Some problems and solutions of fuzzy clustering based data-driven fault diagnosis techniques in practice

This article applies existing fuzzy clustering based fault diagnosis methods to a CSTR process. The following problems are observed: (1) The value of memberships of online data to known patterns are so small that it's hard to compare them;(2) because the data is dynamic, the range of samples needs to be determined when PCA is applied to extracting features; (3) when judging an unknown fault, we need to consider whether it is a new fault or a known fault with increasing intensity. The reasons of these questions are studied from the perspectives of theory and practical application and possible solutions are proposed. The range of samples is determined in PCA, the problem of small values of membership is solved by optimizing the diagnosis process, and the idea of using fault vectors is introduced to improve the accuracy when identifying an unknown fault. Finally, we demonstrate the effectiveness of our proposed method through simulation experiments and verify it can produce better fault diagnosis results.