Using Data-Mining Crisp-DM Methodology to Predict Drilling Troubles In Real-time

Abstract

Abstract Drilling is considered one of the most challenging and costly operations in the oil and gas industry. Several initiatives were applied to reduce the cost and increase the effectiveness of drilling operations. One of the frequent difficulties that faces these operations is unexpected drilling troubles that take place and stops the operation, resulting in losing a lot of time and money, and could lead to safety issues culminating in a fatality situation. For that, the industry is in continues efforts to prevent drilling troubles. Part of these efforts is utilizing the artificial intelligence (AI) technologies to identify troubles in advance and prevent them before maturing to a serious situation. Multiple approaches were tried in the past. However, errors and significant deviations were observed when comparing the prediction results to the actual drilling data. This could be due to improper design of the artificial intelligent technology or inappropriate data processing. Therefore, searching for dynamic and adequate artificial intelligent technology and encapsulated data processing model is very essential. This paper presents an effective data-mining methodology to determine the most efficient artificial intelligent technology and the applicable data processing techniques, to identify the early symptoms of drilling troubles in real-time. This methodology is CRISP-DM that stands for Cross Industry Standard Process for Data Mining. This methodology consists of the following phases: Business Understanding, Data Understanding, Data Preparation, Modeling, Evaluation and Deployment. During these phases, multiple data-quality techniques were applied to improve the reliability of the real-time data. The developed model presented a significant improvement in identifying the drilling troubles in advance, compared to the current practice. Parameters such as hook-load and bit-depth, were studied. Actual data from several oil fields were used to develop and validate this smart model. This model provided the drilling engineers and operation crew with bigger window to mitigate the situation and resolve it, prevent the occurrence of several drilling troubles. In addition to significant time and cost savings, CRISP-DM provided the artificial intelligent experts and the drilling domain experts with a framework to exchange knowledge and increase the synergy between the two domains significantly, leading to a common and clear understanding, and long-term successful drilling and AI teams collaboration. The novelty of this paper is the introduction of data-mining CRISP methodology for the first time in the prediction of drilling troubles. It enabled the development of a successful artificial intelligence model that outperformed other models in predicting drilling troubles.