Aprendizagem de máquina aplicada ao desenvolvimento de sensores virtuais e na detecção e diagnóstico de falhas para processos químicos reativos em linguagem Python

Abstract

As a result of the fast technological progress, the fourth industrial revolution came about, also known as Industry 4.0. It explores digital and physical systems with the combination of ifferent technologies such as artificial intelligence and statistics. With the implementation of these technologies, it’s possible to improve productivity and reduce operational risks. In the chemical process, for economical and quality reasons, sensors are attributed to observe and identify process variables. Soft sensors are seen as an alternative option to the problems related to traditional sensors. The complexity of the industrial process generates uncertainty and instability in the accuracy of the process and with the aim of reducing this problem, the use and applicability of these computational tools are essential. In the Industry 4.0 context, machine learning makes possible the construction of soft sensors and automatic fault detectors. Due to them, it’s achievable to identify the faults previously and measure difficult variables. This work makes the use of Artificial Neural Network and Random Forest applied in a CSTR complex reactive process through the computational tool Python. The database used was the result of math system modelling. The results showed good performance metrics of the technics applied with a shorter sample period. Better results were obtained, for the approach established, with the Artificial Neural Network in the construction of the soft sensors. For detection and faults analysis, Random Forest generated greater results. In conclusion, the tools studied are robust and applicable in complex reactive systems and they are also essential to chemical engineers who are progressively involved in the Industry 4.0 context.