Systematic study of sol-gel parameters on TiO2 coating for CO2 photoreduction

Abstract An optimized sol-gel method for TiO 2 coating was developed for CO 2 photoreduction using a systematic Design of Experiments (DOE) approach and an efficient Plackett-Burman (PB) design. Conducting fifteen experiments, the effects of seven sol-gel process parameters were investigated on four catalytic properties for CO 2 photoreduction: band gap, catalyst coverage, anatase crystallinity and ratio anatase:rutile. The PB design results yielded a wide range for the catalyst’s properties: band gap (1.86–3.35 eV), coverage (11.53–99.63%), anatase crystallinity (0–48.5 nm) and ratio anatase:rutile (1:0 and 0:1). Machine vision using manual global, Otsu global and adaptive thresholding were compared for estimating the coverage of catalyst on a glass slide support.

A C C E P T E D M A N U S C R I P T

Introduction
CO2 photoreduction is a promising route towards lowering CO2 emissions and potentially producing useful chemicals that include CH4 as a fuel. [1,2] TiO2 and its derivatives have often been used for the photoreduction of CO2 due to its band gap, low cost, availability and chemical stability. [3,4] The efficiency of CO2 photoreduction using TiO2 can be effected by the catalyst's properties: band gap [3], crystallinity [3,5] and ratio of anatase:rutile phases [6]. The band gap of TiO2 prepared by sol-gel has been previously modified using phosphorous dopants [7], nitrogen and fluorine dopants [8], addition of SiO2 to N-doped TiO2 and calcination temperature. [9,10] The crystallinity of TiO2 particles isolated from sol-gel have been tuned by calcination holding time [11] and calcination temperature. [9,10,[12][13][14][15][16] The effect of calcination temperature on the ratio of anatase:rutile also has been reported. [9,10,[17][18][19][20][21][22] For the practical application of CO2 photoreduction and flexibility in reactor design, bearing in mind the challenge of dealing with light with directional limitations as an energy source, the catalyst needs to be applied to a support to yield a quality coating. TiO2 coating by sol-gel is attractive, as it requires a simple set up and is relatively low cost. Films of TiO2 have been prepared using sol-gel dip coating and spin coating methods [9,10,12,[23][24][25][26]. Sol-gel preparation parameters have been previously studied, including amount of water, solvent, acidic hydrolyzing catalyst, high molecular weight co-polymer, calcination temperature and holding time. However, there is no example that includes a detailed investigation of all the parameters used in the preparation of a TiO2 sol-gel on the catalyst's properties: band gap, crystallinity, ratio anatase:rutile and coating quality. This is most likely the case as sol-gel procedures have several preparation parameters and it would require a large number of experiments to investigate each parameter's effect on each property. In addition, using published work and consolidating them into a single experiment that yields all of the desired A C C E P T E D M A N U S C R I P T properties for CO2 photoreduction is implausible, due to the variation in reagents, catalysts, experimental settings and setups used.
Design of Experiments (DOE) is a powerful tool that allows for generating efficient systematic experimental designs to screen and optimise parameters on selected responses or in this case catalyst properties. DOE is an efficient tool for optimising yield for the synthesis of catalysts and catalytic processes; a variety of examples using full factorial, central composite and Box-Behnken designs are described in the literature. [27][28][29][30][31][32][33] The preparation of thin films for novel materials and catalytic processes are challenging, as they require the investigation and optimisation of several process parameters. DOE offers a robust and systematic method for investigating and optimising these coating parameters. A D-optimal design was successfully employed for the development of inkjet printed electrochromic thin films. [34] The effect of parameters for the preparation of TiO2 thin films: amount acetic acid, water, ethanol, co-polymer Pluronic P123 and calcination temperature on the activity of photodegradation of methyl orange was investigated using a Taguchi method. [35] In another example, the preparation of TiO2 powder by sol-gel process, a full factorial design was used to investigate the effect of the parameters (amount Titanium tetra-isopropoxide, water, hydrochloric acid and flow feeding rate of reactants )on the mean particle size of calcined TiO2 isolated. [36] In this work, a systematic approach using DOE is presented to screen and optimise seven variables: amount isopropanol (IPA), acetic acid, water and polypropylene glycol (PPG), calcination temperature, heating ramp and holding time in fifteen experiments ( Table 1). A subset design of DOE, a Plackett-Burman (PB) was used, as it is very efficient at screening four or more parameters in as few as fifteen experiments. To the author's knowledge, this is the first example of a very efficient Plackett-Burman design being used for the process development and investigation of the effect of seven sol-gel preparation parameters on several photocatalytic and physical properties of TiO2 for CO2 photoreduction. (www.minitab.com) was used for creating the experimental design points ( Table 1) and analysing the design's responses ( Table 2). Minitab was used to model each parameters effect on a response by fitting a where: , is the fitted response and actual response values respectively 0 is the y-axis intercept is the error residual is the matrix of parameter values, is the number of parameters is the co-efficient of the parameter calculated form the cost function = 1,…, for n parameters The parameter coefficients ( ) and error term ( ) were determined from a cost function that minimizes the error residual ( ) between the actual response value and fitted response ( ) by a least squares method Eq. 3.
Parameters coefficients were significant if the value is shown to be statistically not equal to zero. This was determined by analysis of variance (ANOVA), by calculating the p-value with the null hypothesis that the parameter coefficient was equal to zero. The null hypothesis was rejected if the p-value was less than the significance level (alpha, α). The terms were systematically removed from the linear model starting with the parameter coefficient with largest p-value. If three or less parameters were significant, the model was The effect of the parameters were investigated on the responses: band gap, coverage, crystallinity and phase ratio of anatase:rutile ( Table 2). The models presented were compared to the literature and consolidated to yield a single experimental point using a single coating procedure that yielded a desirable response for band gap, crystallinity, ratio of anatase:rutile and coating coverage. The best design point was further optimized by adjusting the PPG volume equivalents. The use of lower cost lower molecular weights PPG were investigated.
The optimized procedure was used to coat stainless steel mesh supports and benchmarked with P25 TiO2.

