Machine-learning-assisted omnidirectional bending sensor based on a cascaded asymmetric dual-core PCF sensor.

Area of Science:

• Photonics
• Fiber Optics
• Machine Learning

Background:

• Bending sensors are crucial for monitoring structural health and in robotics.
• Conventional fiber optic bending sensors often require complex fabrication and post-processing.
• Accurate omnidirectional bending measurement remains a challenge.

Purpose of the Study:

• To design and demonstrate an omnidirectional bending sensor using cascaded asymmetric dual-core photonic crystal fibers (ADCPCFs).
• To integrate machine learning (ML) for predicting bending curvature and orientation.
• To overcome limitations of conventional methods by utilizing full spectral features.

Main Methods:

• Cascading two ADCPCFs with a lateral rotation angle to create a bending-sensitive structure.
• Employing machine learning algorithms to analyze transmission spectra and predict bending parameters.
• Experimental validation of the sensor's performance in measuring curvature and 360° bending orientation.

Main Results:

• The sensor demonstrated high accuracy in predicting bending direction (99.85% accuracy, 2.7° MAE) and curvature (98.08% accuracy, 0.03 m⁻¹ MAE).
• The ML-integrated system successfully predicted omnidirectional bending within 360° without post-processing.
• The approach effectively utilized global spectral features, overcoming reliance on specific spectral dips.

Conclusions:

• The cascaded ADCPCF sensor combined with ML offers a robust and accurate solution for omnidirectional bending sensing.
• This method eliminates the need for specialized fabrication steps, simplifying sensor development.
• The sensor shows significant potential for applications in structural health monitoring, robotics, medical devices, and wearables.