Prediction of forming limit diagrams for steel sheets with an artificial neural network and comparison with empirical and theoretical models

Abstract

The automotive industry heavily relies on forming limit diagrams (FLDs) as essential tools for ensuring the quality and manufacturability of sheet metal components. However, accurately determining FLDs can be complex and resource-intensive due to the numerous material properties and variables involved. To address this challenge, this research employs an artificial neural network (ANN) model to predict FLDs for sheet metals, explicitly focusing on the automotive sector. The study begins by gathering material properties, including sheet thickness, yield strength, ultimate tensile strength, uniform elongation, hardening exponent, and strength coefficient. These properties serve as crucial inputs for the ANN model. Sensitivity analysis is then conducted to discern how each parameter influences FLD predictions. The ANN model is meticulously constructed, with a 6-15-22-3 structure, and subsequently trained to predict FLDs. The results are promising, as the model achieves an exceptional R-value of 0.99995, indicating high accuracy in its predictions. Comparative analysis is carried out by pitting the ANN-generated FLDs against experimental data. The findings reveal that the ANN model predicts FLDs with remarkable precision, exhibiting only a 3.4% difference for the FLD0 value. This level of accuracy is particularly significant in the context of automotive manufacturing, where even minor deviations can lead to substantial product defects or manufacturing inefficiencies. It offers a swift and reliable way of predicting FLDs, which can be instrumental in optimising manufacturing processes, reducing material waste, and ensuring product quality. In conclusion, this research contributes to the automotive manufacturing sector by providing a robust and efficient method for predicting FLDs

Numerical Optimization of Variable Blank Holder Force Trajectories in Stamping Process for Multi-Defect Reduction

“…Therefore, analytical and numerical methods for determining the FLC have been developed as listed in Table 2 . Researchers considered Punch stroke, oil pressure [ 30 ], forming rates [ 40 ], and chemical composition with temperature conditions [ 28 ] to train artificial intelligence models, while some other authors mainly considered material properties such as YS, UTS, EU, EL, etc., and supplemented them with simple engineering conditions such as R , n , t , etc., to predict FLC [ 29 , 41 , 42 ]. The current work utilized appropriate experimental data and advanced machine learning modeling to enhance predictability, enabling more efficient and cost-effective manufacturing practices by reducing the reliance on extensive physical testing.…”