Machine learning boosts performance of optical fiber sensors: a case study for vector bending sensing

Area of Science:

• Photonics and Optical Sensing
• Machine Learning Applications
• Fiber Optic Sensors

Background:

• Optical fiber sensors (OFS) exhibit complex spectral responses to external perturbations.
• This spectral information holds potential for multi-dimensional sensing applications.
• Extracting meaningful data from these complex spectra often requires identifying distinct features.

Purpose of the Study:

• To propose and demonstrate the use of machine learning (ML) for direct spectral analysis in OFS.
• To establish a statistical relationship between complex OFS spectral responses and measurands.
• To showcase ML's capability in sensor applications without prior feature extraction.

Main Methods:

• Development of a simple heterostructure-based OFS device with a capillary tube.
• Utilizing machine learning algorithms for direct signal analysis of spectral data.
• Experimental validation of the sensor's performance in directional bending detection.

Main Results:

• Achieved directional bending sensing with a simple OFS device and ML analysis.
• Demonstrated a broad dynamic range for the sensing application.
• Showed that stringent sensor interrogator requirements (wavelength, bandwidth) can be relaxed without performance loss.

Conclusions:

• Machine learning offers a powerful, generalizable method for analyzing complex spectral responses from OFSs.
• The proposed technique simplifies sensor interrogation requirements while maintaining high performance.
• This approach is adaptable to various OFSs and measurands, enhancing sensing capabilities.