Análise da relação entre dados de monitoramento online e offline da qualidade da água do NUPLAM

Abstract

The Center for Research in Food and Medicines (NUPLAM) is a supplementary unit of the Federal University of Rio Grande do Norte (UFRN), responsible for the production of medicines for the Brazilian Unified Health System (SUS). In addition to the manufacturing and distribution of medicines, NUPLAM also serves as an academic space for the development of research, extension projects, and teaching activities. This Undergraduate Thesis aimed to investigate the feasibility of predicting the results of offline laboratory measurements of purified water quality based on online monitoring data, focusing on the variable Total Organic Carbon (TOC), one of the main parameters required by health regulations to ensure the purity of water used in the pharmaceutical industry. Offline measurements are performed in laboratories and are accepted by regulatory bodies. Online data, on the other hand, are collected by sensors and are used only as complementary monitoring tools. The challenge proposed by NUPLAM was to establish a statistically significant relationship between these two forms of data collection and, if possible, to build an equation that would allow the estimation of offline values based on online data. To this end, time series and regression techniques were applied, with emphasis on the construction of a regression model with ARIMA errors, which are appropriate for handling autocorrelated data over time. The fitted models indicated that the online data have a significant relationship with the offline data, making it possible to build a predictive equation with good fit, low mean squared error, and satisfactory residuals. The analysis demonstrated that the online variable can be used to predict laboratory values with reasonable accuracy. The selected model presented the lowest error among the tested models, and the residual analysis reinforced the adequacy of the fit. However, it is noted that the inclusion of other variables associated with TOC measurement could further enhance the model's predictive capability. In addition to offering a practical solution to NUPLAM, this work demonstrates the real-world application of statistics in industrial contexts, showing how this science can be useful in process optimization, cost reduction, and supporting decision-making in companies, research centers, and other institutions. The study also highlights the potential of predictive approaches in the continuous monitoring of NUPLAM’s water quality, promoting greater efficiency without compromising compliance with regulatory standards.