Drilling, completions and production activities have received heavy doses of instrumentation.  With the proliferation of IoT devices there is no dearth of data.  It is just not the tremendous volumes of data being produced by these activities, but also the densities of data streams that the deployed sensors are churning out that makes Oil & Gas a fertile ground for the development of a whole new class of insight drawing AI tools and techniques.  With the price of oil under tremendous downward pressure, both operators and contractors are looking for ways to significantly optimize operations, increase savings, reduce non-productive time (NPT) and smoke out every minute of invisible lost time.

At ChrysalisGold we are tackling a number of well known problems in the drilling and production domains within Oil & Gas for our customer SigmaStream, a Houston-based company.  Key among these is the detection of operational states of the rig at any given time, in real-time.  Using multivariate time-series classification the extraction of insights from high density data in real-time, and turning those insights into operational states or "Operational Events" forms the basis of the automation strategy of SigmaStream.  A supervised learning approach using Random Forest based multivariate time-series classification with a few tweaks has yielded an extremely accurate model for use with real-time streaming data.

 A multivariate time series of drilling data. This time series consists of eight variables representing eight mechanical parameters measured at the surface of rig