High Precision Decoupling and Demodulation for Temperature-Strain Measurement of Optical Fiber Fabry–Perot Sensor

To address the challenges of parameter decoupling and low demodulation accuracy in fibre-optic interferometric dual-parameter sensors, we proposes an innovative method combining empirical mode decomposition, discrete wavelet transform, and bidirectional long short-term memory (BiLSTM) networks for dual-parameter demodulation of temperature and strain. A temperature-strain dual-parameter sensor is designed based on the filling method, where polydimethylsiloxane (PDMS) is used for temperature sensing and the air cavity between the fibre ends for strain measurement. The sensor demonstrates sensitivities of 0.56 nm/°C for temperature and 28.3 pm/μϵ for strain. Empirical mode decomposition is employed to decouple the temperature and strain parameters, effectively separating the temperature spectrum and the temperature-strain spectrum without the need for predefined decomposition substrates. The discrete wavelet transform is applied to restrain noise of the signal and reduce the feature dimensionality, enhancing signal quality and compressing the model's parameters, achieving a compression ratio of up to 0.063. The BiLSTM network captures both the continuity of the wavelength sequence and the intensity variations across the full spectrum, significantly improving demodulation accuracy. The model is trained and tested using 2560 samples within a temperature range of 20 to 50°C and strain range of 0 to 750 μϵ The final root mean square error is 0.012 °C and 0.105 μϵ, demonstrating that the demodulation method effectively decouples the temperature and strain parameters, identifies the nonlinear relationship between these parameters and the spectra, and achieves high demodulation accuracy.