Virtual sensing of internal forces of full-scale flexible barrier system with KAN-transformer prediction model

Abstract

Various sensors have been widely employed to measure dynamic responses and assess the performance of flexible barrier systems in mitigating rockfall disasters. However, their real-world applications are hindered by challenges, such as sensor malfunctions, power outages, and other operational disruptions, which may lead to data loss or corruption. To address these limitations, this study presents a virtual sensing approach for estimating internal forces in flexible barrier systems, employing a time series prediction model enhanced by the Transformer- Kolmogorov-Arnold Network (KAN) deep learning model. First, component-specific and impact-related parameters in numerical models of flexible barrier systems are calibrated to extract dynamic responses of critical components under rockfall impacts, forming a virtual sensing dataset. A time series prediction neural network is then trained on this dataset to model the relationship between the measured internal forces of key components and the unmeasured internal forces of other components during rockfall impacts. This virtual sensing model is designed to replace physical sensor measurements in scenarios where long-term sensor deployment is impractical. Comparative analysis with traditional models, conducted using the constructed dataset, demonstrated the superior performance of the proposed method in virtual sensing tasks. Furthermore, the effectiveness and robustness of the proposed method were validated using real sensor data from a full-scale three-span flexible barrier system subjected to rockfall impacts. The results showed that the virtual sensing model can predict internal forces with a peak error within 10% and a mean peak error below 5%, confirming its high accuracy. Overall, the proposed method exhibits significant potential for cost-effective health monitoring of numerous flexible barrier systems in mountainous regions.