Early wheel flat detection: an automatic data-driven wavelet-based approach for railways

Early Identification of Unbalanced Freight Traffic Loads Based on Wayside Monitoring and Artificial Intelligence

“…Moreover, the most relevant information related to the operational and environmental effects is retained in the first axes of the PCA. The number of p components to be discarded is calculated by the common rule that considers the percentage of total variance explained by each component to be equal to 80% [ 27 ]. In both PCA and ARX features, only the first principal component is discarded.…”

“…These methodologies rely on advanced signal processing combined with machine learning techniques and are typically based on the extraction of proper features to distinguish undamaged and damaged situations. Previous studies demonstrate good results in railway defect detection using these approaches, namely, in the detection of train wheel damages, such as flats [ 26 , 27 , 28 , 29 , 30 , 31 ], out-of-roundness [ 32 ], and squats and corrugation [ 33 ]. Typically, these damage identification techniques require several operations including [ 34 , 35 ]: (i) data acquisition, (ii) feature extraction, (iii) feature normalization, (iv) feature fusion, and (v) feature classification.…”

“…This operation converts the time-series data into more condensed information, allowing damage to be more easily observed. Symbolic data, continuous wavelet transform [ 27 , 37 ], principal component analysis (PCA) [ 38 , 39 ], and autoregressive models [ 34 , 40 ] are examples of effective techniques for extracting damage-sensitive features for both static and dynamic monitoring. In applications where the measured quantity is acceleration, autoregressive models (AR) have been widely disseminated, and the autoregressive model with exogenous input (ARX) [ 35 ] shows a higher sensitivity to damage in relation to AR due to its ability to capture cross information between sensors.…”

A Fault-Diagnosis Method for Railway Turnout Systems Based on Improved Autoencoder and Data Augmentation

“…At present, various articles in the literature have put forward the application of artificial intelligence technology to the fault diagnosis of RTS, and the related methods can be summarized as the threshold-based method [ 9 , 10 ], machine learning (ML) based [ 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 ], or deep learning (DL) based [ 23 , 24 , 25 , 26 ]. Huang et al [ 9 ] proposed a fault-detection method by using dynamic time warping based on the turnout current curve, which can detect faults by comparing the distance from the template curve.…”

“…To avoid these drawbacks, various research based on ML aimed to relatively reduced participation of manual settings, and it has been proven to be efficient and accurate in the field of fault diagnosis. Unsupervised learning and supervised learning are two types of ML [ 14 , 15 ]. Based on the feature engineering, several classification unsupervised methods, such as fuzzy cognitive map [ 16 ], support vector machine (SVM) [ 17 ], backpropagation [ 18 ], and ensemble classifier [ 19 ], have been presented in order to find the best combination features/classifiers to make intelligent decisions regarding the presence of RTS.…”