Abstract
In bridge constructions, a large number of prefabricated components must be accurately assembled on-site. However, discrepancies frequently arise between the design specifications and the actual measurements. The placement adjustments that account for those discrepancies are essential. Traditionally, such adjustments have been addressed through trial assemblies conducted in factories, but this approach entails significant human and physical costs, prompting a strong demand for labor-saving solutions. In recent years, numerical pre-assembly systems utilizing three-dimensional measurement data have been developed. However, the placement adjustment process is still performed manually, and improving its efficiency remains a critical issue. Our previous study formulated this placement adjustment task as a multi-objective optimization problem and attempted to automate the process using a multi-objective evolutionary algorithm (MOEA). This study aims to enhance the diversity and quality of solutions by utilizing initial solutions derived from single-objective optimization solvers, employing multiple sets of weights with varying trade-offs among the objectives. Furthermore, considering the potential multimodality of this problem, where multiple distinct solutions in the decision space may yield the same objective function values, this study evaluates the effectiveness of multimodal MOEAs capable of exploring and maintaining such diverse solutions in the decision space.
