Machine learning design for long-term reliability and stress testing
The techniques disclosed herein enable systems to measure the long-term reliability of machine learning agents prior to deployment at a control system. This is achieved through analysis of control system component specifications to determine a useful lifespan of the components such as projected failure rate, hours continuous operation, and so forth. The system can derive parameters for the machine learning agent to interact with the components such as action frequency and action range. From the component lifespan, action frequency, and action range, an accelerated test procedure is constructed to evaluate the reliability of the machine learning agent. From executing the accelerated test procedure, a reliability score can be calculated for the machine learning agent.
1. A method comprising:
extracting a plurality of product specifications from a plurality of components within a physical control system controlled by a machine learning agent;
determining a nominal phase defining a time period of a consistent failure rate and an aging phase defining a time period of an increasing failure rate for each component of the plurality of components based on the plurality of product specifications;
calculating an action frequency for each component of the plurality of components based on the associated nominal phase, the action frequency defining a frequency at which the machine learning agent can actuate each component;
determining an action range for each component of the plurality of components using one or more acceleration factors associated with a stress parameter of the component, the action range defining a range of values configurable by the machine learning agent for each component;
constructing an accelerated test procedure for determining a reliability of the machine learning agent based on the nominal phases, the aging phases, the action frequencies, and the action ranges of the plurality of components;
executing the accelerated test procedure within a simulated environment comprising the machine learning agent and a plurality of simulated components that correspond to the plurality of components to calculate a reliability score quantifying a failure rate of the machine learning agent with respect to the plurality of simulated components;
training the machine learning agent, based on the reliability score, to learn a configuration of the physical control system that maximizes the reliability score by iteratively repeating the accelerated test procedure; and
deploying machine learning agent to the plurality of components of the physical control system to implement the configuration of the control system that maximizes the reliability score in accordance with the nominal phases, the aging phases, the action frequencies, and the actions ranges of the accelerated test procedure.
2. The method of claim 1 , wherein the nominal phase comprises a number of operational hours and a number of power cycles calculated by an offline model based on the product specifications.
3. The method of claim 1 , wherein the aging phase comprises a failure probability calculated using a statistical probability distribution based on the product specifications.
4. The method of claim 1 , wherein the plurality of components are mechanical components, and the stress parameter is mechanical stress.
5. The method of claim 1 , wherein the plurality of components are electrical components, and the stress parameter is voltage.
6. The method of claim 1 , wherein the accelerated test procedure comprises a plurality of corner tests, each corner test defining a sequence of steps for evaluating the machine learning agent under an associated load condition.
7. The method of claim 1 , wherein a maximum action frequency is equal to a power cycle frequency of the plurality of components.
8. A system comprising:
one or more processing units; and
a computer-readable medium having encoded thereon computer-readable instructions that when executed by the one or more processing units cause the system to:
extract a plurality of product specifications from a plurality of components within a physical control system controlled by a machine learning agent;
determine a nominal phase defining a time period of a consistent failure rate and an aging phase defining a time period of an increasing failure rate for each component of the plurality of components based on the plurality of product specifications;
calculate an action frequency for each component of the plurality of components based on the associated nominal phase, the action frequency defining a frequency at which the machine learning agent can actuate each component;
determine an action range for each component of the plurality of components using one or more acceleration factors associated with a stress parameter of the component, the action range defining a range of values configurable by the machine learning agent for each component;
construct an accelerated test procedure for determining a reliability of the machine learning agent based on the nominal phases, the aging phases, the action frequencies, and the action ranges of the plurality of components;
execute the accelerated test procedure within a simulated environment comprising the machine learning agent and a plurality of simulated components that correspond to the plurality of components to calculate a measure of reliability for the machine learning agent reliability score quantifying a failure rate of the machine learning agent with respect to the plurality of simulated components;
train the machine learning agent, based on the reliability score, to learn a configuration of the physical control system that maximizes the reliability score by iteratively repeating the accelerated test procedure; and
deploy machine learning agent to the plurality of components of the physical control system to implement the configuration of the control system that maximizes the reliability score in accordance with the nominal phases, the aging phases, the action frequencies, and the actions ranges of the accelerated test procedure.
9. The system of claim 8 , wherein the nominal phase comprises a number of operational hours and a number of power cycles calculated by an offline model based on the product specifications.
10. The system of claim 8 , wherein the aging phase comprises a failure probability calculated using a statistical probability distribution based on the product specifications.
11. The system of claim 8 , wherein the plurality of components are mechanical components, and the stress parameter is mechanical stress.
12. The system of claim 8 , wherein the plurality of components are electrical components, and the stress parameter is voltage.
13. The system of claim 8 , wherein the accelerated test procedure comprises a plurality of corner tests, each corner test defining a sequence of steps for evaluating the machine learning agent under an associated load condition.
14. The system of claim 8 , wherein a maximum action frequency is equal to a power cycle frequency of the plurality of components.
15. A computer-readable storage medium having encoded thereon computer-readable instructions that when executed by a system cause the system to:
extract a plurality of product specifications from a plurality of components within a physical control system controlled by a machine learning agent;
determine a nominal phase defining a time period of a consistent failure rate and an aging phase defining a time period of an increasing failure rate for each component of the plurality of components based on the plurality of product specifications;
calculate an action frequency for each component of the plurality of components based on the associated nominal phase, the action frequency defining a frequency at which the machine learning agent can actuate each component;
determine an action range for each component of the plurality of components using one or more acceleration factors associated with a stress parameter of the component, the action range defining a range of values configurable by the machine learning agent for each component;
construct an accelerated test procedure for determining a reliability of the machine learning agent based on the nominal phases, the aging phases, the action frequencies, and the action ranges of the plurality of components;
execute the accelerated test procedure within a simulated environment comprising the machine learning agent and a plurality of simulated components that correspond to the plurality of components to calculate a measure of reliability for the machine learning agent reliability score quantifying a failure rate of the machine learning agent with respect to the plurality of simulated components;
train the machine learning agent, based on the reliability score, to learn a configuration of the physical control system that maximizes the reliability score by iteratively repeating the accelerated test procedure; and
deploy machine learning agent to the plurality of components of the physical control system to implement the configuration of the control system that maximizes the reliability score in accordance with the nominal phases, the aging phases, the action frequencies, and the actions ranges of the accelerated test procedure.
16. The computer-readable storage medium of claim 15 , wherein the nominal phase comprises a number of operational hours and a number of power cycles calculated by an offline model based on the product specifications.
17. The computer-readable storage medium of claim 15 , wherein the aging phase comprises a failure probability calculated using a statistical probability distribution based on the product specifications.
18. The computer-readable storage medium of claim 15 , wherein a maximum action frequency is equal to a power cycle frequency of the plurality of components.
19. The system computer-readable storage medium of claim 15 , wherein the plurality of components are electrical components, and the stress parameter is voltage.
20. The system of claim 8 , wherein the accelerated test procedure comprises a plurality of corner tests, each corner test defining a sequence of steps for evaluating the machine learning agent under an associated load condition.