Identifying the Robust Machine Learning Models to Cement Sheath Fatigue Failure Prediction

Abstract

ABSTRACT: The main goal of cementing is to provide reliable zonal isolation throughout the life of a well. The integral of cement protects the casing and prevents fluid formation from migration. This study aims to use the machine learning method to investigate the fatigue behavior caused by cyclic changes in pressure and temperature. The dataset is retrieved from 6 literature papers, including 479 cement fatigue failure cases. Input parameters were selected to be seven cement-related parameters (cement material type, additive type, UCS, Young's modulus, curing pressure, temperature, and curing time) and seven experimental-related factors (highest loading stress, cyclic loading frequency, stress increment rate, temperature cycles, confining pressure, confining rock type and cycles for fatigue failure). We investigated 150 combinations of features for optimization purposes and selected the best model with the highest performance. The machine learning algorithms include Decision Tree (DT), Support Vector Machine method (SVM), and Artificial Neural Networks (ANNs). The best accuracy of the different models is 90.91% (DT), 72.7% (SVM), and 87.5% (ANNs). The data that we have collected is similar to actual wellbore geometries (hollow concentric cylindrical geometry). This makes the model more applicable to practical cases. 1. INTRODUCTION The main goal of cementing is to provide reliable zonal isolation throughout the life of a well. The integral of cement protects the casing and prevents the formation of fluid from migration. Cement quality is influential to wellbore integrity and zonal isolation. Zonal isolation is essential for the longevity and productivity of any oil or gas well. Nowadays, most wells around the world are extending deeper and deeper. Downhole conditions become increasingly complicated. Changes in downhole conditions, in terms of temperature and pressure, expose cement sheath to deductive stress that impacts the integrity [1]. Cement sheath failure is often reported in many deep wells where high pressure and temperature exist. Cement failure is manifested by interzonal annular fluid movement and abnormally high annular pressure at some point behind the casing up to and at the surface [2]. Subsurface geo-mechanical changes can also cause severe defects in oil and gas wells' casing strings and cement barriers. Damages in cement sheath such as voids, cracks, and channels or de-bonding between cement and casing, caused by either a poor cementing operation or due to the thermal/pressure fluctuations in a well, can threaten the integrity of cemented annuli and compromise zonal isolation [3]. Moreover, many factors in the complicated downhole condition during drilling, completion, stimulation, and production operations can arouse mechanical failures within the cemented annulus. Once these voids, cracks, and defects are connected, they can become channels that provide pathways for migration and communication of hydrocarbons between formations, as shown in Figure 1.1b, thus compromising zonal isolation. Cement failure can lead to environmental problems such as contamination of freshwater tables, sustained annular pressure buildup (SAP), leakage in aged wells [4], and the need for extensive/expensive remedial work. The detrimental effects of cement sheath failure could negatively affect hydrocarbon production and the well's productive life. For example, for production wells, the presence of channels in the cement sheath can result in non-productive communication pathways, causing stimulation and completion inefficiency, thus negatively affecting completion effectiveness and well productivity [5]. For abandoned wells, the failure of cemented barriers and plugs can lead to uncontrolled leakage of formation fluids and gas, with vent flows to the surface. In some extreme cases, the failure of a cemented sheath can even result in control accidents that could result in loss of life, economic loss, and environmental damage [6].