Predicting and monitoring railway asset health in real time using the k-means algorithm in the absence of failure-to-failure data

Abstract

A difficult technology, prognosis seeks to improve the dependability and performance of a system or component by precisely predicting and estimating its remaining useful life. However, there is a lack of real-world examples of effective prognostic applications, despite the abundance of literature on the subject of predictive maintenance prognosis. This is because, while substantial flaws are often avoided by frequent maintenance, there is a dearth of run-to-failure data that might be used to construct the prediction model. Instead of using run-to-failure statistics, this research suggests a new methodology to predict when railway maintenance will be needed. In many cases, the primary point is that evaluating the severity of the issue and estimating the remaining time are sufficient for planning maintenance. A deterioration indicator for railway door systems is first created using motor current signals. This indicator is based on the dynamic temporal warping approach, which measures the similarity between typical normal and problematic behavior. The fault severity is then evaluated using the K-means method, and for each fault severity level, a representative time estimate is then provided. You may forecast the remaining time until the critical fault severity level is reached without requiring run-to-failure data thanks to this estimation. Predictive maintenance planning for railway door systems is therefore made possible by the suggested strategy. Better remaining time forecast is also possible after adding operational data to the fault severity threshold. The suggested solution is also applicable to traditional railway assets and other electro-mechanical actuators as the controller or motor drive mainly provides the motor current signals, which do not need any extra sensors.