Neural Approaches to Prognostics and Health Management of Rolling Stock

Railway transportation is a mobility solution that must be both reliable and safe. To this end, the technical field of predictive maintenance focuses on applying data science to maximize the availability of rolling stock assets. This leads to modeling their degradation and minimizing their downtime by preventing service-affecting failures. To this end, Artificial Intelligence (AI) and Machine Learning have proven to be effective techniques for extracting latent patterns from the available data. This dissertation puts the emphasis on Deep Learning, which is the state of the art in neural network research as the leading paradigm in AI and Machine Learning. Additionally, the scope of the work is framed in the multinational industrial context of Alstom, operating worldwide in rail markets, and active in the fields of passenger transportation, signaling and locomotives. The thesis is intended to be an expert reference work at Alstom in the area of predictive maintenance for rolling stock, especially through the use of neural networks for developing advanced maintenance solutions that are reliable and cost-effective. To this end, different environments have been considered, including mixed data types, i.e., continuous and discrete variables, and different predictive objectives such as diagnosis and prognosis. As a result of this research, three journal articles have been published (in addition to some conference papers).