Machine learning assisted high-precision temperature sensor in a multimode microcavity

Abstract

Whispering gallery mode (WGM) microcavities are excellent platforms for ultra-sensitive sensing due to high-quality factor and small mode volume. However, the conventional sensing method by tracking single-mode changes is difficult to fully utilize the sensing information, which limits the measurement precision and dynamical range. Here, we demonstrate a high-precision temperature sensor based on the multimode sensing method in a packaged microbubble resonator (PMBR). Remarkably, a low-cost broadband spectrum source is used as probe light to provide more sensing modes for high-precision measurement. Empowered by a machine learning method, the multimode spectral information are fully utilized, and the true temperature is precisely readout with mean-squared error (MSE) of 0.0138. The detection limit is lower three times than single-mode sensing method, capable of reaching 0.117 °C. In addition, the correlation coefficient (R2) between predictions and truth is as high as 0.9996 within the measurement range of 25–45 °C. With the low-cost laser source and high detection precision, this work provides a new perspective for intelligent optical microcavity sensors and their engineering applications.

Keywords

• WGM microcavity
• Multimode sensing
• Temperature sensor
• Machine Learning