Development of Multiobjective Optimization Using Graphics Processing Unit Computing in Fluid-Film Lubrication Analysis

Many applications in tribological and engineering fields require the solution of multiobjective optimization. Parallel computing is commonly applied to reduce the execution of such an optimization. This is due to the fact that most of the global search algorithms are designed to search multiple locations simultaneously. As the performance of the computing power increased daily the lubrication models are becoming more complex as well. For a multi-factor multiobjective optimization problem the execution time is still a practical barrier in engineering practices. The main purpose of the project is to develop an efficient global search algorithm using new parallel computing technique to facilitate the tribological analysis, especially when the optimum points lie in the boundary of the design space. The genetic algorithm and particle swarm optimization are to be used in this study. The development of an optimization algorithm which favors the search of extremes on the boundary of the design space is proposed. The lubrication model is gap-compensated hydrostatic bearings. The new method is based on dividing rectangle method, but favors the fitness evaluation using the values from the boundaries. The main computing tasks are to be conducted in parallel by graphics processing units to reduce the execution time of the optimization process. The multiobjective optimization can be achieved by incorporating Pareto criterion in each of the optimization process. The approach proposed in this study should be useful in dealing with many tribological design problems. The effectiveness of graphics processing units in computing and the search efficiency of the proposed optimization method are to be evaluated.

Project ID:PB10107-1753
External Project ID:NSC101-2221-E182-023