Innovative Application of Statistical Sound Analysis for Fault Diagnosis in Variable-Speed Induction Machines

The article presents an innovative method leveraging statistical sound analysis to diagnose malfunctions in variable-speed drive-controlled (VSD) induction machines (IM). This early anomaly detection approach strengthens preventive maintenance by preventing major failures. The bell-shaped curve of the Gaussian distribution, integrated into this analysis, reveals a significant concentration of values around the mean, thus facilitating the early detection of abnormal sounds emitted by the machine. Experiments highlight that driving IMs controlled by frequency inverters at low speeds can induce the emergence of resonance frequencies, altering the machine's sound in an unusual and bothersome manner. To address this issue, a statistical characterization of the sound data from IMs is adopted. In this experimental study, the machine is first powered by a zero-frequency inverter, then by frequencies of 16.66Hz, 33.33Hz, and finally at the nominal frequency of 50Hz. The recorded sound signals are then transformed in to data matrices, and a statistical analysis is conducted, calculating different variables and coefficients of high-order statistics (HOS). This approach leads to precise characterization of normal and abnormal sounds, thus improving the machine's reliability and overall operational efficiency.