Structured magnetorheological elastomers (MREs) are composite materials that consist of a soft elastomer matrix with embedded magnetizable particles arranged in a chain-like pattern. For real-world samples, it is infeasible to explicitly resolve the microstructure. Therefore, a multiscale approach is required. Herein, we present a framework based on physics-augmented neural networks (PANNs) [1,2] for the macroscale modeling of structured MREs taking into account their transversely isotropic behavior. The developed PANN macromodel adheres to important physical conditions and ensures objectivity, material symmetry, thermodynamic consistency, and vanishing free energy, stress, and magnetization in the absence of mechanical and magnetic loads [1]. By assuming the absence of an electrical fields and current densities, as well as a quasi-stationary process, both the microscale and macroscale problem can be solved using finite element (FE) simulations based on a variational principle under the restriction of magneto-hyperelasticity [2]. In a first step, data is generated by means of numerical homogenization. For this purpose, sampled magneto-mechanical loads are applied to a representative volume element in FE simulations. The resulting microscale variables are homogenized to create a database for training and testing the PANN macromodel. In a second step, the developed PANN is trained with the generated data by using a Sobolev training approach. During training, the model automatically detects the direction of the particle chains [3]. Using the model trained with the full training dataset allows for the accurate prediction of magnetization, mechanical stress and total stress, within the range of the training data. REFERENCES [1] Kalina, K. A., Gebhart, P., Brummund, J., Linden, L., Sun, W. and Kästner, M. Neural network-based multiscale modeling of finite strain magneto-elasticity with relaxed convexity criteria. Computer Methods in Applied Mechanics and Engineering 421, 2024. [2] Kalina, K. A., Brummund, J., Sun, W. and Kästner, M. Neural networks meet anisotropic hyperelasticity: A framework based on generalized structure tensors and isotropic tensor functions. Preprint, arXiv:2410.03378, 2024. [3] Gebhart, P. Skalenübergreifende Modellierung magneto-aktiver Polymere auf Grundlage energie-basierter Variationsprinzipien. PhD thesis. TU Dresden, 2024.