High-Sensitivity Micro-Newton Force Sensor With Neural Network-Assisted Demodulation of Multimode Fiber Interferometer

Accurate measurement of small forces is essential in fields such as micro-robots, nanotechnology, and biological cells. This paper introduces a high-sensitive optical fiber sensor for micro-newton forces measurement utilizing neural network-assisted demodulation. This sensor employs a singlemode-multimode-singlemode (SMS) multimodal interference structure, in which a vulcanized silicone rubber film is embedded to achieve the force measurements. This design enables mechanical sensing at the micro-newton level. In traditional interference dip wavelength tracking methods, measurement results often exhibit nonlinear fitting, which increases the system demodulation complexity. Therefore, to enhance measurement accuracy, we introduced a Long Short-Term Memory (LSTM) algorithm combined with a Bayesian optimized LSTM algorithm for data processing. This approach effectively captures the sensor’s response characteristics under different external forces and significantly improves fitting accuracy. After optimization, our sensor achieved a measurement accuracy of 0.99982, with errors reduced to 23.87 μN. The manufacturing process of this sensor is simple and compact, providing essential technical support for the advancement of biomedicine and nanotechnology.