This work presents a Gaussian Process (GP) modeling method to predict statistical characteristics of injury kinematics responses using Human Body Models (HBM) more accurately and efficiently. We validate the GHBMC model against a 50\%tile male Post-Mortem Human Surrogate (PMHS) test. Using this validated model, we create various postured models and generate injury prediction data across different postures and personalized D-ring heights through parametric crash simulations. We then train the GP using this simulation data, implementing a novel adaptive sampling approach to improve accuracy. The trained GP model demonstrates robustness by achieving target prediction accuracy at points with high uncertainty. The proposed method performs continuous injury prediction for various crash scenarios using just 27 computationally expensive simulation runs. This method can be effectively applied to designing highly reliable occupant restraint systems across diverse crash conditions.

翻译：本研究提出了一种高斯过程建模方法，用于更准确、高效地预测基于人体模型的损伤运动学响应统计特征。我们针对50百分位男性死后人体替代模型实验验证了GHBMC模型。利用该验证模型，我们创建了多种姿态模型，并通过参数化碰撞仿真生成了不同姿态和个性化D形环高度下的损伤预测数据。随后，我们使用该仿真数据训练高斯过程模型，并采用一种新颖的自适应采样方法以提高精度。训练完成的高斯过程模型在不确定性较高的点位实现了目标预测精度，展现了其鲁棒性。所提出的方法仅需27次计算成本高昂的仿真运行，即可对各种碰撞场景进行连续的损伤预测。该方法可有效应用于设计适应不同碰撞条件下的高可靠性乘员约束系统。