Détection des défauts dans un environnement de maintenance prédictive

Abstract

Failure forecasting is an important function in predictive maintenance. This thesis proposes a methodology capable of predicting a number of faults using signals collected from permanent magnet DC motors. These electric motors are often used in the industrial world, especially in production lines. With the advent of the Industrial Internet of Things, condition monitoring and signal acquisition are now greatly simplified. That is why, fault prediction has become a prevalent research topic in the field of industrial maintenance. In this work, motor signals are generated by simulation using MATLAB and Simulink. These signals represent different motor operating regimes subjected to mechanical and electrical faults. Five (5) types of faults are analyzed using three (3) different engines. They are bearing problems, power supply short circuit, speed, current and voltage sensor problems. In addition, speed setpoints are used as input reference simulating different motor operating conditions such as conveying, transfer and lifting. A total of 9000 signals each having a 5-second length are generated for this project. Four (4) predictive models, FCN, ResNet, Encoder and LSTM, from deep learning techniques are trained and evaluated using the generated signals. Their performances are quantified using classification metrics such as accuracy, precision, recall and F-score. Experimental results indicate that the Encoder is the best model during training with an accuracy of 89.6%, a precision of 90.47%, a recall 90.08% and a F-score of 90.28%. This trained model when applied to new data was able to produce an accuracy of 88.53% and a recall of 88.67%. More interesting still, the Encoder model is able to correctly classify the absence of faults at a rate of 83.14% while the score is 50.4% for FCN, 14.4% for ResNet and 54.4% for LSTM. Finally, source code for this project is available at : https://github.com/alexchartrand/Fault-detection-in-IIoT and MATLAB-Simulink signal generation files are stored in the folder https://github.com/alexchartrand/DCMotor-fault-simulation.

Date	22 Dec 2021
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Tony Wong (Supervisor)