Railway Track Joints and Fasteners Fault Detection using Principal Component Analysis

This works presents a machine learning based fault detection algorithm specifically for condition monitoring of different types of railway tracks. The algorithm relies on one of the most commonly used machine learning algorithms, Principal Component Analysis (PCA), for extracting the patterns of various defected and nondefected railway track components including rail fasteners and joints like fishplate. The algorithm ensures a very fast yet robust feature extraction workflow primarily by virtue of its inherently offered dimensionality reduction resulting in lesser computational burden. The classification task is handled by the Euclidean distance classifier that identifies the nearest neighbor of the test image in the subspace spanned by the most dominant eigenvectors extracted from the training dataset during feature extraction workflow. Two varying railway track datasets, from Bangladesh and Pakistan, have been used in this work to validate the proposed algorithm using standard training to test ratios. Multiple classification scenarios are presented and analyzed in detail for both datasets with supporting results. MATLAB_R2022a has been used for development of the proposed algorithms that offers an overall efficiency of more than ninety percent under varying scenarios.