The essence of manufacturing is predictability and repeatability. However, due to the inherent nature of randomness and heterogeneity in material, and process variations in the manufacturing, the current deterministic design methods create defects or nonconformance, which are very difficult to prevent. Inclusion of uncertainty in the process design and the optimization cycle will bring better understanding of its impact on the product quality. To date, there are very few studies that incorporate risk of defects formation into the design and optimization of multi-stage metal forming process. In this study, uncertainty in process and material parameters is included in the modeling of extrusion (single machine) and hot rolling (multiple machines) processes. These models are demonstrated for process optimization that leads to more reliable design solutions where defects can be greatly reduced because of the insensitivity of the solution to the process variations.
In this dissertation, we propose a computational approach to study the influence of uncertainty on the characteristics of defects in extrusion and hot rolling processes. Calibrated deterministic physical models are built in computer codes that replace the industrial process. They are fundamental tools that utilize the efficiency and accuracy of finite element method (FEM), and provide rich physical information through proper Design of Experiments (DOE) that investigate the mechanism of defect formation. A novel extrusion speed instability model is introduced into the code for investigation of sensitivity of system response to material instability and allows extrusion at high speeds. A new control scheme is proposed, by analysis of simulation results, through an appropriate predictive model as the reliable solution for defects reduction. The most robust setting of extrusion temperature and speed is found on a changed solution space in the predictive model.
In hot rolling, a new method based on the strain-state on the surface of the rolled billet is successfully developed and applied to the investigation of risk of surface cracking in the rough rolling process. Due to the coupling nature of the strain distributions in the multiple step deformation, single machine FEM model with traditional DOE-RSM approach cannot resolve the problem. Instead, specialized multi-machine FEM code – ROLPAS, with new criterion based on surface strain state for estimation of risk of surface cracking, is used to solve the problem for low computational cost. A sequential optimization technique based on the proper selection of DOE, integrated with iteratively updated meta-models, was successfully applied to find the optimal reduction configuration of hot rolling process that greatly reduced the risk of surface cracking. By intentionally incorporating uncertainty into empirical material models and process metal models using Monte Carlo simulation and other reliability analysis methods, the risk of defect formation has been successfully assessed. Finally, the most reliable configurations of the extrusion and hot rolling processes were recommended to the industrial partners. They were found to successfully reduce the defect rate in industrial production.