In this week’s post, we will give you three hints on how to improve your own rig state engine.
For starters, some background information follows. Sentinel RT has a real-time rig state detection model in it. Real-time in the sense that it only uses past data – “i.e., data until that time instant” – to determine the current rig state. We evaluate our model on three aspects – how accurate it is, how quickly it detects the change in rig state, and how infrequently it flags the rig state as “undefined”. It is arguably the best in the industry – a statement that is somewhat difficult to validate, since the rig states determined by various parties are different. For reference, we focus on the rig states outlined in Slide 2. It has to-date worked out to be sufficient for all real-time analysis during drilling. With that background, let us move on to the hints.
First hint: We use a “pseudo” dynamic Bayesian network for our rig state detection model. The use of prior rig states in the detection of our current rig state dramatically improves rig state detection accuracy. Give that concept a try – if you are not already doing it.
Second hint: When it comes to algorithms, two heads are better than one. The rig state engine in Sentinel RT is free (Sentinel RT Lite) – Yes, free – but not open source (yet). Integrate it with your own rig state engine to arrive at a better rig state estimation than either of them alone.
Third hint: Sentinel RT Lite is free, and there is an online version (Check slide 5) that you can try out today. Run your data on it, to automatically label your massive datasets. Correct any errors you find. Then train your own ML model using the labelled data. Click below for slides on this topic:
Hope you found these hints useful. Email us at firstname.lastname@example.org if you have any questions. Next week, we will do a deep dive into how to differentiate between rotary drilling and slide drilling when pipe rocking is involved. So, stay tuned.