基于人工神经网络的铝合金铸锭裂纹倾向预测

Prediction of Al alloy ingot crack tendency based on artificial neural network

在电磁半连续铸造条件下,针对不同工艺参数下铝合金圆铸锭的裂纹倾向,建立一种基于多层前馈神经网络的预测模型。网络的输入变量为铝合金铸锭的尺寸、成分以及工艺参数,输出变量为裂纹的量化值,采用改进后的带动量因子的BP训练算法,计算多组不同工艺条件下的裂纹预测值,并进行真实试铸实验。结果表明:裂纹预测结果的最大相对误差为13.9%,最小相对误差为0;在工艺指标控制范围内,模型的裂纹预测曲线能较好地反映铸锭裂纹的真实倾向。

A prediction model based on multiplayer feed-forward artificial neural networks（ANN） was developed for modeling the correlation among different process parameters and cracks tendency of Al alloy ingot. The input variables were the size, composition and process parameters of ingots. The output variable was the quantified value of ingot cracks. The model was trained by the improved BP algorithm. The results show that the maximal relative error of prediction value is 13.9% and the minimal one is 0. The prediction curve makes a good performance in reflecting the ingot crack tendency.