Magneto-mechanical resonators (MMRs) are a promising new type of passive sensor. Their use requires the solution of a parameter identification problem in real time, in which the frequency and the damping of an oscillatory motion have to be inferred from a time sequence of deflection angles.
In this work, the authors investigate the performance of a simple convolutional neural network (CNN) for this task. The network was trained solely on synthetic data, so no measured training data are needed to set it up.
The results demonstrate that the network reconstructs the natural frequency of the resonator in very little time, but struggles to estimate the rate of damping robustly. The study thus shows where such a fast approach based on machine learning can already support the use of MMRs as sensors and where further development is needed.
The study is a collaboration between the Chair of Computational Mathematics and the Institute for Biomedical Imaging at Hamburg University of Technology, together with the University Medical Center Hamburg-Eppendorf and Fraunhofer IMTE, carried out within the Collaborative Research Centre CRC 1615 (SMART Reactors).
Finn Sommer, Sarah Reiß, Ufuk Akkaya, Sebastian Götschel, Martin Möddel, Tobias Knopp, Daniel Ruprecht (2026). Parameter Identification for Magneto-Mechanical Resonators by a Convolutional Neural Network. Proceedings in Applied Mathematics and Mechanics 26 (4), e70270.