A literature review on the applications of artificial intelligence to European rail transport safety

Abstract In accordance with the current European railway regulations and particularly the two directives relating to the interoperability (Directive (EU) 2016/797) and safety (Directive (EU) 2016/798) of the railway system, this literature review proposes to classify artificial intelligence (AI) applications by distinguishing the structural elements (Infrastructure, Energy, Control‐Command‐Signalling and Rolling Stock) and the functional elements (Operation and Traffic Management, Maintenance and Telematics Applications) of the European railway system. Several “classic” AI techniques are implemented, including machine learning (supervised, semi‐supervised, unsupervised), deep learning such as artificial neural networks (ANN), natural language processing (NLP), case‐based reasoning (CBR), etc. However, the inadequacy of these approaches to capitalize, share and reuse the knowledge involved has oriented research towards the development of new approaches based on ontologies and knowledge graphs. This study shows that the stages of data acquisition, modeling, processing and interpretation pose a crucial problem in rail transport. In addition, with complex models described as “black boxes”, it is difficult to understand how the internal reasoning mechanisms of the AI system impact the solution and predictions. The new explainable AI (XAI) approach can possibly provide an element of response to this problem.