A hybrid of FEM simulations and generative adversarial networks to classify faults in rotor-bearing systems

Abstract

Condition monitoring of rotor-bearing systems using artificial intelligence has great significance to guarantee the reliability and security of mechanical systems. However, in engineering applications, AI model will fail to classify faults with insufficient fault samples owing to complex working condition. A hybrid fault classification approach is presented by combining finite element method (FEM) with generative adversarial networks (GANs) for rotor-bearing systems. Firstly, FEM simulations are employed to calculate simulation fault samples as additional sources of missing fault samples. Secondly, GANs is used to acquire abundant synthetic samples generated from the simulation and measurement samples, which aims to expand fault samples. Finally, the complete fault samples, including simulation, measurement and their corresponding synthetic samples, are utilized as training samples to train typical classifiers, and further to identify unknown faults. High classification accuracies for a rotor-bearing system using different kinds of artificial intelligent (AI) models are obtained, which demonstrates the effective of proposed method. It is noticed that the present idea can be guided to solve insufficient fault samples problem in more complex mechanical system with agreeable fault classification accuracy.