Structural characterization and catalytic sterilization performance of a TiO2 nano‐photocatalyst

Abstract In view of the food safety and hygiene issues caused by pathogenic microorganisms, tetrabutyl titanate was used as a precursor for the preparation of a TiO2 nano‐semiconductor photocatalyst via the sol–gel process. The plate count method was then adopted to investigate the photocatalytic sterilization performance of the synthesized TiO2 nanoparticles toward Escherichia coli, Staphylococcus aureus, and Candida albicans. Subsequently, a backpropagation (BP) neural network model was developed to predict the photocatalytic sterilization performance. The photocatalyst was structurally characterized by the Brunauer–Emmett–Teller method for specific surface area determination, transmission electron microscopy, X‐ray diffraction, and X‐ray photoelectron spectroscopy. The results indicated that the prepared TiO2 nano‐photocatalyst was of high purity with a specific surface area of 76.5 m2/g and the particle size range 15–18 nm. The nanoparticles exhibited characteristic peaks corresponding to the oxide component Ti–O, hydroxyl group ˙OH and oxygen chemisorbed and presented an anatase‐dominated multiphase structure that enhanced the photocatalytic performance. UV irradiation at 254 nm produced better sterilization effects on E. coli, S. aureus, and C. albicans, with elimination rates after 30 min of reaction of 97.8%, 99.4%, and 93.6%, respectively. These results indicated that the TiO2 nano‐photocatalyst is a promising environmentally friendly catalyst with good sterilization performance. The constructed BP neural network also exhibited high training accuracy and good generalization ability, with correlation coefficients between the network‐predicted and experimental target values of 0.9789. These results support research on the intelligent processing of photocatalytic sterilization with TiO2 nanoparticles.


| INTRODUC TI ON
Foodborne pathogenic diseases have become a major food safety problem and the most widespread hygiene issue globally (Dupont, 2007). Such diseases are mainly caused by bacteria Gabriel, Som, Madkour, Eren, and Tew (2007), the most common being salmonellae, pathogenic Escherichia coli, staphylococci, pathogenic streptococci, Clostridium botulinum, Vibrio parahaemolyticus, Campylobacter jejuni, and shigellae (Teunis, Takumi, & Shinagawa, 2004). In addition, pathogenic bacteria present in air-conditioning systems and other confined spaces (various bacilli, molds, legionellae, pathogenic yeast E. coli, Staphylococcus aureus, and influenza viruses) also pose a direct threat to human health (Xu, She, & Xu 2003;Yan & Ding, 2011). Therefore, much emphasis has been placed on the research of antibacterial technologies and materials with broad-spectrum sterilization effects and good sterilization performances.
Currently, common antibacterial agents can be classified into inorganic and organic agents (Zhang et al., 2016). Organic antibacterial agents, which mainly comprise aldehydes, alcohols, phenols, biguanides, iodophors, and surfactants, are the earliest known antibacterial materials. These agents come in wide variety and have low production costs; however, they possess several disadvantages including poor thermal resistance, high specificity, and a narrow antibacterial activity spectrum (Denis-Rohr, Bastarrachea, & Goddard, 2015;Farah et al., 2015;Kenawy, Al-Deyab, Shaker, El-Sadek, & Khattab, 2009). In addition, their toxicity limits the food safety of organic antibacterial agents and promotes the development of antimicrobial resistance. On the other hand, inorganic antibacterial agents provide a higher level of safety and exhibit better thermal resistance, durability, and sustainability as well as a broad antibacterial activity spectrum. Thus, with the increasing demand for environmental quality, the market prospects of inorganic antibacterial materials are extremely promising (Lemire, Harrison, & Turner, 2013;Qin, Li, & Wang, 2017;Kalyani, Venkatraju, Kollu, Rao, & Pammi, 2015). However, inorganic antibacterial agents mainly comprise silver-based formulations, which release endotoxins during the bacteria elimination process. Hence, they are currently unable to achieve the controllable sustained release of silver ions (Charles and Heinig, 1993;Cai, Hashimoto, Itoh, Kubota, & Fujishima, 1991).
In view of these issues, current research has been devoted toward creating novel antibacterial technologies (e.g., plasma sterilization, high voltage electrostatic water treatment, and photocatalysis) and materials (Xu et al., 2003). Following the report by Fujishima and Honda (1972) that TiO 2 enables sustained redox reactions under photoirradiation in solar photovoltaic cells, research on the photocatalytic activity of TiO 2 has been attracting increasing attention.
Although the photocatalytic performance of TiO 2 nanoparticles is closely related to factors such as the crystal structure and specific surface area, reports on the relationship between the structure and catalytic sterilization performance of TiO 2 nanoparticles are relatively scarce. In addition, as the photocatalytic sterilization effects of TiO 2 are influenced by many factors, substantial debugging is required for the determination of the appropriate sterilization conditions. Thus, to conserve research resources, we have harnessed the intelligent nonlinear simulation advantages of neural networks and adopted them in the sol-gel process (Guan, Yang, & Chen, 2015;Yang, Yu, & Guan, 2018). This simple proposed method provided good uniformity for the synthesis of the TiO 2 nano-photocatalyst.
Subsequently, the Gram-negative (G − ) bacterium E. coli, Grampositive (G + ) bacterium S. aureus, and pathogenic yeast Candida albicans were selected as representative contaminant targets to investigate the crystal structure and photocatalytic sterilization performance of TiO 2 . In addition, a neural network was applied, for the first time, to predict the photocatalytic sterilization performance of TiO 2 and thus, facilitate the rapid determination of optimum sterilization conditions.

