Diagnosis of Vibration Systems for Decision Making

Abstract

Defect detection in mechanical systems is of critical importance in industrial applications, particularly for rotating components. This paper presents a model-based methodology for early online detection of mechanical faults using vibration analysis. The proposed approach is implemented in three main phases: model identification, optimal filter synthesis, and fault detection. A key advantage of this method is its ability to detect faults automatically, without requiring expert interpretation, and the possibility of putting it online or offline. This is achieved through a physics-based model of the system and an optimized implementation. Fault detection relies on analyzing the variance of the residual signal, defined as the difference between measured outputs and those estimated by the optimal filter. A fault is flagged when this variance exceeds a predefined threshold. The effectiveness of the method is demonstrated experimentally on a vibratory mechanical test bench equipped with gears. Validation is performed by comparing the results with expert assessments conducted during the experiments and with findings reported in the literature.