Application of data science to study fluorine losses in the phosphate industry

Artificial intelligence has become an attractive science for companies as it allows effective data analysis, which helps to improve the manufacturing processes. The aim of this work is to study fluorine losses in a phosphoric acid unit by applying data science methods to process data. Conductivity was used as an indirect measure of fluorine losses in each recovery cycle. After a pre-processing of the data, a Gaussian Mixture Models (GMM) clustering algorithm was applied. Two clusters were found in the data: one with limited losses, and the other with significant losses. In addition, a ratio (R) was created from measurement data to identify the level of fluorine loss compared to fluorine gain during a time step. This ratio R is used in turn to determine whether the plant generates an acceptable amount of fluorine losses.