day (mtpd)), in areas with low natural gas prices and high oil prices.

More recent interest in FT technology is in small-scale applications to produce renewable fuel from MSW or cellulosic biomass. This involved developing technology that lowered costs whilst improving efficiency. In 2009, Johnson Matthey designed a novel catalyst carrier device to fit inside a tubular reactor that allows for the use of smaller catalyst particles. At the same time, bp developed an improved second generation (Gen2) catalyst formulation (15). Both organisations then worked to combine both the new catalyst and the novel catalyst carrier device, which produced a step change in commercial FT performance (see Figure 3). The CANS^TM catalyst carrier technology received global recognition, winning both the Research Project Award and the Oil and Gas Award at the Institution of Chemical Engineers (IChemE) Global Awards in 2017, and the Rushlight Clean Energy Award and Rushlight Bioenergy Award in January 2020. These accolades demonstrate how advanced FT technology will dramatically impact the chemical engineering industry, with many real-world applications.

Fig. 3

The novel catalyst carrier reactor design (16) combines the advantages of the fixed-bed tubular reactors and the slurry-phase systems. Its modular design enables low-risk scale-up and simple operation, while the smaller catalyst particles offer high productivity and selectivity. The stacked catalyst carriers have a unique design that aids their ability to perform FT synthesis as shown in Figure 4.

Fig. 4

A reactor tube contains 60–80 of the CANS^TM catalyst carriers and effectively creates a series of mini adiabatic radial-flow reactors with interbed cooling. Radial flow through each CANS^TM catalyst carrier means that, although the reactor tubes are 10–15 m long, the effective catalyst bed thickness is only around 15% of the overall tube length. This enables the use of sub-millimetre catalyst particles, which improves selectivity and activity whilst limiting the reactor pressure drop to that of a conventional fixed-bed tubular reactor. Wide-diameter tubes of 75–100 mm are used in the novel reactor, which has the effect of reducing the heat-transfer surface per unit volume of catalyst. However, this is compensated by a larger temperature difference at the wall, where reactants are hottest (as opposed to the centre of the tube in conventional fixed-bed tubular reactors). Combining this structure with a high gas velocity through the narrow annulus between the CANS^TM catalyst carrier body and tube wall results in excellent heat transfer. By separating heat removal from the catalyst bed, good control of the reaction temperature is also achieved without the risk of quenching the reaction. The advanced reactor technology also enables operation with >50% inerts in the reacting gas, allowing a single-stage FT reactor to be used in a recycle loop to maximise overall conversion of carbon monoxide to >90% (Figure 5).

Fig. 5

Compared to conventional fixed-bed tubular reactors, the new CANS^TM catalyst carrier and optimised catalyst reduces the number of reactor tubes by 95%, significantly simplifying the design and fabrication of the reactor, resulting in a reduction of capital expenditure costs of around 50% for the FT unit. There is also a three-fold increase in production for the same size reactor as the catalyst performance is closer to that of a powder, with excellent heat and mass transfer to, from and within catalyst particles. The increased productivity at least halves the catalyst volumes usually required for the same production rate. Additionally, containing the catalyst inside the CANS^TM catalyst carrier removes the requirement to filter the catalyst from the wax product. Instead the catalyst is easily replaced by removing the entire catalyst carrier, meaning there is no interaction with the hazardous cobalt catalyst material. Fundamentally, this makes FT applications possible at both small and large scales, with around 6000 bbl day^–1 (770 mtpd) achievable in a single reactor of around 900 tonnes. For areas with tighter transport restrictions, 2000 bbl day^–1 can be delivered in a single reactor of around 4 m diameter and 250 tonnes in weight.

One of the main challenges to overcome was proving that the concept worked at commercial scale, and so Johnson Matthey has invested in extensive testing to develop the engineering science necessary to implement the novel reactor concept. This involved building customised rigs to validate heat-transfer performance, hydraulics and accurately measure reaction kinetics on high-throughput microreactors. Significant engineering effort has been required to develop models capable of accurately predicting the performance of commercial-scale reactors. The initial proof of concept work was carried out using CANS^TM catalyst carriers manufactured by an experienced prototyper.

