论文部分内容阅读
影响大尺寸高强度U型折弯件回弹的变量众多,使得其弯曲回弹难以控制。提出一种基于约束的高斯过程潜变量(RGPLVM)筛选技术来进行最优变量的筛选和降维。将筛选出的变量作为决策变量,并以最小二乘支持向量机(LSSVM)为基础,构建了大尺寸高强度U型折弯件的最小二乘支持向量机(LSSVM)元模型。分别以支持向量机(SVM),LSSVM和BP神经网络为模型进行预测,并将预测结果与实际工程零件进行对比。结果表明LSSVM模型的预测结果更为接近实际零件的回弹情况,从而验证了所提方法的可行性。
A large number of variables that affect the resilience of a large-size high-strength U-shaped bending part make it difficult to control the bending rebound. A constrained Gaussian process latent variable (RGPLVM) screening technique was proposed to select and reduce the optimal variables. Based on the LSSVM, a LSSVM meta-model of large-size and high-strength U-shaped bending part is constructed using the selected variables as decision variables. SVM, LSSVM and BP neural network are respectively used as models to predict, and the prediction results are compared with the actual engineering parts. The results show that the LSSVM model predicts more close to the actual parts of the springback, which verifies the feasibility of the proposed method.