Study On Critical To Quality Characteristic Identification For Complex Products Based On Feature Selection

Complex products contain a large number of quality characteristics(QCs),and identifying critical to quality characteristics(CTQs)related to product quality is a necessary step for product quality improvement.In manufacturing processes,the measurements of QCs are collected from production lines,which provide the fundament for data analysis.This study establishes CTQ selection methods based on feature selection algorithms according to different features(high dimensionality,data imbalance and QCs are time-ordered)of manufacturing data from complex products.First,two CTQ identification methods with strong dimensionality reduction capabilities are established for balanced data.They are CTQ identification methods based on improved ReliefF and CTQ identification method based on genetic simulated annealing algorithm.In the first method,a novel method deciding the number of CTQs to be selected is developed.The experimental result shows that the improved ReliefF outperforms ReliefF in eliminating noisy or redundant QCs.In the second method,an integrated fitness function combining two measures,QC importance and the number of QCs,is established for feature optimization.The experimental results show the fitness function improves the dimensionality reduction capability of the proposed method.Second,a two-phase CTQ identification framework is established for imbalanced data.In the first phase,a set of candidate CTQ sets is identified by a multi-objective optimization method.In the second phase,the ideal point method(IPM)is adopted to select the best compromise solutions(CTQ sets)from the candidate solutions.Based on the two-phase framework,two identification methods based on multi-objective methods,improved NSGA-II and improved DMS,are developed respectively.In addition,the two methods consider two different types of QC importance measure for imbalanced data,respectively.The experimental results show the two methods perform efficiently in CTQ identification for imbalanced data.Finally,as the QCs corresponding to different manufacturing processes,they are time ordered.Thus,a two-phase CTQ selection method is proposed for imbalanced data,considering the order of QCs.In the first phase,a hybrid multi-objective method combining genetic algorithm and DMS is used to select candidate CTQ sets.In the second phase,the best compromise solutions are selected from the candidate CTQ sets,where the order of QCs is considered to select the QCs in earlier processes.The experimental results show the effectiveness of this method.