Novel ACGAN-Based Framework Fault Detection Technique for Unbalanced Data

Fault detection is of great significance in ensuring the operational safety of industrial equipment. However, in practice, it is difficult to obtain fault data because industrial equipment is in normal operation most of the time, and at the same time, the huge difference in the probability of occurrence of different faults leads to the problem of imbalance of fault data. These problems will affect the generalization ability and robustness of deep learning models. In order to solve the above problems, this paper proposes a multi-generators based ACGAN network. Firstly, the number of generators is determined according to the number of fault categories, to ensure that each generator can generate a single type of fault data. On this basis, this paper proposes the concept of "gene" to represent the category fault information, and uses the category genes to modify the noise input to get the noise with different mean and variance, so that it contains more effective features. Finally, the data imbalance fault detection method proposed in this paper is applied to the fault data set of rail vehicle door system for experimental verification and analysis. The experimental results show that the method in this paper can effectively improve the performance of fault detection and achieve the expected goal.