Preparation of TiO2 sol-gel
Sol-gel mixtures were prepared using different amounts of chemicals as listed in Table 1. 0.6 ml of titanium (IV) n-butoxide was used for all of the experiments. IPA, titanium butoxide and glacial acetic acid were transferred and agitated together in a sealed sample vial. Deionized water was added in a single step and the vial sealed.
To ensure homogeneity, of at times a viscous mixture, the mixture was agitated for eighty minutes. The mixture was stored in a sealed vial to age overnight. PPG was then added in a single step before being agitated for 40 minutes. Samples were stored in a fridge set at 4 °C for a maximum period of one month, to prevent excessive ageing, before being used to coat a glass strip support. Additional experiments were carried out using the same procedure described above for WT2 ( Table 1.) towards testing the effect of molecular weight (400, 1000 and 3000 g.mol -1 ) and volume equivalents of PPG on the sol-gel.

Coating and calcination
Glass strips (15 of 0.5 mm × 25 mm) were cleaned with dilute sulphuric acid in water (10%), rinsed with deionized water washes (3 × 25 ml) and one IPA wash (25 ml). The glass strips were then dried overnight in an oven at 70 °C. Approximately 2.5 ml of the sol-gel prepared was transferred to a HPLC sample vial. The glass strips were immersed into the sol-gel for one minute before being withdrawn using a dip coater motor set at 50 Hz (Te-Sun Industry Works, Taiwan). The coated glass strips and an aliquot of the sol-gel prepared were calcined as shown in Table 1. For CO2 photoreduction tests, stainless steal gauge 20 mesh was coated using a similar procedure described above for the glass strips without a dilute sulphuric acid wash.

UV-vis absorption
The light absorption and electronic band were characterized using a UV-vis spectrometer (Perkin Elmer lamda 950) equipped with a 150 mm integration sphere (Perkin Elmer). The band gap was determined using a direct intercept method (Fig 2) to estimate the adsorption maxima which was then used in Eq. 4. [37,38] = ℎ ×

Image analysis
The coated slides and coverage were analyzed by scanning electron microscopy (SEM) and Energy-dispersive X-ray spectroscopy (EDX

Crystallinity and phase analysis
Powder X-ray diffraction to determine crystallinity and anatase:rutile phase ratio was performed using a Bruker D8 Advance powder diffractometer, operating with Ge-monochromated Cu K-alpha radiation where: is the fraction of anatase phase in the powder and are the X-ray intensities of the anatase (101) and rutile (110)

Band gap response
A range of band gaps were recorded for the fifteen PB experiments (Fig 2). The PB design was converted to a two factor full factorial design and an interaction effect between calcination temperature and holding time was found (Fig. 3c). Increasing the calcination temperature resulted in higher band gap values (Fig. 3b). An increase in the holding time decreased the band gap recorded and had less of an effect relative to calcination temperature (Fig. 3a,b). The interaction main effects plot shown in Fig. 3c showed that at high calcination

