摩擦焊接轴向缩短量神经网络与支持向量机预测

摘要

The accurate control of axial shortening is a key factor for precise inertia friction welding. The inertia friction welding of high-temperature alloy was numerically simulated with ABAQUS finite element (FE) software, and the effects of welding parameters on the axial shortening were investigated. According to the simulated results, two models based on support vector machine (SVM) algorithm and radial basis function (RBF) neural network were developed to predict the axial shortening. By comparing two models, it is found that RBF neural network model showed a better agreement with the FE simulations than the SVM algorithm. Therefore, the RBF neural network could be helpful for FE modeling of inertia friction welding and reducing the time cost.%轴向缩短量是惯性摩擦焊接过程中的关键参量.文中利用ABAQUS有限元软件对高温合金管材惯性摩擦焊接过程进行了模拟,获得并研究了不同焊接工艺参数下的轴向缩短量结果.基于模拟结果,分别建立了支持向量机(SVM)和径向基函数(RBF)神经网络的轴向缩短量的预测模型.两种模型的对比表明,对于该小样本的预测,RBF神经网络比SVM智能预测结果更接近有限元模拟值.因此RBF神经网络模型可以更好的辅助摩擦焊接的有限元模拟,并有效降低模拟的时间成本.