Oversampling adversarial network for class-imbalanced fault diagnosis

The collected data from industrial machines are often imbalanced, which poses a negative effect on learning algorithms. However, this problem becomes more challenging for mixed type of or while there is overlapping between classes. Class- imbalance requires robust system can timely predict and classify the data. We propose new adversarial network simultaneous classification fault detection. In particular, we restore balance in imbalanced dataset by generating faulty samples proposed mixture distribution. designed discriminator our model to handle generated prevent outlier overfitting. empirically demonstrate that; (i) trained with generator generates normal distribution be considered as detector; (ii), quality outperforms other synthetic resampling techniques. Experimental results show that performs well when comparing diagnosis methods across several evaluation metrics; coalescing generative (GAN) feature matching function effective at recognizing samples.