Multi-Scale Multi-Phase Coupling of Low-Temperature Plasma Physics Using LIF Data-Driven Physics-Informed Neural Network

Biomedical and environmental applications of low-temperature plasma rely on the diagnostics and control of reactive species generated in the open air. The species depend on the control input space of the plasma generator and the environmental parameters. However, when evaporation, plasma chemical deposition, turbulence, etc. are also involved in real-world applications, the control becomes difficult. Most plasma diagnostics can detect a limited number of species. Numerical simulations of multi-scale problems, where the nanosecond-micrometer-scale plasma dynamics must be coupled with millisecond-centimeter scale fluid dynamics, are also challenging.

In this work, a data-driven physics-informed neural network (PINN) is used to solve the problem. Based on the control inputs, the PINN predicts the concentrations of more than 100 species at steady state. The error is evaluated using all the fluid conservation laws and the k-epsilon turbulent model, chemical rate equations with evaporation and condensation, in addition to the spatially resolved LIF measurement of OH concentrations. The post-training network is thus able to predict the species concentrations and temperatures in accordance with both the experimental observations and the physical laws. Its accuracy will be tested using the LIF data of other species which are not used in the training. This method is thus the first general solution of the multi-scale and multi-phase coupling between microscopic plasma and macroscopic turbulent flow.