Dynamic execution of predictive models
Disclosed herein are systems, devices, and methods related to assets and predictive models and corresponding workflows that are related to the operation of assets. In particular, examples involve assets configured to receive and locally execute predictive models, locally individualize predictive models, and/or locally execute workflows or portions thereof.
1. A local analytics device configured to monitor operating conditions of an asset, the local analytics device comprising:
an asset interface configured to couple the local analytics device to the asset;
a network interface configured to facilitate communication between the local analytics device and a computing system (i) configured to monitor operating conditions of the asset and (ii) located remote from the local analytics device;
at least one processor;
a non-transitory computer-readable medium; and
program instructions stored on the non-transitory computer-readable medium that are executable by the at least one processor to cause the local analytics device to:
identify a predictive model that is configured to output predictions related to operation of the asset;
based on one or more execution functions corresponding to the identified predictive model, determine whether the local analytics device should execute the predictive model, wherein the one or more execution functions are configured to quantify (i) an expected cost associated with executing the predictive model at the local analytics device that is based at least in part on an expected accuracy of predictions related to the operation of the asset to be output by the predictive model as executed at the local analytics device and (ii) an expected cost associated with executing the predictive model at the computing system that is based at least in part on an expected accuracy of predictions related to the operation of the asset to be output by the predictive model as executed at the computing system;
if it is determined that the local analytics device should execute the predictive model, execute the predictive model based on operating data for the asset received via the asset interface; and
otherwise, transmit to the computing system, via the network interface, (i) an instruction for the computing system to execute the predictive model, and (ii) operating data for the asset received via the asset interface.
2. The local analytics device of claim 1 , wherein the identified predictive model is a first predictive model, and wherein the program instructions stored on the non-transitory computer-readable medium are further executable by the at least one processor to cause the local analytics device to:
before identifying the first predictive model, receive from the computing system, via the network interface, a plurality of predictive models comprising the first predictive model, wherein each of the plurality of predictive models is configured to output a likelihood that a respective failure will occur at the asset within a given period of time in the future.
3. The local analytics device of claim 1 , wherein the one or more execution functions comprise a first performance score function for the local analytics device that is configured to quantify the expected cost associated with executing the predictive model at the local analytics device and a second performance score function for the computing system that is configured to quantify the expected cost associated with executing the predictive model at the computing system, and wherein determining whether the local analytics device should execute the predictive model is based on a comparison between the expected cost output by the first performance score function and the expected cost output by the second performance score function.
4. The local analytics device of claim 3 , wherein determining whether the local analytics device should execute the predictive model comprises:
executing (i) the first performance score function to determine the expected cost associated with executing the predictive model at the local analytics device, and (ii) the second performance score function to determine the expected cost associated with executing the predictive model at the computing system; and
comparing the expected cost output by the first performance score function and the expected cost output by the second performance score function.
5. The local analytics device of claim 4 , wherein determining whether the local analytics device should execute the predictive model comprises:
based on the comparison, determining that the predictive model should be executed by the local analytics device when the expected cost output by the first performance score function is less than or equal to the expected cost output by the second performance score function.
6. The local analytics device of claim 4 , wherein determining whether the local analytics device should execute the predictive model comprises:
based on the comparison, determining that the predictive model should be executed by the local analytics device when the expected cost output by the first performance score function is at least a threshold amount less than the expected cost output by the second performance score function.
7. The local analytics device of claim 4 , wherein the program instructions stored on the non-transitory computer-readable medium are further executable by the at least one processor to cause the local analytics device to:
before determining whether the local analytics device should execute the predictive model, receive from the computing system, via the network interface, the first and second performance score functions, wherein the first and second performance score functions correspond to the identified predictive model and are defined by the computing system.
8. The local analytics device of claim 1 , wherein the operating data for the asset received via the asset interface comprises signal data from (i) at least one sensor of the asset, (ii) at least one actuator of the asset, or (iii) at least one sensor of the asset and at least one actuator of the asset.
9. The local analytics device of claim 1 , wherein the program instructions stored on the non-transitory computer-readable medium are further executable by the at least one processor to cause the local analytics device to:
before determining whether the local analytics device should execute the predictive model, modify the one or more execution functions based on one or more dynamic factors.
