Prognostics and health management system

• US 10,496,791 B2
• Filed: 04/17/2015
• Issued: 12/03/2019
• Est. Priority Date: 04/17/2014
• Status: Active Grant

• Current Assignee: Lockheed Martin Corporation (Martin Marietta Corporation)
• Original Assignee: Lockheed Martin Corporation (Martin Marietta Corporation)
• Inventors: Das, Sreerupa
• Primary Examiner(s): Nguyen, Hiep V
• Application Number:
  US14/689,413
• Publication Number:
  US 20150302163A1
• Time in Patent Office: 1,691 Days
• Field of Search: 705 2
• CPC Class Codes: G05B 23/0213 Modular or universal configuration of the monitoring system, e.g. monitoring system having modules that may be combined to build monitoring program; monitoring system that can be applied to legacy systems; adaptable monitoring system; using different communication protocols
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/02 Neural networks
• CPC Class Codes: G16H 40/67 for remote operation

Abstract

A flexible prognostics and health management (PHM) device is disclosed. Sensor data is generated that quantifies real-time conditions of components in a machine based on sensor signals generated by sensors. A PHM system loads a PHM module that implements externally invokable functions. The PHM system receives an analytics script. The analytics script comprises interpretable commands and predictive learning model parameter values associated with a predictive learning model. Some of the interpretable commands are configured to invoke some of the externally invokable functions with at least some of the predictive learning model parameter values. The PHM module invokes a command interpreter module to interpret an interpretable command and the predictive learning model parameter values and generate first executable bytecodes. The first executable bytecodes execute to generate a notification identifying a predicted future condition of a first component for presentation on a display device.

Claims

1. A method for implementing prognostics health management (PHM) on a machine having a plurality of components, comprising:

generating sensor data that quantifies real-time conditions of the plurality of components based on sensor signals generated by a plurality of corresponding sensors;

loading, by a PHM system, a PHM module generated at a first time and implementing a plurality of externally invokable functions;

receiving, by the PHM system, a first analytics script generated at a second time, the first analytics script comprising a plurality of first interpretable commands and a plurality of predictive learning model parameter values associated with a first predictive learning model, some of the plurality of first interpretable commands configured to invoke some of the plurality of externally invokable functions with at least some of the plurality of predictive learning model parameter values;

invoking, by the PHM module, a command interpreter module to interpret at least one first interpretable command and the at least some of the plurality of predictive learning model parameter values and generate first executable bytecodes based on the at least one first interpretable command and the at least some of the plurality of predictive learning model parameter values, the first executable bytecodes configured to invoke the at least some of the plurality of externally invokable functions with the at least some of the plurality of predictive learning model parameter values identified in the at least one first interpretable command; and

executing the first executable bytecodes to generate a notification identifying a predicted future condition of a first component for presentation on a display device based at least in part on the at least some of the plurality of predictive learning model parameter values and the sensor data that quantifies the real-time condition of the first component.

2. The method of claim 1, wherein generating the sensor data that quantifies the real-time conditions of the plurality of components based on the sensor signals generated by the plurality of corresponding sensors comprises:

receiving, by the PHM system, the sensor signals generated by the plurality of corresponding sensors, the sensor signals indicating the real-time conditions of the plurality of components; and

processing the sensor signals to generate the sensor data that quantifies the real-time conditions of the plurality of components.

3. The method of claim 1, wherein processing the sensor signals to generate the sensor data that quantifies the real-time conditions of the plurality of components comprises filtering the sensor signals and extracting features from the sensor signals to generate the sensor data.

4. The method of claim 1, wherein the first predictive learning model comprises a neural network and the plurality of predictive learning model parameter values comprises neural network weights generated by the neural network during a training phase of the neural network.

5. The method of claim 1, further comprising:

detecting, by the PHM module, the at least one first interpretable command in the first analytics script, the at least one first interpretable command configured to invoke a first externally invokable function of the plurality of externally invokable functions with the at least some of the plurality of predictive learning model parameter values; and

wherein invoking the command interpreter module to interpret the at least one first interpretable command and generate the first executable bytecodes further comprises;

passing, by the PHM module, the at least one first interpretable command and the at least some of the plurality of predictive learning model parameter values to the command interpreter module; and

generating, by the command interpreter module based on the at least one first interpretable command and the at least some of the plurality of predictive learning model parameter values, the first executable bytecodes that invoke the first externally invokable function with the at least some of the plurality of predictive learning model parameter values; and

wherein executing the first executable bytecodes to generate the notification identifying the predicted future condition of the first component for presentation on the display device further comprises;

invoking, by the first executable bytecodes, the first externally invokable function with the at least some of the plurality of predictive learning model parameter values; and

generating the notification identifying the predicted future condition of the first component for presentation on the display device.

6. The method of claim 1, wherein the PHM module is written in a first programming language having a first syntax, and wherein the plurality of first interpretable commands is written in a second interpretable programming language having a second syntax that is different from the first syntax.

7. The method of claim 6, wherein the plurality of first interpretable commands comprises English language words.

8. The method of claim 1, wherein the predicted future condition of the first component comprises a remaining useful life (RUL) of the first component, and further comprising:

predicting, by the first executable bytecodes, based on the sensor data and the at least some of the plurality of predictive learning model parameter values, the RUL of the first component; and

generating, by the first executable bytecodes, the notification for presentation on the display device that indicates the RUL of the first component.

