An improved multi-sensor rail damage detection method based on geometric-numerical adaptive weight fusion and exponential Gini entropy

Railway transportation is a critical pillar affecting the economy, and the timely detection of potential rail damage is essential for its operational safety. However, most existing methods rely on a single sensor, which leads to a limited detection range and poor signal quality stability. This in turn reduces accuracy and reliability, especially when multiple damages are involved. To overcome these challenges, this paper proposes an improved acoustic emission (AE) testing method based on multi-sensor signal geometric-numerical adaptive weight fusion (GNAWF) and exponential Gini entropy (EGE). GNAWF mitigates inter-channel attenuation biases and waveform distortions by combining geometric attenuation weight derived from the attenuation effect model with numerical similarity weight based on similarity metrics, thereby enhancing signal confidence and damage-related information. By combining the Shannon entropy-based distribution sensitivity factor with the Gini index to compute the EGE feature, dual sensitivity to spectral variations is achieved, thereby enhancing detection accuracy. The proposed method is validated using a vehicle-mounted rail damage detection experimental platform. It is also compared with mainstream signal fusion, feature extraction, and damage detection methods. The results demonstrate that the proposed method exhibits significant superiority across all evaluated metrics.