MTTR Prediction of railway rolling stock using regression algorithms

Railway equipment is very important in terms of maintenance, due to the complexity of its mechanical and electrical systems and the number of interchangeable parts. Thus, in order to ensure reliable and safe performance, railway companies must carry out regular maintenance and replace faulty equipment in a timely manner, without disrupting train operations. In addition, maintenance facilities are spread over the entire rail network, so an intelligent solution is needed to predict maintainability following breakdowns and allow technicians to take the necessary measures to ensure the proper functioning of all their equipment. Mean Time To Repair (MTTR) is one of the performance indicators used to determine the maintainability of an asset. Using historical data, in combination with failure data and maintenance action data, we explore several methods: linear and polynomial regression algorithms, Lasso regression, Ridge regression and random forests to build MTTR prediction models for the rail system. The results obtained are very promising and show that the linear and polynomial algorithms outperform the others on the prediction of maintenance performance in terms of MTTR.