• In this post, we will provide some parting takeaways on automatic sensor data validation and cleansing.

    1) Detecting missing data / outliers in time series data is relatively straightforward. Do this before any analysis to improve the quality of your output. Otherwise – Garbage In Garbage Out.

    2) Detecting drifting or biased data is hard. Redundancy is a key enabler to detecting these. Build and deploy analytical models to exploit redundancies.

    3) In the drilling domain, the analytical models that you use will change with drilling activity. The model to use during a static hole condition will be different from the one to use during rotary drilling which in turn will be different from the one to use while slide drilling. It is not an easy task to build redundancy during all activities, but it has been done before. So, it is not impossible either.

    4) Models have uncertainties and so does sensors (generally represented using accuracy and precision) (Slide 2). The important surface sensors to validate are block position, hookload, torque, RPM, flow In, flow out, mud volume and standpipe pressure. The industry already knows that it is not an easy task getting the accuracy and precision of these sensors in a scalable manner. Detecting drift and bias is made difficult due to this.

    5) Cleansed data using models will not always be accepted by end users, as models are not always correct. When a fault is flagged, the end user is more willing to take note, if he/she is shown the actual and the model estimated value side by side (Slide 4).

    6) Finally, build algorithms that work even when data is bad. Because data tends to be bad more often than we wish.

    The slides in this post are self-explanatory and provide some additional takeaways. Download for reference.

    Over the course of this year, we have introduced you to various calculations within Sentinel RT in these posts. We hope you found them educational. There are more of these, but we will pause on the technical aspects for a bit and will go into the long-term vision associated with the development of the Sentinel RT. Stay tuned.

    Click below for slides on the topic:

    Data Quality Summary