FaultFace: Deep Convolutional Generative Adversarial Network (DCGAN) based Ball-Bearing Failure Detection Method

Failure detection is a useful tool in the industry to improve system performance and reduce costs due to unexpected malfunction events. However, it is usual that there are few datasets about nominal or failure operating behavior of a system, which difficult the correct training and validation of automated failure detection methods. This paper proposes a methodology to perform failure detection on Ball-Bearing joints for rotational shafts using deep learning techniques called FaultFace. The vibration time series of the motor shaft is acquired as an indicative signal and is transformed into a 2D image representation called face portrait using Continuous Wavelet Transformation. Considering that the dataset is reduced and unbalanced, a Deep Convolutional Generative Adversarial Network is employed to produce new face portraits of the nominal and failure behaviors to obtain a balanced dataset. Based on the newly generated dataset, a Convolutional Neural Network is trained for fault detection. The confusion matrix is employed as a performance index for the classification algorithm. Obtained results show that this methodology reaches a 98% of accuracy for failure detection on the ball-bearing joint.