Deep Learning Assisted Liquid Level Sensing Based on an MZI by Using Thin-Core Fiber With a Diaphragm

We propose a liquid level sensing scheme utilizing a Mach -Zehnder interferometer (MZI), achieving high accuracy and wide-range measurements through the application of deep learning algorithms. In the scheme, the MZI is constructed by connecting a thin-core fiber (TCF) with a core-offset in the middle of the single-mode fiber (SMF). A cylindrical cavity is fabricated using 3D printing technology, with the MZI optical fiber structure bonded to the diaphragm on the surface of the cavity. Liquid pressure is exerted on the diaphragm, resulting in the deformation of the MZI structure and inducing changes in the optical spectrum. However, traditional spectral dip tracking methods cannot achieve wide-range and high-precision liquid level sensing. By introducing deep learning Long Short-Term Memory (LSTM) and Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) models to analyze the sensing spectra, we overcome the limitations of measurement range and accuracy of the liquid level measurements, extending the measurement range to 450 mm, achieving a coefficient of determination of 0.99986, and reducing the error within 2 mm. The proposed analysis method has been successfully applied in the liquid level sensing field, demonstrating the great potential of deep learning algorithms in signal demodulation.