• Washouts impact drilling hydraulics and can be a major source of non-productive time especially when it leads to twist-offs and ultimately expensive fishing jobs. It is good therefore to be alerted about a potential washout as early as possible. Here we show how we use a Bayesian network for this purpose.

    A washout will generally show itself in the sensor signals. A tell-tale signature is the reduction in pump pressure (P) at constant circulation rate (Q). This trend can be quantitatively captured using a metric called the hydraulic coefficient: h=P/Q^2, where Q can either be the flow in rate or the flow out rate. We can use the hydraulic flow out coefficient to differentiate between a washout and a pump failure, which are often very difficult to distinguish from EDR data. Hydraulic coefficient calculated using flow out should decrease during washout but remains flat during pump failure.

    In the first slide you can see the pump pressure decrease even as flow in and flow out remains constant. This is captured in both the flow in and flow out hydraulic coefficient trends, both of which decrease. Additionally, one could compare the measured pump pressure against model predicted pressure (bit pressure drop + frictional pressure drop + differential pressure), but this does require contextual data to run the hydraulics model. We also need to rule out other sources of pump pressure changes to reduce false alarms. Many of the nodes in the Bayesian network (shown in slide 2) serve that purpose.

    Slides 3 and 4 give you more details about the states in the network that are conducive to a high washout belief. Reach out to us if you want more information on washout detection. Next week we will go into pump failure detection, which as you may have guessed from Slide 2, uses the same network model. We will also discuss some nuances when it comes to using this belief in the field.

    Click below for additional slides on the topic:

    Washout Detection