Development of an AI model for electronic board maintenance decision prediction for railway equipment

Railway equipment is required to work fault-free under rugged conditions such as continuous operating heat loads. One of the most important considerations in railway maintenance is the ability to predict failure, due to aging, early enough such that spare parts can be acquired just-in-time which normally takes a lead time of several weeks to up to around a year from the supplier. This study has conceived and tested the RUS Boost Ensemble machine learning algorithm to predict maintenance decisions of railway electronic boards based on measurable component values. Some traditional approaches like MIL-217 are commonly used for electronic reliability prediction but these types of approaches do not consider load profiles, failure root causes, and practical limitations in the number of available experimental test samples. This study develops a prognostic approach by considering actual load conditions and life limiting factors, and utilises machine learning algorithms to build the model. The model also makes use of real-life test samples from lab ALT (Accelerated Life Testing). After development of the AI model to predict electronic board component maintenance, the test results revealed the predictive accuracy to have up to 95% correlation for the red/urgent category and 94% for the yellow/medium-urgency category.