Dynamic adjustment method of Flex-PLI knee biomechanical performance based on Physics-Data Hybrid model

Abstract

Within the pedestrian protection evaluation system utilizing the Flex-PLI, knee ligament-related metrics serve as the core parameters for injury assessment. However, the pass rate during both static and dynamic calibration of the knee exhibits considerable randomness, and current adjustment practice predominantly relies on inefficient trial-and-error method. To address this, the knee spring elongation is optimized and Physics-Data Hybrid Model (PDHM) through a physics-guided static modeling framework is developed. This framework incorporates a parameter-prediction network and a physics-weight network to dynamically regulate the physical structure in real time. Externally, differentiated data-augmentation strategies are applied to distinct data types, while a multi-scale feature extraction method combined with a feature-fusion module enhances data integration. By processing static calibration curves, the initial spring elongation is inversely validated and dynamic calibration peak values are predicted, thereby enabling precise adjustment of knee biomechanical performance. Simulation and experimental results show that the proposed PDHM improves prediction accuracy by 32.91% compared to existing models, achieving an overall MAE of 0.4286 and MAPE of 4.3%, which confirms its high predictive accuracy and robustness.