Estudo das aplicações de data science e machine learning na engenharia química

Abstract

Today, with the advancement of technology and the ability of computers to store huge amounts of data, companies and industries have sought out professionals trained to deal with the challenge of organizing and extracting meaningful insights from those data and conferring competitive advantages such as demand prediction, fault detection and exploitation of standards. The area that deals with this branch is called Data Science, and employs a combination of statistical, mathematical and computational resources to make inferences about patterns found at the data, which would hardly be detected only by human analysis. Combined with it, the use of algorithms capable of learning and predicting patterns called Machine Learning are driving the industry by introducing artificial intelligence to solve this type of problem. The advancement of computing, as well as the popularization of programming and the emergence of complete mathematical libraries has made the use of these tools increasingly everyday for chemical engineers, especially in the areas of process analysis and optimization. In this context, the work presented here aimed to perform a bibliographic and theoretical review of the current context of data in chemical engineering to elucidate the theory behind supervised machine learning algorithms, their advantages, limitations and applications before 3 case studies relevant to chemical engineering. The results showed that the accuracy of the models studied vary according to the application given to them, having, in general, more flexible algorithms (Tree Models and Artificial Neural Networks) performed better for nonlinear and classification problems, while more rigid algorithms (Simple and Multiple Linear Regression) performed better in experiments in which the nature of the relationship between the input and output variables was already known to be linear. The knowledge of physical-chemical relationships of the experiments under analysis also contributed to the construction of more accurate models.