Sensor-Fusion based Autonomous Railway Track Fault Detection Bot using YOLOv5s and Ultrasonic Sensing

This paper presents an efficient, low-cost, and Sensor-Fusion Based Autonomous Railway Track Fault Detection Bot Using YOLOv5s and Ultrasonic Sensing capable of identifying surface anomalies using a hybrid deep learning and sensor-based system. Conventional inspection techniques often require a lot of manual effort and can be difficult to carry out in hard-to-reach areas. Our proposed system integrates ultrasonic sensors for surface fault detection and a Raspberry Pi camera module to capture track images for advanced surface fault detection analysis using the YOLOv5s object detection model. The dataset used for training comprises 1,869 labeled images resized to 416×416 pixels, with a batch size of 4, trained over 50 epochs in 1.456 hours. Testing was performed on 251 unlabeled images using the best.pt weights. The system achieved a mean average precision ([email protected]) of 91.4%. The results demonstrate the model’s effectiveness in real-time fault detection with potential for deployment across diverse railway environments.