Artificial neural network (ANN) was developed to predict the morphology of TiO2 nanotube prepared by anodization. The collected experimental data was simplified in an innovative approach and used as training and validation data, and the morphology of TiO2 nanotube was considered as three parameters including the degree of order, diameter and length. Applying radial basis function neural network to predict TiO2 nanotube degree of order and back propagation artificial neural network to predict the nanotube diameter and length were emphasized in this paper. Some important problems such as the selection of training data, the structure and parameters of the networks were discussed in detail. It was proved in this paper that ANN technique was effective in the prediction work of TiO2nanotube fabrication process.
Titanium dioxide (TiO2) nanotube has attracted great attention in recent years owing to its remarkable potential applications in the areas of electronics, gas-sensing materials, optics, and biotemplating (Mor et al., 2006 and Varghese et al., 2003). Some experiments indicated that the geometrical feature of the nanotube affects its photocatalysis and sensing properties impressively which are crucial for applications (Valota et al., 2009 and Vega et al., 2007). DC current electrochemical anodization is widely used to fabricate highly ordered TiO2 nanotube (Mor et al., 2006 and Varghese et al., 2003). In order to optimize its properties in applications, a lot of theoretical and experimental works have been carried out to investigate the relation between anodization conditions and geometrical parameters of the nanotube. Works focusing on the mechanistic aspects successfully explained the formation process of the tube in general, but failed in predicting the geometrical parameters of the tube at a certain anodization condition precisely (Macak et al., 2007 and Yasuda et al., 2007). The influence of some important anodic conditions on the structure and morphology of TiO2 nanotube was investigated through experiment (Oh et al., 2008 and Prakasam et al., 2007), but the result was unsatisfied and the process was both costly and time-consuming. Till now there are still lack of descriptions in detail about the relationship between anodization parameters and morphology of TiO2 nanotube.
Artificial neural network (ANN), a powerful tool for modeling complex processes, is finding growing acceptance in the field of aerospace, automotive, electronic, manufacturing, robotics, telecommunication, etc. (Guessasma and Coddet, 2004 and Parthiban et al., 2007). ANN method can reveal certain correlations based on analyzing experimental data in biological mode without knowing their underlying physical mechanism. It can handle non-linearity, imprecise and fuzzy information and simulate the input/output mapping at any accuracy in theory (Zhang, Yang, & Evans, 2008). Most importantly, ANN method can provide users the prediction power but not being confined to input/output fitting. Now it is almost a standard modeling technique based on statistical approach. ANN has been successfully applied to simulate the structures (Khanmohammadi, Garmarudi, Khoddami, Shabani, & Khanlari, 2010) and physical properties (Dutta et al., 2010 and Kandjani et al., 2010), such as adsorption efficiency, photocatalytic efficiency, for various kinds of nanostructures. As to the formation process of self-organized TiO2 nanotube, which contains plenty of experimental data and lacks precise physical model, ANN method is considered to be suitable for the prediction work.
In this paper, two different ANN models was innovatively employed to the prediction work of TiO2 nanotube preparation process with the aim for architecting suitable ANNs to simulate morphology of TiO2 nanotube under a certain fabricating condition. Some data simplification and network training process were discussed in detail. Finally, optimized network structures and mature networks with satisfied prediction accuracy were obtained, and some further usages were also discussed.
The ANN model was used to predict the morphology of TiO2 nanotube fabricated by anodization. Based on the experimental data, BPANN and RBF neural network were successfully applied to the prediction on diameter, length and distribution of TiO2 nanotube. The network parameters were optimized step by step and finally 3 neural networks with satisfied prediction accuracy were obtained, in which the percentage of significative nanotube distribution predictive values reached 93.48% and the R value of the testing data reached 0.9806 and 0.9997. Meanwhile, some important and innovative works were carried out during this attempt. With regard to the data processing, an innovative approach to simplify the complex experimental data used for ANN input/output was explored and five degrees of order were introduced to describe the nanotube distribution. Further more, a complete and effective approach of applying BPANN for prediction was carried out, through which the nanotube diameter and length were successfully predicted. It was also proved in this study that ANN could be employed in the prediction work of TiO2 nanotube fabrication process. More in-depth works on this field could be done based on the present paper. As a powerful mathematical tool, ANN would be quite promising for other application in the future.
