The Application of Kernel Ridge Regression for the Improvement of a Sensing Interferometric System

Area of Science:

• Optoelectronics
• Machine Learning
• Sensor Technology

Background:

• Interferometric sensors offer high sensitivity but are limited by short measurement ranges using traditional linear methods.
• Existing techniques for determining measurand parameters in interferometric systems often rely on linear sensitivity, restricting their effective operational span.
• Multilayer interferometric sensors are valuable but face challenges in achieving broad measurement ranges with conventional analysis.

Purpose of the Study:

• To investigate the application of Kernel Ridge Regression (KRR), a machine learning technique, for improving the measurement range of multilayer interferometric sensors.
• To demonstrate how spectral features from interferometric systems can be leveraged using KRR for enhanced parameter estimation.
• To evaluate the effectiveness of different kernel functions within KRR for temperature sensing applications.

Main Methods:

• Utilized spectral features, specifically wavelength positions and peak amplitudes from interference spectra, as input for the machine learning model.
• Applied Kernel Ridge Regression (KRR) with four distinct kernel functions to estimate temperature based on spectral features.
• Focused on transforming spectral features using kernel functions to enable non-linear regression for improved sensor performance.

Main Results:

• Kernel Ridge Regression (KRR) successfully estimated temperature using spectral features from multilayer interferometric sensors.
• Implementation of KRR with a Gaussian kernel achieved a root-mean-square error of 0.094 °C for temperature measurements.
• The measurement range for temperature was extended by a factor of eight, from a limited range to 4.5–50 °C, compared to traditional methods.

Conclusions:

• Kernel Ridge Regression (KRR) is an effective machine learning method for expanding the measurement range of interferometric sensors.
• The use of spectral features and kernel functions in KRR provides a robust approach for accurate temperature estimation in sensing systems.
• This study demonstrates a significant advancement in interferometric sensor technology by overcoming traditional range limitations through advanced machine learning.