Research:
The scope of the research is the monitoring of the health conditions of railway assets, specifically the study of predictive diagnostic methods (PHM Prognostics and System Health Management) for the assessment of the maintenance status of a railway vehicle. The aim of the research is to create a framework that links the state of integrity of the asset to the useful life of a set of its components. The prediction of the residual useful life of the component is linked to the assessment of the risk associated with its failure. The risk measurement is the indicator that supports the asset manager in deciding when and how to perform maintenance activities on the vehicle. The relevant information to feed the decision model derives from the definition of the most relevant components with respect to safety and availability, in relation to the operating model. The components are identified through the functional breakdown of the vehicle, the analysis of their impact in terms of safety, availability and cost of repair and/or replacement.
From the analysis of the time series of failures, real cases, and from the RAMS clauses of the project, four components have been identified for the study: wheelset, brake disc, brake lining and pantograph. FMECA analyses of these components are carried out, from which the failure metrics are linked with the operating model in order to define algorithms for predicting their wear mode. In the study, data analysis models based on machine learning are used, with neural networks informed by physics, analysis of fault images with convolution network methods and the use of LLMs for the analysis of work relationships. To provide a decision support method, the results of the analyses are linked to the study of the causes of failures, FRACAS method, performed by expert personnel and compared with what is obtained through the support of data analysis methods, via machine learning and LLM. This will allow to create and calibrate a standard decision support interface, a prescriptive maintenance model. Through the decision model it will be possible to have an automatic assessment of the best maintenance decision to be taken by the fleet manager, based on a dynamic implementation of the risk assessment, operating model and maintenance costs. The framework built defines the relationships between degradation functions, and will also be able to consider their mutual interaction. It allows to introduce additional components, depending on the impact it has on the vehicle architecture and with respect to the operating model; therefore, it aims to be scalable for different types of vehicles or assets.