Modelagem do sistema de bombas centrífugas submersas sob escoamento multifásico usando bond graphs

Author(s)

Felipe de Castro Teixeira Carvalho

• Document type: Doctoral Thesis
• Press: Campinas, SP.
• Institution: Universidade Estadual de Campinas (UNICAMP). Faculdade de Engenharia Mecânica
• Defense date: 2024-01-19

Examining board members

Alberto Luiz Serpa; Ricardo Augusto Mazza; Alberto Costa Nogueira Junior; Luiz Carlos Sandoval Goes; Diego Eckhard

• Advisor: Alberto Luiz Serpa

Abstract

Electrical Submersible Pumps (ESPs) are extensively utilized in industries requiring high flow rates and boosting pressures. In the oil and gas sector, ESPs frequently handle multiphase flows, including oil-water emulsions. The non-Newtonian behavior of the emulsions can lead to system instabilities, resulting in a dynamic behavior. However, existing research primarily targets the steady-state behavior of ESPs under emulsion conditions. The primary objective of this research is to develop and validate a dynamic model for ESPs using bond graph theory, to conduct identifiability analysis, and to estimate the system parameters with a limited number of sensors. The adopted methodology encompasses collecting experimental data under both stationary and dynamic conditions, modeling the ESP using bond graphs, and using Physics-Informed Neural Networks (PINNs). The ESP model equations were obtained using a library developed in this work in Julia. The identifiability analysis conducted on the model considered the structural and the practical. The PINNs were employed to address the inverse problem and evaluated using simulated and experimental data. The model was validated by comparing experimental data with numerical simulations. This study presents a bond graph-based dynamic model for ESPs that incorporates mechanical and hydraulic subsystems, resulting in a set of Ordinary Differential Equations (ODEs). In steady-state conditions, the model yielded a high coefficient of determination and relatively small error bounds, which attested to the model's reliability. In the dynamic scenario, the fine-tuning of parameters enhanced the model's capability in capturing pressure dynamics, although minor deviations in the pressure spike are observed. Through local structural identifiability analysis, twelve parameters were identified as uniquely determinable; however, practical identifiability was achieved with only eight. The application of PINNs demonstrated effectiveness in estimating parameters and states, particularly in low water cut conditions. The proposed method exhibited limitations in high water cut and noisy environments, being areas for future investigation. The model with the PINN can be used in control, condition monitoring, fault detection, and optimization, even in challenging scenarios of unknown fluid properties. While the model accurately captures system dynamics, limitations emerge due to assumptions on the viscosity and the pump-pipe coupling, requiring future work on better modeling the pressure wave velocity in the system. For the inverse problem, a robust PINN algorithm would enhance the unknown parameter estimation. The methodology lays a foundation for modeling more complex ESP systems with applications beyond the oil industry (AU)

• FAPESP's process: 19/14597-5 - Modelling the dynamical behavior in Electrical Submersible Pump (ESP) for liquid-liquid flow
• Grantee: Felipe de Castro Teixeira Carvalho
• Support Opportunities: Scholarships in Brazil - Doctorate