Particle swarm based simplex optimization implemented in a nonlinear, multiple-coupled finite-element-model for stress grading in generator end windings

Due to highly nonlinear material characteristics in combination with electrical-thermal coupled partial differential equations and the complex geometry the design of stress grading systems for large rotating machines is a difficult and time consuming process. In order to accelerate this process, a finite element model is developed. The model takes the nonlinear electrical and thermal coupled material properties into account. Furthermore it is able to calculate the electric and thermal behavior of a painted or taped stress grading system. The goal of this work is to present strategies to determine optimal stress grading-configurations for a minimization of the electrical as well as the combined electrical-thermal stress caused by the potential grading. Therefore, several numerical, global bounded optimization algorithms are implemented in the finite-element-model and analyzed regarding efficiency and effectiveness. As a result a self developed partial swarm based simplex optimization algorithm (PSBSO), is introduced which obtains the best result for this special optimization problem. This hybrid-algorithm combines the positive features of particle swarm optimization (PSO) and globalized bounded nelder-mead algorithm (GBNM).