The use of numerical simulations for tool tryout and process control is becoming increasingly prevalent. In this work, the deep drawing process of a car inner door panel of DC06 mild steel is numerically analyzed and compared with industrial process results. Five batches of DC06 material were analyzed mechanically and tribologically. Diverse tribological models were developed based on experimental strip drawing tests, where a Coefficient of Friction (CoF) was obtained as a function of contact pressure, sliding velocity, and amount of lubricant. A topography analysis was defined to compare material batches and to replicate industrial tool conditions. The simulation was fed with three tribological models: constant (CoF 0.15), Filzek pressure and velocity dependent, and TriboForm with lubrication zones. Thinning, Forming Limit Diagram (FLD) and draw-in were used as indicators for the comparison. Using the industrial tool, both FLD and draw-in were measured and compared with the numerical models. The constant model predicted the most conservative strain state and also differed most from the experimental results. The P-v-dependent and TriboForm models more accurately predicted the experimental results. This work highlights the importance of considering more complex tribological models to feed numerical simulations to yield results closer to real process conditions.

The quality and complexity demands of manufactured parts in sectors such as automotive and aeronautics lead to narrower process windows. This affects the repeatability and stability of the process, where material properties and process variations have a major impact. In bending processes, the bending angle is affected by variability in mechanical and microstructural properties, especially in high-strength materials. To address this, mechanical and microstructural characterization is crucial. This study conducted mechanical and microstructural characterization on five high-strength steels from different suppliers: three DP980 and two CP980. These materials are currently used by an industrial company in the automotive sector to manufacture a real product by means of U-bending, where a real issue of variability exists. Tensile tests were performed to quantify mechanical fluctuations. Microstructural analysis was also performed to determine the grain size and volume fractions of martensite and ferrite in the case of DP980, and ferrite, bainite, and retained austenite in the case of CP980. The largest variations were found for the hardening exponent, mean grain size, and elongation. To analyze their variability in an industrial process, U-bending tests were carried out using the five materials and the bending angle after the springback was measured. A total of 250 pieces were bent for the different materials and press strokes. Variations up to 1.25° in bending angle were found between the five batches for the same press stroke. A quantitative correlation analysis was performed to estimate the influence of the different parameters on the bending angle, where sheet thickness and tensile strength were shown to be two of the most influential parameters. Knowing this influence based on the variability of the properties, a control approach can be developed to reduce defects.

The increased complexity of geometries and the improved properties of sheet metal components result in narrower process windows, highlighting the need for better process control to minimize deviations and to ensure the production of high-quality parts. In this context, this study focuses on controlling the bending angle of a seat rail component manufactured by a renowned TIER1 company. This angle changes due to material, process fluctuations and post-forming springback. Two types of material, a cold-rolled Dual Phase DP980 steel and a Complex Phase CP980 high-strength steel, are both employed interchangeably when manufacturing this component. Variations in the mechanical properties and thickness of these two materials result in significant differences in post-springback bending angle. To tackle this challenge, various control strategies have been developed including a classical controller and a controller enhanced with a metamodel-based feedforward term. For the latter, two approaches were used: a simulation-based metamodel and an experimental data-based metamodel. Heuristic-based disturbances, reflecting both material variability and process changes (tool mounting variations, tool wear, gap changes and temperature variations), have been considered. To calibrate the new controller parameters and gains, a constrained-based genetic algorithm approach has been utilized together with a numerical virtualization of the process. After this virtual set-up, the new controllers have been tested experimentally in a real environment, using an industrial U-bending tool and a 4000 kN servomechanical press. The new controllers have proven to be an efficient method for enhancing the process robustness. A classical controller, employing a feedback control system, enabled consideration of part-to-part variations. On the other hand, the addition of a metamodel-based feedforward term facilitated anticipation of material properties and sheet thickness changes, thereby preventing scrap production.

Hybrid twins and residual-learning strategies are increasingly used to reconcile the broad coverage of physics-based simulations with the fidelity of production measurements. In industrial stamping, however, truly matched simulation-experiment cases are scarce, while large-scale monitoring data are clustered around a narrow operating window. This work proposes an industrially practical hybrid metamodel in which the discrepancy (ignorance) model is trained exclusively from approximate residuals computed at in-domain production inputs, considering a surrogate of a thermal enabled AutoForm Sigma Design of Experiments (DoE). To prevent uncontrolled extrapolation, learning and evaluation are restricted to an explicitly validated domain defined through multivariate kNN support in standardized space derived from the DoE cloud. The residual model is selected through five-fold cross-validation on 10,446 in-domain production samples and then retrained on the full approximate-residual set. A small set of five matched cases is kept as an external check based on true residuals. The resulting hybrid predictor enables the generation of synthetic, experiment-informed corrected data while retaining the DoE coverage required for downstream modelling tasks.