Temperature prediction of Bragg grating sensing based on a one-dimensional convolutional neural network

Area of Science:

• Optoelectronics
• Artificial Intelligence
• Sensor Technology

Background:

• Traditional temperature demodulation for fiber Bragg gratings (FBG) faces limitations in accuracy and speed, particularly in critical applications like subway tunnel fire monitoring.
• Existing methods, such as fitting and peak detection, can be insufficient for complex spectral data and real-time analysis.

Purpose of the Study:

• To introduce a new, highly accurate, and fast method for demodulating FBG sensing spectra using a one-dimensional convolutional neural network (1DCNN).
• To address the limitations of conventional temperature demodulation techniques in demanding environments like subway tunnels during fires.

Main Methods:

• Development and implementation of a 1DCNN model integrated with an FBG temperature measurement experimental device.
• Training the 1DCNN model using 1,800 experimental spectral data samples.
• Utilizing Adam's random optimization algorithm for efficient model training and temperature prediction.

Main Results:

• Achieved a high prediction accuracy of 99.95% with a root-mean-square deviation (RMSE) of 0.0832°C.
• Demonstrated superior performance compared to traditional maximum peak methods, as well as GRU and LSTM algorithms.
• Validated the effectiveness of the 1DCNN approach for improving measurement accuracy in FBG sensing.

Conclusions:

• The proposed 1DCNN-based demodulation method significantly enhances the accuracy and speed of FBG temperature sensing.
• This approach provides a viable high-speed demodulation solution for FBG sensing systems, meeting the demands of large-scale, real-time monitoring.
• The 1DCNN method represents a substantial advancement for applications requiring precise temperature monitoring, especially in hazardous environments.