A fuzzy instance-based model for predicting expected life: a locomotive application

Abstract

The behavior of complex electromechanical assets, such as locomotives, tanks, and aircrafts, varies considerably across different phases of their lifecycle. Assets that are identical at the time of manufacture will 'evolve' into somewhat individual systems with unique characteristics based on their usage and maintenance history. Utilizing these assets efficiently requires a) being able to create a model characterizing their expected performance, and b) keeping this model updated as the behavior of the underlying asset changes. This paper outlines a fuzzy peer-based approach for performance modeling combined with an evolutionary framework for model maintenance. A series of experiments using data from locomotive operations were conducted and the results from this initial validation exercise are presented. The approach of constructing local predictive models using fuzzy similarity with neighboring points along appropriate dimensions is not specific to any industry or asset type. This approach is not limited to the locomotive domain and may be applied in any problem where the premise of historical similarity along chosen dimensions implies similarity in future behavior.