Adaptive fault diagnosis model for high-speed railway turnout using deep convolutional neural networks

Safety is crucial for high-speed railway transportation. Sensor gadgets monitor train elements to ensure safety and reliability. Accurate fault diagnosis is essential for reliable operation. Manual feature extraction is time-consuming and prone to errors. Intelligent fault diagnostics face challenges in extracting features from railway track images and identifying failures in turnout systems. This paper proposes a deep convolutional neural networks-based adaptive fault diagnosis model (DCNN-AFDM) using the Kaggle Railway Track Fault Detection dataset. DCNN-AFDM incorporates automatic feature extraction, fault type recognition, and comprehensive fault classification. It achieves rapid fault localisation by analysing 2D greyscale images of turnout current signals. The model enhances accuracy and reduces training time. Results show the DCNN-AFDM model has a 96.67% accuracy, 96.11% precision, 98.43% F1-Score, and 95.33% fault detection ratio compared to other approaches. Copyright © 2023 Inderscience Enterprises Ltd.