Machine learning identifies liquids employing a simple fiber-optic tip sensor.

Area of Science:

• Optoelectronics
• Chemical Sensing
• Machine Learning

Background:

• Accurate liquid identification is crucial in various scientific and industrial applications.
• Traditional methods can be complex, time-consuming, or require significant sample volumes.
• Developing simple, rapid, and accurate liquid identification techniques remains an ongoing challenge.

Purpose of the Study:

• To propose and validate a simple fiber-optic tip sensor system for liquid identification.
• To integrate droplet evaporation analysis with machine learning for enhanced classification.
• To demonstrate the efficacy of this combined approach for distinguishing different liquids.

Main Methods:

• A fiber-optic tip sensor was designed using a single-mode fiber and a pendant liquid droplet forming an extrinsic Fabry-Perot interferometer (EFPI).
• The evaporation process of the liquid droplet was monitored using a single-wavelength light source, capturing changes in optical interference.
• Time-transient response data was converted to image data via continuous wavelet transform and used to fine-tune pre-trained convolutional neural networks (CNNs).

Main Results:

• The fiber-optic tip sensor successfully captured the entire evaporation event of pendant liquid droplets.
• Continuous wavelet transform effectively transformed evaporation data into image representations suitable for machine learning.
• Machine learning classification models, specifically fine-tuned CNNs, achieved over 98% accuracy in identifying different liquids.

Conclusions:

• The proposed fiber-optic tip sensor system offers a simple yet highly effective method for liquid identification.
• Combining optical measurements of droplet evaporation with machine learning provides a powerful tool for chemical sensing.
• This technique demonstrates significant potential for rapid, accurate, and label-free liquid analysis.