目的 虚拟装配在工业中可节约生产成本，提升机械部件生产效率。现有的虚拟现实引擎缺乏自动建立碰撞体功能，无法完整还原实际装配过程中的物理属性；通用化构建零件网格实体是提升虚拟装配实用性、精确性及通用性的重要途径。方法 以批量农机部件为样本对象，设计批量预处理算法及改进采样相关算法，通过构建三维模型样本的图片数据集，训练人工智能分类检测网络，从图片中分类并检测相关参数，实现自动构建碰撞体功能。结果 经过优化算法处理训练得到的分类检测网络从图片分类零件种类的精度在98%以上，从图片检测零件各项碰撞体构建参数的精度在98%以上；未经优化处理训练的网络不收敛。结论 本研究方法可以有效提升人工智能分类检测网络的识别精度及训练效率，结合碰撞体参数化构建程序，可提升碰撞体建模精度。
Objective Virtual assembly can save production cost and improve the production efficiency of mechanical parts in industry. Due to the lack of the function of automatically creating collision body, the existing virtual reality engine can not completely restore the physical properties in the actual assembly process. Universaly building mechanical part grid entity is the important approach for improving practicability, accuracy and universality of virtual assembly.Method The batch preprocessing algorithm and the improved sampling algorithm were designed for the batch agricultural machinery parts sampling. The image data set of three-dimensional model sample was constructed to train the artificial intelligence classification detection network, which can classify and detect the relevant parameters from the image samples, and realize the function of automatic construction of collision body.Result The accuracy of the classification network trained by the optimized algorithm was more than 98% for the classification of parts from pictures, and more than 98% for the construction parameters of collision bodies from pictures. However, the network without optimized training did not converge.Conclusion This method can effectively improve the recognition accuracy and training efficiency of the artificial intelligence classification detection network, and improve the modeling accuracy of the collision body by combining with the collision body parameterization construction program.