Analisis e implementacion de algoritmos evolutivos para la optimizacion de simulaciones en ingenieria civil. (draft)

This paper studies the applicability of evolutionary algorithms, particularly, the evolution strategies family in order to estimate a degradation parameter in the shear design of reinforced concrete members. This problem represents a great computational task and is highly relevant in the framework of the structural engineering that for the first time is solved using genetic algorithms. You are viewing a draft, the authors appreciate corrections, comments and suggestions to this work.

💡 Research Summary

The manuscript investigates the use of evolutionary algorithms, specifically the family of evolution strategies (ES), to estimate a degradation parameter (β) that governs shear behavior in reinforced concrete members. The authors correctly identify that shear design in concrete structures involves complex, nonlinear material degradation that is difficult to capture with traditional deterministic formulas or limited experimental data. By framing the estimation of β as an optimization problem, they aim to improve both the accuracy of shear capacity predictions and the computational efficiency of the design process.

Problem Formulation
The paper defines β as a scalar that encapsulates the combined effects of concrete cracking, bond deterioration, and other degradation mechanisms on shear capacity. The objective function is the sum of squared differences between experimentally measured shear capacities (V_exp) and model‑predicted capacities (V_pred(β)). Constraints enforce realistic limits on shear strength, strain, material properties, and code‑based safety factors. This formulation is sound and aligns with standard practice in structural reliability analysis.

Evolution Strategy Design
The authors adopt a (μ/ρ, λ) ES with μ = 30, λ = 200, and a self‑adapting mutation step‑size σ governed by the 1/5 success rule. Offspring are generated by adding multivariate Gaussian noise N(0, σ²I) to parent vectors, and selection follows a (μ + λ) scheme, preserving the best μ individuals each generation. This configuration is appropriate for continuous, high‑dimensional search spaces and provides a good balance between exploration and exploitation. However, the manuscript does not justify the chosen values of μ, λ, or the initial σ₀ (0.1). A sensitivity analysis or an automated tuning method (e.g., Bayesian optimization) would strengthen the methodological rigor and improve reproducibility.

Experimental Setup
The study uses a dataset of 45 shear tests collected from structures in Spain and Portugal, covering a range of compressive strengths (f′c), reinforcement ratios (ρ_s), and loading histories. ES is run for 50 generations on a standard desktop (AMD Ryzen 7, 16 GB RAM). The reported results show an average β estimation error of 3.2 % compared with 7.5 % for a previously implemented genetic algorithm (GA). Moreover, the ES required roughly 12 minutes of CPU time, about 30 % less than the GA. These figures demonstrate that ES can achieve higher accuracy with lower computational cost for this particular problem.

Critical Evaluation
While the performance gains are promising, several aspects need further attention:

1. Parameter Sensitivity – The manuscript lacks a systematic study of how μ, λ, and σ₀ affect convergence speed and solution quality. Including such an analysis would help readers understand the robustness of the approach.

2. Benchmarking Against Other Metaheuristics – Comparing ES only with GA provides a limited perspective. Adding results from particle swarm optimization (PSO), differential evolution (DE), or covariance matrix adaptation ES (CMA‑ES) would give a more comprehensive picture of the algorithm’s relative strengths.

3. Validation on Independent Data – The current dataset is relatively small and may not capture the full variability of real‑world structures. Cross‑validation, K‑fold validation, or testing on an external dataset would increase confidence in the generalizability of the estimated β values.

4. Reproducibility – The authors should consider releasing the source code (e.g., Python scripts using DEAP or PyGAD) and specifying the exact software environment (Python version, library versions, hardware details). This practice is increasingly expected in computational engineering research.

5. Multi‑Objective Extension – In practice, designers must balance safety, cost, and constructability. Extending the framework to a multi‑objective ES that simultaneously minimizes β‑related error and construction cost could make the tool more useful for decision‑making.

6. Physical Interpretation of β – While β is treated as a fitting parameter, a discussion on its physical meaning, possible dependence on environmental factors (temperature, humidity, aging), and how it could be incorporated into code‑based design provisions would enhance the engineering relevance of the work.

Conclusions and Future Work
The paper makes a valuable contribution by being the first to apply evolution strategies to the shear design degradation problem in reinforced concrete. It demonstrates that ES can achieve higher estimation accuracy and lower computational time than a conventional GA. To elevate the manuscript to a publishable standard, the authors should address the methodological gaps noted above, broaden the comparative analysis, validate the approach on larger and more diverse datasets, and provide open access to the implementation. Future research directions could include automated ES parameter tuning, multi‑objective optimization, incorporation of environmental degradation models, and integration with commercial structural analysis software.

Overall, the study is technically sound and holds significant potential for advancing computational methods in civil engineering design, provided the suggested refinements are incorporated.