Multi-Objective Optimization for Critical Dimension Identification in Biomass Boiler CAD Models Based on Genetic Algorithms
Аннотация
Identifying critical dimensions in computer-aided design (CAD) models of large welded structures, such as biomass boilers, often relies heavily on manual expertise and lacks unified discriminative criteria. To address these issues, this study proposes a multi-objective optimization method based on an improved Non-dominated Sorting Genetic Algorithm II (NSGA-II) framework for critical dimension identification under unlabeled or sparsely labeled conditions, where pseudo-labels are used only for search guidance. A four-driver rule system—incorporating functionality, assembly relevance, geometric dominance, and tolerance sensitivity—was constructed to evaluate the dimensional importance in typical biomass boiler components, including pressure-bearing parts, flange interfaces, and welded joints. Knowledge-guided encoding, adaptive crossover, and feasible-solution repair mechanisms are further integrated to enhance convergence efficiency and solution feasibility. A bi-objective model was designed to simultaneously optimize recognition performance and computational cost. Experimental evaluations of representative parts, assemblies, and complex support structures demonstrate that the proposed method achieves high-quality recognition with pseudo-label guidance for search and manual annotations for final evaluation, improving precision by an average of 2.7% over the standard NSGA-II and significantly accelerating convergence. The results indicate consistent performance across boiler CAD models of increasing complexity and suggest potential transferability to similar welded assemblies, providing a practical approach for parameter identification and design optimization in engineering applications.