Rapid Machine-Learning-Enabled Multiobjective Planar Antenna Design With Intermittent Local Design Enhancement
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
This research introduces an innovative methodology for fast electromagnetic (EM)-based multi-objective (MO) optimization of planar antennas. The presented approach is implemented within a machine learning (ML) routine that integrates neural network (NN) metamodels with intermittent gradient-based local refinement. Multiple designs are produced iteratively through optimization of the NN surrogate. The underlying search mechanism is a multi-objective evolutionary algorithm, followed by local enhancement based on approximate antenna response Jacobians derived from NN predictions. The EM data acquired in the process are then applied to iteratively enhance the model. This hybrid strategy—alternating between ML-based exploration and gradient-driven local improvement—ensures faster convergence and a more accurate reconstruction of the Pareto front than conventional ML-based techniques. The method is thoroughly validated with the help of several planar antennas. The average optimization cost is around 280 EM simulations, offering over 80% savings compared to one-shot surrogate-based methods and a notable speedup over standard ML frameworks without local refinement. Moreover, the inclusion of intermittent gradient corrections significantly enhances both convergence efficiency and Pareto set accuracy.
Fields of Study & Subjects
Advanced Multi-Objective Optimization Algorithms
Antenna Design and Optimization
Antenna Design and Analysis
Publication details
- Journal / source
- IEEE Transactions on Antennas and Propagation
- Volume
- 74
- Issue
- 6
- Pages
- 5024–5036
- Publication date
- 27 March 2026
- Language
- EN