基于逐次神经网络近似技术的轿车简化模型多学科设计优化

提出一种空间形变技术与神经网络近似技术及多学科可行性方法相结合的多学科设计优化方法,基于此方法对轿车简化模型进行多学科设计优化。该方法使用逐次递归近似技术与空间缩放技术,使得收敛空间逐步减小,近似精度逐步提高,利用此高精度的数学代理模型替代整车碰撞学科与NVH学科的有限元模型,最后,基于此代理模型进行车身多学科设计优化,使整车耐撞性得到提高,改善了车身的一阶振动特性,减轻了整车质量。

Multidisciplinary design optimization of simplified vehicle model based on sequence neural network approximation technique

This paper provides the multidisciplinary design optimization（MDO）method which combines the space deformation technique and neural network approximate technique together with MDF method.The simplified vehicle model is optimized based on the MDO method.Due to the application of sequence approximate technique and space deformation technique,the convergence is reduced,the accuracy of the approximate is improved,and the crush discipline and NVH discipline with the high-accuracy agent model are replaced.In the end,the vehicle is optimized based on the agent model,and the optimization result shows that the crashworthiness and vibration characteristics are improved and the weight of vehicle is decreased.