Unorthodox Optimization Tasks for Achieving Well-Informed Design for an Automobile Industry

Engineering optimization problems from practice come with a number of complexities, which make generic and pedagogical optimization algorithms difficult to handle them in a straightforward manner. First, the problems usually involve a large number of variables and constraints. Second, the optimization task must, most often, be performed for a number of conflicting objectives. Third, practitioners need to find a near-optimal or a better-than-existing solution quickly, despite the fact that solution evaluations can be computationally expensive. Fourth, a curious practitioner is also often interested in not one, but multiple diverse yet equally good solutions, in order to compare a set of choices before selecting a single solution. Fifth, before the final acceptance of a preferred solution, they are often interested in performing a sensitivity analysis around the chosen solution, so that they are satisfied with any trade-off advantage which a neighboring solution may provide. Sensitivity is usually considered from a number of different issues, such as uncertainty in parameters and variables in implementing the solution, noise in objective and constraint functions, discrepancies between simulation models for objectives and constraints and actuality, etc. In this paper, the sensitivity of a solution is considered from the point of constraint violation. It helps the practitioners to know the maximum possible gain in critical objectives for a conceivable relaxation of the original constraints. Advanced optimization methods are used for meeting some of the above practicalities through a large-scale engineering design problem obtained from an automobile industry. Such approaches allow industrial practitioners to not only have an optimized solution at the end, but also to gather useful knowledge about the problem, which will help them lead in their profession and use the knowledge for future applications. Methods suggested in this paper are generic and can be applied to other similar engineering optimization problems.