A hybrid modeling framework for quality prediction in sheet hydroforming process: application of regression analysis and machine learning algorithm

In recent years, the sheet hydroforming process has received widespread attention as one of the new and efficient methods in metal forming. The complexity of material behavior at different temperatures and strain rates has made the need for accurate modeling and reliable prediction of results more apparent than ever before. In the current research work, using the Johnson-Cook material model, the temperature and strain-rate dependent plastic behavior was simulated in the Abaqus software environment to more accurately represent the sheet deformation obtained from the hydroforming process. Three key process variables—including fluid pressure, sheet blank-holder force, and die edge radius—were selected at three different levels, and 27 numerical simulations were performed based on a full-factorial design of experiment. To evaluate the process performance, two quality indices, maximum thinning and thickness uniformity, were considered as the main quality criteria. The relative importance of each index was then determined using the Shannon entropy method (thinning about 65% and uniformity about 35%). Next, two prediction models—including a regression model and a K-nearest neighbor machine learning algorithm—were employed to identify the relationships between inputs and outputs and predict intermediate values. Based on the FE simulation outcomes, the optimal thinning and uniformity rates were improved by about 47% and 80%, respectively, compared to the worst case. The simultaneous study of two key indicators, combined with the application of the aforementioned methods, can lead to more accurate predictions of the quality of hydroformed components.