System and method for the detection of faults in a multi-variable system utilizing both a model for normal operation and a model for faulty operation

• US 9,366,451 B2
• Filed: 12/23/2011
• Issued: 06/14/2016
• Est. Priority Date: 12/24/2010
• Status: Active Grant

• Current Assignee: Commonwealth Scientific & Industrial Research Organisation (Commonwealth Of Australia As Represented By Department Of Industry, Innovation, Climate Change, Science, Research And Tertiary Education)
• Original Assignee: Commonwealth Scientific & Industrial Research Organisation (Commonwealth Of Australia As Represented By Department Of Industry, Innovation, Climate Change, Science, Research And Tertiary Education)
• Inventors: Guo, Ying, Li, Jiaming, West, Sam, Wall, Joshua, Platt, Glenn
• Primary Examiner(s): Hill, Stanley K
• Assistant Examiner(s): Bejcek, II, Robert H
• Application Number:
  US13/336,153
• Publication Number:
  US 20120185728A1
• Time in Patent Office: 1,635 Days
• Field of Search: 706/12
• US Class Current: 1/1
• CPC Class Codes: F24F 11/30 for purposes related to the operation of the system, e.g. for safety or monitoring
• CPC Class Codes: F24F 11/32 Responding to malfunctions or emergencies
• CPC Class Codes: F24F 11/38 Failure diagnosis
• CPC Class Codes: F24F 11/63 Electronic processing
• CPC Class Codes: F24F 2140/00 Control inputs relating to system states
• CPC Class Codes: G05B 2219/2614 HVAC, heating, ventillation, climate control
• CPC Class Codes: G05B 23/0254 based on a quantitative model, e.g. mathematical relationships between inputs and outputs; functions: observer, Kalman filter, residual calculation, Neural Networks
• CPC Class Codes: G06F 18/214 Generating training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06N 20/00 Machine learning

Abstract

A method for detecting faulty operation of a multi-variable system is described. The method includes receiving operational data from a plurality of components of the multi-variable system and processing the operational data in accordance with a plurality of dynamic machine learning fault detection models to generate a plurality of fault detection results. Each fault detection model uses a plurality of variables to model one or more components of the multi-variable system and is adapted to detect normal or faulty operation of an associated component or set of components of the multi-variable system. The plurality of fault detection results are output.

Claims

1. A method for detecting faulty operation of a HVAC system, the method including:

receiving operational data from a plurality of components of the HVAC system and feeding the operational data to a plurality of dynamic machine learning fault detection models;

automatically mapping the HVAC system by processing the operational data using the plurality of dynamic machine learning fault detection models;

training one or more of the fault detection models to learn patterns of normal operation of the HVAC system wherein faults are detected as deviations in the normal operation and training one or more of the fault detection models to learn patterns of faulty operation of the HVAC system wherein normal operation is detected as deviations in the faulty operation;

generating a plurality of fault detection results noting the presence or absence of a fault through the detection of deviations in the patterns of normal operation or faulty operation of the HVAC system;

each fault detection model using a plurality of variables to model one or more components of the HVAC system and being adapted to detect normal or faulty operation of an associated component or set of components of the HVAC system; and

outputting the plurality of fault detection results noting the presence or absence of a fault through the detection of deviations in the patterns of normal operation and faulty operation of the HVAC system.

2. The method according to claim 1, wherein at least two fault detection models are adapted to detect normal or faulty operation of different associated components or sets of components.

3. The method according to claim 1, wherein the plurality of fault detection models includes:

at least one general fault detection model adapted to detect normal or faulty operation of a general set of components of the HVAC system; and

at least one specific fault detection model adapted to detect normal or faulty operation of a specific component or set of components of the HVAC system, wherein

each specific fault detection model is associated with a general fault detection model, and

the component or set of components associated with the specific fault detection model is a subset of or related to the set of components of the associated general fault detection model.

4. The method according to claim 3, wherein one of the at least one general fault detection models is adapted to detect normal or faulty operation of the HVAC system as a whole.

5. The method according to claim 3, wherein the operational data is only processed in accordance with the at least one specific fault detection model if processing the operational data in accordance with the associated general fault detection model indicates the existence of a fault.

