Enhanced Engine Vibration by Computer Vision Methods for Predictive Maintenance

Abstract

Predictive maintenance focuses on the continuous monitoring of the system, with the goal of identifying potential degradations in advance, and commonly uses vibration sensors. However, this method can become costly when dealing with very large networks. As an alternative, studies have emerged that use high-speed cameras to capture video footage of motion and analyze vibration through computer vision processing and motion magnification algorithms. In this context, this work proposes a vibration analysis system based on video, following three main approaches: spectrogram analysis, vibration magnification, and classification of the equipment's operational condition using PCA and Random Forest. For algorithm validation, numerical data was collected using a vibration sensor, and video recordings were made using a GoPRO camera. Experiments were conducted in stages, and the results showed that the algorithm was able to distinguish between the motor’s operating condition classes and classified these states with an accuracy of at least 86%.