Keywords: track maintenance, machine learning techniques, track geometry prediction, longitudinal level defects, isolated defects, Long Short-Term Memory (LSTM).

The ability to anticipate future variations in track geometry plays an important role in railway asset management. Reliable prediction of longitudinal level irregularities supports earlier recognition of vulnerable locations, improves the timing of maintenance interventions, and helps reduce operational and maintenance costs. This study investigates machine-learning techniques for forecasting isolated longitudinal-level (LL) defects using EM120 inspection vehicle data. As a case study, almost 17 km of the recently reconstructed Leixões Line in Portugal was considered. For each defect, a local window between ±25 m was extracted to evaluate the behaviour of the fault zone. In this study various prediction machine learning techniques were compared: Artificial Neural Network (ANN), XGBoost, Long Short-Term Memory (LSTM), and Ridge regression and their performance was assessed using R², Mean Absolute Error (MAE), and Root Mean Square Error (RMSE). The results show that LSTM regression model provided the most consistent overall performance across the isolated defects, demonstrating strong predictive capability.

download the full-text of this paper (PDF, 13 pages, 924 Kb)

go to the previous paper
go to the next paper
return to the table of contents
return to the volume description