基于KBRF算法的镍基690合金应力腐蚀裂纹扩展速率预测模型

镍基690合金广泛用于压水堆核电站核岛主设备关键部件及焊缝,高温高压水环境应力腐蚀开裂(SCC)是其潜在的失效机理。由于SCC行为影响因素多达二十余种,因此存在参数化模型预测精度不高的问题。本文通过融合随机森林机器学习算法(Random Forest, RF)与基于领域知识的MRP-386参数化模型,建立了镍基690合金SCC裂纹扩展速率KBRF(Knowledge-Based Random Forest)预测模型,结果表明,领域知识的引入增强了KBRF模型的鲁棒性,准确性较MRP-386参数化模型和RF等机器学习模型显著提高,预测结果与实验值较为接近,将应用于我国压水堆核电站镍基690合金部件及焊缝在反应堆冷却剂中的应力腐蚀裂纹扩展工程预测。

Prediction model of stress corrosion crack growth rate of nickel-based alloy 690 based on KBRF algorithm

Stress corrosion cracking (SCC) as a potential failure mechanism endangers structural integrity of the nickel-base 690 alloy components and welds that are widely used in the high temperature and high pressure water environment in pressurized water reactors (PWRs). Due to the complexity of the interweaving influences, the existing parameterized prediction models developed for SCC are limited for engineering assessment by rather lower accuracy. In this study, a Knowledge-Based Random Forest (KBRF) model was developed for predicting the SCC growth rate of the nicked-base 690 alloy through combining random forest machine learning algorithm (RF) with domain knowledge-based MRP-386 parameterized model. It was found that the robustness and accuracy of the KBRF model were significantly improved, in comparison with the MRP-386 parameterized model and the RF machine learning model by introducing domain knowledge into the machine learning modeling. The results demonstrate potential engineering application of the presented model on SCC growth rate prediction of nicked-base 690 alloy components and welds in PWRs.