Abstract:

Continuous mass production not only requires enterprises to identify the state of the production process more accurately and efficiently， but also needs to identify potential quality problems in the production process， so as to avoid quality risks in advance. The existing control chart abnormal pattern recognition method can monitor whether the production process is abnormal in the short term， but it is difficult to identify and analyze the potential quality problems in the long-term production process. In order to overcome this limitation， this study proposes a method， using relative entropy to identify the quality problems through distribution pattern. Firstly， five training sets and test sets of distribution anomaly patterns are generated by simulation. Then， the probability distribution of the mass characteristic values of the samples under the test set or the probability density after fitting by kernel density estimation is used as the input of the model. Finally， the relative entropy is used to quantify the similarity and divergences between the distribution of the actual production sample and each distribution of the estimated abnormal pattern， the production state classification index is output， and further pattern recognition is completed by constructing two quality pattern judgement criteria. It is shown that the proposed method in this paper can accurately identify the abnormal quality patterns in both the discrete and continuous states of the quality characteristic parameters. Through comparative analysis with the correlation method， it is found that our method has a higher classification accuracy， and thus it can recognition the quality abnormal pattern of the process more effectively.

Key words: relative entropy, kernel density estimation, quality abnormal pattern recognition, quality control