基于RBF人工神经网络的喷丸成形工艺参数预测方法  

Prediction of Shot Peening Parameters Based on the RBF Artificial Neural Network

作　　者：王伟[1] 娄相芽[1] 杨永红[1] 王俊彪[1] 

机构地区：[1]西北工业大学机电学院,陕西西安710072

出　　处：《组合机床与自动化加工技术》2008年第8期43-45,49,共4页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：青年科技创新基金(W016222)

摘　　要：提出一种采用RBF(Radial Basis Function,径向基函数)人工神经网络方法近似建立薄壁金属零件喷丸变形效果与喷丸工艺参数之间的非线性不可逆关系,以预测喷丸成形所需工艺参数。根据成形试验中喷丸成形工艺参数影响喷丸成形效果的规律性分析,建立了预测喷丸成形工艺参数的RBF人工神经网络模型。选取一定的喷丸成形试验数据为样本,利用交替梯度算法对所建立的网络进行训练。利用训练后的RBF网络对试验数据进行预测,预测结果数据与实际数据之间的最大误差为4.36%,能够有效预测所需的成形工艺参数。通过与BP网络进行比较表明:基于RBF人工神经网络的喷丸成形工艺参数预测方法训练时间短、预测精度高、稳定性好。A RBF （Radial Basis Function） ANN （Artificial Neural Network） method is proposed to simulate the nonlinear irreversible relations between the shot peened deformation of sheet metal part and the shot peening parameters. The RBF network is modeled based on the experimental rules drawn from shot peen forming experiments. Sampled data is selected from peen forming experiments to train the RBF network model with an alternating gradient method. Experiments of parameter prediction are carried on with the trained RBF network exclusive those used as training samples. The predicted parameters are compared with those used in experiments, the max error of which is 4.36%, which proves the RBF method effective in predicting of shot peen forming parameters. The RBF method is compared to BP （Back-propagation） ANN method, which shows that the RBF method is more precise, stable and less time consuming on predicting of shot peen forming parameters.

关 键 词：喷丸成形 工艺参数预测 RBF人工神经网络 

分 类 号：V260.5[航空宇航科学与技术—航空宇航制造工程]