Research On Data Dimensionality Reduction And Entropy Analysis For Process Quality Control

Process Quality Control(PQC) can effectively monitor the formation of product quality in the productive phase of the product. The measurement of the quality of the process mainly includes two aspects: first, whether the process quality is stable; second, weather the stability of the process capability meets the technical requirements. Statistical process control(SPC) can be used to judge the stability of process quality; Process capability analysis(PCA) is used to evaluate the actual processing capacity of the productive process. With the development of intelligent manufacturing, a large number of quality related data can be collected on process, assembly, testing, inspection and other aspects, of which Data Mining can make full use. However, the high data dimension will increase the complexity of data mining. As for PCA, it is assumed that the process parameters follow uniform distribution, which is not always satisfied in actual production, thus the process capability analysis cannot reflect the actual ability of the process.In order to apply data mining to SPC more efficiently and assess production process capability more accurately, it is necessary to effectively control the dimension of quality related data, and to solve the hypothetical problem of the process capability analysis. So this paper mainly focuses on these two issues:(1)Aiming at the problem of high dimension of data, feature extraction method is applied to reduce data dimension. Based on principal component analysis(PCA), the sliding window model and outlier detection are introduced to incrementally update the new arriving data and remove outliers, which can improve the efficiency of principal component analysis and reduce influence of outliers on the dimensionality reduction of data. Thus the dimension of quality parameters could be reduced quickly and accurately. The effectiveness of the proposed algorithm is verified by simulation analysis.(2)To overcome the shortcomings of the traditional process capability analysis that the process distribution is uniform, the entropy method is introduced to detect the changes of data distribution. The sample entropy method could only detect the change of time series variance before, whereas in this paper the input is modified such that it cannot only detect the variance change of the time series data, but also can detect the change of the mean value. The proposed algorithm is validated by simulation analysis; the accuracy of the proposed algorithm is proved through simulation. At the end the percentile of sample entropy ratio under varying variance and mean in separate data segments is calculated, which provides a reliable reference for the identification of the change in the distribution of productive process.On the basis of research on process quality control, this paper has proposed a solution to the problem that exists in the application of data mining in SPC and in that of process capability analysis in practical use. The solution is then verified by simulation. Therefore, this paper is of certain value as a reference for the application of data mining, big data and other information technology in the process quality control of production.