Otimização do processo de produção de cloro e soda cáustica com aplicação de técnicas de machine learning.

Abstract

In the chlorine, caustic soda and hydrogen production industry in Brazil, three electrolytic cell technologies are used commercially for this purpose: mercury cell, diaphragm and membrane. In cells with diaphragm technology, this plays a decisive role in cell efficiency. Its importance ranges from the energy efficiency of the cell to aspects related to operational safety. In this work, machine learning (ML) models were built to predict the performance of electrolytic cells, based on industrial data relating to diaphragm cells produced by the UCS (Chlorine Soda Unit) of the Braskem S/A factory, located in the state of Alagoas. The trained models showed satisfactory performance in predicting cell performance based on diaphragm manufacturing and cell operation data. The Random Forest model achieved superior performance in relation to the other models, with an accuracy greater than 90%, an important result that serves as a basis for the search for improving the performance of the diaphragm. This result confirms the feasibility of applying machine learning techniques in the chlorine and caustic soda production industry, enabling improvements in production processes. Another important result of the modeling developed was the obtaining of the most relevant variables for the models, making it possible to control these variables in the manufacturing and operation processes of the electrolytic cells, contributing to an increase in their performance.