Machine Learning Assisted Fast Demodulation of Large Dynamic Range Dual-Parameter Optical Fiber Sensors

The mutual constraint between sensitivity and measurement range has always been a problem that restricts the development and application of optical fiber interference sensors. We propose a new high-sensitivity tapered dual-parameter optical fiber sensor packaged with polydimethylsiloxane (PDMS) in this article. The use of the multilayer perceptron (MLP) network for demodulation of spectrum without preprocessing improves the constraint relationship between the measurement range and sensitivity of the interferometric sensor, ensuring high sensitivity while expanding the dynamic range. By introducing principal component analysis (PCA) to reduce the dimensionality of the original data, the computational efficiency is improved, and the practicality of the model is improved. As far as we know, this is the first time that PCA and MLP have been used simultaneously to demodulate the spectrum. The detection system has advantages, such as low cost, high sensitivity, easily prepared, fully packaged, and remote real-time monitoring ability. This demodulation method can also be easily extended to demodulation of other multiparameter sensors.