Development of Input Selection Methods for Predictive Modelling in the Health Monitoring of Gas Turbine Engines

点击翻译按钮获取中文摘要

The objective of this research project is to develop efficient input selection methods intended for predictivemodelling applications as part of industrial partner's (TECSIS Corporation) ongoing projects that deal with thequantification of the health monitoring of a gas turbine (GT) engine using data analytics tools. TECSISprovides product development and research and development services, and has an active research portfolio inPrognostics and Health Management (PHM) system development. Going forward as part of their continuousproduct advancements, TECSIS needs a methodology to automatically select the most dominant inputs thathave significant influence on outputs like exhaust gas temperature (EGT) and power which are major indicatorsfor health monitoring of gas turbines. The proposed input selection methods will be developed utilizingadvanced machine learning techniques by the research team from the University of Waterloo in closecollaboration with the technical experts and engineers from the industrial partner. The benefits of the proposedinput selection methods include improved prediction accuracy, faster and more cost-effective predictivemodels, better interpretations of constructed models, and cost savings on the next round of data collection dueto fewer inputs involved. These methods also have significant implications for developing predictive modeling,classification, and clustering applications in other mechanical, electrical, and software systems that TECSISworks in. Incorporation of the proposed input selection methods into its predictive modeling and other patternrecognition tools will help TECSIS to expand its applications areas. The success of this project will enable theindustrial partner to create new source of revenue generation and reach out to new clientele.