Behavior Anomalies Detection in Drilling Time Series Through Feature Extraction

The industry has focused mainly on extracting key performance indicators (KPI) from its operational processes that aggregate data in different forms. The computation of average values has been one of the most important ways to measure process performance. However, averages or any other aggregate measures do not give a mechanism to identify how to improve a process. To detect abnormal activities with longer durations than normal (time anomalies), or not following the standard process (behavior anomalies), a drilling engineer has to manually review the drilling parameters individually. It is therefore essential to implement an automated mechanism to identify failures or anomalies within the process in real-time and provide feedback to field personnel in an efficient and easy-to-understand fashion to develop improvement plans. To automatically identify time and behavior anomalies from the real-time surface data, the proposed workflow consists of three steps. First, the time sequence signal is split into drilling activities through data segmentation. Next, we extract features from the segmented activities and statistically convert the feature score into probability. Based on that, the system automatically judges whether it is an anomaly or not. The algorithm has successfully demonstrated its applicability in the field data with better interpretability.