Showing 6 results for Machine Learning
A. Kaveh, M. R. Seddighian, N. Farsi,
Volume 13, Issue 2 (4-2023)
Abstract
Despite the advantages of the plastic limit analysis of structures, this robust method suffers from some drawbacks such as intense computational cost. Through two recent decades, metaheuristic algorithms have improved the performance of plastic limit analysis, especially in structural problems. Additionally, graph theoretical algorithms have decreased the computational time of the process impressively. However, the iterative procedure and its relative computational memory and time have remained a challenge, up to now. In this paper, a metaheuristic-based artificial neural network (ANN), which is categorized as a supervised machine learning technique, has been employed to determine the collapse load factors of two-dimensional frames in an absolutely fast manner. The numerical examples indicate that the proposed method's performance and accuracy are satisfactory.
Pooya Zakian, Pegah Zakian,
Volume 14, Issue 2 (2-2024)
Abstract
In this study, the support vector machine and Monte Carlo simulation are applied to predict natural frequencies of truss structures with uncertainties. Material and geometrical properties (e.g., elasticity modulus and cross-section area) of the structure are assumed to be random variables. Thus, the effects of multiple random variables on natural frequencies are investigated. Monte Carlo simulation is used for probabilistic eigenvalue analysis of the structure. In order to reduce the computational cost of Monte Carlo simulation, a support vector machine model is trained to predict the required natural frequencies of the structure computed in the simulations. The provided examples demonstrate the computational efficiency and accuracy of the proposed method compared to the direct Monte Carlo simulation in the computation of the natural frequencies for trusses with random parameters.
M. Ilchi Ghazaan, M. Sharifi,
Volume 15, Issue 2 (4-2025)
Abstract
This paper introduces a novel two-phase metamodel-driven methodology for the simultaneous topology and size optimization of truss structures. The approach addresses critical limitations in computational efficiency and solution quality. The framework integrates the Flexible Stochastic Gradient Optimizer (FSGO) with adaptive sampling and machine learning to minimize the number of structural analyses (NSAs), while achieving lighter, high-performance designs. In Phase One, FSGO employs a dual global-local search strategy governed by Extensive Constraints (EC), a dynamic constraint relaxation mechanism to balance exploration of unconventional topologies and exploitation of optimal member sizes. By creating adaptive margins around design constraints, EC enables broader exploration of the design space while ensuring feasibility. Phase Two focuses on precision size optimization, leveraging pruned metamodels trained on critical regions of the design space to refine cross-sectional areas for the finalized topology. Comparative evaluations on benchmark planar and spatial trusses demonstrate the method’s superiority: it reduces NSAs by 22–79% compared to state-of-the-art approaches and achieves 0.04–0.7% lighter designs while eliminating up to 31% of redundant members. Results validate the framework as a paradigm shift in truss optimization, merging computational efficiency with structural innovation.
Pooya Zakian, Pegah Zakian,
Volume 15, Issue 4 (11-2025)
Abstract
This study employs Monte Carlo simulation together with a deep feedforward neural network to predict the natural frequencies of truss domes under uncertainty. Material and/or geometric properties of these structures are modeled as random variables, and their influence on the natural frequencies is examined. Monte Carlo simulation is applied to perform stochastic eigenvalue analyses of the finite element models. To reduce computational cost, a deep neural network is trained to predict natural frequencies in place of repeated eigenvalue solves, accelerating the overall simulation. Bayesian optimization is used to tune the network hyperparameters. Numerical examples show that the proposed approach substantially improves computational efficiency and predictive accuracy compared with direct Monte Carlo simulation for domes with random inputs.
P. Hosseini, M. Paknahad, A. Kaveh,
Volume 16, Issue 1 (1-2026)
Abstract
Concrete mixture design optimization has evolved from traditional trial-and-error approaches to sophisticated computational methods. This paper presents a comprehensive review of optimization techniques applied to concrete mixture proportioning, covering statistical methods (Response Surface Methodology, Taguchi method), particle packing models, machine learning algorithms (Artificial Neural Networks, Random Forest, XGBoost, Support Vector Regression), and metaheuristic optimization techniques (Particle Swarm Optimization, Genetic Algorithms, EVPS, SA-EVPS). The review synthesizes findings from over 180 published studies, with detailed analysis of recent advances in artificial intelligence applications for multi-objective optimization of mechanical properties, cost, workability, durability, environmental sustainability, and structural performance. Key findings indicate that ensemble machine learning methods achieve superior prediction accuracy (R² > 0.95) for compressive strength, while metaheuristic algorithms effectively handle multi-objective trade-offs generating Pareto frontiers. The review also identifies critical research gaps including the need for standardized datasets, interpretable AI models, integration of life cycle assessment, and field validation of optimization results. Recent developments in self-adaptive algorithms (SA-EVPS) demonstrate improved convergence and solution quality for both material and structural optimization problems.
Mr V. Jabbari, Dr H. Azizian, Dr R. Sojoudizadeh, Dr L. Rahimi,
Volume 16, Issue 2 (4-2026)
Abstract
Structural design seeks to achieve optimal performance with minimum cost while meeting code requirements. Evaluating optimized designs usually depends on finite element analysis, which is computationally expensive. Recently, surrogate models have been developed to predict structural behavior more efficiently. Among these, Support Vector Machine (SVM) has become a reliable tool in civil engineering. However, the predictive power of SVM is highly dependent on proper parameter tuning. This study introduces the Improved Electric Eel Foraging Optimization Algorithm (I-EEFO) for training SVM to estimate the response of steel frames. Two benchmark structures, a 2‑story and a 7‑story steel frame, were analyzed, and the results were compared with other metaheuristic algorithms. The proposed method achieved very high accuracy: mean squared errors of 1.11E‑13 for the 2‑story frame and 2.99E‑07 meters for the 7‑story frame over 10 runs. The root mean square errors for displacement prediction on test data were 2.67E‑07 and 7.23E‑04 meters, respectively, confirming reliable estimates. Convergence curves demonstrated that I‑EEFO converges faster and more effectively than competing methods. These findings highlight the potential of the proposed approach as a robust and computationally efficient alternative to traditional simulations, offering engineers a practical tool to reduce costs in structural design without compromising accuracy.
