Causal Discovery for Railway Health Condition Monitoring - A Case Study

Abstract

Integrating knowledge of causal relationships into machine-learning models might help to overcome the problem of limited interpretability of these models. Causal methods can thus be expected to enable innovative data-driven approaches for transparent and reliable diagnostics in safety-critical domains such as railway health condition monitoring. Based on available interlocking data, this paper exemplarily demonstrates how causal discovery can be applied to learn the causal influences of certain weather parameters on the insulation resistance of the electric installation of electronic interlockings.