Drilling operations involve substantial planning and execution to achieve safe and cost effective well delivery. To improve drilling efficiency by minimizing costly non-productive time, active surveillance programs leverage large volumes of data generated at various stages of the drilling process. With increased reliance being placed on real-time quantitative measurements at the drill site, there has been recognition of the limited use of the large amount of valuable unstructured data generated throughout lifecycle of the workflow. Generated data in the form of communication messages and daily reports include substantial information and recordings of activities. However, such worthy sources of insight are largely untapped by conventional, established tools for monitoring and alert.
We present a proof of concept for mining daily drilling reports for three wells drilled between 2008 and 2010. There were occurrences of drilling impacts and NPT events ranging from slight operational delay to other events that introduced delays to the critical path. Employing concept extraction and pattern frequency techniques, we were able to track and monitor reported symptoms of the observed behavior to help identify root cause and compound factors leading to such an event.
In an effort to bring a fundamental understanding of the unstructured data relevant to the process, whilst simultaneously reducing the time required collating and processing this data, we applied unsupervised learning methods Results reveal associations between extracted patterns of some key issues and relative progress over a period of time which were not apparent to drilling engineers at first glance. In addition, the technology provides enhanced capability around the interpretation and visual representation of largely untapped collection of documents. The process can be automated to integrate pattern extraction and visualization with existing systems to empower engineering staff and technical specialists involved in the monitoring and analysis of well construction.