A reinforcement learning approach for real-time adjustment of thermal imaging parameters in thermoplasmonic applications

Abstract

The application of thermoplasmonics faces challenges related to precise temperature control distribution for managing heat in heterogeneous materials. A hybrid Silicon Carbide (SiC) and Aluminium (Al) paste was developed for effective temperature control in thermoplasmonic heating-based applications. The thermal images of this hybrid paste of SiC-Al is examined for multimodal parameters to estimate the plasmonic heat.

Reinforcement learning (RL) is implemented over the SiC-Al composite's thermal images using the estimated parameters for effective heat distribution, employing optimized laser power density, irradiation time, and heating period for thermoplasmonic applications. Optimizing the RL technique reduced the temperature in the central region from 75 to 70 °C, and increased it in the peripheral area from 45 to 55 °C.

Laser intensity was changed from 100 mW to 75 mW at the center to avoid overheating the tissue, the exposure time was altered from 30s to 45s, and the positioning was 5 cm away (rather than 2 cm) from the center to allow better heat conductivity.

The structural changes thereof were verified through Raman spectroscopy by altering the vibrational modes of the samples. The heating methodology employed enabled a decreased thermal gradient of up to about 10 °C. Thus, the hybrid paste SiC-Al exhibits significant thermoplasmonic material applications.