9. The method of claim 8, further comprising:

receiving a second analytics script generated at a third time that is later than the second time, the second analytics script replacing the first analytics script, the second analytics script comprising a plurality of second interpretable commands and a plurality of updated predictive learning model parameter values associated with the first predictive learning model, some of the plurality of second interpretable commands configured to invoke some of the plurality of externally invokable functions with at least some of the plurality of updated predictive learning model parameter values;

invoking, by the PHM module, the command interpreter module to interpret at least one second interpretable command of the plurality of second interpretable commands and the at least some of the plurality of updated predictive learning model parameter values to generate second executable bytecodes;

executing the second executable bytecodes;

predicting, by the second executable bytecodes, based on the sensor data and the at least some of the plurality of updated predictive learning model parameter values, an updated RUL of the first component that is different from the RUL; and

generating, by the second executable bytecodes, a notification for presentation on the display device that indicates the updated RUL of the first component.

10. The method of claim 9, wherein the first component comprises one of a fuel pump and an alternator.

11. The method of claim 1, further comprising:

processing, by the PHM module, the first analytics script; and

assigning, by the PHM module based on the first analytics script, each corresponding sensor of the plurality of corresponding sensors a corresponding port value, and wherein the plurality of first interpretable commands is configured to access the sensor data associated with each corresponding sensor based on the corresponding port value.

12. The method of claim 1, wherein receiving the first analytics script generated at the second time comprises:

interfacing with a remote device; and

receiving the first analytics script generated at the second time from the remote device.

13. A prognostics health management (PHM) system for implementing PHM on a machine having a plurality of components, comprising:

a sensor interface configured to be coupled to a plurality of sensors;

a communication interface configured to communicate with a remote device;

a controller comprising a processor coupled to the sensor interface and the communication interface, and configured to;

generate sensor data that quantifies real-time conditions of the plurality of components based on sensor signals generated by a plurality of corresponding sensors;

load a PHM module generated at a first time and implementing a plurality of externally invokable functions;

receive a first analytics script generated at a second time, the first analytics script comprising a plurality of first interpretable commands and a plurality of predictive learning model parameter values associated with a first predictive learning model, some of the plurality of first interpretable commands configured to invoke some of the plurality of externally invokable functions with at least some of the plurality of predictive learning model parameter values; and

invoke, via the PHM module, a command interpreter module to interpret at least one first interpretable command of the plurality of first interpretable commands and the at least some of the plurality of predictive learning model parameter values and generate first executable bytecodes based on the at least one first interpretable command and the at least some of the plurality of predictive learning model parameter values, the first executable bytecodes configured to invoke the at least some of the plurality of externally invokable functions with the at least some of the plurality of predictive learning model parameter values identified in the at least one first interpretable command; and

execute the first executable bytecodes to generate a notification identifying a predicted future condition of a first component for presentation on a display device based at least in part on at least some of the plurality of predictive learning model parameter values and the sensor data that quantifies the real-time condition of the first component.

14. The PHM system of claim 13, wherein to generate the sensor data that quantifies the real-time conditions of the corresponding components based on sensor signals generated by the plurality of corresponding sensors, the controller is further configured to:

receive, by the PHM system, the plurality of sensor signals generated by the plurality of corresponding sensors, the plurality of sensor signals indicating the real-time conditions of the plurality of components; and

process the plurality of sensor signals to generate the sensor data that quantifies the real-time conditions of the plurality of components.

15. The PHM system of claim 14, wherein to process the sensor signals to generate the sensor data that quantifies the real-time conditions of the plurality of components, the controller is further configured to filter the sensor signals and extract features from the sensor signals to generate the sensor data.

16. The PHM system of claim 13, wherein the controller is further configured to:

detect, by the PHM module, the at least one first interpretable command in the first analytics script, the at least one first interpretable command configured to invoke a first externally invokable function of the plurality of externally invokable functions with the at least some of the plurality of predictive learning model parameter values; and

wherein to invoke the command interpreter module to interpret the at least one first interpretable command and generate the first executable bytecodes, the controller is further configured to;

pass, by the PHM module, the at least one first interpretable command and the at least some of the plurality of predictive learning model parameter values to the command interpreter module; and

generate, by the command interpreter module based on the at least one first interpretable command and the at least some of the plurality of predictive learning model parameter values, the first executable bytecodes that invoke the first externally invokable function with the at least some of the plurality of predictive learning model parameter values; and

wherein to execute the first executable bytecodes to generate the notification identifying the predicted future condition of the first component for presentation on the display device, the controller is further configured to;

invoke, by the first executable bytecodes, the first externally invokable function with the at least some of the plurality of predictive learning model parameter values; and

generate the notification identifying the predicted future condition of the first component for presentation on the display device.

17. The PHM system of claim 13, wherein the PHM module is written in a first programming language having a first syntax, and wherein the plurality of first interpretable commands is written in a second interpretable programming language having a second syntax that is different from the first syntax.

18. The PHM system of claim 17, wherein the plurality of first interpretable commands comprises English language words.

19. The PHM system of claim 13, wherein the first component comprises one of a fuel pump and an alternator.

20. A method for implementing prognostic health management (PHM) on a machine, comprising:

receiving, by a PHM system comprising a processor, a plurality of sensor signals from a plurality of corresponding sensors, the plurality of sensor signals being indicative of real-time conditions of a plurality of components;

accessing, by a PHM module implementing a plurality of externally invokable functions, a first analytics script generated at a point in time, the first analytics script comprising a plurality of first interpretable commands and a plurality of predictive learning model parameter values associated with a first predictive learning model, some of the plurality of first interpretable commands configured to invoke some of the plurality of externally invokable functions; and

in response to a real-time condition of a first component of the plurality of components, interpreting, by the PHM module, a respective first interpretable command and generating respective first executable bytecodes based on the respective first interpretable command, and causing execution of the respective first executable bytecodes to generate a notification identifying a predicted future condition of the first component for presentation on a display device based at least in part on at least some of the plurality of predictive learning model parameter values and the real-time condition of the first component.