FT catalysis is strongly influenced by cobalt crystallite size, support properties and catalyst treatments. Selection of the cobalt crystallite size is critical to obtaining the required performance. Larger cobalt crystallites result in a less active catalyst due to the lower surface area to volume ratio, and thereby require higher temperatures to achieve a target conversion. Alternatively, if cobalt crystallites become too small the chemistry favours chain termination (methane formation) over chain growth (C–C coupling). While the desirable cobalt crystallite (17, 18) size is in the narrow range of 8 nm to 10 nm for FT synthesis, the activity of a good catalyst can be significantly reduced by suboptimal treatments, such as the reduction stage of the cobalt oxide to the active metallic phase. Cobalt-based FT catalysts are normally made by impregnating the support with a cobalt salt, calcination to give cobalt oxide and subsequent reduction under hydrogen in the plant to give the active cobalt metal phase. The catalyst reduction process is defined in Equation (ii), which highlights the significant levels of water that are produced throughout the catalyst bed, and this in turn can sinter, reoxidise or damage the catalyst significantly if not fully catered for under process conditions. Catalyst bed profile effects are also significant as the bottom sections of the bed are exposed to the water produced at the top of the bed, while higher pressures required commercially also lead to higher water partial pressures in the catalyst pores.

(ii)

bp originally developed the Gen2 catalyst formulation as a drop-in for conventional fixed-bed tubular reactors and this formulation has been adapted to the CANS^TM technology by Johnson Matthey to produce sub-millimetre size catalyst particles at scale. Developing the new formulation to achieve improved activity, selectivity and stability, while optimising the catalyst activation has required thousands of hours of testing at laboratory scale in high throughput and pilot plant test units. This has been supported by a state-of-the-art FT unit, with exceptional online analytics for all products up to C18 and analytical capabilities such as in situ X-ray diffraction and temperature programmed reductions which enable catalyst evaluation under process conditions (19, 20).

While the hydrocarbons and oxygenates that were identified are known compounds formed during the low temperature, cobalt catalysed, FT process the combination of the multiple analysis techniques used has allowed a level of detail to be gained on the FT product composition that is seldom reported (9). Typically, the long-chain 1-alcohols and carboxylic acids were found to be present at levels of one tenth and one thousandth that of hydrocarbons of equivalent carbon chain length respectively. Additionally, hydrogen-1 nuclear magnetic resonance (¹H-NMR) and carbon-13 nuclear magnetic resonance (¹³C-NMR) analyses were used to quantify the average class compounds concentration of 1-olefin, cis- and trans-2-olefins, 1-alcohol and aldehyde as appropriate for the technique used. The 1-olefin:n-paraffin ratio in the hydrocarbon liquid and wax products was found to decrease significantly with increasing carbon chain length in both phases and much more so than those of the 2-olefin or 1-alcohol.

Catalyst activity and selectivity is only part of the process however, with stability, robustness to process events and life duration also playing a vital role in a commercial catalyst. Johnson Matthey Davy and bp built on their extensive experiences of the Gen1 catalyst in the Nikiski demonstration plant to optimise this further for the Gen2 catalyst. This included several catalyst life tests which operated for many thousands of hours at steady FT process conditions. This included the catalyst formulation used in CANS^TM catalyst carriers operating with exceptional performance over an 18,000 h life test. The gradual drop in catalyst activity over this period was compensated by an increase in operating temperature within the reasonable limits of a commercial reactor. Despite frequent shutdowns and other challenges associated with laboratory-scale operation, the catalyst was still showing good activity and selectivity at the end of this test. This is a result of the process having been designed to be robust and operate in chemically stable conditions.

The CANS^TM catalyst carrier concept has been successfully demonstrated at commercial scale on a pilot plant at Johnson Matthey’s research and development (R&D) facilities in Stockton-on-Tees, UK. It is not practical to test a full-length commercial reactor tube at these facilities, due to limitations on gas supply and product storage capacity, so a creative approach was required. Flexible design of the pilot plant enabled testing of multiple commercial-size CANS^TM catalyst carriers in a much shorter tube; by recycling gas, liquid products and produced water to simulate the full range of conditions and flowrates present in a commercial reactor tube. This, coupled with raising steam in the reactor cooling jacket, has enabled full demonstration of the catalyst, CANS^TM catalyst carriers, hydraulics and heat transfer at commercial conditions, flows and tube diameters.

