Crack prediction in sheet forming of zirconium alloys used in nuclear fuel assembly by support vector machine method

Abstract

Due to the poor formability of zirconium alloy, crack is one of its most common failure modes when the alloy is used for strips of spacer grids of nuclear fuel assembly during the sheet forming process. In this paper, the relationships between processing parameters (including punching speed, stamping clearance and blank-holder force (BHF)) and the crack risk of sheet forming process are discussed. A continuous forecasting model is constructed by using support vector machine regression (SVR) method. The maximum thinning ratio (TMTR), which is usually used as an index to quantitatively evaluate the crack risks during sheet forming, serves as the output while the processing parameters as the input variables of the forecasting model. The performances of the SVR model are assessed by six different regression metrics and results are compared with five other machine learning models. The evaluation results demonstrate that the SVR model has the best performance (with a relative mean error of 2.76% and the value of R2 0.939) and the capacity to accurately predict the maximum thinning ratio of the sheet forming process of zirconium alloys. The coupled and interactive effects of the processing parameters on the sheet forming are also analyzed through the results of the forecasting model.