Determining the Causes of Train Delays - An Automatic Fuzzy Matching Approach

Abstract

In order to manage the health of assets, knowledge of conditions on the system level is often required. One of the most common approaches to determine the system conditions is to check the frequency of unplanned or corrective maintenance of considered systems. However unplanned or corrective maintenance usually means the issue has reached an intolerable degree and therefore leaves very limited room to improve the health condition of assets. To depart from the conventional approach of asset health management based on unplanned maintenance, Dutch Railways has decided to focus on operational disturbances, i.e., train delays, due to technical issues of trains in order to determine which system deterioration to investigate at the early stage. This work introduces the framework of the cause determination system for train delays that has been implemented within Dutch Railways. The cause determination is composed of two parts. One is an automatic fuzzy matching system that will match a delay to the most probable service request assigned by the support center. The causes of 40% of delays can be automatically identified by this method with an accuracy of 95%. This significantly reduces the human hours spent in identifying delay causes. The rest of cause determination is manually carried out by the Delay Analysis by Calling project. In this project, the calling team will first call train drivers to ask and record the encountered issues and handling procedures, and then the reliability engineers will determine the causes of delays based on these feedbacks. The calling team therefore serves as a feedback channel for drivers and support center and enables us to analyze on the delay handling and advice given by the support center for continuous improvement and development. Another advantage of integrating a highly accurate and robust fuzzy matching system into an interactive cause identification framework that requires input from various business units is that it intrigues interests and builds up trust in data science technology within the organization. This helps us to smoothly introduce this culture change which often is a critical point in transforming into a data-driven organization, especially for the maintenance industry. Based on the identified delay causes, Dutch Railways has built a Delay Analysis Dashboard which can provide a good overview of system conditions for various fleets, and also provide more possibilities to avoid operational disturbances.