| Chemicals
Brilliant green lactose bile broth, lactose bile fermentation broth, Sabouraud liquid medium, and plate count agar were purchased from Guangdong Huankai Microbial Sci. and Tech. Co., Ltd.; E. coli (GIM 1.355), S. aureus (GIM 1.221), and C. albicans (GIM 2.130) were purchased from the Guangdong Microbial Culture Collection Centre; low-sodium salt was purchased from the Guangdong Salt Industry Group Co. Ltd. Anhydrous ethanol, nitric acid, and acetic acid were of AR grade, while tetrabutyl titanate was of CP grade. All the water used in the experiments was fresh ultrapure water.

| Preparation of the TiO 2 Nano-photocatalyst
The TiO 2 nano-photocatalyst was prepared via the sol-gel process (Yang, Huang, Ye, & Chen, 2010) with tetrabutyl titanate (TBT; C 16 H 36 O 4 Ti) as the precursor. The detailed preparation procedure is as follows: 10 ml TBT was mixed with 30 ml anhydrous ethanol.
Next, a solution mixture comprising 50 ml ethanol, 3 ml water, and 1.5 ml nitric acid was added dropwise into the TBT-ethanol solution to obtain a light-yellow gel. The gel was then mixed thoroughly for 12 hr and left to stand at room temperature for 24 hr. Subsequently, the gel sample was heated in a water bath at 80°C to evaporate the solvent. After drying, the gel was ball-milled for 2 hr to obtain a fine powder, which was calcined in a temperature-programmed muffle furnace, under inert N 2 atmosphere, at 450°C for 2 hr. Finally, the calcined powder was milled again to produce the TiO 2 nano-photocatalyst, which was used in the subsequent experiments.

| Brunauer-Emmett-Teller (BET) analysis
BET analysis of the specific surface area (TriStar II 3020, Micromeritics Instrument Co.) was performed at the constant temperature 77 K using liquid nitrogen. Prior to testing, the samples were degassed at 573 K for 3 hr. The adsorbed volumes under different adsorption pressures were then measured, and the specific surface area of each sample was calculated using the BET equation.

| X-ray diffraction (XRD) analysis
XRD (D/MAX IIIA, Rigaku Co.) was performed under the following conditions: room temperature; Cu K α X-ray source, K α radiation generated with a Cu target; tube voltage, 30 kV; tube current, 30 mA; scan range, 10-60°(2θ); and scan speed, 4°/min. The average particle size of the crystallites in the powdered sample was calculated using the Scherrer equation (Equation 1): where β is the full width at half maximum (FWHM) of the strongest diffraction peak of the sample, λ is the X-ray wavelength, and θ is the diffraction angle.

| X-ray photoelectron spectroscopy (XPS) analysis
XPS (VG Multilab 2000, Thermo Fisher Scientific) was next performed to analyze the elemental components and valence states on the catalyst surface. The following parameters were employed: energy resolution, 0.48 eV; imaging spatial resolution, 3 μm; minimum analysis area, 15 μm; X-ray source, Mg K α, hv = 1,253.6 eV; full spectrum range, 0-1,000 eV; and C 1s calibrated binding energy, 284.6 eV.

| Preparation of the microbial suspensions
Based on the requirements of aseptic operations, the starter cultures of E. coli, S. aureus, and C. albicans were separately transferred onto sterile growth media on a clean bench and subsequently cultured in a biochemical incubator at 37°C for 24 hr. The C. albicans culture was cultured at 25°C to enable activation. The activated culture was then inoculated into the nutrient broth and incubated at constant temperature, with shaking, to reach the exponential growth phase. Prior to further use, each of the initially formulated microbial suspensions was subjected to a series of saline dilutions to concentrations in the range 1.0 × 10 5 -1.0 × 10 6 cfu/ml (expected colony count, 30-300 cfu/plate).

