An Algorithm for Updating the Electronic Structure of a Product Based on Evolutionary Programming with Non-Dominated Sorting

Abstract

The relevance of this work stems from the regular need to revise the composition of complex technical products during their life cycle. Replacing discontinued components, updating specifications, and accounting for budget constraints require an automated search for compromise solutions instead of labor-intensive and error-prone manual scanning of options. The aim of this work is to develop an algorithm for the multi-criteria synthesis of a sequence of operations for editing the electronic structure of a product (ESP), allowing one to obtain from the original configuration a new one close to the target ESP based on a set of criteria for improving characteristics, completeness of assembly, and cost non-exceedance. The scientific novelty lies in the development of the specialized algorithm for evolutionary programming of electronic structure of a product – a specialized version of NSGA-II in which the crossover operator is replaced by a mutation operator operating directly on the ESP tree and the sequence of ESP tree editing instructions, as well as in the post-processing procedure, including the removal of redundant instructions and ranking of Pareto-optimal solutions using the TOPSIS method with entropy weights. In addition, a specialized version of the MOEA/D algorithm, adapted to the same representation of ESP editing programs, was developed and compared with the proposed specialized NSGA-II. The conducted numerical experiment which involved automated reconfiguration of computational assemblies using the specialized algorithm demonstrated the algorithm’s convergence with respect to the hypervolume metric, the algorithm generated 17 non-dominated Pareto-front solutions. The algorithm for redundant instruction removal reduced the number of instructions in the ESP tree editing programs belonging to the constructed Pareto-front by an average of 32% of the original number of instructions. The comparison with the specialized MOEA/D showed that the specialized NSGA-II achieves a higher hypervolume value under matched computational budgets, with the advantage most pronounced for small population sizes (up to 32% relative gain at n=10) and diminishing as the population size grows (about 3% at n=100). The practical significance of the work is that the proposed approach automates the task of reconfiguring products using a limited set of components: instead of manually exploring the space of all available configurations, the designer receives a ranked list of specific ESP editing programs with their numerical scores for each criterion. The method is applicable in product lifecycle management (PLM/PDM) systems to support decision-making when updating product specifications and when planning maintenance or component replacement.