Study On Product CTQ Identification Based On Feature Selection

Critical-to-Quality Characteristics (CTQ), which can substantiate the demandsof customers, is defined as quality characteristics which are both highly related tocustomer demands and of good capability to be measured.The purpose of the thesis is to handle the problem of CTQ identification inhigh-dimensional manufacturing processes. High-dimensional processes usuallyhave an extremely large number of characteristics, and there are always muchcomplicated correlations among them. Therefore, it is of great difficulty toaccomplish the CTQ identification task only in product design. As a result, theauthor tries to make use of some historical data of manufacturing process, to identifyCTQs by the means of feature selection methodology, which has been thoroughlystudied in the field of computer science, and integrates the classifier buildingalgorithms and validation algorithms to score the identification results.Additionally, in order to identify the CTQs in industrial complicated processes,the author proposes a new hybrid method to combine the advantages of Partial LeastSquares Regression and wrapper feature selection. To solve the problem of dataimbalance, the author proposes another hybrid feature selection method based onG-Mean index. The new methods are evaluated from the two different perspectives:capability of dimension reduction, and the classification performance, and theexperiment achieves satisfactory results, which can further prove that the featureselection techniques, after proper modification, can be used in the area of qualityengineering.