66 / 2025-03-30 12:13:25

Large Language Model Driven Fault Diagnosis of Rotating Devices with Cross-Modal Transfer

Rotating machinery, Fault diagnosis, Large language model, Transfer learning, Cross-modal

Rotating machinery is pivotal in the manufacturing industry, underpinning the reliability and stability of production systems. Precise fault diagnosis and condition monitoring of such machinery hold considerable theoretical and practical value. Nevertheless, complex and dynamic working conditions often diminish the accuracy of fault diagnostics, and the acquisition of stable industrial fault data presents significant challenges, leading to pronounced overfitting in diagnostic models. To overcome these obstacles, this study introduces a fault diagnosis approach for rotating machinery that leverages pre-trained large language models with a cross-modal transfer learning framework. The methodology employs a pre-trained language model architecture, incorporating a diagnostic model with GPT-2 as the core component. Cross-modal knowledge transfer is facilitated within specific layers of GPT-2 through transfer learning, thus bolstering the model’s ability to diagnose faults across different modalities. The efficacy of this method is corroborated through its application to rolling bearings, a vital element of rotating machinery, across various operational scenarios.