Digitalization and artificial intelligence applied to modeling and analysis of processes in chemical engineering

Abstract

The digital transformation of the process industry has promoted the incorporation of advanced data analysis tools in chemical process engineering. Although traditional phenomenological models have enabled the design and optimization of reactors with a high degree of reliability, their application may present limitations related to computational time and adaptability to variable operating conditions. This study aimed to evaluate the potential of machine learning for the prediction and optimization of chemical reaction systems. The research was conducted in 2026 at the Ignacio Agramonte University (Cuba). A case study was conducted analyzing a continuous stirred tank reactor in which a first-order irreversible reaction was considered. In the first stage, a phenomenological model based on mass balances and chemical kinetics was formulated and implemented in MATLAB to simulate the behavior of the reactor and generate synthetic data under different operating conditions. Subsequently, the generated data were used to train a feedforward artificial neural network capable of predicting the outlet concentration of the reactor. Validation was performed using Statgraphics Centurion and applying analysis of variance and confidence intervals. Results show that the model accurately reproduces the reactor behavior, demonstrating that the integration of phenomenological models, machine learning techniques, and statistical analysis constitutes an effective approach for the optimization of chemical processes. The feedforward neural network model showed a high level of accuracy and generalization capability, validating its applicability as a surrogate for the phenomenological model.