Simple Non Regressive Informed Machine Learning Model for Prescriptive Maintenance of Track Circuits in a Subway Environment

A Track Circuit (TC) System enables automatic train detection and protection functions. A typical TC failure is the False Occupancy (FO), i.e., a TC results occupied for a certain period of time when there is no train on it. FO can be resolved by manually driving a train on the faulty TC with a waste of time and resources. After a FO, operators, based on their experience, need to decide if there is an actual need for maintenance or if the problem is due to other temporary causes. In this paper, we propose a Simple Informed Machine Learning (ML) based model able to automatically prescribe maintenance after a FO fully leveraging on the operator experience. However, ML models in modern industrial MLOps pipelines demand continuous data collection, model re-training, testing, and monitoring, creating a large technical debt. In fact, one of the main requirements of these pipelines is to not be regressive, i.e., not simply improve average performance but not introduce mistakes previously not present (negative flips). In this work we face this problem by empowering the proposed ML with Non Regressive properties. Results on real data coming from a portion of an Italian Metro managed by Hitachi Rail STS will support our proposal.