Hybrid multi-encoder transformer and case-based reasoning for intelligent decision support in high-speed railway vehicle maintenance

Abstract

Despite the recent development of artificial intelligence-based methods for predicting high-speed railway malfunctions, the actual maintenance procedure for fixing vehicle breakdown remains highly dependent on human expertise. However, delays in identifying faults and determining appropriate maintenance procedures cause considerable disruptions to railway traffic. In this study, we propose a new solution to enhance the efficiency of fault identification and maintenance decisions during breakdowns in high-speed railway vehicles. We develop a text-to-text generative model that handles handwritten breakdown data and uses a multi-encoder transformer model with case-based reasoning (CBR) to propose a maintenance strategy based on existing knowledge and expertise. Our experimental tests demonstrate the superiority of the proposed method in terms of using historical maintenance actions and fault sentences concurrently compared with baseline approaches. Our findings hold strong potential for improving decision support systems in high-speed railway vehicle maintenance by harnessing the synergy between CBR and generation models to derive actionable insights from maintenance data records.