Multi‐fidelity EM simulations and constrained surrogate modelling for low‐cost multi‐objective design optimisation of antennas

Abstract

In this study, a technique for low‐cost multi‐objective design optimisation of antenna structures has been proposed. The proposed approach is an enhancement of a recently reported surrogate‐assisted technique exploiting variable‐fidelity electromagnetic (EM) simulations and auxiliary kriging interpolation surrogate, the latter utilised to produce the initial approximation of the Pareto set. A bottleneck of the procedure for higher‐dimensional design spaces is a large number of training data samples necessary to construct the surrogate. Here, the authors propose a procedure that allows us to confine the model domain to the subset spanned by the reference points, including the extreme Pareto‐optimal designs obtained by optimising the individual objectives as well as an additional design that determines the Pareto front curvature. Setting up the surrogate in the constrained domain leads to a dramatic reduction of the required number of data samples, which results in lowering the overall cost of the optimisation process. Furthermore, the model domain confinement is generic, i.e. applicable for any number of design goals considered. The proposed technique is demonstrated using an ultra‐wideband monopole antenna optimised with respect to three objectives. Significant reduction of the design cost is obtained as compared to the reference surrogate‐assisted algorithm.

Fast Multi-Objective Optimization of Antenna Structures by Means of Data-Driven Surrogates and Dimensionality Reduction

“…Additional reduction of the computational cost can be obtained by using variable-fidelity EM simulations. One of possible realizations is to construct the surrogate at the level of coarse-discretization EM model [54], [55]. This renders considerable savings (acquisition of the training data is the single most expensive component of the optimization procedure) but requires an additional refinement step in order to account for the misalignment between the low-and high-fidelity models.…”

“…Unfortunately, design of modern antennas often requires handling at least medium number of parameters (>10), where rendering reliable surrogate within the entire space is not possible. A viable alternative is an appropriate confinement of the search space along with the employment of variable-fidelity EM simulations [54], [55]. The keystone here is an identification of the region encapsulating the Pareto front and a construction of the replacement model therein.…”

“…The keystone here is an identification of the region encapsulating the Pareto front and a construction of the replacement model therein. This approach demonstrably yields significant computational benefits [55] despite the fact that the space reduction techniques usually provide only a rough approximation of the Pareto front.…”

Machine Learning-assisted Antenna Design optimization: A Review and the State-of-the-art

“…The primary challenge with machine learning-assisted multi-fidelity antenna optimization methods has always been how to efficiently ensure the uniqueness of the parameter extraction from the low-fidelity design space to the highfidelity design space [26]- [28]. In recent times, a number of methods have been proposed to adequately overcome this bottleneck for various use cases [26], [27], [29]. Some of the latest methods are discussed as follows based on their innovations.…”

Improved-Efficacy Optimization of Compact Microwave Passives by Means of Frequency-Related Regularization

“…Given the aforementioned challenges, it is no surprise that the development of methods for accelerating EM-driven design procedures has been widely researched over the last decades. The available techniques include gradient-based routines expedited by adjoint sensitivities [32], [33] or sparse Jacobian updates [34], [35], as well as surrogate-assisted algorithms involving approximation models [36]- [38] and variable-fidelity simulations [39]- [41]. A representative example of the latter is space mapping [42] widely used in microwave engineering [43].…”