基于多保真度融合的电池包底部碰撞侵入量预测研究OA
Prediction of Battery Pack Intrusion Depth Under Underbody Impact Using Multi-Fidelity Fusion
针对电动汽车动力电池包底部碰撞安全性评估中存在的试验成本高、仿真效率低等问题,提出一种基于多保真度融合与自适应采样的电池包底部碰撞侵入量预测方法.通过动力学分析建立圆锥形障碍物模型,结合非线性变形理论推导出侵入量与碰撞参数的解析关系.采用拉丁超立方采样策略构建多保真度数据集,并基于高斯过程回归(GPR)建立两级多保真融合模型(EMFGP),引入主动学习优化.数值算例验证表明,该方法在二维数学算例中归一化均方根误差较传统多保真神经网络(MFNN)降低62.7%.工程实例中,模型对8维参数(碰撞速度、角度、底板厚度等)的侵入量预测误差低于5%,且计算耗时仅为MFNN的10%.研究揭示了侵入量与障碍物尖锐度呈正相关,与底板厚度、加强筋高度呈负相关的规律,为电池包底部防护设计提供了高效数字化工具.
To address the challenges of high experimental costs and low simulation efficiency in safety assessment of electric vehicle battery packs under underbody impact,this study proposes an intrusion depth prediction method based on multi-fidelity fusion and adaptive sampling.A conical obstacle model was established through dynamic analysis,and the analytical relationship between intrusion depth and collision parameters was derived using nonlinear deformation theory.Latin hypercube sampling was employed to construct a multi-fidelity dataset,while a two-level enhanced multi-fidelity Gaussian process(EMFGP)model was developed and further optimized through active learning.Numerical validation show that,for 2D benchmark problems,the proposed method reduces the normalized root-mean-square error by 62.7%compared to conventional multi-fidelity neural networks(MFNN).In the engineering application,the model predicts intrusion depth for an 8-dimensional parameter space,including impact velocity,impact angle and underbody panel thickness,with an error of less than 5%,while requiring only 10%of the computational time of the MFNN.The results reveal that intrusion depth is positively correlated with obstacle sharpness and impact angle,and negatively correlated with underbody panel thickness,providing an efficient digital tool for the design of battery pack underbody protection.
鲁俊;蒋胜;杨鑫;史方圆;白芳华;毛溶洁;刘学文
重庆师范大学 国家应用数学中心,重庆 401133重庆师范大学 国家应用数学中心,重庆 401133招商局检测车辆技术研究院有限公司 安全及可靠性测试研究事业部,重庆 401329重庆长安汽车股份有限公司 智能汽车安全技术全国重点实验室,重庆 400023招商局检测车辆技术研究院有限公司 安全及可靠性测试研究事业部,重庆 401329重庆长安汽车股份有限公司 智能汽车安全技术全国重点实验室,重庆 400023重庆师范大学 国家应用数学中心,重庆 401133
交通工程
动力电池包底部碰撞机器学习多保真度融合
power battery packunderbody collisionmachine learningmulti-fidelity fusion
《汽车工程学报》 2026 (5)
745-757,13
重庆市自然科学基金创新发展联合基金项目(CSTB2024NSCQ‐LZX0042)
评论