Real-time dynamic strain sensing in optical fibers using artificial neural networks

Area of Science:

• Optical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Wavelength-scanning coherent optical time domain reflectometry (WS-COTDR) is crucial for measuring dynamic strain in optical fibers.
• Traditional data analysis methods for WS-COTDR are computationally intensive, limiting real-time applications.
• Improving the speed and accuracy of strain measurement is essential for advanced fiber sensing.

Purpose of the Study:

• To introduce artificial neural networks (ANNs) for enhanced data interpolation and signal shift computation in WS-COTDR.
• To demonstrate the superiority of ANNs over standard correlation algorithms for strain measurement.
• To enable real-time strain prediction and improve the performance of fiber optic sensing systems.

Main Methods:

• Training ANNs with synthetic data to predict signal shifts from wavelength scans.
• Applying domain adaptation techniques to align ANN predictions with real-world measurement data.
• Comparing ANN performance against standard correlation algorithms for accuracy, speed, and robustness.

Main Results:

• ANNs reduced data analysis time by over two orders of magnitude, enabling real-time strain prediction.
• Significant improvements in strain noise reduction and linearity of sensor response were observed.
• ANNs demonstrated superior performance with low signal-to-noise data, extended distance ranges, and coarser sampling settings.
• Successful demonstration of distributed ground movement measurement along a telecom fiber.

Conclusions:

• ANNs offer a powerful and efficient solution for raw measurement data interpolation and signal shift computation in fiber sensing.
• The proposed ANN-based approach significantly enhances the capabilities of WS-COTDR for dynamic strain measurement.
• These techniques are broadly applicable to various correlation and interpolation challenges in optical fiber sensing and beyond.