Machine learning based equipment failure prediction
A method comprises receiving a time series of data values for a time window of each operational parameter of a number of operational parameters of equipment; calculating a time derivative feature that comprises a change of the data values of a first operational parameter of the number of operational parameters over the time window; and classifying, using a machine learning model and based on the time derivative feature, an operational mode of the equipment into different failure categories.
1. A method comprising:
receiving a time series of data values for a time window of each operational parameter of a number of operational parameters of equipment;
calculating a time derivative feature that comprises a change of the data values of a first operational parameter of the number of operational parameters over the time window;
encoding the time derivative feature of the first operational parameter to generate a time derivative encoded value based on a rate of change of the time derivative feature over some period of time during the time window relative to a set of threshold values;
training a machine learning model using the time derivative encoded value to learn failure prediction patterns for the equipment; and
classifying, using the machine learning model and following the training of the machine learning model, an operational mode of the equipment in real time based at least in part on the time derivative encoded value.
2. The method of claim 1 , further comprising:
calculating a gradient feature that comprises a change of the data values of a second operational parameter of the number of operational parameters relative to a change of the data values of a third operational parameter of the number of operational parameters over the time window;
encoding the gradient feature to generate a gradient encoded value;
training the machine learning model using the gradient encoded value to learn failure prediction patterns for the equipment; and
classifying, using the machine learning model and following the training of the machine learning model, the operational mode of the equipment in real time based at least in part on the gradient encoded value.
3. The method of claim 2 ,
wherein classifying the operational mode comprises classifying, using the machine learning model and based on the encoded time derivative value and the encoded gradient value, the operational mode of the equipment into different failure categories.
4. The method of claim 3 , wherein the different failure categories comprise at least one of stable, unstable, pre-failure, and failure.
5. The method of claim 2 , wherein encoding the gradient feature comprises,
in response to an increase of the change in the data values of the second operational parameter relative to the change in the data values of the third operational parameter being greater than a large gradient increase threshold, encoding the gradient as a major gradient increase;
in response to a decrease of the change in the data values of the second operational parameter relative to the change in the data values of the third operational parameter v being greater than a large gradient decrease threshold, encoding the gradient as a major gradient decrease;
in response to the increase of the change in the data values of the second operational parameter relative to the change in the data values of the third operational parameter being less than a small gradient increase threshold, encoding the gradient as a minor gradient increase;
in response to the decrease of the change in the data values of the second operational parameter relative to the change in the data values of the third operational parameter being less than a small gradient decrease threshold, encoding the gradient as a minor gradient decrease; and
in response to the change of the data values of the second operational parameter relative to the change in the data values of the third operational parameter changing less than a constant gradient threshold, encoding the gradient as a constant.
6. The method of claim 1 , wherein encoding the time derivative feature comprises,
in response to the change over time of the value of the first operational parameter increasing greater than a drastic time increase threshold, encoding the time derivative feature as a drastic time increase;
in response to the change over time of the value of the first operational parameter decreasing greater than a drastic time decrease threshold, encoding the time derivative feature as a drastic time decrease;
in response to the change over time of the value of the first operational parameter increasing less than a minor time increase threshold, encoding the time derivative feature as a minor time increase;
in response to the change over time of the value of the first operational parameter decreasing less than a minor time decrease threshold, encoding the time derivative feature as a minor time decrease; and
in response to the change over time of the value of the first operational parameter changing less than a constant time threshold, encoding the time derivative feature as a constant.
7. The method of claim 1 , further comprising:
determining outlier features of data values for the time window,
wherein classifying the operational mode of the equipment comprises classifying, using the machine learning model and based on the outlier features, the operational mode of the equipment.
8. The method of claim 1 , wherein the equipment comprises an electrical submersible pump.
9. The method of claim 1 , further comprising: modifying the operation of the equipment in response to the classifying the operational mode of the equipment.
