Classification of Rail Switch Data Using Machine Learning Techniques

Abstract

Rail switches are critical infrastructure components of a railroad network, that must maintain high-levels of reliable operation. Given the vast number and variety of switches that can exist across a rail network, there is an immediate need for robust automated methods of detecting switch degradations and failures without expensive add-on equipment. In this work, we explore two recent machine learning frameworks for classifying various switch degradation indicators: (1) a featureless recurrent neural network called a Long Short-Term Memory (LSTM) architecture, and (2), the Deep Wavelet Scattering Transform (DWST), which produces features that are locally time invariant and stable to time-warping deformations. We describe both methods as they apply to rail switch monitoring and demonstrate their feasibility on a dataset captured under the service conditions by Alstom Corporation. For multiple categories of degradation types, the baseline models consistently achieve near-perfect accuracies and are competitive with the manual analysis conducted by human switch-maintenance experts.

On the Fault Detection and Diagnosis of Railway Switch and Crossing Systems: An Overview

“…Four papers used a shallow learning algorithm as an FD algorithm with application in railway S&C systems. The first one [96] was based on a one-class SVM (OCSVM) algorithm, the second paper [97] exploited a self-organizing map (SOM) technique, the third one [98] applied two machine learning classifiers, a Gaussian Naïve Bayes (GNB) classifier and a neural network classifier (multilayer perceptron), and the fourth one [99] explored two recent machine/deep learning frameworks for classifying various switch degradation indicators including the featureless recurrent neural network called a Long Short-Term Memory (LSTM) architecture and the Deep Wavelet Scattering Transform (DWST). In the first paper, the OCSVM algorithm used a similarity measure of 'edit distance with real penalties' to classify the S&C health conditions as healthy or faulty.…”

“…They only measured the pressure (for the Clamplock) or current for the two others, where two fault modes were considered in this study, switch hard to release fault mode and fails to make detection, i.e., sensor fault mode. Finally, two recent machine/deep learning frameworks, the LSTM architecture and the DWST, were explored and they were evaluated for their feasibility on a dataset captured under the service conditions by the Alstom Corporation [99].…”

“…Data-driven approaches FD methods [86][87][88][89][90][91][92][93][94][95][96][97][98][99] SVM with Gaussian kernel [90]; OCSVM [96]; SOM technique [97]; PCA [94]; data-mining techniques [86]; harmonic regression and VARMA [93]; K-S test [95]; DTW [89]; GNB [98]; LSTM and DWST [99].…”

A Fault Detection Method for Electrohydraulic Switch Machine Based on Oil-Pressure-Signal-Sectionalized Feature Extraction

“…Further, de Aguiar [ 5 ] used the set-membership concept derived from the adaptive filter theory in the type-1 and singleton/nonsingleton fuzzy logic systems so that the model convergence speed was improved and computation complexity was reduced, and then he demonstrated that the upper and lower singleton type-2 fuzzy logic system was a more effective classifier for an electric switch machine fault [ 6 ]. The long- and short-term memory (LSTM) and the deep wavelet scattering transform (DWST) were explored for classifying various switch degradations, and the feasibility of a dataset captured under the service conditions was demonstrated in [ 7 ]. A hybrid fault diagnosis (HFD) method was adopted to identify a fault based on the current curves of a railway switch machine in [ 8 ].…”

“…Zhou [ 23 ] applied grey correlation theory to the intelligent fault detection of turnouts based on the oil pressure signal of a ZYJ7 electrohydraulic switch machine. For the existing fault detection, the expert system [ 24 ] is hard for acquiring knowledge and needs a lot of prior knowledge of railway staff; the Kalman filtering method [ 25 ] can only be successful in a part of the dataset; a reliable and reasonable prior probability has to be provided for a Bayesian network [ 26 ], in which determination is very difficult; a support vector machine [ 3 ] is a binary classifier in principle, which is very sensitive to feature selection; and a neural network [ 7 ] needs numerous samples for training to avoid misdetection. However, unsupervised clustering methods can support multiple fault detections and effectively improve performance, which do not need to be trained and be provided with many prior parameters.…”