Simulação e estimação em um processo de hipertermia com nanofluidos utilizando redes neurais informadas por física e filtro de partículas

Abstract

Hyperthermia is a promising technique for cancer treatment, attracting significant interest from the scientific community. The use of metallic nanoparticles enables enhanced heat deposition in tumors when exposed to external energy sources, such as lasers. However, there are still challenges in accurately modeling heat transfer and estimating state variables, such as temperature and heat sources, during treatments. This study investigates the heating of a nanofluid in a simulated experiment, where a nanofluid containing palladium-ceria oxide (PdCeO2) nanoparticles is heated by a near-infrared diode laser. The study proposes and an alyzes two complementary models to describe the heating process. The first model describes heat transfer in a two-dimensional domain and employs Physics-Informed Neural Networks (PINNs) trained under different architectures, along with the finite volume method, using an implicit formulation for temporal interpolation and central differences for spatial gradients. The results are verified using COMSOL software and validated against experimental data, ensuring the accuracy of the approach. The second model represents the transient average temperature increase and combines a PINN with a particle filter for state estimation. The PINN solves the heat transfer model and acts as the state evolution model in the particle filter. Synthetic and real temperature measurements, obtained from nanofluid heating experiments, are used to solve the state estimation problem. The results demonstrate that the PINN-based approach accurately predicts various experimental conditions. Furthermore, the combination of PINNs and particle filters emerges as a promising tool for modeling and controlling thermal processes in biomedical applications, such as cancer thermotherapy