Applications for Data Analytics using JRU Logs from Autonomous Trains

Abstract

In Alstom’s autonomous railway systems used for mining operations, vast amounts of operational data are recorded by onboard Juridical Recording Units (JRUs). However, these logs are complex and difficult to interpret. This thesis addresses the challenge of parsing and structuring JRU log data to enhance its readability and enable advanced data analysis. A custom log parser was developed to convert raw logs into a structured, readable format. To explore the potential of this data for predictive analysis, an LSTM Autoencoder neural network was trained for anomaly detection based on temporal patterns. The results demonstrate the feasibility of using machine learning for operational insights, and suggest promising future applications in automated fault detection and predictive maintenance.