Evaluation Protocol Sensitivity in Frequency-Constrained Truss Optimization: A Comparative Study of Adaptive and...

Area of Science:

• Structural Optimization
• Computational Intelligence
• Engineering Optimization

Background:

• Metaheuristic algorithm benchmarking in frequency-constrained truss optimization (TPFC) is known to be sensitive to evaluation protocols.
• This sensitivity has not been systematically investigated, leading to potential biases in performance comparisons.

Purpose of the Study:

• To introduce and utilize a dual protocol benchmarking framework to quantify protocol-induced ranking bias in TPFC.
• To evaluate the impact of different evaluation protocols on the performance rankings of five metaheuristic algorithms.

Main Methods:

• A dual protocol benchmarking framework was developed for TPFC.
• Five metaheuristic algorithms (Polar Fox Algorithm, NSM-LSHADE-CnEpSin, NSM-LSHADE-SPACMA, NSM-MadDE, NSM-BO) were evaluated on five truss benchmarks.
• Two FE-based protocols with varying MaxFEs (10,000 × D and 20,000 × D) were employed.

Main Results:

• Under a standardized protocol, LSHADE variants ranked highest, while the Polar Fox Algorithm ranked lowest due to non-uniform FE allocation.
• Under an extended protocol, the Polar Fox Algorithm's performance gap reduced by 44-79%, improving its ranking.
• Ranking reversals were observed on larger-scale problems, highlighting protocol dependency.

Conclusions:

• Algorithmic rankings in TPFC are strongly dependent on evaluation protocols, not solely intrinsic algorithmic quality.
• Fixed FE ceilings can disadvantage adaptive and exploration-intensive algorithms.
• The dual-protocol framework offers a basis for protocol-aware benchmarking in constrained structural optimization.