基于神经网络-遗传算法的回流焊参数设定  

The Setting Approach for Reflow Soldering Parameter Based on the Improved ANN-GA

作　　者：郭瑜[1] 孙志礼[1] 袁哲[2] 潘尔顺[3] 

机构地区：[1]东北大学机械工程与自动化学院,沈阳110819 [2]沈阳建筑大学交通与机械工程学院,沈阳110168 [3]上海交通大学机械与动力工程学院,上海200240

出　　处：《机械科学与技术》2013年第8期1211-1214,1220,共5页Mechanical Science and Technology for Aerospace Engineering

基　　金：国家自然科学基金项目(50875168);教育部新世纪优秀人才支持计划项目(NCET-11-0328)资助

摘　　要：针对回流焊工艺过程中输入参数难以设定的问题,对多重质量特征值间非线性映射关系进行分析,提出一种基于改进遗传算法的输入参数设定方法。采用实数编码形式,直接表达了各基因表示的意义。应用改进的遗传算子和染色体重启机制,提高了搜索准确程度,避免了早熟现象和陷入局部最优的可能。再结合神经网络预测方案建立完整的参数设置与生产预测模型。以某公司实际生产数据为例,MAPE评估显示预设参数满足企业生产误差精度要求,因此所提出的设定方法可以有效地进行回流焊生产,为企业回流焊生产工艺规划提供指导。Currently, experiment plays the prime method in soldering reflow profile forecasting. The prime difficul- ty in reflow processing is input parameters setting. According to the nonlinear relationship between the multi input and output, an improved genetic algorithm （GA）-based input parameters setting method is proposed. Real-coding form expressed each gene＇s significances directly. Usage of improved genetic operators and chromosome restart mechanism avoids the premature phenomenon and possibility of fell into a local optimum, and improves search ac- curacy as well. The genetic operation combined with neural network prediction program built the complete parame- ter settings and production forecasting model. MAPE （ mean absolute percentage error） assessment is carried to compare GA prediction and the production data of a company. Result shows that predicted error meets the required precision. In conclusion, BP neural network is effective and efficient in reflow profile prediction.

关 键 词：回流焊 遗传算法 实数编码 适应度函数 算术交叉 

分 类 号：TN41[电子电信—微电子学与固体电子学]