基于深度学习的商用车前轴结构拓扑优化应用
Application of Deep Learning in Topological Optimization for Structure of Commercial Vehicle Front Axle
张东东 1张乐迪 2李亮亮 3姬晨阳 2赵礼辉1
作者信息
- 1. 上海理工大学 机械工程学院,上海 200093;机械工业汽车机械零部件强度与可靠性评价重点实验室,上海 200093;上海市新能源汽车可靠性评价专业技术服务平台,上海 200093
- 2. 上海理工大学 机械工程学院,上海 200093
- 3. 义和车桥有限公司技术中心,山东诸城 262200
- 折叠
摘要
汽车轻量化是促进节能减排最重要的手段之一.前轴作为商用车的簧下零部件,轻量化设计的节能减排效果尤为明显,对提升整车的市场竞争力具有重要意义.以商用车前轴为对象,利用深度学习技术进行拓扑优化,实现轻量化设计.首先,确定前轴轻量化设计目标,并借助Optistruct平台进行拓扑优化可行性分析;然后,基于Unet网络,构建适用于多工况输入的Res-CB-Unet卷积神经网络;接着,考虑前轴载荷、尺寸变化,以静力学分析结果构建网络输入数据集,以对应的拓扑构型构建标签数据集;最后,采用Adam算法训练得到用于预测前轴拓扑构型的网络模型.算例模型评估结果表明,提出的方法能够快速有效地生成前轴的拓扑构型,可获得比传统拓扑结果更优的构型,计算效率显著提升.该方法可为汽车典型结构的快速轻量化设计提供技术支持和实施路径.
Abstract
Promoting energy efficiency and emission reduction is crucial in the automotive industry,and lightweight design is a key strategy.Commercial vehicles benefit significantly from the lightweight designof the front axle,employedas a critical suspension component.This paper focuses on deep learning for topological optimization to achieve lightweight design for front axles.The approach involves establishing lightweight design objectives and performing a feasibility analysis using topological optimization through the Optistruct platform.A Res-CB-Unet convolutional neural network,based on the Unet architecture,is then constructed.The network is trained using input parameters derived from static analysis results of the initial front axle structure,considering factors such as load and dimensional variations.A corresponding dataset of labeled topological configurations for the front axle is generated.The results of this study demonstrate that the proposed method rapidly and effectively predicts the front axle's topological configuration,potentially outperforming traditional methods,thus significantly enhancing computational efficiency.This approach provides valuable technical support and a practical pathway for efficient lightweight designof typical automotive structures.
关键词
前轴/轻量化设计/深度学习/拓扑优化/Res-CB-Unet网络Key words
front axle/lightweight design/deep learning/topological optimization/Res-CB-Unet network引用本文复制引用
出版年
2024
NETL
NSTL
万方数据