Avaliação de eficiência energética em sistemas de água gelada para refinarias de óleo vegetal utilizando mineração de dados

Abstract

Energy efficiency is an ageless subject in industrial plants, mainly in refrigeration systems. Overall, these systems work by refrigerant fluid compression. Hence, it is requested power from the engine – an electric source – to compress, condense, expand and evaporate the fluid, returning to its outset state. For this thermodynamic cycle occurs in its minimal efficiency, it is necessary that the cooling capacity (kW) be higher than the provided electrical power. Nevertheless, this type of indicator is not static, depending on several process variables. Thus, the increasing use of digital technologies for equipment measurement instruments has become a competitive condition for the analyse and prediction of a huge volume of data in current time. Thereby, the main objective of this work was mining data from a chill water refrigeration circuit (Chiller) in order to take out knowledge from the system. Once the plant was fully operating, the methodology used was CRISP-DM, in which the seeming variables that influence the cooling capacity were listed, collected and treated. The target variable calculated was OCP (Operational Coefficient of Performance), which measure the ratio between heat load and the supplied energy to the cycle. Then, the OCP data was split in two operational ranges labelled “Bad” and “Regular”, in which the other variables were classified. For that, three machine learning based models were tested: Decision Tree, Random Forest and XGBoost, with all three models owning more than 80% in score metrics. XGBoost was the best model tested, which pointed out the most feature importance for OCP: inlet and outlet temperatures from evaporator, head loss in the evaporator, outlet condenser temperature, head loss in the condenser, superheat discharge temperature, inlet condenser temperature, electrical current in the engine and compression camara volume. The project aims to bring a real industrial improvement case connected with a modern resolution approach, using computational languages, like Python and SQL, highlighting which is the chemical engineer profile searched in industries nowadays.