AI and neural networks are commonly associated with learning from large amounts of data. Especially in complex applications in fluid mechanics, however, only limited data or parameter variations are usually available. More recent developments—so-called Physics-Informed Neural Networks (PINNs)—make use of fundamental physical equations on the basis of which the neural network is trained. Within the scope of the presentation, two industrial applications of PINNs for solving fluid-mechanical problems are presented.

The efficient removal of solid particles and liquid droplets is of crucial importance in many industrial processes. The main objective is to achieve a high separation efficiency while simultaneously minimizing the associated pressure loss. Due to the microscopic nature of filtration, gaining deeper insight through experiments is only possible to a limited extent and requires costly experimental setups. Conventional numerical approaches in fluid mechanics (CFD) require a high computational effort, which is economically impractical. In this work, a modelling approach for simulating the filtration process is presented, which couples a Physics-Informed Neural Network (PINN) with the Smoothed Particle Hydrodynamics (SPH) method.

Inline multistage rotor–stator mixers have become established as an industrial solution for rapidly and efficiently mixing highly viscous fluids while minimizing the temperature rise during the process. Classical Computational Fluid Dynamics (CFD) methods are prone to numerical diffusion and can lead to an overestimation of mixing efficiency. Therefore, a model based on Physics-Informed Neural Networks (PINNs) is presented as an alternative solution to conventional Computational Fluid Dynamics (CFD) methods.