Optimized Design of Micro-Joint Motors Using MPSO With Embedded ELM Indirect Surrogate

This paper develops a modified particle swarm optimization algorithm with an embedded extreme learning machine (ELM-MPSO), improving accuracy and efficiency for global optimization of a micro-jointed motor with multiple modes. Different from conventional optimization, ELM acts as an indirect surrogate model (indirect-SM) to drive the MPSO algorithm online. To enhance the generalization ability of ELM, a fitness grading strategy is formulated. By just training optimized samples, indirect-SM can be built into the algorithm to form an online interaction mechanism. With this mechanism, optimized particles are pre-filtered and pre-supplemented online according to grade, thereby enhancing their quality. Sobol sampling and double updating techniques are introduced to ensure particle global and local characteristics. In addition, fitness is evaluated according to its complexity during optimization, which reduces function calls. Optimization evaluation results show that the ELM-MPSO algorithm improves convergence and diversity for global optimization. An optimal candidate design is obtained by applying the proposed algorithm to representative points (RPs) optimization of a micro-joint motor. Finally, the prototype experiments demonstrate that the entire optimization strategy is effective.