Surrogate assisted artificial neural network and NSGA II framework for continuous multi-objective optimization of sheet metal blanking

Abstract

Sheet metal blanking optimization is traditionally performed using design-of-experiments (DOE) techniques that evaluate only discrete parameter combinations, limiting the identification of optimal operating conditions within the continuous design space. This study proposes a surrogate-assisted Artificial Neural Network (ANN) and NSGA-II framework for continuous multi-objective optimization of sheet metal blanking. A unified ANN surrogateab model was developed using 100 experimental observations obtained from Taguchi L25 orthogonal array experiments on four engineering materials (SS316 steel, C110 copper, AA1100 aluminium, and C26000 brass). The model predicts burr height from sheet thickness, punch–die clearance, and material type, while analytical formulations were integrated to evaluate blanking force, shearing energy, and specific work, resulting in a four-objective optimization framework. A systematic architecture search involving 420 ANN models identified the optimal 6-12-1 network, achieving an MSE of 6.71 × 10−6, RMSE of 0.00259 mm, and R2 of 0.995. Model robustness was further verified using repeated five-fold cross-validation, sensitivity analysis, and comparative evaluation against six alternative surrogate learning models. The trained ANN was coupled with NSGA-II to generate 140 continuous Pareto-optimal solutions beyond the discrete DOE grid, enabling identification of balanced knee-point solutions through normalized Euclidean distance from the utopia point. Experimental validation of twelve representative Pareto solutions demonstrated excellent agreement between predicted and measured burr heights, with a mean absolute error of 0.0023 mm and a mean absolute percentage error of 3.31%. The proposed framework provides an accurate and computationally efficient decision-support methodology for continuous multi-objective optimization of sheet metal blanking and establishes a surrogate-assisted optimization strategy that can be extended to intelligent manufacturing optimization problems following appropriate retraining and validation.