Performance Analysis of Different Graphs in Graph-Based Remaining Useful Life Prediction Model

Condition-based maintenance has progressed substantially through the integration of IoT and advanced sensing technologies. Prognostics and Health Management (PHM) has emerged as a pivotal strategy for enhancing reliability and minimizing downtime by facilitating early fault detection and optimizing maintenance planning. Predicting the Remaining Useful Life (RUL) is a vital component of prognostics and health management for machinery. It effectively prevents catastrophic failures and reduces the costs associated with unplanned maintenance. Numerous Deep Learning (DL) methodologies have been applied for predicting RUL owing to the versatility of their architectures. This paper investigates the use of Graph Neural Networks (GNNs) for RUL estimation. GNNs are adept at modeling complex relationships within equipment data by representing components and their connections as nodes and edges in a graph, thereby capturing intricate dependencies and operational patterns. Building on recent advancements, this study conducts a comparative analysis of various methods for constructing adjacency matrices used in GNNs, crucial for accurate RUL estimation. The study leverages the Graph Attention Model (GAT) to assess the effectiveness of different graph structures in predicting RUL, especially under conditions of machinery degradation. Our findings emphasize the importance of edge quality and quantity in determining the performance of GAT models. This paper concludes with an in-depth discussion on the impact of various graph construction methods on RUL estimation, providing insights for future research and practical applications in industrial settings.