应用神经网络技术预测应力腐蚀开裂

利用已有的应力腐蚀开裂数据训练人工神经网络，对奥氏体不锈钢在含有氯离子和氧的溶液中的应力腐蚀开裂敏感性进行了预测。所用的网络结构为三层结构，氯离子和氧含量作为网络输入，腐蚀开裂敏感性作为输出。学习算法为反向传播算法，以预测精度作为标准，训练得到网络的优化结构为2-6-1。结果表明，该劂络对应力腐蚀的预测比较准确，用神经网络技术预测应力腐蚀开裂敏感性是可行的。

APPLICATION OF NEURAL NETWORK TECHNIQUES TO PREDICTING SCC OF AUSTENITIC STAINLESS STEELS

An artificial neural network with three layers was developed to predict SCC risk of austenitic stainless steels in Cl~-, O_2-containing solution. Concentrations of Cl~- ions and O_2 were taken as the inputs of the network, and the output was the SCC risk. The learning algorithm was back propagation algorithm. The network was trained with the given data, and the neuron of hidden layer was obtained which equals to 6. Then, the SCC risk of austenitic stainless steels in the solutions containing different concentrations of Cl~-, O_2 was predicted. Results showed that it was an appropriate method for SCC prediction.