User-Centric Anomaly Detection Enabled by XAI : Enhancing Predictive Maintenancein the Railway Industry

Sammanfattning

As sensor data in trains is sent and collected by train operators, utilising this data with anomaly detection systems could improve predictive maintenance of these trains. While reactive and preventive maintenance are the most common forms of maintenance in the railway sector, predictive maintenance with the use of artificial intelligence (AI) technologies is becoming increasingly important. This study proposes a framework that enhances predictive maintenance in the railway industry. By applying the Design Science Research (DSR) methodology, this framework was designed, implemeted, demonstrated and evaluated. The implemented artifact consists of a long-short-term-memory (LSTM) model, an autoencoder for anomaly detection, and explainable artificial intelligence (XAI) tools. It was not only created to enhance predictive maintenance, but also to help users understand its mechanisms and to improve trust. The LSTM model predicts the values of a particular train part, then these values are sent to the autoencoder to detect anomalies, and the explainable AI algorithms make it clear on what basis the LSTM predicted these anomalies.