采用两种基于人工神经网络（ANN）的经验学习方法，即双层感知器（DLP）模型和Elman反馈（EF）模型，分析应力腐蚀破裂（SCC）数据，预测奥氏体不锈钢在高温水（HTW）中的SCC敏感性。对304不锈钢（SS）和316SS的两组SCC数据，DLP模型经过长时间的训练周次并不收敛，而EF模型在有限的时间内收敛到一稳定值。304SS和316SS的SCC敏感性依赖于温度（T）、溶解氧浓度（DO）、氯离子浓度（[Cl-]）以及电位（E）。采用EF模型，待预测样本数据被包含在训练数组里(方法Ⅰ)比不包含(方法Ⅱ)的情况有更高的预测率。用于EF模型的SCC阈值（ThV）影响预测率，当ThV≤0.6时，对304SS而言，预测率的范围大约是0.66～0.90（方法Ⅰ），0.60～0.79（方法Ⅱ）；对316SS，预测率范围约为0.81～0.98(方法Ⅰ)，0.78～0.90（方法Ⅱ）,从预测率平均值来看，预测率服从正态分布，0.5应为最佳阈值。EF模型对定性预测ASS在高温水中的SCC行为有较高的预报率，是一个很有用的工具。
Two kinds of empirical learning methods based on artificial neural network(ANN),i.e., double layer perceptron(DLP) model and Elman feedback(EF) model,have been used to analyze SCC data and predict the SCC susceptibitlity of austenitic stainless steels in high temperature water(HTW). The results indicated that DLP model could not converge after long training epochs while EF model could reach a steady value within limited training epochs for the SCC data of stainless steels(SS).The SCC susceptibility fo 304SS and 316SS in HTW depends on the parameters such as temperature(T), dissolved O2 content(DO), chloride ion content ([Cl-]) and electrode potential(E). The threshold value(ThV) for SCC used in the EF model affected the prediction ratios. For THV<=0.6, the ranges of prediction ratio were ca.0.66-0.90for method Ⅰ(including the data to be predicted) and 0.60～0.79 for method Ⅱ(excluding the data to be predicted) for 304SS, ca.0.81～0.98 for method Ⅰ and 0.78～0.90 for method Ⅱ for 316SS. The curves of mean value of prediction ratios show that the prediction ratios have the characteristics of normal distribution and the best ThV is 0.5. The EF model is a very useful tool for qualitatively predicting the SCC behaviour of austenitic stainless steels in HTW.
参考文献
| [1] |
- 下载量()
- 访问量()
- 您的评分：
-
10%
-
20%
-
30%
-
40%
-
50%