Over 20,000 h of testing under commercial flowsheet conditions has demonstrated the performance of the CANS^TM catalyst carrier and the Gen2 catalyst with a confirmed product slate and stable catalyst life. A C5+ selectivity of around 90% and C5+ productivities in excess of 300 g l^–1 h^–1 have been demonstrated on the pilot plant. The crude FT product consists of a wax stream which is liquid at reaction conditions and solid at ambient temperature and a light hydrocarbon condensate stream which is liquid at ambient temperature. Figure 6 shows both these products are high quality, clean and catalyst-free.

Fig. 6

The International Energy Agency, France, has measured the share of global energy-related CO₂ emissions from transport at 23%, with aviation contributing 2–3% of worldwide anthropogenic CO₂ emissions (21). There is great potential for this figure to be reduced by using synthetic fuels from sustainable feedstocks, and this makes fuels produced via the FT process an attractive alternative to current aviation fuels. Synthetic fuels also burn cleaner, due to the absence of sulfur and aromatics, while also producing fewer particulates (22). As a result, FT fuels lead to increased combustion and turbine life, while the enhanced thermal stability reduces deposits on engine components and fuel lines. This results in a dual advantage for the aviation industry, in terms of both improved fuel economy and less maintenance of aviation equipment.

The scale of the market is substantial, with air transportation alone expected to consume at least 500 million tonnes per year (11 million bbl day^–1) of fuel by 2050 (23). Practical limitations on the supply of waste feedstocks or local, low-cost renewable power for hydrogen production typically limit the scale of each project to less than 5000 bbl day^–1. With tens of thousands of filled CANS^TM catalyst carriers required for each project, the ability to consistently and efficiently produce both the CANS^TM catalyst carriers and the FT catalyst which they contain is crucial for successful commercial deployment of the technology.

To address this, Johnson Matthey has collaborated closely with a company skilled in delivering sustainable engineered solutions for vehicle exhaust after treatment systems, to develop a mechanical design for the CANS^TM catalyst carriers that is economical to make, can be easily filled with catalyst and meets the functional specifications developed by Johnson Matthey. This has been confirmed by testing of commercial prototypes.

A production line has been constructed and commissioned for mass manufacture and catalyst filling of the CANS^TM catalyst carriers, and the first charge has now been produced for the first commercial project. The production line is fully automated to allow the safe filling of the cobalt-containing catalyst and contains state of the art equipment and in-line quality control to assure the CANS^TM catalyst carriers meet the required functional specifications. The functional specifications were established during the development of the CANS^TM catalyst carriers from concept to prototype with testing performed on in-house built rigs at Johnson Matthey in Teesside, UK. Identifying these upfront allowed Johnson Matthey and the manufacturer to work together to ensure the resulting production line would safely, efficiently and consistently produce high-quality units.

A collaborative team of engineers from Johnson Matthey and the manufacturer worked closely during initial commissioning of the line to work through the various challenges associated with scale-up to mass manufacture of a novel process. This knowledge will be invaluable as further improvements and optimisations are implemented.

In order to achieve high-quality performance, challenging activity and selectivity targets were set for the FT catalyst. Scale-up of the chosen formulation took place at Billingham in the UK. Catalyst preparation was initially at laboratory scale with short-term and long-term testing of development samples conducted using both micro-reactors and CANS^TM catalyst carriers, facilitating accurate modelling of the performance of a full-scale FT reactor.

As scale-up continued, preparation of the FT catalyst moved into the Manufacturing Science Centre (MSC) where appropriate technologies were identified for each of the steps involved in catalyst production. The technical risk of scale-up was minimised by using down-scaled versions of full-scale production equipment.

A fully developed catalyst manufacturing process was transferred from the MSC to a dedicated production asset located at Clitheroe in the UK (Figure 7). Careful attention was paid to the specification of the raw materials used. To ensure a proper understanding of the impact of trace impurities on long term FT catalyst performance, a series of experiments were conducted in which the FT catalyst was doped with different FT poisons.

Fig. 7

Every production batch of FT catalyst has been tested against an agreed quality assurance specification. Conforming product was loaded into CANS^TM catalyst carriers as it was manufactured, thus minimising the overall production timeline.

The Johnson Matthey Davy/bp FT technology incorporating CANS^TM catalyst carriers offers benefits to both small- and large-scale operations with good economics, opening up the prospect of exciting future applications.

