Maintenance prédictive pour une turbine de puissance en utilisant des algorithmes d’apprentissage machine

Abstract

Today, automation and machine learning technologies are making their way into the industry, allowing companies to get the most out of their production chains. One of the greatest strengths of this technological advancement is that it enables equipment to function longer without failure. This is a consequence of the predictive maintenance applied to the industrial system and carried out with data analysis systems. Predictive maintenance is about predicting machine failures before they happen and avoiding unscheduled downtime. One of the main techniques generally used in predictive maintenance is the extraction of vibration data from machines. The information extracted from this data allows us to identify, predict, and prevent rotating machinery breakdowns. In this context, this thesis proposes a method of diagnosis and prognosis of the imbalance fault of a power turbine using the techniques of artificial intelligence and machine learning by putting their advantages over the old diagnostic procedures, mainly temporal analysis and frequency analysis of signals. This research study is designed to collaborate with our industrial partner, Siemens Energy, based in Montreal. It aims to improve and automate its machine condition maintenance process. Therefore, this work aims to propose a predictive maintenance approach to monitor and detect in real-time whether the power turbine is subject to an imbalance or it is in normal operating condition using the techniques of artificial intelligence. For this, a comparative study of different analytical approaches and several machine learning algorithms, such as SVM (Support Vector Machine), RF (Random Forest), and KNN (K Nearest Neighbors), was made to choose the most efficient method in terms of defect detection.

Date	17 Dec 2021
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Amin Chaabane (Supervisor)