基于BP神经网络的模塑封材料疲劳寿命预测

Prediction of fatigue life of packaging EMC material based on BP neural networks

根据模塑封材料(EMC)疲劳实验,针对BP神经网络[反向传播神经网络(BPNN)]拟合误差与预测误差关系不稳定的应用问题,结合主成分分析法,"主动"改善网络结构,建立了基于BP神经网络的EMC材料疲劳寿命预测模型,进行了分析,并与一般的BP神经网络模型作了比较。结果表明,该方法得到的BP神经网络经过训练后能稳定表征EMC材料的各种参数与疲劳寿命间的内在关系。当网络拓扑结构为2-4-1时,预测结果稳定,预测误差平方和(SSE)为0.5623～0.0271,拟合误差(MSE)为0.0906～0.0278,具有实用性。

Based on the fatigue experimental results of packaging epoxy molding compound （EMC） material, with a focus on the application problem of the instability on fitting and prediction error of BP neural network, the prediction model of fatigue lifetime for EMC materials was established. The network structure was improved with ＂initiative way＂ by principal component analysis, And then, the relevance network model was established and analyzed and compared with general BP neural network model. The results show that the trained BP neural network can stably obtain the characterization of implied relationship between the various parameters and fatigue life of packaging EMC Material. It also show very stable forecasting results and in good utility when network topology is in the state of 2-4-1, MES is between 0.090 6～0.027 8, SSE is between 0.562 3～0.027 1.