Spatial–Temporal Predictive Maintenance Using GNN–TCN Hybrid Learning for Industrial Bearing Fault Diagnosis
Abstract
Industry 4.0 is the complete automation based on digital technologies, where the Industry Internet of Things (IIoT) pays the major part. One of the finest applications of IIoT in the Industrial operation is predictive maintenance, where it helps to obtain the fault conditions of the machines and to determine the proactive actions that actually save time and cost of the production processes. In predictive maintenance, various sensors are used where their relationship actually helps in making accurate decisions. On the other hand, the temporal degradation factor of the machine needs to be analyzed to make the predictions on time for its corrective measures on maintenance activity. Hence, the proposed work uses the Graph Neural Network for Spatial relationship learning between the sensor modalities and the Temporal Convolution Network (TCN) for the temporal progression degradation factor in the machines. This helps in attaining spatial and temporal learning to take proactive actions in the maintenance of the machines. The proposed work attains 96.86% accuracy with an F1 score of 95.9% using the Paderborn University Bearing Dataset.