GAN-AE BASED FAULT DIAGNOSIS METHOD FOR ROTOR MACHINE

Abstract

The rotor fault diagnosis plays an important role in rotating machinery system by early detection and avoiding dangerous situations of rotating ma chinery system. Accurate fault diagnosis of rotor machine ensures reliability and security of rotor mechanical systems. There are many existing methods for the rotor fault diagnosis, in these methods have a problem of data defi ciency. To solve the problem of data deficiency and improving efficiency of model, introduced a method is called GAN-AE based rotor fault diagnosis. The data deficiency problem of existing methods are solved by using a Gen erative Adversarial Network (GAN) model, GAN model generating a series of new synthetic samples from the original data samples and they are similar to the original data sample and it aim to expand the original raw sample . The generated synthetic samples are utilized as the training samples to train the classifier and to identify the unknown faults. The GAN generated signal combined to the original dataset and then it given to the Auto Encoder (AE) model. These complete data is given to the auto encoder model and it extracts the signal features and it provides better accuracy than the Auto Encoder model. In this work, GAN model is combining with AE model for the rotor fault diagnosis. GAN-AE method solve insufficient fault samples problem in more complex mechanical system with agreeable fault classification accuracy. The GAN-AE method can offer better capability of extracting features, and the accuracy of fault diagnosis of rotating machines. MAFAULDA database is used as the base experimental dataset. Experimental results and analysis show that this GAN-AE based rotor fault diagnosis has better performance