Dual-Grained Clustering with Concurrent Evaluation of Static and Dynamic Slow Features for Instantaneous Product Quality Assessment

Abstract

Product quality assessment (PQA) aims to detect the quality of products through real-time physical measurement statistics. This problem is essentially a multivariate statistical analysis problem, which is prevalent in complex industrial process. However, there are few researches focusing on this problem. Besides, most existing PQA methods depend on the sense organs of evaluation experts or consumers, which is too subjective and the accuracy of evaluation results is often difficult to guarantee. In this article, we propose an instantaneous PQA method which performs two-step evaluation on dual grains with the concurrent static and dynamic slow features of quality data obtained by slow feature analysis. First step is the coarse-grained evaluation step, which is the analysis with static slow features and then four quality levels can be achieved. For each quality level, to perform finer evaluation, the second step, which is fine-grained evaluation step, analyzes dynamic slow features to get the range of normal fluctuation characteristics. Finally, the online evaluation procedure and judging rules are designed for new products. To evaluate the feasibility of the proposed method, a case which is to assess the quality of a batch of cigarettes produced by an actual cigarette factory is studied and the result conforms to the experience of cigarette experts and demonstrates that the proposed method realizes instantaneous quality assessment.