Neural network modeling of flow hydrodynamics in packed bed reactors

Abstract

This paper presents the results of determination of the hydrodynamic characteristics of the packed bed reactor using an artificial neural network. Experimental and model values of the pressure drop are obtained, a graphical dependence of the hydraulic resistance of the packed bed on the fictitious gas velocity is constructed. In recent years, the processes of complex separation of gas and liquid mixtures have attracted considerable attention, both in academic research and in industrial applications, since they combine the advantages of flow recirculation and cost reduction by reducing equipment units. However, the design of separation processes, especially when it comes to the packed bed reactor, is still a difficult task for chemical engineers due to the difficulties associated with obtaining process models that can reliably describe the mutually influencing processes and phenomena, including simultaneous separation reactions, heat exchange in the conditions of the hydrodynamics of the packed bed reactor. The complex behavior of the process forced to look for a very reliable and powerful tool for modeling and simulating the dynamics of the packed bed reactor. Neural networks are one of the strategies proposed for solving such problems, since they can be trained to process and describe complex phenomena. This paper presents an approach using a neural network and the laws of physics with regression analysis. The efficiency and reliability of these method are proved by numerical experiment in comparison with experimental data taking into account various initial/boundary conditions. Thus, in this paper, neural networks are developed and modeled using Wolfram Mathematica to predict the resistance of the packed bed of the column.