Artificial Neural Network based Vapor-Liquid Equilibrium Modeling for Simulation of Transcritical Multiphase Flows

The requirement of high power outputs and high efficiencies of combustion engines such as rocket engines, diesel engines, and gas turbines has resulted in the incremented of the system pressure close to the thermodynamically critical point. This increase in pressure often leads to the fluids becoming either transcritical or supercritical in state. This has led to increased interest in both the multi-component phase change phenomena as well as their chemical reactions. In this work, an artificial neural network (ANN) aided VLE model is coupled with a fully compressible computational fluid dynamics (CFD) solver to simulate the transcritical processes occurring in high-pressure liquid-fueled propulsion systems. The ANN is trained on Python using the TensorFlow library, optimized for inference (i.e., prediction) using ONNX Run-time (a cross-platform inference and training machine-learning accelerator), and coupled with a C++ based fully compressible CFD solver. This plug-and-play model/methodology can be used to convert any fully compressible and conservative CFD solver to simulate transcritical processes using only open-source packages, without the need of in-house VLE-based CFD development. The solver is then used to study high-pressure shock-droplet interaction in both two- and four-component systems where qualitative and quantitative agreement is shown with results based on both direct evaluation and the state-of-the-art in-situ adaptive tabulation (ISAT) method. The ANN model is faster than the direct evaluation method and the ISAT model by 4 times for the four-component shock-droplet interaction. The ANN model also shows implicit load balancing as long as the MPI decomposition is performed uniformly amongst the number of cores chosen, as the inference time for ANN predict does not change with the change in thermodynamic state, unlike traditional VLE solvers. Regarding the parallel scalability of this model, good strong scaling characteristics with number of processors is also observed.