Remote condition monitoring of rail tracks using distributed acoustic sensing (DAS): A deep CNN-LSTM-SW based model

Railroad condition monitoring is paramount due to frequent passage through densely populated regions. This significance arises from the potential consequences of accidents such as train derailments, hazardous materials leaks, or collisions which may have far-reaching impacts on communities and the surrounding areas. As a solution to this issue, the use of distributed acoustic sensing (DAS)-fiber optic cables along railroads provides a feasible tool for monitoring the health of these extended infrastructures. Nevertheless, analyzing DAS data to assess railroad health or detect potential damage is a challenging task. Due to the large amount of data generated by DAS, as well as the unstructured patterns and substantial noise present, traditional analysis methods are ineffective in interpreting this data. This paper introduces a novel approach that harnesses the power of deep learning through a combination of CNNs and LSTMs, augmented by sliding window techniques (CNN-LSTM-SW), to advance the state-of-the-art in the railroad condition monitoring system. As well as it presents the potential for DAS and fiber optic sensing technologies to revolutionize the proposed CNN-LSTM-SW model to detect conditions along the rail track networks. Extracting insights from the data of High tonnage load (HTL)- a 4.16 ​km fiber optic and DAS setup, we were able to distinguish train position, normal condition, and abnormal conditions along the railroad. Notably, our investigation demonstrated that the proposed approaches could serve as efficient techniques for processing DAS signals and detecting the condition of railroad infrastructures at any remote distance with DAS-Fiber optic cable setup. Moreover, in terms of pinpointing the train's position, the CNN-LSTM architecture showcased an impressive 97% detection rate. Applying a sliding window, the CNN-LSTM labeled data, the remaining 3% of misclassified labels have been improved dramatically by predicting the exact locations of each type of condition. Altogether, these proposed models exhibit promising potential for accurately identifying various railroad conditions, including anomalies and discrepancies that warrant thorough exploration.