A Wavelet Transform-Based Transfer Learning Approach for Enhanced Shaft Misalignment Diagnosis in Rotating Machinery

Rotating machines are vital for ensuring reliability, safety, and operational availability across various industrial sectors. Among the faults that can affect these machines, shaft misalignment is particularly critical due to its impact on other components connected to the shaft, making it a key focus for diagnostic systems. Misalignment can lead to significant energy losses, and therefore, early detection is crucial. Vibration analysis is an effective method for identifying misalignment at an early stage, enabling corrective actions before it negatively impacts equipment efficiency and energy consumption. To improve monitoring efficiency, it is essential that the diagnostic system is not only intelligent but also capable of operating in real-time. This study proposes a methodology for diagnosing shaft misalignment faults by combining wavelet transform for feature extraction and transfer learning for fault classification. The accuracy of the proposed soft real-time solution is validated through a comparison with other time-frequency transformation techniques and transfer learning networks. The methodology also includes an experimental procedure for simulating misalignment faults using a laser measurement tool. Additionally, the study evaluates the thermal impacts and vibration signature of each type of misalignment fault through multi-sensor monitoring, highlighting the effectiveness and robustness of the approach. First, wavelet transform is used to obtain a good representation of the signal in the time-frequency domain. This step allows for the extraction of key features from multi-sensor vibration signals. Then, the transfer learning network processes these features through its different layers to identify the faults and their severity. This combination provides an intelligent decision-support tool for diagnosing misalignment faults, enabling early detection and real-time monitoring. The proposed methodology is tested on two datasets: the first is a public dataset, while the second was created in the laboratory to simulate shaft misalignment using a laser alignment tool and to demonstrate the effect of this defect on other components through thermal imaging. The evaluation is carried out using various criteria to demonstrate the effectiveness of the methodology. The results highlight the potential of implementing the proposed soft real-time solution for diagnosing shaft misalignment faults.