• It is time again for a Bayesian network. This time a simpler one – a naïve Bayesian network. Last week, we mentioned that if we are able to filter out noises from the “total mud volume” channel, we can determine more accurately and with lesser false alarms, whether the hole is flowing or not. One of the many noises to track is the external addition of mud to the pits. Slide 1 shows the network we use to identify mud addition.

    There are four nodes assisting this task. The two nodes on the right help rule out the likelihood of a pump activity flow back or a connection flow back. Another node tracks the rate of change of mud volume (mud additions usually show up as a linear increase in mud volume). Finally, if driller memos are available, those can be used as an input to the network as well. Slide 2 summarizes how all these factors influence the estimation of a mud addition event. An example is shown in Slide 3.

    Slide 4 shows examples of memos that are typically typed into an EDR.  The first suggested task in the utilization of these memos is to organize them into groups, so that they are easier to extract information from. Given the size and many varieties of drilling memos, a semi-supervised method is suggested here, to develop a classifier to categorize the drilling memos into groups. Once they are grouped, they are much easier to use in networks such as the one in Slide 1.

    Slide 6 shows an example where using memos helped detect a mud addition event quicker than otherwise. Like the example shown in these slides, memos help identify mud removal and pump activity as well. We recommend paper SPE-206340 if you are interested in learning more.

    We are now at a stage where we can apply what was discussed last week and this, to detecting kicks and losses. Stay tuned.

    For slides on this topic click below:

    Mud Volume and NLP