Abstract
Soil corrosion is always a critical concern to corrosion engineering because of the economic influence of soil infrastructures as has been and has recently been the focus of spent nuclear fuel canisters. Besides corrosion protection, the corrosion prediction of the canister is also important. Advanced knowledge of the corrosion rate of spent nuclear fuel canister material in a particular environment can be extremely helpful in choosing the best protection method. Applying machine learning (ML) to corrosion rate prediction solves all the challenges because of the number of variables affecting soil corrosion. In this study, several algorithms of ML, including series individual, boosting, bagging artificial neural network (ANN), series individual, boosting, bagging Chi-squared automatic interaction detection (CHAID) tree decision, linear regression (LR) and an ensemble learning (EL) merge the best option that collects from 3 algorithm methods above. From the performance of each model to find the model with the highest accuracy is the ensemble stacking method. Mean absolute error performance matrices are shown in Fig. 15. Besides applying ML, the significance of the input variables was also determined through sensitivity analysis using the feature importance criterion, and the carbon steel corrosion rate is the most sensitive to temperature and chloride.