The potential of using simulation and optimization techniques in the design of sheet metal forming processes has been investigated. Optimization has been used in a variety of sheet metal forming applications. This usage has given ideas and guidelines on how to formulate an optimization problems, how to ensure its rapid convergence and how to reach a feasible process design. Furthermore one method to include the uncertainty and variation in the governing parameters has been evaluated.

For each forming process many formability problems exist which usually are associated with fracture, wrinkling and springback. The aim of a good process design is to avoid these problems. By the use of simulation based design the physical trial- and error iterations can to a large extent be replaced by virtual iterations. When optimization techniques are used in the design process the number of iterations can be very large. In this work effort has been made to develop an optimization method, Space Mapping (SM), which is less computing intensive. It has been shown that SM can drastically reduce the required computing time and that its optimal solution is close to the optimal solution obtained by the more computing intensive Response Surface Methodology (RSM). The drawback of SM is that the method is less robust compared to RSM, and that it requires a better initial design for convergence.

Each industrial sheet metal forming process exhibits a certain degree of stochastic behavior due to uncertainties and variations in material properties, geometry and other process parameters. The forming process must be designed such that it is insensitive to these uncertainties and variations, i.e. the process must be robust. One possibility to consider uncer tainties and variations in a simulation and optimization based design process is to use the Monte Carlo method, in particular in combination with response surfaces as metamodels. It has been shown in this work that a linear response surface can successfully be used to identify the important design variables and to give an estimate of the probabilistic response. It has also been found that quadratic surfaces are required for a more accurate evaluation of the response.