Redes neurais e SVM aplicadas no desenvolvimento de sensores virtuais e detecção e diagnóstico de falhas em processos reacionais complexos

Abstract

The evolution of industry throughout the centuries has been marked by revolutions that have transformed the way products are manufactured and processes are managed. The Fourth Industrial Revolution, also known as Industry 4.0, represents a significant leap towards intelligent and connected production systems, where the integration of cutting-edge technology plays a crucial role in increasing the efficiency, productivity, and safety of industrial processes. In this context, the development of virtual sensors and fault detection and diagnosis techniques in processes stands out as a vital area of research, with the potential to improve process control and safety. In this regard, this work aimed to develop computational algorithms in Python with the objective of modeling a complex reaction system focusing on the creation of virtual sensors and fault detection and diagnosis models, using Artificial Neural Networks (ANN) and Support Vector Machines (SVM). The mathematical modeling of a CSTR reactor, operating with the Van de Vusse reaction, was implemented in Python and used to generate historical process data that served as the basis for the development of virtual sensors and fault detection and diagnosis models. The results indicate that the ANN models showed high accuracy and effectiveness in estimating the concentration of product B, while the SVM models proved unsuitable for this task. In fault detection and diagnosis, the ANN model achieved 97% accuracy, with superior performance in fault detection compared to SVM. On the other hand, SVM showed better performance in fault diagnosis, correctly identifying the specific types in more than 99.98% of cases. A possible improvement in the process would be the integration of the techniques, using ANN for detection and SVM for identifying the type of fault. The proposed models and strategies can be applied to enhance the control and monitoring of complex reaction processes and optimize reactor performance.