Fulcrum BioEnergy, USA, is the first to licence the Johnson Matthey Davy/bp FT technology in its Sierra BioFuels Plant, located near Reno, Nevada (Figure 8). The Sierra plant will be the first in the USA to produce a renewable low-carbon transportation fuel from MSW or household garbage. The plant will first sort the waste to recover recyclables and remove material not suitable for processing, so is not in competition with recycling processes. The remaining material will be processed into a feedstock before being fed into a gasification system to produce a syngas. This is then converted into hydrocarbons by the FT technology for the production of renewable fuels. The Sierra plant construction is approaching completion and when operational will convert approximately 175,000 tonnes of MSW into approximately 42 million litres of renewable FT product each year.

Fig. 8

Multiple projects are being developed in the USA and Europe, which can make a significant contribution to meeting the demand for renewable transportation fuels in the next decade.

In order to meet greenhouse gas emissions reduction targets, especially for aviation (24), production of sustainable fuels will have to substantially increase. There are a range of sustainable fuels potentially available, but limitations on sustainable feedstocks and viable technology routes mean that diesel and jet fuel production via FT synthesis will need to form a key part of this industry. Producing fuels via FT synthesis is not new. However, cost of production was always a barrier, with existing large-scale producers of FT fuels unable to economically scale down to match the size of the waste facilities that feed them. The CANS^TM technology addresses this problem, offering an economic and efficient solution at the scales required by the industry.

The proposed solvent plant is co-located with a Kraft paper and pulp mill in China with throughput as defined in Table I. Figure 1 outlines the Kraft process, which conventionally directs weak black liquor to multi-effect evaporators, producing strong black liquor which is combusted in a Tomlinson boiler to produce steam (4). This steam makes the mill self-sufficient in steam and electrical energy. Importantly, the cooking chemicals (NaOH and Na₂S) are recovered and recycled to the pulping process.

Fig. 1

As previously mentioned, investments in heat integration have freed up a portion of the weak black liquor coproduct for alternative uses. This study explores the opportunity of utilising this excess coproduct, taken as 25% of total production, for isopropanol and acetone production through aerobic fermentation in an integrated solvent plant as outlined in Figure 1.

Given black liquor has no economic value as a product, it is costed at its utility value. This is calculated based on its conventional use for renewable electricity generation, requiring capital investment in increased steam turbine capacity. The foregone net present value (NPV) associated with this conventional use is used as the utility value for the black liquor feedstock.

In the proposed solvent plant (Figure 1), weak liquor undergoes SCWG to carbon dioxide and hydrogen. A challenge, however, is the efficient recovery of the cooking chemicals from the SCWG reactor and their recycle to the pulp mill digestor. Loss of these salts would result in a significant cost to the pulp mill. Under supercritical conditions, the properties of water change from polar to apolar, where the solubility of inorganic salts is very low (31). Cao et al. described the precipitation of alkali sodium salts in SCWG, reporting a neutral pH for the reactor effluent, suggesting that under supercritical conditions the salts largely precipitate from the solution (32). However, this precipitation can cause issues with plugging and fouling within the reactor (33). In this study the salts are removed prior to entering the SCWG reactor, in a manner similar to supercritical water desalination (34, 35) and modelled for SCWG of black liquor (33).

The solvent plant’s mass and energy balance was informed by experimental data from continuous gas fermentation (28), and rigorous process simulation using Aspen HYSYS v11. The lignin content in black liquor was modelled as guaiacol, a model compound for lignin (36), as principal feed to the solvent plant. The weak black liquor is further diluted prior to entering the SCWG reactor, as lower biomass concentrations promote superior thermal cracking and yields greater hydrogen and carbon dioxide owed to the increased water concentration favouring the forward water-gas shift reaction (37).

The simplified flow diagram (Figure 1) outlines the six plant sections of the solvent plant, whilst Figure 2 presents a detailed process flow diagram and operating conditions for upstream and downstream processing. The unit operations included in each of the six plant sections are summarised in Table II. Table III summarises the scale-up of the experimental gas fermentation data for the process simulation, which recognises the oxygen mass transfer limitations associated with the safety requirement to maintain non-flammable operating conditions. The heat integration between the low temperature exothermic gas fermentation and the high temperature endothermic SCWG is facilitated using a heat pump with isopentane as the working fluid (28).

