The trained neural network model is tested for interpolation capability using seven warpage values (referred to as check data set) which are not used in training process. The results predicted from the ANN model are compared with those obtained by FE analyses in Table 5 for seven check sets. It is seen from Table 5 that ANN prediction is in good agreement with the FE results. This indicates that the developed ANN model has good interpolation capability and can be used as an efficient predictive tool for warpage. ANN prediction can be enhanced further by increasing the hidden layer number and the neuron number per layer if higher computational cost in network training process can be sacrificed.
參考答案:利用训练过程中没用过的七个翘曲值检验训练的神经网络模型的插值能力。 来自ANN模型预测结果与所得有限元分析做了比较（如表5）。 从表5看出,ANN的预测与有限元的分析结果符合的很好。 这表明ANN模型具有良好的插值能力,可作为一种对于翘曲预测的有效工具 。如果牺牲网络训练过程中的高运算成本，通过增加隐层的神经元数目可以提高ANN预测能力。