An Inference Spatiotemporal Machinery Prognostics Approach Based on Graph Learning

Abstract

Machinery prognostics facilitates predictive maintenance, minimizing downtime and operational expenses. Nonetheless, challenges persist due to low signal-to-noise ratio and non-stationary signals. Spatiotemporal feature extraction through recurrent and convolutional neural networks has shown promise in addressing these challenges. Nevertheless, the traditional convolutional learning algorithm, which is based on Euclidean distances between the learned features, can increase the model uncertainty. Moreover, traditional feature fusion techniques can weaken the model's performance. This study proposes a novel inferential spatiotemporal approach. Two independent networks based on long short-term memory and a graph convolutional network are designed to extract the influential spatiotemporal features. Then an adaptive neurofuzzy inferential network is introduced to calculate the remaining useful life based on the extracted spatiotemporal features. Experimental validation using a benchmark bearing dataset under various operational conditions demonstrates that the proposed approach outperforms existing state-of-the-art methods by 59.34 %.