Improving Recommendation System by using a knowledge Graph Database for Maintenance of Rolling Stock

Abstract

Capitalization and sharing of knowledge play an essential role in the management of railway rolling stock maintenance. It is an activity often associated with collaborative decision-making systems. In this paper, we build a maintenance recommendation system on TigerGraph Cloud. We use the Failure tree graph constructed in previous work. We built a recommendation system that recommends the top-10 actions to a technician based on the rating prediction. We train the recommender model with a gradient descent algorithm. In the last part of this work, we compare the accuracy of this model based on Knowledge graph with the Collaborative filtering model. Preliminary results indicate that graph-based recommendation systems perform better than baseline methods. This study contributes to the sharing of knowledge on repair methods for rolling stock. The added value of our study is the use of a real dataset in the field of maintenance of railway rolling stock.