LAUSR

Fault detection is crucial for ensuring the safety of the drilling process. Without proper detection and treatment of bit bounce, it could aggravate the wear of drill string and lead to another serious incident. Yet the data distribution of drilling parameters will shift as drilling depth increases, which may result in poor fault detection performance. The alarm optimization strategy of the fault detection model is also rarely discussed in the drilling process. Hence a bit bounce detection and alarm optimization approach based on domain generalization is proposed for the drilling process. The conditional invariant domain generalization (CIDG) is utilized to learn the optimal invariant representation from the historical data in the previously drilled well section. The Bayesian classifier is performed for fault detection. The corresponding alarm optimization strategy is designed based on the alarm performance evaluation. The data from an actual drilling field is utilized for the experiment to illustrate the efficiency of the proposed method. The application of the proposed bit bounce detection method will contribute to maintaining a satisfactory fault detection performance across different drilling depths.