Machine learning-assisted high-accuracy and large dynamic range thermometer in high-Q microbubble resonators

Area of Science:

• Optics and Photonics
• Sensing Technologies
• Machine Learning Applications

Background:

• Whispering gallery mode (WGM) resonators offer high sensitivity and fast response for precise measurements.
• Traditional WGM sensing methods often ignore valuable data from multiple resonances, limiting performance.
• Developing novel sensing strategies is crucial to fully leverage WGM resonator capabilities.

Purpose of the Study:

• To demonstrate that multimode sensing in WGM resonators contains more Fisher information than single-mode tracking.
• To investigate the potential of multimode sensing for achieving superior measurement performance.
• To develop an intelligent optical sensing system based on WGM resonators.

Main Methods:

• Utilizing a microbubble resonator as the sensing platform.
• Implementing an automated experimental setup to collect multimode spectral signals.
• Employing a generalized regression neural network (GRNN) machine learning algorithm for data analysis and temperature prediction.

Main Results:

• The multimode sensing method achieved an average temperature error of 3.8 × 10⁻³ °C within the 25.00°C to 40.00°C range.
• Demonstrated that multimode sensing contains more Fisher information compared to single-mode tracking.
• Analyzed the impact of data resources, including training data volume and temperature range variations, on prediction accuracy.

Conclusions:

• Multimode sensing significantly enhances the performance of WGM resonator-based sensors.
• The developed GRNN-based temperature detection system exhibits high accuracy and a large dynamic range.
• This work establishes a foundation for intelligent optical sensing utilizing WGM resonators.