6. The method according to claim 3, wherein if both a general fault detection process and a specific fault detection process associated with that general fault detection process indicate faulty operation of the HVAC system, the fault is diagnosed as a fault in the component or set of components associated with the specific fault detection process.

7. The method according to claim 1, wherein at least one of the plurality of fault detection models is selected from a group including dynamic Bayesian network based fault detection models and hidden Markov model based fault detection models.

8. The method according to claim 1, further including:

processing the plurality of fault detection results in accordance with a data fusion process to generate a fused fault detection result, the fused fault detection result including information as to whether a fault is perceived in the operation of the HVAC system; and

outputting said fused fault detection result.

9. The method according to claim 8, wherein the data fusion process is a Dempster-Shafer based data fusion process.

10. The method according to claim 1, further including:

for each of the plurality of fault detection results calculating an associated confidence level; and

using the calculated confidence levels to select which of the plurality of fault detection results is or are likely to be correct.

11. The method according to claim 10, further including fusing said plurality of confidence levels to provide a fused confidence level result;

and

using said fused confidence level result along with said fused fault detection result to determine the existence of and diagnose a fault in the HVAC system.

12. The method according to claim 10, wherein said confidence levels are belief masses calculated according to a Dempster-Shafer based process.

13. The method according to claim 10, further including:

selecting one or more of the fault detection results based on the confidence level associated with the fault detection result;

generating a set of variables, the set of variables including the variables modelled by the fault detection models which generated the selected fault detection results; and

outputting said set of variables for use in diagnose a fault in the HVAC system.

14. The method according to claim 1, further including:

processing the operational data in accordance with at least one of the fault detection models multiple times to generate multiple results from said at least one fault detection model, and

clustering the multiple results in accordance with a clustering algorithm in order to increase the reliability of the fault detection result.

15. The method according to claim 14, wherein the clustering algorithm is a K-means clustering algorithm.

16. The method according to claim 1, wherein if two fault detection results are contradictory, the method further includes:

flagging one or both of the fault detection models yielding the contradictory results; and

modifying or retraining each flagged fault detection model.

17. The method according to claim 1, wherein if a fault detection result is identified as being a false alarm, the method further includes:

flagging the fault detection model yielding the result identified to be a false alarm; and

modifying or retraining the flagged fault detection models.

18. The method according to claim 14, wherein if two fault detection results are contradictory, the method further includes:

flagging one or both of the fault detection models yielding the contradictory results;

modifying or retraining each flagged fault detection model; and

using said flagged fault detection models to modify the clustering algorithm in order to improve the clustering of the multiple search results.

19. A method for detecting faulty operation of a multi-variable system, the method including:

receiving operational data from a plurality of components of the multi-variable system and feeding the operational data to a plurality of dynamic machine learning fault detection models;

automatically mapping the multi-variable system by processing the operational data using the plurality of dynamic machine learning fault detection models;

training one or more of the fault detection models to learn patterns of normal operation of the multi-variable system wherein faults are detected as deviations in the normal operation and training one or more of the fault detection models to learn patterns of faulty operation of the multi-variable system wherein normal operation is detected as deviations in the faulty operation;

generating a plurality of fault detection results noting the presence or absence of a fault through the detection of deviations in the patterns of normal operation or faulty operation of the multi-variable system;

each fault detection model using a plurality of variables to model one or more components of the multi-variable system and being adapted to detect normal or faulty operation of an associated component or set of components of the multi-variable system; and

outputting the plurality of fault detection results noting the presence or absence of a fault through the detection of deviations in the patterns of normal operation and faulty operation of the multi-variable system.

20. The method according to claim 19, wherein at least two fault detection models are adapted to detect normal or faulty operation of different associated components or sets of components.

21. The method according to claim 19, wherein the plurality of fault detection models includes:

at least one general fault detection model adapted to detect normal or faulty operation of a general set of components of the multi-variable system; and

at least one specific fault detection model adapted to detect normal or faulty operation of a specific component or set of components of the multi-variable system, wherein

each specific fault detection model is associated with a general fault detection model, and

the component or set of components associated with the specific fault detection model is a subset of or related to the set of components of the associated general fault detection model.

22. The method according to claim 21, wherein one of the at least one general fault detection models is adapted to detect normal or faulty operation of the multi-variable system as a whole.