| Influence of UV irradiation combined with the TiO 2 nano-photocatalyst (TiO 2 -UV) on the elimination rate
For each of the prepared microbial suspensions, 0.01 g TiO 2 nanoparticles were added to 100 ml suspension. The nanoparticle-containing suspensions were then added to a series of plate cultures using the coating method, and photocatalysis was carried out under UV irradiation at 254 nm using a 16 W UV lamp. After 30 min of reaction, the plates were incubated in a biochemical incubator at 37°C (the C. albicans cultures were incubated at 25°C). The colonies were then counted using the plate count method and the sterilization performances of the proposed TiO 2 -UV method with the three pathogens after 30 min reaction were investigated. Blank (reaction performed in the absence of TiO 2 nanoparticles and UV irradiation) and control (reaction performed in the presence of either TiO 2 nanoparticles or UV irradiation) groups were established for the experiment. In addition, to ensure data reliability, two independent parallel experiments were performed each time and each experiment was performed in triplicate to obtain the average values. The elimination rate of each type of microbe by the TiO 2 nano-photocatalyst was calculated using Equation (2)

| Influence of the TiO 2 concentration on the elimination rate
The sterilization performances of the TiO 2 nano-photocatalyst at different concentrations (0.01, 0.1, and 1.0 g/L) toward the G − and G + bacteria and C. albicans (UV irradiation, 254 nm; reaction time, 30 min) were next investigated. A blank group (reaction performed in the absence of TiO 2 nanoparticles and UV irradiation) was established for this experiment, and the elimination rates were calculated using Equation 2.

| Intelligent optimization of the neural network
Backpropagation (BP) neural network algorithms account for 80%-90% of the total neural network algorithms used by researchers (Guan et al., 2015;Yang et al., 2018). In this study, experimental analysis was combined with a three-layer BP neural network model constructed using MATLAB 7.1. The input layer parameters used in the model were the influencing factors of the experiments (TiO 2 -UV, reaction time, and TiO 2 concentration), while the output layer parameter was the elimination rate; the number of hidden layer parameters, as determined by simulation, was eight.

| BET analysis
The BET results revealed that the specific surface area of the prepared TiO 2 nano-photocatalyst (76.5 m 2 /g) was higher than that of commercial P25 TiO 2 nanoparticles prepared via the gas-phase method (50 ± 15 m 2 /g) (Wen, Yi, Pei, & Xu, 2017). This marked increase implied enhanced catalytic activity of the prepared catalyst over that exhibited by P25 (Mulakov et al., 2020;Yang et al., 2010). Figure 1 illustrates the transmission electron micrograph of the TiO 2 nanoparticles prepared by calcination at 450°C. The TiO 2 nanoparticles displayed a relatively uniform particle size in the range 15-18 nm, which was smaller than that of the P25 TiO 2 nanoparticles. These results are consistent with the BET results and reveal that the TiO 2 nanoparticles formulated in this study had higher specific surface areas than those of the P25 TiO 2 nanoparticles.  & Lei, 2006) have revealed that anatase TiO 2 has a higher catalytic activity than rutile or brookite TiO 2 , while TiO 2 with a mixed crystal structure provides better catalytic effects than TiO 2 with a single crystal structure. In addition, the average TiO 2 particle size in the sample, preliminary calculated using Equation 1, was 16.9 nm, which is consistent with the TEM results.   (Cao, Li, Zhang, & Wang, 2010  thereby leading to bacterial and fungal deactivation and even death (Tian et al., 2009). In addition, XPS analysis revealed that the TiO 2 samples prepared in this study contained ˙OH radicals with activation energies ≤402.8 MJ/mol. This enabled the rapid and effective decomposition of the organic cellular components and organic nutrients (necessary for growth and reproduction), thereby causing protein denaturation and ultimately death. and 97%, respectively. Under sole UV irradiation, without the presence of the photocatalyst, the elimination rates of both bacteria only exceeded 90% after 30 min and gradually leveled off at 100% after 50 min reaction. For C. albicans, the elimination rate only exceeded 90% after 40 min and eventually leveled off at ~93%. Based on the XRD data, the particle size of the prepared TiO 2 nano-photocatalyst was 16.9 nm. These ultrafine TiO 2 nanoparticles could be rapidly and completely adsorbed onto the pathogen surfaces, thereby enabling rapid photocatalytic activity upon UV irradiation. In this case, the generated e − -h + pairs and ˙OH radicals could penetrate and damage structures such as the cell walls within the pathogen through oxidation, thereby achieving good sterilization effects in a relatively short time. pairs and ˙OH radicals) upon UV irradiation. In addition, these sites were not effectively utilized, thereby leading to lower elimination rates. These results suggested that a higher TiO 2 concentration would produce more active oxidants and thus, enhanced elimination rates. In fact, the increase in TiO 2 concentration, from 0.01 to 0.1 g/L, led to a significantly positive correlation between the elimination rate and TiO 2 concentration, and the elimination rates of the G − and G + bacteria and C. albicans increased to 97.8%, 99.4%, and 93.6%, respectively. However, when the TiO 2 concentration was further increased to 1.0 g/L, the elimination rates of the three types of representative microbes merely increased by 0.8%, 0.2%, and 0.5%, respectively, when compared to the elimination rates achieved with a TiO 2 concentration of 0.1 g/L. An excessively high

