Answers are grounded in this paper and cite their sources.
In modern industry, bearing fault diagnosis is important for maintaining the reliability and safety of rotating machinery. With the widespread deployment of the industrial internet of things, large volumes of sensor data have become available, supporting the application of end-to-end deep learning models in this field. The limited physical interpretability of purely data-driven models can constrain their reliability under changing operating conditions. The multi-scale nature of bearing vibration signals also makes it difficult for conventional deep learning methods to capture non-local dependencies across different time segments. This paper develops a physics-informed hierarchical graph neural network (PI-HGMN) for bearing fault diagnosis. PI-HGMN constructs a dual-dimensional physics-constrained representation framework that embeds time-domain bearing-dynamics priors and frequency-domain fault characteristic-frequency priors into feature learning. A hierarchical graph construction strategy with edge-level physical residuals is designed to form a weighted graph whose relative connectivity reflects both feature similarity and physical consistency. A multi-scale hierarchical graph network then integrates local fault details with broader operational patterns. Experiments on five public benchmark datasets and one industrial dataset show competitive and stable diagnostic performance under identical-condition, cross-condition, and noisy settings. The physical residual is evaluated with canonical normalized coefficients on a latent feature grid and is interpreted as a dynamic-consistency regularizer rather than as a direct estimator of dimensional bearing parameters.
No takes yet. Share an insight, caveat, or question.
Jia et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: