Knowledge-enhanced optimization of a mobile parallel robot considering topological structure and dimensions

Abstract

Mobile parallel robots, formed by mounting parallel mechanisms on AGVs, combine merits of both automatic transporting and high-precision operation of large-scale parts. Optimal design of mobile parallel robots faces two key challenges: (1) co-optimizing topology of AGV frame and dimension of parallel module, (2) difficulty in embedding domain knowledge to the conventional “black-box” optimization. To address these, we propose a knowledge-enhanced optimization framework. Kinematic, stiffness, and mass models of parallel module, as well as surrogate model of AGV frame are established. A knowledge base is built by explicit knowledge from engineering and knowledge extraction through cluster analysis. Hard and soft constraints are defined, in which soft constraints are embedded to objective. Initial population of optimization searching is determined by preference-domain cluster, and a knowledge-guided multi-objective algorithm is developed. Comparative study to the optimization without knowledge injection show that Pareto front solutions of our method concentrate densely in designer-preferred regions. It achieves 36.45 % higher preference-aligned performance and shows accelerated convergence with stabilized HV progression curves. The framework effectively addresses co-optimization challenges by knowledge fusion and preference guidance.