This research builds upon research related to utilization of vibration and acoustic waves for continuous anomaly detection. Current industrial standards for vibration monitoring utilize threshold-based algorithms which require specialized labor to initiate, require inordinate amounts of data to be continuous, and are inaccurate. The research developed the vibration and acoustic resonance spectrometry (VARS) preprocessing algorithm and a framework to apply ML for successfully detecting failures in electro-mechanical systems. Whereby we have automated much of the specialized labor, implemented dimension reduction without loss of information, and improved on the status quo of univariate threshold monitoring.

The next step for this research is to automate fault classification of both type and severity. This will be accomplished by testing a library of signals with labeled fault signatures across several fault types and severities for each type. Over the course of the second phase of this project, we are conducting experiments for determining an optimal method for automated mechanical fault classification. The research includes signal processing and data feature extraction, parametric comparison testing of feature input and AI accuracy, increasing generality of models to differentiate between simultaneous faults.  and test the ML or AI models.