Fig. 2

Isopropanol and acetone are produced in both the aqueous and vapour phase of the bioreactors. The solvents in the vapour phase are recovered via gas absorption through mass integration using internal process streams, i.e. the isopropanol product was utilised to recover acetone, and water to recover isopropanol. For the isopropanol in the aqueous phase, azeotropic distillation is required due to the homogeneous minimum boiling point azeotrope formed between isopropanol and water (38). Conventionally, this azeotrope is broken using an entrainer, historically benzene (39). However, owed to its carcinogenic properties, alternative entrainers such as cyclohexane have been adopted (40). An alternative azeotropic separation technique is pressure swing distillation, taking advantage of the composition differences in the azeotrope at different pressures (41). In this work, pressure swing distillation was employed with the coproduct acetone acting as an unconventional entrainer. Further detail of the separation train is presented in Figure 2.

A U-loop bioreactor, similar to the one used by Peterson et al., is used in this work (42). The benefit of a U-loop bioreactor is that high mass transfer coefficients can be achieved without the need for mechanical agitation, leading to greater oxygen transfer rate and a reduced power requirement compared to conventional stirred tank reactors (42). The oxygen mass transfer coefficient calculation associated with the solvent plant’s mass balance is presented in Table S1 in the Supplementary Information (available with the online version of this article), falling at the lower end of the range of mass transfer coefficients reported by Peterson et al. (42). Details of the experimental gas fermentation data is presented in Table III; a more detailed explanation of the experimental procedure can be found in Bommareddy et al. (28).

Significant heat integration makes the solvent plant self-sufficient in electricity and both low and medium pressure steam. Furthermore, process water recovered from distillation and the steam condensate is recycled to reduce the water make-up burden.

The process flow diagram for conventional renewable electricity generation, used to value the black liquor, is presented in Figure 3. An additional steam turbine is required to produce the renewable electricity for sale, relying upon the existing multi-effect evaporators, air compression and Tomlinson boiler. Superheated steam at 9000 KPa and 480ºC is used in the steam turbine (44). The medium pressure steam exiting the turbine is used in the multi-effect evaporators to concentrate the excess black liquor to 75% and to preheat the auxiliary air supplied to the Tomlinson boiler. Similarly, the associated electricity demand for the air compressor and pump is provided by the electricity generated. Resultantly, through conventional renewable electricity generation, the excess black liquor produces 138 GWh year^–1 for sale to the grid.

Fig. 3

The mass and energy balance associated with the rigorous process simulation informs the capital cost, fixed operating cost and variable operating cost estimation. For the capital cost estimation, major equipment purchase costs were estimated using the models from Seider et al. (45), with the exception of the turbo-expander (46). Three different methods are used to calculate the FCI, owed to differences in the estimation methods. These three methods are designated as: the NREL method outlined in the 2011 NREL report (27), the Towler and Sinnott (TS) method taken from Chemical Engineering Design (47) and the Hand method detailed in Sustainable Design Through Process Integration (48). The calculation basis of the three methods is presented in Table IV.

For all three methods, the calculated equipment purchase costs are multiplied by an installation factor to obtain the inside battery limit (ISBL) installed costs. Both the NREL and Hand methods use installation factors dependant on the equipment type, whereas the TS method uses a universal multiplier. All installed equipment costs were adjusted to 2019 costs using the Chemical Engineering Plant Cost Index of 607.5 (49). A location factor of 0.51 was used for China (using indigenous materials), based on the 2003 location factor of 0.61 (47), updated to 2019 via the Chinese Yuan to US dollar exchange rate.

Three methods were used to calculate the fixed operating costs as summarised in Table V. As before, the NREL method (27) and the TS method (47) were employed. However, as the Hand method is solely for FCI, the third was the taken from Coulson and Richardson Volume 6 (50). Variable operating costs were estimated based on the costs detailed in Table VI, subject to annual inflation as outlined in Table VII.

Time series analysis was used to forecast the long-term average price of isopropanol and acetone. Takens’ theorem was used as the basis for this analysis (53). Takens’ theorem states that for a deterministic system, the underlying state variables that created the time series are embedded within the data. Using this theorem a deterministic, dynamic system can be reconstructed based on the observed time series. Forecast models constructed using the embedded state variables assume that the market drivers underpinning the trajectory of the state variables in phase space remain largely unchanged. An embedding dimension of 10 was used to reconstruct the isopropanol and acetone price models from monthly average price data obtained from the Intratec database (54). In this work, a radial basis function neural network (RBFNN) containing eight neurons was used as a model to predict the future commodity prices. The network was trained as a one step ahead predictor by minimising the mean square error of the difference between the actual and predicted prices. Once trained, the network was evaluated (tested) in free run mode, where successive predicted prices (outputs) become inputs to the RBFNN. The confidence limits corresponding to the trained RBFNN were calculated as a reliability measure of the prediction as per the work undertaken by Leonard, Kramer and Ungar (55). The benefit of using an RBFNN is that the resultant forecast price is an impartial product of the dataset’s underlying state variables.

