Real-time learning for real-time data: online machine learning for predictive maintenance of railway systems
Minh-Huong Le-Nguyen, Fabien Turgis, Pierre-Emmanuel Fayemi, Albert Bifet
Abstract
In the railway, an unexpected fault reduces service availability and may cause fatalities. Therefore, maintenance must be timely. Nowadays, the omnipresence of onboard sensors generates data enormously, thus enabling data-driven predictive maintenance. For this, machine learning has come into prominence. Traditionally, a machine learning model is trained on a batch of data. However, this approach lags behind on fast-paced data streams. For real-time learning on real-time data, we propose a pipeline that employs online machine learning to address predictive maintenance of sensorized railway systems. We showcase the implementation and experimental results of two modules automating data preprocessing to demonstrate the potentials of online learning. We also discuss the intuition of another module using online clustering to monitor the evolution of system health.