Remaining Useful Life Prediction Combining Advanced Anomaly Detection and Graph Isomorphic Network

Abstract

Condition monitoring (CM) has garnered extensive attention in the era of Industry 4.0 and digital manufacturing. It is crucial to monitor the health status of machinery to ensure reliability, safety, production quality, and effectiveness. Advanced predictive maintenance strategies increase equipment availability and effectiveness by accurately predicting failures, thus facilitating maintenance engineering decisions and preventing unplanned machinery breakdowns. In this research, a novel predictive maintenance strategy is proposed by integrating anomaly detection and fault prognostics, two significant challenges in smart maintenance, into one CM system. For anomaly detection, we developed an intelligent methodology based on the skip convolution generative adversarial network (SCGAN). This network combines a convolutional autoencoder (CAE), generative adversarial network (GAN), and skip connections, forming a robust system to construct health indicators (HIs), effectively and efficiently tracking the degradation status of rolling element bearings with fault identified using the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$3\sigma $ </tex-math></inline-formula> criterion. Validation on real experimental datasets demonstrates that the developed HIs show a stable trend during the healthy stage and a marked increase when deterioration is detected. Subsequently, we employ an advanced graph isomorphic network (GIN) for remaining useful life (RUL) prediction. GIN utilizes graph data and graph convolutions (GCs) to map complex relationships between degradation evolution and RUL. This approach outperforms existing deep learning models, such as convolutional neural networks (CNNs), long short-term memory (LSTM) networks, and CNN-LSTM, providing more accurate RUL prediction.

Topics & Concepts