Elnaz Farahmandi - Institute of Petroleum Engineering, Chemical Engineering Department, College of Engineering,University of Tehran, Tehran, Iran

The oil and gas drilling industry encounters various difficulties, especially the risk of casingcollapse, which can result in substantial financial and safety issues. This paper examines theapplication of geomechanics principles and machine learning techniques for the analysis ofcasing collapse incidents. The article utilizes a comprehensive dataset obtained from wellbores inthe Marun oil field to develop and verify Random Forest regression, XGBoost, and linearregression models. These models are employed to forecast the maximum horizontal stress (σH)that leads to casing collapse. The initial stage involves partitioning ۲۲,۳۲۳ data records obtainedfrom a collapsed wellbore into two subsets: a training set consisting of ۱۷,۸۵۸ records (۸۰%) andan independent testing set consisting of ۴,۴۶۴ records (۲۰%). The efficacy of these models isassessed by statistical metrics, which demonstrate their ability to accurately characterize thehazards associated with casing-collapse. The findings underscore the capacity of machinelearning models to provide a rapid, precise, and cost-effective substitute for conventionalgeomechanical models. This data can greatly improve the evaluation of the likelihood of casingcollapse and the planning of well operations in oil drilling activities. These models are used toidentify areas with high risk and create plans for casing and cementing that can reduce dangersand withstand strong shear forces. In summary, this study offers significant information for thestrategic organization of drilling programs and the reduction of casing collapse hazards in oil andgas drilling activities.

Wellbore casing collapse, Maximum horizontal stress, Random Forest regression,XGBoost, linear regression models