Drilling Through Data: Automated Kick Detection Using Data Mining

Abstract

Abstract This paper details using advancement in data analytics and the huge amount of data generated while drilling to develop an automated system to detect kicks while drilling. Detecting kicks in early stages gives the crew additional time to control it resulting in a safer and more efficient drilling operation. Five models were developed and evaluated to optimize kick detection they are: Decision Tree, K-Nearest Neighbor (KNN), Sequential Minimal Optimization (SMO) Algorithm, Artificial Neural Network (ANN), and Bayesian Network. The models were trained to detect kicks based on actual kick cases. The models are predicting kicks using only surface parameters such as: pressure gauges, flow meters, hook load, rate of penetration, torque, pump rate, and weight on bit. The performance of the five models is then evaluated and compared. Best two models were Decision Tree and K-Nearest Neighbor.