An Online Diagnosis Method for Sensor Intermittent Fault Based on Data Driven Model

The signal acquisition and analysis of the structure vibration can effectively diagnose the operation status and fault type. The visual inspection method has the advantage of nondestructive testing and providing full-field vibration information for monitoring the vibration of rotating machinery. However, the accuracy requirements for measuring the small displacements generated by the low-amplitude vibration of high-speed rotating machinery are extremely high. It remains a challenge to achieve accurate, precise, and fast extraction of high-frequency and small-amplitude micro-vibrations using visual-based displacement measurement methods. Therefore, this article proposes the object detection model, the vibration displacement detection network (VDDNet), for measuring vibration displacement of rotating bodies. VDDNet reduces the required computational complexity of the actual detection target and optimizes the quality of vibration displacement measurement. Additionally, by introducing an attention module and adopting different learning strategies to share rotating body target features, the real-time and accuracy of vibration monitoring are ensured. Furthermore, the evaluation method used in this article comprehensively considers detection accuracy, real-time performance in deep learning methods, and displacement capture accuracy in vibration displacement monitoring. Actual vibration experimental results demonstrate that the proposed method is minimally affected by illumination and can achieve a better real-time capture accuracy of rotor micro-displacements, further proving its effectiveness and feasibility, which is helpful in advancing related research in this field.