Hybrid neural network for drilling anomaly detection
The present disclosure relates to systems, non-transitory computer-readable media, and methods for detecting a washout or other anomaly event in a wellbore. In particular, in one or more embodiments, the disclosed systems receive a plurality of measurements including a measured flow rate into the wellbore, a measured weight on a drill bit in the wellbore, a measured depth of the drill bit in the wellbore, and a measured pressure at a standpipe of the wellbore. In one or more embodiments, the disclosed systems estimate one or more parameters of a physical model for determining a theoretical estimate of the standpipe pressure. In one or more embodiments, the disclosed systems determine a probability that the washout or other anomaly event is occurring in the wellbore based at least partially upon the measurements and the theoretical estimate of the standpipe pressure.
1 . A method for detecting a drilling anomaly, the method comprising:
receiving a plurality of physical measurements related to a wellbore, wherein the plurality of physical measurements include a flow rate measurement of a flow rate of drilling fluid into the wellbore, a weight on bit (WOB) measurement of a weight on a drill bit in the wellbore, and a depth measurement of the drill bit in the wellbore;
based at least partially upon the plurality of physical measurements, estimating a plurality of estimated parameters of a physical model for estimating a drilling characteristic;
determining a theoretical estimation of the drilling characteristic based on the physical model using the plurality of physical measurements and the plurality of estimated parameters, wherein determining the theoretical estimation includes:
multiplying the depth measurement with a first parameter of the plurality of estimated parameters;
exponentiating the flow rate measurement by raising the flow rate measurement to a power of a second parameter of the plurality of estimated parameters; and
multiplying the WOB measurement with a third parameter of the plurality of estimated parameters;
determining, utilizing a drilling event classification neural network, an anomaly probability that a drilling anomaly event is occurring based on the theoretical estimation of the drilling characteristic and a measurement of the drilling characteristic; and
using the anomaly probability, decreasing a drilling operation setting by adjusting operation of subterranean drilling equipment, wherein the drilling operation setting includes at least one of a flow rate decrease in the flow rate, a WOB decrease of the weight on the drill bit, or a pressure decrease in a standpipe pressure of the drilling fluid at a standpipe.
2 . The method of claim 1 , wherein:
determining the theoretical estimation of the drilling characteristic comprises determining a standpipe estimation of the standpipe pressure; and
determining, utilizing the drilling event classification neural network, the anomaly probability that the drilling anomaly event is occurring comprises determining a washout probability of a washout event.
3 . The method of claim 2 , further comprising activating a trigger in response to determining that the washout probability exceeds a threshold value, wherein the trigger initiates decreasing the drilling operation setting via the flow rate decrease, the WOB decrease, and the pressure decrease.
4 . The method of claim 2 , further comprising physically measuring the flow rate measurement, the WOB measurement, and the depth measurement.
5 . The method of claim 1 , further comprising:
generating, via a graphical user interface, a display of one or more of the measurement of the drilling characteristic or the theoretical estimation of the drilling characteristic; and
indicating a start of the drilling anomaly event on the display.
6 . The method of claim 1 , further comprising training the drilling event classification neural network utilizing a loss function comprising a binary cross-entropy term weighted according to a first uncertainty, a mean squared error term weighted according to a second uncertainty, and a regularizing term.
7 . The method of claim 6 , wherein utilizing the loss function comprises nullifying the mean squared error term during a ground truth anomaly event.
8 . The method of claim 1 , wherein estimating the plurality of estimated parameters of the physical model comprises utilizing a convolutional neural network to generate estimations of the plurality of estimated parameters.
9 . A system comprising:
a memory device; and
at least one processor configured to cause the system to:
receive a plurality of physical measurements related to a wellbore, wherein the plurality of physical measurements include a flow rate measurement of a flow rate of drilling fluid into the wellbore, a weight on bit (WOB) measurement of a weight on a drill bit in the wellbore, and a depth measurement of the drill bit in the wellbore;
based at least partially upon the plurality of physical measurements, estimate a plurality of estimated parameters of a physical model for estimating a drilling characteristic;
determine a theoretical estimation of the drilling characteristic based on the physical model using the plurality of physical measurements and the plurality of estimated parameters, wherein determining the theoretical estimation includes:
multiplying the depth measurement with a first parameter of the plurality of estimated parameters;
exponentiating the flow rate measurement by raising the flow rate measurement to a power of a second parameter of the plurality of estimated parameters; and
multiplying the WOB measurement with a third parameter of the plurality of estimated parameters;
determine, utilizing a drilling event classification neural network, an anomaly probability that a drilling anomaly event is occurring based on the theoretical estimation of the drilling characteristic and a measurement of the drilling characteristic; and
using the anomaly probability, decrease a drilling operation setting by adjusting operation of subterranean drilling equipment, wherein the drilling operation setting includes at least one of a flow rate decrease in the flow rate, a WOB decrease of the weight on the drill bit, or a pressure decrease in a standpipe pressure of the drilling fluid at a standpipe.
