A physics‐constrained neural electrical resistance tomography framework for computationally efficient monitoring of multiphase separation systems

Reliable monitoring of phase distribution in multiphase separation systems is crucial for ensuring the efficiency of separation and operational stability in equipment such as hydrocyclones, gravity separators, and flotation devices. Electrical resistance tomography (ERT) offers a low‐cost, non‐intrusive, and non‐invasive approach for monitoring of multiphase separation systems. However, owing to the strong dependency on the mesh resolution and the choices of regularization considered, conventional methods for obtaining phase distributions are often computationally expensive and ill‐suited for real‐time applications. In this paper, a fast and robust physics constrained neural reconstruction framework is proposed for real‐time monitoring of multiphase flow systems using ERT. In this approach, the conductivity distribution is parameterized using a single hidden layer feed‐forward neural network with in a physics‐constrained optimization framework, resulting in a mesh‐independent representation and when combined with a smart initialization strategy, the computational cost is significantly reduced. To preserve the sharp phase boundaries that are commonly encountered in multiphase separations, a sparsity constraint on the conductivity gradient is incorporated. The efficacy of the proposed method over existing methods is demonstrated using several numerical studies involving hydrocyclone and a stratified flow systems, and an experimental case study on hydrocyclone is employed to demonstrate the practical applicability of the method. From the results, it was observed that the proposed method achieves sharper phase boundary reconstruction with nearly constant computational cost with increasing mesh resolution, making it suitable for online monitoring of industrial flow systems.