Machine Health Assessment Based on an Anomaly Indicator Using a Generative Adversarial Network

In prognostics and health management, the absence of fault data is a challenge that hinders practical applications in the field. When an absence occurs, the only option is to build a proper health indicator for anomaly detection. While there have been numerous traditional approaches toward this end, they have had drawbacks in one way or another. In this study, a new approach is proposed to develop an anomaly indicator that overcomes previous limitations by using a generative adversarial network (GAN). GANs have recently drawn attention as a means to generate virtual samples resembling the original distribution. Two examples—the bearing and train door system—are considered to examine the approach’s capabilities. The data acquired for the normal condition are used to train the GAN, the health is monitored over time using the trained GAN indicator, and the anomaly is successfully detected by identifying a decrease at a point in time.