Estimating Collapse Pressure of Centralizer Subs From Machine Learning Models

Abstract

Abstract Centralizer subs are run in conjunction with the casing strings in the oil/gas wells to ensure that the casing is centralized while it is installed down hole. Centralizer subs are fabricated of stronger material than the casing strings and designed such that it can sustain a higher collapse pressure than the attached tubing string. A typical centralizer sub is a tube with some complex geometrical features, so the collapse pressure of a centralizer sub can only be estimated by conducting a finite element analysis or subjecting it to a collapse pressure test. Both the options are time consuming and expensive. In this work, a machine learning based regression model is used to derive a parametric equation for calculating the collapse pressure of a centralizer sub. The data needed to train and cross validate the regression model is obtained from finite element analysis (FEA). This machine learning based equation provides a closer estimate of the collapse pressure of the centralizer subs to the results obtained from the FEA than the existing collapse prediction equations from API RP 1111. This machine learning based estimation of collapse pressure will help in correctly predicting the collapse rating of the centralizer sub without performing FEA or testing for each individual subs. This approach of building machine learning models from data generated from FEA can be used for analysis of other equipment as well. With the availability of past data collected/generated through years, the recent advances in machine learning can be used to save time and resources.