01698378

Component

Predicting railroad track defects is of great importance for safe and effective railway transportation. Using foot-by-foot raw track geometry, rail defects and tonnage data, this paper develops a new machine learning based approach not only to identify the track geometry parameters that most contribute to rail defects occurrence but also to predict the rail defects occurrence. Taking the huge volume of the data into account, Singular Vector Decomposition (SVD) and a recursive feature elimination algorithm are applied to reduce the dimensions. In addition, to capture more knowledge from the geometry data, some features, including Track Quality Index (TQI), energy, and time-trend are extracted. This, in turn, facilitates the learning and predicting process. Moreover, since there exists a very limited number of rail defects, the Adaptive Synthetic Sampling Approach (ADASYN) is applied to overcome the issue of imbalance in the dataset. In terms of machine learning algorithms, the proposed approach benefits from an extreme gradient boosting (XGBoost) algorithm in which the hyperparameters are optimized using a Bayesian optimization method. Finally, this approach is applied to the data on 100 miles of class I railroad to demonstrate its applicability and efficiency.

This paper was sponsored by TRB committee AR060 Standing Committee on Railway Maintenance.

19-05217

English

5p

2019

Transportation Research Board 98th Annual Meeting

Digital/other

References; Tables

Data files; Geometric segments; Machine learning; Mathematical prediction; Railroad rails; Railroad tracks; Structural deterioration and defects

Data and Information Technology; Maintenance and Preservation; Railroads

Transportation Research Board Annual Meeting 2019 Paper #19-05217

TRIS, TRB, ATRI

Dec 7 2018 9:53AM