SGO-Based Optimized Lstm for Performing Vibration Analysis to Predict Fault in High-Speed Rail Tracks

Train bogies are vital components of trains since they are in close contact with the rails and support the weight of the train. Train bogie vibrations are dynamic loads caused by a variety of processes, including wheel and rail unevenness, impact at rail joints, and wheel flatness. In the frequency range of 1–80Hz, vibration is created which causes fault and uncomfortable for passengers. Numerous methods were introduced to predict the fault based on vibrations or some other signals. Yet those models are poor due to improper training, complex design, and poor fit for real-time utilization. To overcome these issues optimized LSTM combined with an auto encoder approach is developed to predict fault based on mechanical vibration in an effective way. Acoustic sensors are attached to a train bogie to capture the vibrations from the passing trains. After that, the encoder and decoder are the two main parts of the auto encoder, which is regarded as an unsupervised learning technique that denoises the original signal vibration. Finally, an optimized LSTM classifier is employed to classify fault and no-fault railway tracks in the pre-signals. Through the application of SGO optimization, which appropriately selected the learning rate and hidden neurons of LSTM, the performance of LSTM was enhanced. The suggested approach's effectiveness is evaluated utilizing performance metrics including specificity, recall, accuracy, and precision; the findings reveal 94.2 %, 92.5 %, 94.2 %, and 95 %. This analysis demonstrate that the suggested approach can anticipate a track's defect with accuracy using vibration data in a real-time scenario. Major Findings: The optimized LSTM combined with an autoencoder effectively predicts faults in train bogies based on mechanical vibrations, addressing issues from previous methods. Acoustic sensors capture train vibrations, and the autoencoder denoises the signal, enhancing fault detection accuracy.