Analysis and Prediction of Track Geometry Degradation: Case study of Addis Ababa Light Rail Transit
No Thumbnail Available
Date
2021-06
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Development of good maintenance policy requires better understanding of the long-term behavior of railway track systems. Addis Ababa Light Rail Transit is new system which has been in operation for the last five years. Understanding the degradation pattern of such system will help in scheduling and controlling maintenance activities, which is a problem to many railway operating entities. Track safety and maintenance effectiveness are improved by clearly understanding future condition of the track.
This paper analyses and predicts the track geometry degradation of Addis Ababa Light Rail Transit. Track geometry degradation models developed by different researchers were critically reviewed to determine suitable model for AALRT. Factors which contribute to track geometry degradation were analyzed to determine the relationship between the factors and track degradation. Factors such as Number of Trips, Curve Radius, Speed, Grade, Tonnage and Track Surface were analyzed. All analyzed factors except Track Surface were found to have influence on rate of degradation for curved and straight sections of AALRT. The selected influencing variables and longitudinal leveling changes were used to develop track degradation model. Multiple regression model was developed in SPSS while Artificial Neural Network model (ANN) was developed on MATLAB software. The performances of two models were evaluated for straight and curved sections of North South line.
North South line is a double line, model development was performed per each direction (Uplink line and Downlink line) and whole line regardless of direction. Six multiple regression models were developed. For straight sections three models were developed in which Downlink line displayed tremendous performance (R2=98%) compared to Uplink line (R2=85%) and North South line (R2=88%). For curved sections performance of 91% was observed on Uplink line and Downlink line while North South line had performance of 88%. Six ANN models were also developed as a complex model to predict future condition of track. Performance of 97% R-squared was obtained for both curved and straight sections of Uplink line while performance of 96% was observed in straight section and 93% performance was observed in curved sections of Downlink line. North South line display performance of more than 90% for both straight and curved sections of line. Comparison of two models showed that both can predict the longitudinal levelling changes with very low error. Although the ANN model has better performance than Regression model.
Description
Keywords
Track Geometry Degradation, Regression model, ANN model, AALRT