| Summary: | 碩士 === 大同大學 === 化學工程研究所 === 91 === Application of classical experiment design methods to find optimal operating conditions for a process, the testing regions of the control factors are usually located in limited spans. However, if the testing regions are broad and the process is nonlinear, how to design the experiments efficiently and build a representative model based on the experiment data is the main objective of this work. The uniform design method can design a total of q experiments for a process with n factors and q levels in each factor. If the cost of each experiment is high, low partitioned levels are usually proposed first to do the experiments. However, if a reliable artificial neural network (ANN) model cannot be built based on the obtained experimental data, we developed a sequential pseudo-uniform design method to locate additional experiments in the experimental region. Furthermore, information analysis was used to confirm the identified ANN model is consistent with the process. Applications of the proposed method to a benchmark problem, namely minimization of the modified Himmelblau function, a penicillin fermentation process and a series reaction in batch reactor to obtain an optimal operating trajectory were examined. The results prove that the product and process development based on the proposed method requires a reasonable number of experiments.
|
|---|
