灰色双重残差修正的多芯片组件寿命预测

针对利用传统灰色模型进行多芯片组件寿命预测时存在的精度不足,以及预测精度随时间跨度增加而显著降低的问题,提出马尔科夫-尾段双重残差修正的多芯片组件寿命灰色预测方法。将在灰色GM(1,1)模型预测值基础上经马尔科夫法优化后的残差作为尾段灰色残差模型的输入值,实现双重残差修正。以对热循环加速试验条件下得到少量试验数据的影响多芯片组件寿命的电阻值进行寿命预测为例,试验结果表明,相较于传统灰色模型和神经网络预测方法,所提出方法在小样本条件下平均残差分别减小了80.469%和68.53%,预测精度得以提高,结果更加可靠,能够更准确地预测多芯片组件的寿命。

Double residual correction for life prediction of multi-chip module

Aiming at the problems of inadequate accuracy in the prediction of multi-chip module life using traditional gray model and the significant decrease in prediction accuracy with the increase of time span, a Markov-tail double residual correction method for multi-chip component life prediction based on grey model is proposed. Based on the predicted value of grey GM ( 1,1) model, the residual optimized by Markov method is used as the input value of the tail segment grey residual model to achieve double residual correction. An example is given to predict the life of multi-chip module by using resistance values which influence the life of multi-chip module with a small amount of test data obtained under the condition of accelerated thermal cycle test. The experimental results show that compared with the traditional grey model and neural network prediction methods, the average residuals of the proposed method are reduced by 80. 469% and 68. 53% under the small samples condition, the prediction accuracy is improved, the result is more reliable and the life prediction of multi-chip modules can be more accurate.