Eventos Anais de eventos
COBEM 2023
27th International Congress of Mechanical Engineering
GRAPH NEURAL NETWORK APPLIED TO BEARING FAULT DIAGNOSIS
Submission Author:
Alan Lopes , PR , Brazil
Co-Authors:
Luiza Scapinello Aquino, Alan Lopes, Laio Oriel Seman, Viviana Mariani, Leandro dos Santos Coelho
Presenter: Alan Lopes
doi://10.26678/ABCM.COBEM2023.COB2023-1574
Abstract
Bearing fault diagnosis is a critical task in industrial maintenance to ensure the reliable and efficient operation of machinery. With the advent of graph neural networks (GNNs), there has been increasing interest in leveraging the power of graph-based representations for bearing fault diagnosis. Bearing fault diagnosis is an essential task in the maintenance of rotating machinery, and machine learning and deep learning approaches have shown promising results in this area. In this paper, the Case Western Reserve University (CWRU) bearing dataset for benchmarking bearing fault diagnosis using GNNs is presented. The dataset consists of vibration signals collected from four bearings under different operating conditions, including normal, inner race fault, outer race fault, and roller fault. The signals are acquired from accelerometers placed on the bearing housings, and they are sampled at a frequency of 12 kHz. The dataset also includes labels indicating the health condition of each bearing at different time steps, allowing for supervised training of GNN models. A graph-based representation of the bearing fault data is proposed, where each bearing is represented as a node in a graph, and the vibration signals are used to define edge weights between nodes. This representation captures the spatial and temporal dependencies among the bearings and their fault conditions, enabling GNNs to learn meaningful features for fault diagnosis. To benchmark the performance of GNNs for bearing fault diagnosis, an extensive experiment on the CWRU bearing dataset was conducted. The experimental results demonstrate the effectiveness of GNNs for bearing fault diagnosis, achieving state-of-the-art performance on the CWRU bearing dataset. The benchmarking results provide valuable insights into the capabilities of GNNs for bearing fault diagnosis and can serve as a reference for researchers and practitioners in the field of industrial maintenance. The CWRU bearing dataset and the benchmarking results contribute to the advancement in accurately diagnosing bearing faults using the proposed graph-based representation of the data, with the potential to impact real-world industrial applications by enabling more accurate and efficient bearing fault diagnosis systems.
Keywords
Graph Neural Network, Bearing Fault Diagnosis, Case Western Reserve University