The long-term average price for renewable electricity sales was taken as US$0.109 kWh^–1 as per the biomass subsidy in China (56). This is used to inform the renewable electricity project to value the black liquor and for the excess electricity generated by the solvent plant.

The cost models from Section 2.3 and the product price forecast models from Section 2.4 inform the investment analyses. The black liquor is costed at its utility value, calculated as the foregone NPV from generating renewable electricity. Resultantly, the NPV for the solvent plant is calculated by subtracting the NPV of renewable electricity generation. The investment analysis parameters used are detailed in Table VII.

A sensitivity analysis was conducted using a Monte Carlo simulation based on the cost parameters in Table VIII, creating an uncertainty framework. The cost parameters were taken from (47), with the exception of renewable electricity sale price where the upper limit for the long-term average price was capped at the current biomass subsidy in China, US$0.109 kWh^–1. This limit was applied due to the decreasing trend in renewable electricity subsidies (58). In contrast, the long-term average prices for isopropanol and acetone were varied ±30% from the forecast price. This provides a stochastic counter to the assumption used to determine the forecast prices: that the deterministic market drivers underpinning the trajectory of the state variables remain largely unchanged. However, given that market drivers are subject to change, the long-term average price may be banded with an equal likelihood of being higher or lower than the forecast price.

A uniform distribution for these parameters was used and varied for the solvent plant and conventional renewable electricity generation (used to value the black liquor). All the cost parameters in Table VIII, other than labour and electricity, were varied independently. 2000 simulations were run, stochastically varying the parameters within the defined lower and upper limits to produce a probability distribution of the solvent plant’s NPV.

A cradle-to-gate LCA model was developed using the ecoinvent 3.6 inventory database, following ISO Standards 14040 (59) and 14044 (60). GHG emissions were calculated based on the most recent Integrated Pollution Prevention and Control 100-year global warming potential (GWP) factors to quantify GHG emissions in terms of carbon dioxide equivalents (CO_2eq) (61). Functional units were defined as 1 kg isopropanol, 1 kg acetone and 1 kWh of electricity. In line with the investment analysis, the LCA model considers the net electricity output of solvent plant by subtracting the foregone electricity from combustion of black liquor at the pulp mill. Life cycle environmental impacts are allocated between these three products using both economic and energy allocation. The GHG emission rate for the external process inputs: cooling water, process water and ammonia were taken from the ecoinvent 3.6 inventory database using the allocation at the point of substitution system model (62), whereas electricity was taken as the 2018 China electricity mix (63). The bio-based solvents isopropanol and acetone sequester biogenic carbon dioxide and hence are credited with a negative GHG emission based on their carbon content. Downstream activities, including the use and end-of-life of isopropanol and acetone products are not considered. These activities are assumed to be identical to those of conventional isopropanol and acetone, given that they are chemically and functionally identical, and therefore have no influence on the relative GHG emissions of renewable and conventional solvent products.

The solvent plant (Figure 2) produces three products, summarised in Table IX. The contribution of each product to the plant’s income is also presented. Whilst isopropanol contributes to almost half the solvent plant income the renewable electricity fraction is the second highest contributor, highlighting the significant amount of renewable electricity generated by the solvent plant.

The investment analyses for the solvent plant and conventional renewable electricity generation are detailed in Tables S9 and S12, as per the investment analysis parameters presented in Table VII. The NPV for conventional renewable electricity generation represents the utility value of the black liquor, valued at US$73 million (Table S12). This is subtracted from the NPV of the solvent plant (US$115 million) to produce the cumulative NPV presented in Figure 9. For the nominal TEA model inputs, the solvent plant’s net cumulative NPV is US$42 million.

Fig. 9

Given the conceptual stage of the TEA, a Monte Carlo simulation was undertaken as per the uncertainty framework outlined in Table VIII. The produced probability distribution in Figure 10 avoids making an investment decision based solely on nominal TEA inputs. The cumulative probability curve presents a 70% probability that the solvent plant will achieve a net cumulative NPV between US$35 million and US$85 million, noting that no negative outcomes are predicted.

