BP-AM与RBF网络在LENS成形高度预测中的应用

The Prediction of the Building Precision in LENS Based on LS-SVM Network

金属零件的激光近净成形(Laser Engineered Net Shaping,LENS)是一种新的先进制造技术,但是其成形高度的控制一直是个难题。根据LENS工艺参数与单道成形高度之间复杂的非线性本质关系,采用了自学习率与动量因子改进的Back Propagation(BP)与径向基函数(Radial Basis Function,RBF)神经网络对其成形高度进行预测,详细对比分析了两种网络的训练速度、收敛速度与泛化能力等性能。结果表明:BP网络的训练时间较长、收敛速度较慢与泛化能力较差,而RBF网络相应性能较好,其预测结果与试验结果基本吻合,因此RBF网络更适合LENS成形高度的预测。

Laser Engineered Net Shaping （LENS） is an advanced manufacturing technology,but it is difficult to accurately control its building height.Based on the non-linear intrinsic relationship between technology parameters and the building height in LENS,the Back Propagation （BP） based network improved with Adaptive learning rate and Momentum coefficient （AM） algorithm and Radial Basis Function （RBF） networks are adopted to predict the building height.Then,their training time,convergence rate and generalization ability are comparatively analyzed.The result shows that training time of the BP-AM network is longer with slower convergence rate and poor generalization ability,but these performances are better,and its predicted results agree with experimental results.So the RBF network is more suitable for the prediction of the building height in LENS.