Search for a command to run...
Abstract Drilling operations generate vast amounts of textual information through Daily Drilling Reports (DDRs), which record operations throughout the well development process. However, these activity lines are manually inputted by Drilling Supervisor (DSV). Even though the system that DSV uses to write DDR is already structured and standardized with activity and phase code for every operation, they vary greatly in description across DSVs. The granularity of the write-up depends on DSV. Some DSV consolidate activities into one, some define activity with different codes. We cannot rely solely on the activity and phase codes to do detailed activity timing analysis for DDR. This leads to difficulties in estimating duration for each specific activity, inconsistent benchmarking for offset well analysis, unreliable time estimation for new wells, and limited visibility of Invisible Lost Time. Detailed activity timing analysis of DDRs is needed to extract meaningful data such as Invisible Loss Time (ILT), Non-Productive Time (NPT), activities granularity together with their timing. This analysis can take engineers up to ten working days per well, delaying project planning and reducing efficiency. Thus, we need something to understand the DDR to classify them into correct pre-defined activities and facilitate engineers in doing detailed activity timing analysis. This paper presents an artificial intelligence solution developed collaboratively between 3 parties to automate the classification of drilling activity logs and carry out the detailed time activity analysis from DDR. The system employs a knowledge-based AI engine that combines Natural Language Understanding (NLU) with domain-specific reasoning to interpret over 482 unique drilling activities according to standardized activity taxonomy. Each record is analyzed and mapped to its corresponding drilling phase and will be ingested into a Drilling Time Estimator calculator ecosystem. This enables consistent data ingestion and benchmarking. When applied to over one thousand historical wells, the system reduced processing time from 10 working days to 10–15 minutes per well, achieving 80–90% classification accuracy. The automated outputs are seamlessly integrated into Drilling Time Estimator, allowing engineers to retrieve average activity durations across multiple offset wells in real time. The implementation represents a significant step toward digital drilling transformation, improving data reliability, engineering efficiency, and decision-making across upstream operations.