Application of neural networks in creating hardware schemes of membrane plants

The scheme and operating principle of a membrane unit characterized by separation of polarization layerare considered. Units of this type, in addition to permeate removal, realize separate extraction of the concentrated polarization layer formed on the membrane, and flux core with a lower concentration from the central part of the unit. Analysis of the hardware scheme of the plant with a parallel arrangement of such units shows its insufficient efficiency and low productivity. The main purpose of designing new plants is to intensify the process of separating low-molecular and high-molecular components of the mixture. For this purpose, it is advisable not to mix flows with different concentrations. In order to determine the optimal configuration of the membrane plant to meet the above conditions, a mathematical model developed on the basis of an artificial neural networks was used. The neural networks were developed in the interactive development environment Google Colaboratory in the Python programming language with the use of the PyTorch framework. As a result, two networks were modeled, one of which predicted the dry matter content of the concentrate, and the other predicted the dry matter content of the lean stream.To confirm the results of the developed mathematical model, a program for calculating the concentrations of the components obtained using the material balance was proposed. The results obtained on the basis of the model were used to create the hardware scheme of the plant. An algorithm, a block diagram and a program for calculating the operating time of a membrane plant and the necessary number of membrane units for obtaining a required concentration of the final product are proposed. A certificate of state registration of the computer program has been received for the calculation program. The program interface is described.