Drilling report free text analysis - AI Foresight use case

Background

Drilling contractors produce a very detailed record each day of the operations on a working drilling rig. It consists of a number of structured data entries such as well information, activity codes, drilling depths, Non Productive Time (NPT) related information and much more. Typically, each record also contains additional information registered by the drillers regarding their observations, written in a very concise form with a high frequency of technical symbols and idioms, mistyping, abbreviation of technical terms, and many incomplete sentences. This makes it very hard to use ‘out-of-the-box’ Natural Language Processing models that are trained on natural English vocabulary and syntax, to obtain satisfactory insights from this text data.


Seen the complexity of feeding driller’s notes into an automated DSS system and its manual understanding is very time-consuming, driller’s notes are typically just stored and very rarely analysed, except maybe to help with the investigation of a major HSE incident. Moreover, the notion that their memos will rarely be consulted, will not encourage drillers to spend a lot of time and effort to meticulously write down their observations.


Data & AI solutions

A bespoke NLP solution, based on industry-standard methods and models but highly optimised and trained for a very specific technical environment, extracts and classifies information and finds meaning in difficult to read memos. Once this work is performed, the driller’s notes can be automatically classified according to a predefined classification scheme such as symptoms, actions and events that can help greatly to improve the understanding of the sequences of events and actions taken during the life-cycle of a particular well.


The raw driller’s notes are used as input data for the NLP system, if needed correlated with structured data fields present in the DDRs for further situational awareness.


Continuous skip-gram (Word2Vec) for the word embedding and a Long Short-Term Memory Network (LSTM) or Convolutional Neural Network (CNN).


Business case & benefits

The driller’s notes contained in the Daily Drilling Reports (DDRs) contain data that is provided by a specialist who is on-site, knows the actual circumstances of any event happening at any particular time and his/her interpretation of what is actually going on. This data provide a treasure trove of information that should allow to make any decision support system or incident analysis process smarter and more informed.


Our specially trained NLP models allow for very specific technical documents and notes to be automatically processed and classified to gain superior insight into the drilling process.