2D Temperature Field Reconstruction Using Optical Frequency Domain Reflectometry and Machine-Learning Algorithms.

Area of Science:

• Optoelectronics
• Thermal imaging
• Machine learning for sensor data

Background:

• Accurate 2D temperature field reconstruction is crucial for various applications.
• Optical backscatter reflectometry (OBR) offers a promising non-contact sensing method.
• Existing OBR methods often require detailed information about fiber placement and material properties.

Purpose of the Study:

• To develop and evaluate methods for reconstructing 2D temperature fields on a sensor panel using OBR.
• To assess the performance of linear regression and feed-forward neural networks for this reconstruction task.
• To demonstrate a reconstruction approach that does not rely on precise fiber trajectory or material data.

Main Methods:

• Experimental setup utilizing an optical backscatter reflectometer to measure distributed frequency shifts.
• Development and training of a linear regression model and a feed-forward neural network.
• Training data generated by varying the temperature field and capturing corresponding thermal images of a 250 × 250 mm sensor panel.
• Reconstruction algorithms were designed to be independent of fiber trajectory, panel material properties, and fiber temperature sensitivity.

Main Results:

• The feed-forward neural network achieved a mean absolute error (MAE) of 0.086 °C.
• The linear regression model achieved an MAE of 0.118 °C.
• Both methods successfully reconstructed the 2D temperature field without prior knowledge of specific experimental parameters.

Conclusions:

• Feed-forward neural networks offer superior accuracy for 2D temperature field reconstruction using OBR compared to linear regression.
• The developed methods demonstrate the feasibility of accurate temperature mapping without detailed sensor or fiber characterization.
• This approach has potential for applications requiring non-invasive, high-resolution thermal monitoring.