A magnetoelastic effect-guided design for micro stress/strain detector using deep learning

Stress/strain detectors with high accuracy, low hysteresis and wide range are important requirements in many application fields, such as healthcare, microfabrication, and human-machine interfaces. Here we develop a stress/strain detection method based on deep learning to efficiently estimate the value of the external stress/strain field from the magnetic domain configuration in the ferromagnetic Tb _0.27 Dy _0.73 Fe ₂ with strong magnetoelastic couplings. By using micromagnetic simulation, it is observed that the stress-driven magnetic response in Tb _0.27 Dy _0.73 Fe ₂ possesses characteristics of small hysteresis, which helps the design of high-performance sensors for dynamic detection. More importantly, our deep learning model can accurately distinguish different magnetic domains under the strain variation of ~ 1.25×10 ^-5 , realizing the measurement of microstrains on the order of ~10 ^-5 . Further analyses of feature maps demonstrate that our network can effectively extract subtle discrepancies among the different magnetic domains, thereby accurately inferring the relationship between magnetic domains and stresses/strains. This work provides a new design approach for stress detection, which is desirable for remote and nano-micrometer detection of stress/strain in special scenarios.