Detecting Mechanical Failure with mmWave Radar Using Deep Learning

Abstract

Mechanical failure detection is an important subject in the field of predictive maintenance. In this paper, we focus on the detection of an asynchronous motor malfunction by a Frequency-Modulated Continuous Wave (FMCW) radar using Machine Learning (ML) algorithms. We propose an experimental set-up that will enable us to acquire data from the radar while the motor is running. Two scenarios are proposed. The first in normal motor operation. The second in the presence of a bearing malfunction. We then acquire the dataset, and train our ML models. We find the best accuracy results for LSTM and ResNet, at 93.41% and 93.2% respectively.