Detecting Unconventional Bit Balling in Real-Time Using a Bayesian Network and Mitigating Invisible Lost Time

Abstract While conventional bit balling is readily identifiable, the deep formations of Egypt's Western Desert present a unique and subtle variation: unconventional bit balling occurring at depths up to 15,000 ft in high-frequency interbedded lithologies. This paper details the culmination of a five-year investigation challenging prevailing misconceptions regarding bit balling in High-Performance Water-Based Mud (HPWBM) applications, where standard indicators—such as a completely balled bit face—are often absent. We present the deployment of an AI system using a pioneering edge-deployed machine learning model capable of early detection of balling events. By analyzing micro-trend fluctuations, the system isolates the balling signals from lithological noise and ROP control effects. Upon identification, mitigation involves pumping specialized anti-balling pills and employing specific on- and off-bottom drilling practices to clean the bit in situ, thereby reducing invisible lost time (ILT) and avoiding unnecessary bit trips. Furthermore, the integration of real-time monitoring with historical offset data marked a step-change from reactive to proactive optimization. This enabled the operator to pinpoint exact dysfunction depths in offset wells and pump precautionary pills ahead of these zones. The result is a proven capability to predict the precise initiation of balling, enabling intervention before premature bit failure occurs. This strategy not only decreased balling frequency but also enhanced ROP and eliminated time-consuming, after-the-fact recovery operations. The method was successfully implemented across the operator's assets, yielding record-breaking drilling runs in several fields.