Acoustics Resonance Spectrometry (ARS) for Advanced Streaming Prognostics of Drivetrain System

Failure prognostics is a crucial research direction in modern industrial practices. However, predicting failures from measured data remains challenging due to the high-frequency sampling rates in the time domain and the low sensitivity of statistical moments and frequency-domain analysis to subtle degradation. To address this problem, this paper proposes the use of Acoustic Resonance Spectrometry (ARS) as a machine learning-based approach for predictive monitoring of acoustic signals. By leveraging ARS, real-time analysis becomes feasible, as the high-frequency sampling rate is effectively reduced. First, the algorithmic framework of ARS is introduced to generate a multivariate dataset. Then, the generated data is fed into a machine learning model known as the Multivariate State Estimation Technique (MSET) for anomaly detection in time-series data. Finally, experimental studies validate the proposed approach, demonstrating its effectiveness in accurately and efficiently predicting failures.