Abstract
Physical chemistry relies on physical and chemical principles, such as the conservation of mass, momentum, and energy, that are described in the mathematical structure of differential equations, whose solution simulates the behaviour of chemical processes. From a computational perspective, such differential equations present some challenges for both reaching the required level of calculation accuracy and the numerical techniques utilised to approximate them, especially in presence of stiffness and/or non-linearities. The use of deep neural network to handle these problems is gaining popularity especially in the industrial applications of physical chemistry such as processes modelling and control of chemical reactions. The paper focuses on popular systems of differential equation in chemical kinetics and, through some examples, offers guidelines for choosing a suitable artificial neural network architecture and training strategy (i.e. numbers of layers, number of neurons, activation function, and number of epochs). Indeed, the paper shows how the choice of neural network architecture for chemical engineering process modelling depends on various factors, including primarily the nature of the process under consideration, the quantity and kind of input and output variables, and, finally, the unique features of the physico-chemical process. The experiments carried out in this study were performed on a notebook MacBook Pro (Monterey 12.7.5), with a Dual-Core Intel Core i5 processor of 2.7 GHz with 1 processor, and 2 cores, just to give the reader an idea of what a user who does not have computers dedicated to artificial intelligence techniques can do.