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

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.