Industrial Fault Detection Based on C-Vine Copula Model and Transfer Learning Strategy

Abstract

Fault detection is of great significance for industrial processes as it ensures the stable operation of systems and the safety of personnel. However, factors such as equipment aging and environmental changes often cause data deviations in industrial data that cannot be accurately detected by ordinary models. The copula function can clearly describe the relationship between random variables and has a simple structure that is suitable for transferring knowledge. Therefore, this paper proposes a transfer learning method based on the C-vine copula. The method first determines the structure and parameters of the C-vine copula based on data from the source domain, and then fine-tunes with a small amount of data from the target domain. Experimental results show that the proposed model has higher detection accuracy and can express the relationship between variables more clearly than machine learning and deep transfer models.