10. A system comprising:
downhole equipment to be positioned in a wellbore;
a number of sensors that are to measure a number of operational parameters of the downhole equipment;
a processor; and
a computer-readable medium having instructions stored thereon that are executable by the processor to cause the processor to,
receive a time series of data values for a time window of each operational parameter of the number of operational parameters;
calculate a time derivative feature that comprises a change of the data values of a first operational parameter of the number of operational parameters over the time window;
encode the time derivative feature of the first operational parameter to generate a time derivative encoded value based on a rate of change of the time derivative feature over some period of time during the time window relative to a set of threshold values;
train a machine learning model using the time derivative encoded value to learn failure prediction patterns for the equipment; and
classify, using the machine learning model and following the training of the machine learning model, an operational mode of the equipment in real time based at least in part on the time derivative encoded value.
11. The system of claim 10 , wherein the instructions comprise instructions executable by the processor to cause the processor to:
calculate a gradient feature that comprises a change of the data values of a second operational parameter of the number of operational parameters relative to a change of the data values of a third operational parameter of the number of operational parameters over the time window,
encode the gradient feature to generate a gradient encoded value;
train the machine learning model using the gradient encoded value to learn failure prediction patterns for the equipment; and
classify, using the machine learning model and following the training of the machine learning model, the operational mode of the equipment in real time based at least in part on the gradient encoded value.
12. The system of claim 11 ,
wherein the instructions to classify the operational mode of the equipment comprises instructions executable by the processor to cause the processor to classify, using the machine learning model and based on the encoded time derivative feature and the encoded gradient feature, the operational mode of the equipment into different failure categories.
13. The system of claim 10 , wherein the instructions comprise instructions executable by the processor to cause the processor to:
determine outlier features of data values for the time window,
wherein the instructions to classify the operational mode of the equipment comprises instructions executable by the processor to cause the processor to classify, using the machine learning model and based on the outlier features, the operational mode of the equipment.
14. The system of claim 13 , wherein the different failure categories comprise at least one of stable, unstable, pre-failure, and failure.
15. The system of claim 10 , wherein the equipment comprises an electrical submersible pump.
16. The system of claim 10 , wherein the instructions comprise instructions executable by the processor to cause the processor to modify the operation of the equipment in response to the classifying the operational mode of the equipment.
17. A non-transitory, computer-readable medium having instructions stored thereon that are executable by a processor to perform operations comprising:
receiving a time series of data values for a time window of each operational parameter of a number of operational parameters of equipment;
calculating a time derivative feature that comprises a change of the data values of a first operational parameter of the number of operational parameters over the time window;
encoding the time derivative feature of the first operational parameter to generate a time derivative encoded value based on a rate of change of the time derivative feature over some period of time during the time window relative to a set of threshold values;
training a machine learning model using the time derivative encoded value to learn failure prediction patterns for the equipment; and
classifying, using the machine learning model and following the training of the machine learning model, an operational mode of the equipment in real time based at least in part on the time derivative encoded value.
18. The non-transitory, computer-readable medium of claim 17 , wherein the operations comprise:
calculating a gradient feature that comprises a change of the data values of a second operational parameter of the number of operational parameters relative to a change of the data values of a third operational parameter of the number of operational parameters over the time window
encoding the gradient feature to generate a gradient encoded value;
training the machine learning model using the gradient encoded value to learn failure prediction patterns for the equipment; and
classifying, using the machine learning model and following the training of the machine learning model, the operational mode of the equipment in real time based at least in part on the gradient encoded value.
19. The non-transitory, computer-readable medium of claim 18 ,
wherein classifying the operational mode comprises classifying, using the machine learning model and based on the encoded time derivative feature and the encoded gradient feature, the operational mode of the equipment into different failure categories.
20. The non-transitory, computer-readable medium of claim 17 , wherein the operations comprise:
determining outlier features of data values for the time window,
wherein classifying the operational mode of the equipment comprises classifying, using the machine learning model and based on the outlier features, the operational mode of the equipment.