摘要
在实际生产中,SMT回流焊工艺通常以实验方法预测温度曲线,其高成本低效率是目前亟待解决的问题.针对温度曲线输入参数与曲线多重特征值间的非线性映射关系,提出基于BP神经网络技术的温度曲线预测模型.针对训练中出现的不足,改进了误差计算方法和权值调整方式,消除了预测样本次序对网络的影响,提高了网络训练速度.利用MAPE评估方法将网络预测结果与某公司实际生产数据进行对比,结果显示预测值满足企业生产误差精度要求,因此所建立的神经网络可以有效地进行温度曲线预测,为企业回流焊生产工艺规划提供指导.
Experiment is always the prime method in forecasting SMT soldering reflow profile,while its high cost and low efficiency make the company hard to develop.According to the nonlinear relationship between the multi input and output,reflow profile forecast model based on BP neural network was proposed.The deficiencies in training,such as error calculating and weight adjusting,are improved to eliminate the sample order's impact on the network.Moreover,the network training speed is enhanced rapidly.MAPE assessment approach is carried out to compare network prediction with the production data of a company.The results shows that predicted error meets the demand of required precision.In conclusion,BP neural network is effective and efficient in the reflow profile prediction.
出处
《东北大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
2011年第12期1749-1752,共4页
Journal of Northeastern University(Natural Science)
基金
国家自然科学基金资助项目(50875168)
关键词
回流焊
温度曲线
神经网络
BP算法
动量-自适应学习率
reflow soldering
reflow profile
neural network
BP algorithm
momentum-adaptive learning rate