Use of Logging While Drilling Data and Machine Learning to Predict Maximum Borehole Diameter for Multistage Completion

Abstract

Abstract Stress-induced borehole wall failure during oil and gas wells drilling can lead to over-gauged wellbore. Borehole instability leading to excessively enlarged wellbore can not only increase risk of drilling difficulties but can also cause completions integrity issues. For instance, there is a high probability of poor cementing or compromised zonal isolation across over-gauged wellbore sections in cased and cemented completions or it can result in loss of packers sealing in open hole multi-stage (MSF) completions. Both of these inadequacies of poor cementing and insufficient packers sealing can jeopardize well stimulation operations. Hence, knowing accurate wellbore shape and diameter is critical for multistage well completion requirements. While wireline multi-arm mechanical caliper tools can be run in open hole after drilling to get direct measurement of hole size, not always such measurements are possible due to prevailing drilling difficulties. Under such circumstances, making completion decisions become even more difficult with higher risk of failure. Moreover, a solo wireline logging run to acquire caliper data requires additional cost including rig time. The objective of this paper is to predict synthetic caliper using logging while drilling (LWD) measurements (notably azimuthal density and petrophysical properties) with the help of artificial intelligence (AI) and machine learning algorithms that can be used in place of wireline caliper data thereby saving rig and logging costs and hence reducing overall field development expenditure. This paper describes a machine learning algorithm to predict the mechanical caliper logs using LWD data. Available LWD data for different wells were used to build a robust machine learning algorithm in Python. Different logging parameters from LWD were tested to quantify their sensitivity towards the caliper data and then predict the maximum hole diameter (Cmax) for different wells. The LWD data combines different parameters that have direct effect on the caliper log with machine learning model combining them for different wells to train the model and then predict caliper log based on these parameters. The parameters used in the prediction are gamma ray, four azimuthal photoelectric factors, sonic data, porosity, formation bulk density, four azimuthal formation densities, and mineralogy. The methodology is tested for horizonal wells drilled using 5-7/8" drill-bit. The predicted caliper is compared with measured caliper data so that it satisfies completion requirements for both MSF and cased hole completions.