Density modelling of chlorinated brines on symbolic regression

The study of brine density is a relevant topic in the mining industry and many other areas of engineering. Although there are extensive studies that propose different models for density calculation, there is still no conclusive model, and they are possible for improvement. AI allows you to generate mathematical expressions based on training data using Symbolic Regression, which differentiates it from most AI models that are black boxes. With the help of this technology, thousands of density models made up of sensible equations are trained considering temperature and molality in four different brines, NaCl, KCl, MgCl_2, and CaCl_2,with data obtained from published databases. The results show that all the models obtained present an error (%AAD) less than 1.0 and that many of them are solid and compact and show similarities with equations published in previous works. The residue of the regressions generated is studied, concluding that the models show the need to incorporate new attributes. It is concluded that the use of AI would facilitate the obtaining of these expressions, showing the base structure and how the attributes are related. This would allow new brines or combinations of them that require mathematical expressions to understand the behaviour of the attributes and serve as a basis for their rigorous formulation.