10. The local analytics device of claim 9 , wherein the local analytics device and the computing system are communicatively coupled via a wide-area network, and wherein the one or more dynamic factors comprise at least one of (i) an amount of time since an update to the one or more execution functions, (ii) an amount of time since an update to the predictive model, or (iii) one or more network conditions of the wide-area network.
11. A non-transitory computer-readable medium having instructions stored thereon that are executable to cause a local analytics device to:
identify a predictive model that is configured to output predictions related to operation of an asset to which the local analytics device is coupled via an asset interface of the local analytics device;
based on one or more execution functions corresponding to the identified predictive model, determine whether the local analytics device should execute the predictive model, wherein the one or more execution functions are configured to quantify (i) an expected cost associated with executing the predictive model at the local analytics device that is based at least in part on an expected accuracy of predictions related to the operation of the asset to be output by the predictive model as executed at the local analytics device and (ii) an expected cost associated with executing the predictive model at a computing system located remote from the local analytics device that is based at least in part on an expected accuracy of predictions related to the operation of the asset to be output by the predictive model as executed at the computing system;
if it is determined that the local analytics device should execute the predictive model, execute the predictive model based on operating data for the asset received via the asset interface; and
otherwise, transmit to the computing system via a network interface of the local analytics device, (i) an instruction for the computing system to execute the predictive model, and (ii) operating data for the asset received via the asset interface.
12. The non-transitory computer-readable medium of claim 11 , wherein determining whether the local analytics device should execute the predictive model comprises:
executing (i) a first performance score function to determine the expected cost associated with executing the predictive model at the local analytics device, and (ii) a second performance score function to determine the expected cost associated with executing the predictive model at the computing system; and
comparing the expected cost output by the first performance score function and the expected cost output by the second performance score function.
13. The non-transitory computer-readable medium of claim 12 , wherein determining whether the local analytics device should execute the predictive model comprises:
based on the comparison, determining that the predictive model should be executed by the local analytics device when the expected cost output by the first performance score function is less than or equal to the expected cost output by the second performance score function.
14. The non-transitory computer-readable medium of claim 12 , wherein determining whether the local analytics device should execute the predictive model comprises:
based on the comparison, determining that the predictive model should be executed by the local analytics device when expected cost output by the first performance score function is at least a threshold amount less than the expected cost output by the second performance score function.
15. The non-transitory computer-readable medium of claim 11 , wherein the instructions are further executable to cause the local analytics device to:
before determining whether the local analytics device should execute the predictive model, modify the one or more execution functions based on one or more dynamic factors.
16. The non-transitory computer-readable medium of claim 15 , wherein the local analytics device and the computing system are wirelessly communicatively coupled via a wide-area network, and wherein the one or more dynamic factors comprise at least one of (i) an amount of time since an update to the one or more execution functions, (ii) an amount of time since an update to the predictive model, or (iii) one or more network conditions of the wide-area network.
17. A method comprising:
identifying, by a local analytics device, a predictive model configured to output predictions related to operation of an asset to which the local analytics device is coupled via an asset interface of the local analytics device;
based on one or more execution functions corresponding to the identified predictive model, determining whether the local analytics device should execute the predictive model, wherein the one or more execution functions are configured to quantify (i) an expected cost associated with executing the predictive model at the local analytics device that is based at least in part on an expected accuracy of predictions related to the operation of the asset to be output by the predictive model as executed at the local analytics device and (ii) an expected cost associated with executing the predictive model at a computing system located remote from the local analytics device that is based at least in part on an expected accuracy of predictions related to the operation of the asset to be output by the predictive model as executed at the computing system;
if it is determined that the local analytics device should execute the predictive model, executing, by the local analytics device, the predictive model based on operating data for the asset received via an asset interface of the local analytics device; and
otherwise, transmitting, by the local analytics device to the computing system via a network interface of the local analytics device, (i) an instruction for the computing system to execute the predictive model, and (ii) operating data for the asset received via the asset interface.
18. The method of claim 17 , further comprising:
before determining whether the local analytics device should execute the predictive model, modifying the one or more execution functions based on one or more dynamic factors.
19. The method of claim 17 , wherein determining whether the local analytics device should execute the predictive model comprises:
executing (i) a first performance score function to determine the expected cost associated with executing the predictive model at the local analytics device, and (ii) a second performance score function to determine the expected cost associated with executing the predictive model at the computing system; and
comparing the expected cost output by the first performance score function and the expected cost output by the second performance score function.