Research on General Foundation Model for Intelligent Fault Diagnosis for Rotating Machinery

Given that existing intelligent fault diagnosis methods for rotating machinery often lack generalizability and are typically limited to specific models, structures, operating conditions, measurement points, and load scenarios, a universal fundamental model for intelligent fault diagnosis tailored to rotating machinery is developed. By mining massive volumes of state monitoring data from various types of rotating machinery, a multi-source data structure with a multi-scale adaptive alignment method is proposed. A multi-level state fusion intelligent diagnosis model is constructed, and a universal fundamental model with strong applicability to typical rotating machinery is established. Additionally, a method for individualized customization and adaptation of the diagnosis model is introduced. The proposed method is validated on extensive state monitoring datasets for rotating machinery. Experimental results show that the universal intelligent diagnosis model can directly detect anomalies in unknown measured equipment, achieving an overall diagnosis accuracy of 88.5% without any supervised fine-tuning. With minor fine-tuning using a small amount of measured data, the model rapidly adapts to new equipment and achieves a diagnosis accuracy of up to 98.6%. Furthermore, the proposed data preprocessing method enables cross-equipment signal amplitude normalization while preserving the relative amplitude distribution between healthy and faulty states within the same equipment, effectively retaining key amplitude-based feature differences. These findings demonstrate the strong engineering potential of the proposed method and its promise for widespread application in real-world industrial scenarios.