Finite element model selection using Particle Swarm Optimization

This paper proposes the application of particle swarm optimization (PSO) to the problem of finite element model (FEM) selection. This problem arises when a choice of the best model for a system has to be made from set of competing models, each developed a priori from engineering judgment. PSO is a population-based stochastic search algorithm inspired by the behaviour of biological entities in nature when they are foraging for resources. Each potentially correct model is represented as a particle that exhibits both individualistic and group behaviour. Each particle moves within the model search space looking for the best solution by updating the parameters values that define it. The most important step in the particle swarm algorithm is the method of representing models which should take into account the number, location and variables of parameters to be updated. One example structural system is used to show the applicability of PSO in finding an optimal FEM. An optimal model is defined as the model that has the least number of updated parameters and has the smallest parameter variable variation from the mean material properties. Two different objective functions are used to compare performance of the PSO algorithm.

💡 Research Summary

The paper introduces a novel application of Particle Swarm Optimization (PSO) to the problem of selecting the most appropriate finite element model (FEM) from a set of competing candidates. Traditionally, FEM selection relies heavily on engineering judgment and statistical validation, which often fail to simultaneously address model complexity (i.e., the number of parameters to be updated) and the deviation of material properties from their nominal values. To overcome these limitations, the authors propose representing each candidate model as a particle within a PSO swarm. The particle’s position vector encodes the values of the model’s parameters (such as stiffness, mass, damping coefficients), while a binary flag indicates whether a given parameter is active for that model, thereby allowing models of differing dimensionalities to coexist in a common search space.

Two objective functions are defined. The first is a composite metric that penalizes both the number of updated parameters and the absolute deviation of each parameter from the mean material property, effectively encouraging parsimonious yet physically realistic models. The second is a conventional residual‑based function that minimizes the discrepancy between simulated responses and experimental measurements. By running PSO with each objective, the authors assess how well the algorithm balances model simplicity against fidelity to observed data.

A simple two‑dimensional frame structure serves as the test case. Several a priori FEMs are generated, each differing in which parameters are allowed to vary. Experimental dynamic response data are used as the reference. The PSO swarm (with carefully chosen particle count, inertia weight, and cognitive/social coefficients) iteratively updates particle velocities and positions according to the standard velocity‑position equations, guided by each particle’s personal best (pbest) and the global best (gbest). Convergence is monitored over thousands of iterations.

Results show that the composite objective function reliably identifies a model that uses the fewest parameter updates while keeping material property variations within a narrow band around the nominal values. This model also achieves an acceptable residual error, demonstrating that PSO can simultaneously satisfy multiple design criteria. In contrast, optimizing solely for residual error tends to select more complex models with many updated parameters, raising the risk of over‑fitting. Sensitivity analyses reveal that larger swarms improve global search capability at the cost of computational effort, while higher inertia weights promote exploration but slow convergence; appropriate tuning of these hyper‑parameters is essential for balanced performance.

The authors acknowledge limitations: as the dimensionality of the parameter space grows, the search space expands exponentially, potentially degrading convergence speed. Moreover, strong inter‑parameter correlations may cause the linear velocity update rule to become trapped in local minima. They suggest future work incorporating dimensionality‑reduction techniques, adaptive weight strategies, or hybrid meta‑heuristics (e.g., PSO combined with genetic algorithms) to mitigate these issues.

In conclusion, the study demonstrates that PSO offers a robust, stochastic search mechanism for FEM selection that can handle multi‑objective criteria—model parsimony and physical realism—more effectively than traditional single‑objective approaches. The methodology is validated on a benchmark structural system and shows promise for extension to more complex multi‑physics, multi‑scale engineering problems, where rapid, automated model discrimination is increasingly valuable.