Development of a Mathematical model for polymer selection for drilling fluids using Python

• Autor(a) principal: Magalhães, Leonor Moutinho
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/164140
• topic: Drilling fluids; Polyacrylamide; Marsh viscosity; Brookfield viscosity; Predictive model; Python, Neural Networks; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Química

This dissertation addresses the growing interest in utilizing polymeric systems for foundation element excavation in the construction industry. It develops and applies a comprehensive model to drilling muds, providing a practical tool for predicting and understanding fluid rheology, potentially improving construction timelines. Several experiments were carried out considering the effect of shear stress on the polymer chain of anionic polyacrylamides by analyzing the viscosities before and after the application of shear stress and considering the effect of adding clay to polymer solutions by evaluating Marsh and Brookfield viscosities (initial and final) as well as density, electrical conductivity, pH, and temperature in for subsequent organization into a database and introduction into Python. A neural network was built in Python, using the Pandas, NumPy, Keras, and sklearn libraries. The purposed model used a deep neural network with multiple layers, where each layer is designed to capture important information about the polymers, including characteristics such as molecular weight, anionicity, and other properties. Furthermore, it was implemented an expert network approach to deal with scenarios, where some variables are unknown, allowing the model to provide robust predictions even when incomplete data is provided. At last, two different models were achieved for the two tests carried out experimentally. The drilling fluid behavior is possible to predict considering polymer concentration, initial Brookfield viscosity, temperature, and polymers with different anionicity degrees and molecular weight. When shear stress is applied to the solution as a function of time, the predictive model reached a R2 equal to 0.97, a root mean squared error of 11 s/quart, and a percentage error of 8% and 12% for the initial and final Marsh viscosity, respectively. As for adding clay to the solution, a predictive model with R2 of 0.988, a root mean square error equal to 103 cP and a percentage error equal to 16% for the Brookfield viscosity after clay addition, were achieved. The assessment of predictive model performance with new data has revealed varying percentage errors for distinct polymers. For one polymer, the errors in predicting initial and final Marsh viscosities were 0.4% and 16.3%, respectively, while the other exhibited errors of 14.9% and 22.7%. Additionally, for Brookfield viscosity after clay addition, one polymer displayed a 10% error, while the other exhibited a 5.5% error.