基于全排列组合训练优选的建模参数,建立了铸态Mg-APCa系合金的人工神经网络晶粒尺寸预测模型。对比传统试探法参数所建模型,该模型具有更高的平均相关系数和更低的平均误差,对检验数据的平均预测误差为6%。此外,通过模型预测了不同Al、Ca含量对Mg-Al-Ca系铸态合金晶粒尺寸的影响。结果表明,当A1含量在2.0%~3.0%、Ca含量在2.5%~3.5%时,可获得晶粒较小的Mg-Al-Ca系铸态合金,其尺寸约为150μm。预测结果和实验结果相吻合。
An artificial neural network model was built for predicting grain sizes of as-cast Mg-Al-Ca alloys based on the modeling parameters optimized by using all permutations and combinations training. Compared with the model, whose parameters are obtained from conventional heuristic, this model has higher average correlation coefficient and lower average error. The average error of the model is 5. 5% for predicting the test data. In addition, the effects of different amount of Al and Ca on the grain size in as-cast Mg-Al-Ca alloys were investigated by using the model. The results show that when the amount range are 2. 0-3. 0% for Al and 2. 5%-3. 5% for Ca, the as-cast Mg-Al-Ca alloy with small grain can be obtained, whose grain size is about 150tma. The predicted value is in agreement with the experimental results.