A Machine Learning Perspective for Vibration Sensing and Identification of Modal Parameters of Electromechanical Equipment Using a Mach-Zehnder Interferometer

Vibrations in electromechanical machines pose a risk of performance deterioration and mechanical failures, stressing the need for precise all-weather vibration detection and identification of modal parameters for predictive and proactive maintenance. Using an experimental approach, a dataset of interferograms is generated from an optical sensor with labeled vibration amplitudes corresponding to frequencies ranging from 50 Hz to 250 Hz through voltages of 10 V and 15 V, respectively. The experimental setup integrates a Mach-Zehnder interferometer (MZI) with a vibrating motor to capture minute displacements induced by vibration frequencies and record them as fringe images via a CCD camera. The k-nearest neighbor (k-NN) machine learning and FFT algorithms are employed for analysis. The vibration modes and resonant frequency of the motor are determined from the fringe images using the FFT technique. The dataset is split into a 70% training set and a 30% validation set. Computer vision techniques are applied to extract the features of a local binary pattern (LBP) from the training fringe images. The machine learning model is trained to accurately detect the vibration amplitudes based on the LBP in each fringe image. The proposed approach achieves 98.5% accuracy in detecting the motor vibration frequency. Consequently, MZI has a potential for monitoring the real-time vibrations in electromechanical equipment.