10 . The system of claim 9 , wherein the at least one processor is further configured to cause the system to:
receive, at periodic time intervals, measurement updates to the plurality of physical measurements related to the wellbore;
determine, at the periodic time intervals, estimation updates to the theoretical estimation of the drilling characteristic based on the physical model using the measurement updates and the plurality of estimated parameters;
receive, at the periodic time intervals, characteristic updates to the measurement of the drilling characteristic; and
determine, at the periodic time intervals and utilizing the drilling event classification neural network, anomaly updates to the anomaly probability that the drilling anomaly event is occurring based on the estimation updates and the characteristic updates.
11 . The system of claim 10 , wherein the at least one processor is further configured to cause the system to generate estimated parameter updates of the plurality of estimated parameters, at the periodic time intervals and based at least partially upon the measurement updates.
12 . The system of claim 11 , wherein the at least one processor is further configured to cause the system to:
generate, at the periodic time intervals and via a graphical user interface, a display of one or more of the estimation updates or the measurement updates; and
indicate a start of the drilling anomaly event on the display.
13 . The system of claim 9 , wherein the drilling event classification neural network comprises a long short-term memory neural network and one or more convolutional neural network layers.
14 . The system of claim 9 , wherein the at least one processor is configured to cause the system to estimate the plurality of estimated parameters of the physical model by processing the plurality of physical measurements through a convolutional neural network to generate the plurality of estimated parameters.
15 . A non-transitory computer-readable medium comprising instructions that, when executed by at least one processor, cause the at least one processor to:
receive a plurality of physical measurements related to a wellbore, wherein the plurality of physical measurements include a flow rate measurement of a flow rate of drilling fluid into the wellbore, a weight on bit (WOB) measurement of a weight on a drill bit in the wellbore, and a depth measurement of the drill bit in the wellbore;
based at least partially upon the plurality of physical measurements, estimate a plurality of estimated parameters of a physical model for estimating a drilling characteristic;
determine a theoretical estimation of the drilling characteristic based on the physical model using the plurality of physical measurements and the plurality of estimated parameters, wherein determining the theoretical estimation includes:
multiplying the depth measurement with a first parameter of the plurality of estimated parameters;
exponentiating the flow rate measurement by raising the flow rate measurement to a power of a second parameter of the plurality of estimated parameters; and
multiplying the WOB measurement with a third parameter of the plurality of estimated parameters;
determine, utilizing a drilling event classification neural network, an anomaly probability that a drilling anomaly event is occurring based on the theoretical estimation of the drilling characteristic and a measurement of the drilling characteristic; and
using the anomaly probability, decrease a drilling operation setting by adjusting operation of subterranean drilling equipment, wherein the drilling operation setting includes at least one of a flow rate decrease in the flow rate, a WOB decrease of the weight on the drill bit, or a pressure decrease in a standpipe pressure of the drilling fluid at a standpipe.
16 . The non-transitory computer-readable medium of claim 15 , wherein the instructions, when executed by the at least one processor, cause the at least one processor to:
determine, utilizing the drilling event classification neural network, the anomaly probability by determining a stuck pipe probability of a stuck pipe event.
17 . The non-transitory computer-readable medium of claim 15 , wherein the instructions further cause the at least one processor to train the drilling event classification neural network utilizing a loss function comprising a binary cross-entropy term weighted according to a first uncertainty.
18 . The non-transitory computer-readable medium of claim 17 , wherein the instructions, when executed by the at least one processor, cause the at least one processor to estimate the plurality of estimated parameters of the physical model by utilizing a convolutional neural network to generate the plurality of estimated parameters.