基于进化神经网络的激光熔覆层质量预测  

Quality prediction of laser cladding layer based on improved neural network

作　　者：徐大鹏[1] 周建忠[1] 郭华锋[1] 季霞[1] 

机构地区：[1]江苏大学机械工程学院,镇江212013

出　　处：《激光技术》2007年第5期511-514,共4页Laser Technology

摘　　要：为了有效地控制激光熔覆层质量,采用反向传播(BP)算法建立了激光熔覆层质量(熔覆层宽度、熔覆层深度和稀释率)随激光功率、光斑直径和扫描速度变化的进化神经网络预测模型。针对BP算法存在收敛速度慢、容易陷入局部极小值及全局搜索能力弱等缺陷,采用遗传算法训练BP神经网络,取代了一些传统的学习算法,设计了基于进化神经网络的学习算法。经过理论分析和实验验证,在神经网络的输出端得到期望的线性输出,并在训练样本之外,选取了5组工艺参数检验神经网络模型的可靠性,预测结果与相应的实验结果的最大相对误差为2.14%。结果表明,运用该模型可以方便、准确地选择激光工艺参数,提高激光熔覆层的加工质量。Artificial neural networks were introduced in the area of laser cladding forming. The prediction model of surface quality in laser cladding parts, including the width, depth of cladding layer and dilution, was proposed based on the improved learned arithmetic. The model combined the global optimization searching performance of the genetic algorithm and the localization of the back propagation（BP） neural networks. Five technical parameters were selected to test the reliability of the model. The simulation and experimental results show that the evolutionary neural network based on genetic algorithm can effectively overcome the problem of falling into local minimum point. This method can get higher accuracy of prediction. It improves the measurement precision with the maximum relative error 2.14% between the predicted content and the real value.

关 键 词：激光技术 激光熔覆成形 熔覆层质量 人工神经网络 遗传算法 

分 类 号：TG156.9[金属学及工艺—热处理] TP183[金属学及工艺—金属学]