Skip to content

Physics-informed hierarchal graph neural network with multi-scale feature extraction for bearing fault diagnosis

Oct 2026 · Measurement science and technology · 0 citations

Abstract

In modern industry, bearing fault diagnosis is of crucial significance for ensuring the reliability and safety of rotating machinery. With the widespread deployment of the Industrial Internet of Things (IIoT), massive amounts of sensor data have become available, enabling the extensive application of end-to-end deep learning models in this field. However, the lack of interpretability inherent in such models raises concerns regarding their reliability and constrains the further advancement of intelligent fault diagnosis. Moreover, due to the multi-scale nature of bearing vibration signals, most deep learning methods exhibit limited capability in capturing non-local dependencies across different time segments. To address these challenges, this paper proposes a Physics-Informed Hierarchical Graph Neural Network (PI-HGMN) for bearing fault diagnosis. Specifically, the PI-HGMN constructs a dual-dimensional physics-constrained representation framework that embeds both time-domain bearing dynamics and frequency-domain fault characteristic priors into the deep feature learning process. Furthermore, a hierarchical graph construction strategy with embedded physics-based residuals is meticulously designed to establish a weighted graph structure reflecting the intrinsic physical relationships of the bearing system. Finally, through a multi-scale hierarchical graph convolution network, local fault details and global operational patterns are effectively integrated. Extensive experiments conducted on five publicly available benchmark datasets and one proprietary industrial dataset confirm that the proposed method achieves consistently superior performance under both identical-domain and cross-domain operating conditions.

View source

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.