Fig. 10

Figure 11 summarises the outcome of the cradle-to-gate LCA for the solvent plant, compared to the conventional fossil derived processes; using both economic and energy allocation for the isopropanol, acetone and renewable electricity products.

Fig. 11

Both solvents achieve negative GHG emissions when produced via the solvent plant using economic and energy allocation. The GHG emission for the two allocation methods are comparable, indicating the price per unit energy (US$ MJ^–1) is similar for all three products. The negative emissions are an intrinsic outcome of the cradle-to-gate framework, which excludes the end use for the products. As the total GHG emissions of the solvent plant are lower than the overall biogenic carbon sequestered, negative GHG emissions are achieved for the solvent products.

The negative GHG emissions compare favourably to the conventional isopropanol (hydration of propene) and acetone (oxidation of cumene) processes. Additionally, the GHG emissions associated with the excess renewable electricity from the solvent plant also compare favourably to the electricity mix in China 2018). Furthermore, as the end use for the solvents remains the same regardless of the production method, the relative GHG emissions are valid beyond the cradle-to-gate framework.

As highlighted in the Introduction, the commercial implementation of gas fermentation is largely dominated by anaerobic fermentation. Therefore, it is important to compare the results to a best-in-class technology. In addition to successfully commercialising ethanol production via gas fermentation, LanzaTech have also investigated gas fermentation to produce acetone, a precursor to isopropanol (64). As such, LanzaTech’s investigation undertaken for the US Department of Energy (US DOE), in collaboration with Oak Ridge National Laboratory, USA, is used as a benchmark anaerobic process (65).

As highlighted previously, the primary differences between anaerobic and aerobic fermentation technologies are inherent to the C1 fixation metabolic pathways. Strictly respiratory (aerobic) cell factories require air to be continuously fed into the bioreactor to facilitate carbon fixation. In addition, owed to the intrinsic thermodynamic inefficiency of the Calvin-Benson-Bassham cycle employed by aerobic bacteria, an excess of low temperature heat is produced. As such, a conventional process flowsheet for aerobic fermentation employs operationally costly compressors and chillers. In contrast, for anaerobic fermentation there is a reduced chiller requirement and the compressor duty is less pronounced. Moreover, owed to the presence of oxygen, aerobic fermentations require the use of more costly stainless steel reactors and more complex process control systems. Whilst the latter is an intrinsic requirement of aerobic fermentations, in this work we have reconciled the increased utility demand of aerobic fermentation through process integration (28). This integration employs a heat pump to utilise the low temperature heat generated by aerobic fermentation to heat the SCWG reactor feed, removing the cooling water burden required by the bioreactors. Additionally, the compressor duty is fully supplied through the electricity generated upon letting down the SCWG reactor’s high-pressure gas product. As a result, the economic and LCA outcomes for the solvent plant should be comparable to anaerobic fermentation technology.

LanzaTech’s anaerobic study achieved a combined selectivity of 94.7% for ethanol and acetone, of which 57.3% was acetone (65). LanzaTech disclosed that by selling acetone at market prices they are able to sell coproduced ethanol at or below the US DOE’s 2022 target of US$3 GGE^–1 (GGE = gallon of gasoline equivalent) (66). Therefore, in this study, the price per GGE was calculated for the solvent products as a biofuel mix, with renewable electricity sold at the current market price. A value of US$2.87 GGE^–1 (Figure 12) was obtained, below the US DOE’s target, highlighting the competitiveness of the heat integrated aerobic solvent plant. Notably, neither isopropanol nor acetone are typically used for their fuel value, highlighted by their higher market prices. As such, the solvent plant is profitable as either a biofuel or commodity chemical facility.

Fig. 12

For LanzaTech’s anaerobic process, the cradle-to-gate LCA using energy allocation produced a calculated GHG emission of –1.9 kg_CO2eq kg^–1_{acetone + ethanol} for a heat integrated scenario (see Table S13 for calculation). In Figure 12, the LCA for the solvent plant is presented, indicating a net GHG emission of –2.04 kg_CO2eq kg^–1_{isopropanol + acetone}, which is in line with LanzaTech’s study (Figure 12). Resultantly, from both the TEA and LCA results, the greater thermodynamic efficiency of the anaerobic Wood-Ljungdahl C1 fixation pathway over the aerobic Calvin-Benson-Bassham Cycle is not evident for the heat integrated solvent plant.