INTERPRETABLE PHYSICS-EMBEDDED NEURAL NETWORK FOR DYNAMIC LOAD IDENTIFICATION

This paper presents a novel approach called the Interpretable Physics-Embedded Neural Network (IPENN) to address the problem of load identification in structures with unknown parameters. In contrast to traditional methods, the IPENN method embeds physical parameters into the neurons, providing clear modeling rules and complete interpretability. Case studies demonstrate that the IPENN method not only exhibits innovation in its model architecture but also demonstrates superior generalization capability, computational accuracy, and efficiency. Additionally, the IPENN is applicable not only for load identification but also for modal parameter identification, making it versatile for addressing inverse problems in structural dynamics. By incorporating physical relevance and interpretability, the proposed method overcomes the limitations of existing artificial neural network-based load identification methods, providing a powerful tool for load identification. Consequently, this work holds significant theoretical significance and practical value for the problem of dynamic load identification.