Coverage response
Machine vision provides a powerful tool for analysing images. [42] SEM images were taken of the coated glass slides and a Python script was written to perform manual global, Otsu global and adaptive thresholding -see supplementary info for the script used.
The PB design yielded a variety of coatings ranging from poor clumped coatings (Fig. 4a,b) to a thin uniform film, using WT2 settings, with high surface coverage of 98.94% using adaptive thresholding (Fig. 4f). The Otsu method worked the best out of the three Python methods employed as it showed the greatest variation (Otsuσ² = 432.67, Adaptiveσ² = 227.91 and Manualσ² = 35.15) and showed the best estimation of coverage when compared by eye with the original images ( Table 2). The Python script yielded satisfactory thresholding results, although it must be noted that confirmation of the thresholding still required confirmation by eye.
The amount of water showed the greatest effect on the percentage coverage with an increase in water decreasing the percentage coverage (Fig. 5a,b). Rezaee et al found that for the preparation of yttria-doped zirconia sol-gels, the coating showed more cracking by SEM when excess water was used. [43] Using more water for the preparation of the sol-gels may have caused agglomeration, as water was added in one portion that may have resulted in very fast and complete hydrolysis and polycondensation producing agglomerates.
Moemeni et al reported increased cracking with higher calcination temperature and proposed this was due to differences in expansion co-efficients between the coating and support used. [9,10]

Crystallinity response
The PB design could be converted to a two factor full factorial design for the crystallinity of the anatase phase response as only one factor, calcination temperature, was found to have a statistically significant effect on the crystallinity response (Fig. 5c). Amount of PPG was included as it improved the model by increasing the random distribution of residual vs. fitted values. It is potentially an interesting factor to consider in future studies. Again, as per the reasons given when discussing the ratio of anatase to rutile response (Section 3.1.5), the temperature of the low design points and midpoints were taken at 400 and 525 °C, respectively. The experimental design showed an increase in the crystallinity with higher calcination temperature and the amount of PPG used for the preparation of the sol-gel. An increase in calcination temperature has been reported to yield higher TiO2 crystallinity. [9,10,[12][13][14][15] A C C E P T E D M A N U S C R I P T

Phase ratio response
The PB design showed that the amount of IPA and water used for the preparation for the sol-gel along with the calcination temperature and heating rate were statistically significant effects on the ratio of anatase to rutile phase (Fig. 6a). The results presented use the amount of anatase as a response, if rutile was used the same factors would have been statistically significant and the main effects plots would have been the inverse of that shown in Fig. 6b. It can be seen that an increase in the amount of IPA, calcination temperature and heating rate favored a lower ratio of anatase, whilst the opposite was true for the amount of water used (Fig.   6b). The effect of temperature on the phase change from anatase to rutile is well documented. [9,10,[17][18][19][20][21][22] It is also worth noting that according to Hanoar et al, the transformation of the anatase to rutile phase is a complex process that could include the following factors: particle size, particle shape (aspect ratio), surface area, atmosphere, volume of sample, nature of sample container, heating rate, and inclusion of impurities. [21] The inclusion of organic crystallization impurities in this work may have resulted from the inclusion of IPA in the crystals formed. Ocana et al showed that the inclusion of organic impurities encouraged the formation of the rutile phase on the catalyst's surface. [46] 3.2 Optimisation of best design point

Optimal design point
Due to the nature of the wide design space explored for a PB design, there is a high probability that one of the points will yield a useful response. For this PB design, WT2 showed a favorable percentage coverage (98%), band gap (3.35 eV), crystallinity anatase phase (28.01 nm) and anatase : rutile ratio (25:75). WT2's properties would be of interest to benchmark against the commercially available P25 TiO2 photocatalyst.
Using this point's settings (WT2 , Table 1), the effects of PPG molecular weight and volume equivalents were investigated and are presented in the next section.

