| dc.description.abstract | As the railways are a significant section of transportation infrastructure, it is crucial to use good maintenance procedures for railway networks. Condition-monitoring of railway wheels and tracks is especially important where extremely unfortunate failure happens. So regular examination of railway tracks and wheels health is required to maintain safe and reliable train operations. The traditional method of manually inspecting rail tracks is inefficient and prone to human error and bias. This study aims to improve the aging railway system by using an automated solution to address these difficulties.
A series of experimental tests were carried out before data collection. A LabVIEW application was made for recording data and tested with a function generator and Data Acquisition (DAQ) device. The provided Cemit Data Collection (CDC) time-series data were converted into frequency domain through Fast Fourier Transform (FFT) and Wavelet Transform (WT) in MATLAB. Multivariate data analysis was done through FFT and WT data for fault detection using Principal Component Analysis (PCA) and Partial Least Square Regression (PLS-R).
During PCA, the score plot of FFT data depicted the important samples present in it. However, the score plot of WT data illustrated that important samples were absent. Hence, after further analysis of FFT data using PLSR, the calibration model was built which was validated from test set validation. From this, Y predicted was almost equal to the Y reference, hence the model performance can be significant.
From comparing the score plots of FFT and WT data, it is concluded that the WT data is not useful for discovering faults in the tracks using multivariate data analysis because no relevant samples for further investigation were detected. To get better results, it's also a good idea to analyze multi-channel or more sensor data. | |
