Prescriptive analysis of NSGA-2 variants for performance optimization in constrained truss systems

Multi-objective truss optimization has garnered relatively less research attention compared to single-objective scenarios. This paper presents a prescriptive and predictive analysis of nine recent multi-objective algorithms based on the Non-dominated Sorting Genetic Algorithm-2 (NSGA-2) framework. The algorithms under consideration are NSGA-2, Dynamic Neighborhood NSGA-2 (DN-NSGA-2), Decomposition-based NSGA-2 (DNSGA-2), NSGA-2 with Adaptive Real-coded Simulated Binary Crossover (NSGA-2-ARSBX), NSGA-2 with Dual Tournament Inheritance (NSGA-2-DTI), NSGA-2 with Conflict Handling (NSGA-2-conflict), Reference Point Driven NSGA-2 (RPD-NSGA-2), Grid-based NSGA-2 (g-NSGA-2) and Restricted NSGA-2 (r-NSGA-2). The study aims to evaluate the performance and effectiveness of these algorithms in terms of solution quality, convergence, coverage and diversity. Six benchmark truss optimization problems of varying complexity, ranging from 10-bar to 120-bar trusses (10-, 25-, 37-, 60-, 72- and 120-bar), are considered. The performance of the optimizers is assessed using various indicators, including Generational Distance (GD), Inverted Generational Distance (IGD), Spacing to Extend (STE) and Hypervolume (HV). The design problem targets the minimization of both structural mass and compliance while adhering to stress constraints. Experimental results reveal that NSGA-2, NSGA-2-ARSBX, r-NSGA-2 and g-NSGA-2 consistently outperform the other algorithms, providing high-quality, diverse Pareto-optimal solutions with commendable convergence properties. These algorithms also exhibit robust performance in handling truss optimization problems.