115 / 2024-04-20 19:52:02

Laser Welding Design Optimization with a Multiobjective Evolutionary Algorithm Based on Incremental Support Vector Regression and Local Search

Nonlinear multiobjective optimization,Evolutionary algorithm,Support vector regression,Irregular cutting/packing problem

In the manufacturing industry, laser welding is an important production process, that is influenced by a variety of factors. Providing designers or engineers with appropriate optimal designs is critical to improving the efficiency and competitiveness of the manufacturers. Laser welding design optimization is an interesting multiobjective nonlinear optimization problem. A multiobjective nonlinear programming model is formulated for the laser welding design problem to minimize the length of the welds and the material consumption of the workpieces. An encoding mechanism is proposed to represent the decision variables. A non-dominated sorting genetic algorithm based on local search and incremental support vector regression (NSGA-II-ISVR), as a multiobjective evolutionary algorithm, is proposed. Candidate sets are periodically selected for local search based on decomposition to locate promising areas of the feasible region for further exploitation. The trained SVRs are used as surrogate models to find approximate fitness values of the solutions in the offspring population and a selection rule is used to select the well-performing solutions in the offspring population into the candidate set to reduce the function evaluations of the actual fitness values. The experimental results using four instances of laser-welded blanks demonstrate that the proposed NSGA-II-ISVR is very effective in finding good solutions as compared to four state-of-the-art baseline procedures.