Purpose The purpose of this study is to provide an accurate approach for calculating the magnetic field generated by a permanent ring magnet with a trapezoidal cross-section. Recognizing the limitations of traditional analytical and numerical methods for complex geometries, this research compares conventional approaches with machine learning techniques to evaluate their effectiveness in magnetic field prediction. Design/methodology/approach The method based on fictitious magnetic sources is applied, which allows the determination of the magnetic field around a permanent magnet with a trapezoidal cross-section. Using this approach, the magnetic field is calculated at various points in space surrounding the magnet, taking into account its specific geometry. Based on this data, different machine learning models are used to accurately predict the magnetic field distribution: linear regression, polynomial regression, random forest and kernel ridge regression. These models are trained on data obtained from a semi-numerical approach, enabling a comprehensive analysis and comparison between the results from machine learning predictions and those derived using semi-numerical method. This approach facilitates efficient and accurate modeling of the magnetic field, which is a key component of the research in the context of optimizing magnet design and its applications in various industrial fields. Findings The findings demonstrate that machine learning models, particularly ensemble and kernel-based methods, could outperform traditional methods in terms of accuracy and adaptability to complex magnetic configurations. These results underscore the potential of machine learning to address challenges posed by intricate geometries that analytical methods struggle to model effectively. Originality/value A permanent ring magnet with a trapezoidal cross-section was modeled, and its surrounding magnetic field was calculated, opening up new possibilities for applying magnets of this shape. This paper introduces an innovative use of machine learning techniques for calculating the magnetic field of a trapezoidal cross-section magnet, providing valuable insights into the potential of data-driven methods in electromagnetic analysis. The comparative evaluation presented in the study offers useful guidance on selecting appropriate methods for advanced magnetic field applications, especially in cases where traditional approaches may not be sufficient.
Vučković et al. (Thu,) studied this question.