Research on Fault Diagnosis Method for Key Components of Rotating Machinery Based on Multi-domain Generalized Feature Optimal Transmission and Distribution Alignment Transfer

Rotating machinery such as flue gas turbines, centrifugal compressors, steam turbines, and centrifugal pumps serve as the “power heart” of petrochemical equipment. If a failure occurs leading to an unplanned shutdown, it can result in production interruptions at best, or severe damage and fatalities at worst. Achieving intelligent diagnosis of common typical faults in key components of the above units (such as rotor systems and rolling bearings, etc.) is crucial for improving device reliability, enhancing operational stability, reducing risks and maintaining costs. The fault diagnosis method based on multi domain generalized feature optimal transmission and distribution alignment transfer (GFOT-DAT) is proposed to address challenges such as the difficulty in extracting early-stage weak faults, poor noise interference resistance, confusion of signal features across different fault states, and low learning performance of common features in cross-condition data. Firstly, typical fault training data from multiple source domains of different equipment and operating conditions of rotating machinery are obtained. Secondly, the raw vibration signals are decomposed, filtered, reconstructed, and enhanced to extract commonality and individual features from the enhanced signals. Next, the source domain feature samples are used to train the multi-kernel twin classifiers. An optimal transport solver is constructed based on the Wasserstein distance to quantify the measures between cross-domain feature samples. Then, multi-kernel weighted maximum mean discrepancy is employed to reduce inter-domain distribution differences, and manifold learning is used to align the cross-domain feature space. Finally, the optimal transfer mapping matrix is input into the weighted optimal classifier to obtain the category labels of the test data, achieving fault diagnosis. Additionally, the effectiveness of the GFOT-DAT method is validated using eight transfer diagnosis tasks, which are constructed based on 37 fault cases from both laboratory and real engineering data. A comparison with six cross-domain fault diagnosis methods from the published literature clearly demonstrates the superiority of the proposed method.