CFENN: A Deep Learning Framework for Image-Based Pantograph Contact Force Monitoring

Abstract

In electrified railways, the pantograph makes contact with the catenary. Accurate monitoring of this interaction is critical for reliable power delivery and safety. This paper proposes a non-contact, image-based deep learning method, the Contact Force Estimation Neural Network (CFENN), designed to estimate contact force directly from video sequences. CFENN integrates convolutional layers, a feature enhancement module, and an adaptive Long Short-Term Memory (LSTM) module, to capture spatial-temporal dynamics without relying on the model-based contact force synthesis approaches. The proposed method is validated using a hardware-in-the-loop platform under realistic conditions, including additional faulty pantograph scenario, environmental disturbances, and varying video frame rates. Experimental results demonstrate CFENN’s high robustness and accuracy, highlighting its strong potential for real-time predictive maintenance in modern railway systems.