23. The method according to claim 21, wherein the operational data is only processed in accordance with the at least one specific fault detection model if processing the operational data in accordance with the associated general fault detection model indicates the existence of a fault.

24. The method according to claim 21, wherein if both a general fault detection process and a specific fault detection process associated with that general fault detection process indicate faulty operation of the multi-variable system, the fault is diagnosed as a fault in the component or set of components associated with the specific fault detection process.

25. The method according to claim 19, wherein at least one of the plurality of fault detection models is selected from a group including dynamic Bayesian network based fault detection models and hidden Markov model based fault detection models.

26. The method according to claim 19, further including:

processing the plurality of fault detection results in accordance with a data fusion process to generate a fused fault detection result, the fused fault detection result including information as to whether a fault is perceived in the operation of the multi-variable system; and

outputting said fused fault detection result.

27. The method according to claim 26, wherein the data fusion process is a Dempster-Shafer based data fusion process.

28. The method according to claim 19, further including:

for each of the plurality of fault detection results calculating an associated confidence level; and

using the calculated confidence levels to select which of the plurality of fault detection results is or are likely to be correct.

29. The method according to claim 28, further including fusing said plurality of confidence levels to provide a fused confidence level result;

and

using said fused confidence level result along with said fused fault detection result to determine the existence of and diagnose a fault in the multi-variable system.

30. The method according to claim 28, wherein said confidence levels are belief masses calculated according to a Dempster-Shafer based process.

31. The method according to claim 28, further including:

selecting one or more of the fault detection results based on the confidence level associated with the fault detection result;

generating a set of variables, the set of variables including the variables modelled by the fault detection models which generated the selected fault detection results; and

outputting said set of variables for use in diagnose a fault in the multi-variable system.

32. The method according to claim 19, further including:

processing the operational data in accordance with at least one of the fault detection models multiple times to generate multiple results from said at least one fault detection model, and

clustering the multiple results in accordance with a clustering algorithm in order to increase the reliability of the fault detection result.

33. The method according to claim 32, wherein the clustering algorithm is a K-means clustering algorithm.

34. The method according to claim 19, wherein if two fault detection results are contradictory, the method further includes:

flagging one or both of the fault detection models yielding the contradictory results; and

modifying or retraining each flagged fault detection model.

35. The method according to claim 19, wherein if a fault detection result is identified as being a false alarm, the method further includes:

flagging the fault detection model yielding the result identified to be a false alarm; and

modifying or retraining the flagged fault detection models.

36. The method according to claim 34, wherein if two fault detection results are contradictory, the method further includes:

flagging one or both of the fault detection models yielding the contradictory results;

modifying or retraining each flagged fault detection model; and

using said flagged fault detection models to modify the clustering algorithm in order to improve the clustering of the multiple search results.

37. The method according to claim 19, wherein the multi-variable system is selected from a group including:

a HVAC system;

a refrigeration system;

an energy distribution system;

an automated lighting system;

a security/alarm system;

a smoke/fire detection system; and

a gas detection system.

38. A method for detecting faulty operation of a multi-variable system, the method including:

receiving operational data from a plurality of components of the multi-variable system and feeding the operational data to a plurality of dynamic machine learning fault detection models;

automatically mapping the multi-variable system by processing the operational data using the plurality of dynamic machine learning fault detection models;

training one or more of the fault detection models to learn patterns of normal operation of the multi-variable system, wherein faults are detected as probabilistic deviations in operational data and normal operation is detected as probabilistic matches in operational data, and training one or more of the fault detection models to learn patterns of faulty operation of the multi-variable system, wherein normal operation is detected as probabilistic deviations in operational data and faulty operation is detected as probabilistic matches in operational data;

generating a plurality of fault detection results noting the presence or absence of a fault through the detection of deviations in the patterns of normal operation or faulty operation of the multi-variable system;

each fault detection model using a plurality of variables to model one or more components of the multi-variable system and being adapted to detect normal or faulty operation of an associated component or set of components of the multi-variable system; and

outputting the plurality of fault detection results noting the presence or absence of a fault.

39. The method according to claim 38 wherein generating the plurality of fault detection results includes noting a type of fault and the outputting the plurality of fault detection results includes noting the type of fault.