基于神经网络的Sn-Cu焊接腐蚀性预测与分析

Prediction and Analysis of Sn-Cu Welding Corrosion Based on Neural Network

基于神经网络技术和试验数据,构建了Sn-Cu焊接腐蚀性预测模型,并进行了试验验证和生产线应用。结果表明,该神经网络预测模型各输出参数的相对训练误差均小于3.5%、相对预测误差均小于4%;神经网络预测的焊接参数(卤化物含量0.06%、预热温度110-130-155℃、走板速度(1.6±0.1)m/min、焊接温度(263±2)℃和焊接时间4 s),可以满足无铜镜腐蚀、无铜板腐蚀和耐盐雾腐蚀性能好的现场技术要求,实现耐蚀性优异的Sn-Cu焊接,焊点的质量损失率降低了93.1%。

Based on neural network technology, according to test data, the predictive models of the corrosivity of Sn-Cu solder were built, and the test validation and application of production line was carried out. The results show that the output parameters relative training error using the neural network forecasting model is less than 3.5%, and the relative prediction error is less than 4%. The neural network was used to predict welding parameters ,halide content of 0.06%, preheat temperature of 110-130-155℃, the velocity of walk plank of（1.6±0.1）m/rain, soldering temperature of （263±2）℃ and the soldering time of 4s, meeting the technical requirements of bronze mirror corrosion , copper plate corrosion and salt spray corrosion resistance, the resistance excellent welding of the Sn-Cu solder in erosion resistance can be achieved. The mass loss rate of the solder joints decreases by 93.1%.