Effect of PPG molecular weight and volume used
Building on the PB design's results and using the best design point WT2, the coating process was further investigated for the effect of the molecular weight and amount of the PPG used. PPG with molecular weights: 400, 1000 and 3000 were tested. High molecular weight PPG costs more, and therefore, lower molecular weights were investigated to reduce the cost of the sol-gel preparation. Five volume equivalents PPG relative to the starting volume of titanium (IV) n-butoxide (0.6 ml) was used for the preparation of WT2. To optimise the amount of PPG used, 0.83, 1.66 and 2.50 volume equivalents PPG relative to the starting reagent titanium (IV) n-butoxide were investigated for the preparation of TiO2 sol-gel. When the molecular weight of PPG increased from 400 to 3000, using five volume equivalents PPG, the rutile:anatase ratio increases (Fig. 7). The biggest increase in rutile phase observed for PPG 3000. Some rutile was observed for PPG 1000 and this does warrant further investigation in the future. The anatase crystalline size increased with higher molecular weight (Fig. 7). The mean values recorded for band gap, fraction phase and crystallinity of the anatase phase using 0.83, 1.66 and 2.50 volume equivalents of PPG is shown in Fig. 8. Although a small increase was observed for band gap when increasing volume equivalents of PPG, a one way analysis of variance (ANOVA) indicated that there was no statistically significant difference in means for band gap, anatase phase, rutile phase and crystallinity of anatase phase. The p-value (p) is the probability that the F-statistic (F) is larger than the computed test-statistic value. A p-value that is more than the significance level (α), in this case α = 0.05, indicates that the means are not significantly different (Fig. 8). This supports the output from the PB model, where the volume of PPG had no statistically significant effect on these responses. volume equivalents of PPG, whilst P25 TiO2 exhibited a Type II adsorption isotherm (Fig. 9a) (Fig. 9b) PPG samples, an increase in CO2 adsorption capacity correlated with an increase in photocatalytic activity for both CO and CH4 production (Fig. 11). ). This is likely because a higher CO2 adsorption capacity would increase the probability of CO2 photoreduction as CO2 would be in closer proximity to active sites, especially the (001) anatase facets, on the catalyst's surface. [48] A C C E P T E D M A N U S C R I P T

CO2 photoreduction results
The reaction was attempted twice as Run 1 and Run 2 using 1.66 equivalents of PPG (Fig. 10). For Run 1 and 2 using 1.66 volume equivalents PPG, similar trends were observed for the production of CO and CH4, with Run 1 producing more CO initially and later matching Run 2's CO production. This was confirmed with paired two sample t-tests, where if the calculated t-statistic is greater than the critical value (cv) of the student's tdistribution table, using the degrees of freedom (df) and significance level (α), the means are significantly different. Run 1 and 2 yielded statistically equal CH4 production over the entire time range (α = 0.05, df = 142, t-statistic = -0.5735, cv = 2.576). Run 1 and 2 yielded statistically equal CO production from 204 minutes (α = 0.05, df = 40, t-statistic = 0.9016, cv = 2.704). The coated meshes prepared by the sol-gel method were compared to a mesh coated with reference P25 TiO2. For both CO and CH4 production, P25 TiO2 coated mesh performed better (Fig. 11). The difference in performance could be due to the differences in crystallinity and band gap ( Table 4). Increase in band gap has been reported with decreasing crystal size, where the rutile crystallite size for the 1.66 and 0.83 volume equivalents PPG samples having much larger rutile and slightly bigger anatase crystallite sizes compared to P25. [5,49] The larger band gap and resulting increase in the number of photons generated by P25 would favor CO2 photoreduction. [3] The difference in production could also be due to the difference in coating methods used where the P25 TiO2 was applied to the stainless steel mesh using a slurry coating method versus a dip coating method for the optimised samples. There are examples in the literature that report the influence of carbon impurities introduced during the catalyst's preparation and/or from the coating procedure on the production of both CH4 and CO. [50][51][52][53] Future work would include investigating the effects of the coating procedure on the potential false positive production of CH4 and CO. It is also important to note that this was a benchmarking exercise where the sol-gel process was developed to yield catalyst properties that were benchmarked against P25 TiO2. In addition, although not as efficient as P25, the advantage this sol-gel procedure is its coating flexibility, where it could be potentially applied to different supports on a large scale using a spray paint process. Mesh coated with 1.66 volume equivalents PPG produced more CO and CH4 compared to the mesh coated with 0.83 volume equivalents PPG.
This is most likely due to the synergistic effects of rutile and anatase where an increase in rutile phase exhibits improved photocatalytic activity. [

Conclusions
A systematic approach using a PB design was used to efficiently screen seven sol-gel parameters towards a catalyst coating with a high coverage and useful properties for CO2 photoreduction. It was possible using a DOE methodology to develop a sol-gel procedure that could tune band gap, coverage, crystallinity and anatase:rutile ratio for CO2 photoreduction. Moreover, the PB design yielded similar observations made in several publications in as few as fifteen experiments. The best design point (WT2) was further optimized to use less PPG. The lower activity of the sol-gel prepared catalyst coatings to P25 was likely due to larger crystallite size, lower band gap values and the potential effects of false positives from the different coating procedures used. This PB design and protocols developed could be employed towards a variety of catalyst coatings with a wide range of properties.

Acknowledgments
The authors thank the financial support provided by the Engineering and Physical Sciences Research Council (EP/K021796/1), the Research Centre for Carbon Solutions (RCCS) and the James Watt Scholarship