| Intelligent optimization of the neural network
Using E. coli as the microbial species, a BP neural network for photocatalytic sterilization by TiO 2 nanoparticles was next constructed ( Figure 8). The training accuracy of the model was set at 0.001 and the functions used in the model were as follows: transfer function of the output layer: pureline; transfer function of the hidden layer: tansig; training function: trainbfg; learning function: learngd; performance function: mse. In addition, to reduce the differences between the different parameters and enhance the operation accuracy of the network, the mapminmax function in the Neural Network Toolbox in MATLAB was used for the standardization of the experimental data.
In this study, 17 sets of experimental data were randomly allocated to the training (Kalyani et al., 2015) and prediction (five sets of data) datasets. Figure 9 illustrates the training processes and results when the BP neural network was run. It reveals that the constructed BP neural network converged rapidly with the specified training accuracy limit and the system output values were close to the experimental values, with all errors being within ±5%. This revealed that the constructed neural network provided a good approximation of the variation patterns of the training dataset and could reflect the sterilization performances of the TiO 2 nano-photocatalyst toward E. coli under different parameter conditions.
To evaluate the generalization ability of the network, five sets of prediction data were used as the test data ( Figure 10). The results indicated a close match between the network-predicted and actual values with relatively small errors, demonstrating that the network had a good generalization ability and could provide good estimates from untrained data. Therefore, we concluded that the BP neural network constructed in this study can be used to predict the oxidative sterilization effects of the TiO 2 nano-photocatalyst on E. coli under different parameters. This contributes to resource and cost savings and thus increases the potential of this method for real application.

| CON CLUS IONS
In this study, the sol-gel method was adopted to prepare a TiO 2 semiconductor photocatalyst. When the prepared catalyst was calcined at 450°C, nano-sized particles with an anatase-dominated mixed crystal structure and good photocatalytic effects under UV irradiation were obtained. Subsequently, the sterilization performances toward the pathogenic bacteria E. coli and S. aureus and the pathogenic yeast C. albicans under the influence of different factors (TiO 2 -UV, reaction time, and TiO 2 concentration) were investigated. Under UV irradiation combined with 0.1 g/L TiO 2 , the elimination rates of E. coli, S. aureus, and C. albicans after 30 min reaction reached 97.8%, 99.4%, and 93.6%, respectively. However, after 40 min reaction, the elimination rates of the two bacterial species stabilized at 100%, while the elimination rate of C. albicans stabilized at ~97%. The prepared nano-photocatalyst is non-toxic, does not generate secondary pollution, and produces good sterilization effects toward the representative G − and G + bacteria and pathogenic yeasts. Thus, this proposed technology shows great potential for application and popularization in the field of food hygiene. A BP neural network was employed to achieve both energy and resource savings. The sample data training and prediction results indicated that the network presented a good generalization ability, with a correlation coefficient of 0.9789 between the network-predicted and experimental values. This demonstrates the ability of the network to well predict the photocatalytic sterilization effects of the TiO 2 nanoparticles under different parameter conditions. Therefore, this model is of great engineering significance and shows great potential for further application and popularization in the relevant fields.

ACK N OWLED G M ENTS
This work was financially supported by Guangdong Provincial Key Laboratory of Atmospheric Environment and Pollution Control, Department of Education of Guangdong Province, Department of Science and Technology of Yangjiang and Yangjiang Polytechnic.

CO N FLI C T O F I NTE R E S T
The authors declare that they do not have any conflict of interest.

E TH I C A L A PPROVA L
This study does not involve any human or animal testing.