In the age of artificial intelligence, when computers seem to be able to do everything, should the material research conducted every day also change dramatically? Should the research methods we've been using be changed with the help of powerful computers? These problems are worthy of our consideration. Material research involves many influencing factors, such as material system itself and material preparation process, which is a typical multi-factor system problem. Unfortunately, it is difficult for us to study the non-linear multi-factor complex objects immediately. The univariization method that each factor is studied one by one has been used. Such research results have great limitations. For example, the interaction between factors cannot be revealed. Orthogonal experiments and uniform experimental design are important methods for conducting efficient multi-factor studies. However, the limitations of later data processing of this method cannot meet the requirements of high-level material research. This paper discusses the problem of multi-factor research methods. On the basis of multi-factor uniform experimental design, the importance design methods of additional experimental are proposed. The steps and effects of this method is illustrate with a tow-dimensional sample.
First, a uniform experimental design was used to perform the first round studies with the multi-factors adjusted simultaneously. Multivariate cubic spline method was used to process the obtained multivariate data for obtaining the multivariate function. Then, the multivariate function is used as the weight function to conduct the importance experiment design for the second round of additional experiment design. Importance sampling can achieve more efficient experimental design based on the experimental data of the first round uniform sampling. Finally, multivariate nonlinear mathematical models can be obtained by using artificial neural network for all multidimensional experimental data. With the obtained model, factor optimization and properties prediction can be performed, and also the influence of any single factor can be discussed by fixing the other factors. This method can obtain the relationship model for multifactorial system including interactive influence with less experimental amount. The model can be obtained directly by multi-factor researches, rather than integrating from the univariate results.
The powerful mathematical tools MATLAB is suggested to obtain the multivariate nonlinear mathematical models, solving the weight function problem of importance sampling and the final multivariate data processing problem. A new way for improving the efficiency and level of multi-factor material research is provided.