Data driven prediction of hot-Axles: A case study of prognostics and health management for locomotives

As the most cost- and energy-efficient means of mass transportation, the railway plays a crucial role in the modern society. It is thus essential to continuously improve the safety and operation proficiency of locomotives to allow a sustainable development. A promising solution toward this goal is to take advantage of the fast-growing machine learning techniques to perform Prognostics and Health Management (PHM) on the locomotive fleet. As a first step toward a systematic PHM engineering of locomotives, this work proposes a data driven hot-axle prediction and analysis framework. The framework exploits a vector autoregression predictor to produce online temperature as a dynamic reference for failure prediction. To handle the complex dynamics under varying working and failure conditions, Hidden Markov Model and symbolic regression based techniques are also integrated to identify system states and forecast states under failure conditions.