Data Driven Prognosis approach for Multi Sensor Degradation Data Analysis Using machine learning

Abstract

newlineABSTRACT newline newlineRecent scientific and technological advancements have led to substantial newlinegrowth of complex engineering systems that practically changed every aspect of newlineour lives. Even Industrial Systems have become more efficient with the automated newlinecomplex machinery over traditional ones. As improved automation became newlinepopular, unplanned stops had direct implications, including death, environmental newlinedamage, and enormous financial expenditures. Unscheduled shutdowns as a result newlineof machine and subsystem failures pose significant operational and usage newlinechallenges over their life cycles. Assuring the safety and performance of complex newlineand safety-critical systems is a significant challenge. Prognostics and Health newlinemanagement (PHM) play a crucial role in increasing safety and facilitating newlineoperations planning. However, the growth of data processing and its crucial newlineimpact on remaining useful life predictions need continuous development in order newlineto achieve greater performance levels. Frequently, there is a disconnect between newlinethe sufficiency of prognostic pre-processing and the accuracy of prediction newlinemethods. One strategy to close this gap is to develop an adaptive data processing newlinesystem capable of filtering multidimensional condition monitoring data by newlineidentifying relevant information from numerous data sources. Due to lack of newlineunderstanding of multidimensional failure mechanisms and the collaboration newlinebetween data sources, present prognostic approaches are unable to deal effectively newlinewith complex interdependence and multidimensional condition monitoring data. newlineThe methodology described in this thesis addresses these shortcomings by newlineincorporating a data driven prognosis approach that analyzes the multiple sensor newlinedegradation data to forecast the life expectancy of systems using machine learning newlinemodels. newlineThe proposed methodology describes a feature selection technique for newlinepredicting remaining useful life based on the notion of cause and effect. Together newlinewith data filtering approach and remaining useful life estimation, the