Advanced Corrosion Classification Utilizing Machine Learning and Deep Learning Algorithms

Abstract

Abstract One of the most critical elements in petroleum production engineering is downhole casing integrity. Thus, monitoring downhole casing corrosion is an important element as it ensures the safety and integrity of well assets. Corrosion logging is one important tool that provides valuable information on casing metal loss, that is used as part of a comprehensive monitoring program. In this paper, a new methodology that utilizes advanced Machine Learning (ML) and Deep Learning (DL) to classify downhole casing corrosion integrity status is presented. This method provides valuable additional information and insight that can improve safety. The proposed methodology was to develop an intelligent system using ML & DL that automatically classifies casing corrosion and provides a predicted well downhole corrosion classification to engineers. Firstly, the proposed system actively fetches previously conducted downhole corrosion classification data. Secondly, an advanced pool of ML algorithms was created, and trained on fetched corrosion data. Thirdly, the ML pool evaluated and tested to be uploaded into the system. Finally, newly acquired data for unlogged or old log wells are fed to the advanced ML model to automatically classify downhole casing corrosion based on classes from low to high to engineer and notify them about wells with predicted high corrosion. After finalizing the advanced ML system, it was evaluated on its performance to accurately classify downhole casing corrosion of well and provided system users with targeted classification results. In addition, performance of the system on classification, mitigation and mapping attributes were evaluated using ROC-AUC performance matrix which is a probability curve. After that, testing and evaluating the ML model showed a promising outcome scoring accuracy exceeding 85 % indicating the high efficiency of the model to accurately classify casing corrosion status instantaneously. The developed ML system enabled production engineers to proactively monitor downhole corrosion status reliably and securely. It's worth noting that by implementing such a system have yielded significant impact on our operation leading to both cost, time and recourses optimization. Moreover, the developed corrosion model optimized of thousands of casing corrosion logs conducted through classifying of downhole casing corrosion for unlogged ones, to better optimize resources and prioritize logging highly classified wells to be logged. The proposed system leads to a fast and substantial improvement in acquiring a desired result field-wise in no time. Also, the system provides a detailed description and analysis of the downhole corrosion status to engineers. The developed downhole casing corrosion system has yielded promising results in prediction of wells with higher metal loss. This promotes safety by improving the existing comprehensive well integrity surveillance program.