Detection of Rail Defects Caused by Fatigue due to Train Axles Using Machine Learning

Railways are one of the most important modes of transportation globally, carrying millions of passengers and tons of freight every day. One of the most crucial elements of railway transit is the rail. The safety and reliability of railway operations heavily rely on the condition of the tracks. Over time, rail tracks can develop various defects due to factors such as wear, fatigue, corrosion, and environmental conditions. Because of the intensity inhomogeneity, low contrast, and noise, real-time fault detection of the railway track is a crucial and difficult task. This study introduces a rail visual detection method (RVDM) for detecting tail surface flaws and focuses on a number of important RVDM concerns. Regular railway track inspections are essential for ensuring safe and reliable train operations. Cracks, burnt wheels, ballast difficulties, rail breaks and discontinuities, loose nuts and bolts, misalignment, and super elevation generated on the rails as a result of fatigue due to train axles, pre-emptive inspections, non-maintenance, and delayed identification causes a serious threat to rail transport safety. We gathered dataset including defective and non-defective images for training and validation purposes of the model. Using image classification model from Keras from Tensorflow, we created a pre-trained model with optionally loaded weights from ImageNet. The model correctly identified 99.6% images from training dataset and 100% images from validation dataset. After application of several layers of optimization, we trained our model. It correctly predicted defective and non-defective images chosen at random. It also presents training and validation accuracy and loss graphs using Numpy library. The proposed model can present defective and non-defective parts with an accuracy of 91%.