System and method for detecting and excluding outlier sensors in sensor-based monitoring

• US 20050065744A1
• Filed: 09/02/2004
• Published: 03/24/2005
• Est. Priority Date: 09/05/2003
• Status: Active Grant

• Current Assignee: Siemens Corp. (Siemens AG)
• Original Assignee: Siemens Corporate Research Incorporated (Siemens AG)
• Inventors: Fang, Ming, Neubauer, Claus, Yuan, Chao, Cataltepe, Zehra
• Granted Patent: US 7,096,159 B2
• Time in Patent Office: Days
• US Class Current: 702/104
• CPC Class Codes: G05B 17/02 electric
• CPC Class Codes: G05B 23/0221 Preprocessing measurements, e.g. data collection rate adjustment; Standardization of measurements; Time series or signal analysis, e.g. frequency analysis or wavelets; Trustworthiness of measurements; Indexes therefor; Measurements using easily measured parameters to estimate parameters difficult to measure; Virtual sensor creation; De-noising; Sensor fusion; Unconventional preprocessing inherently present in specific fault detection methods like PCA-based methods

Abstract

A system for detecting one or more faulty sensors in a multi-sensor monitor includes a partitioning module for partitioning sensor values generated by the multi-sensor monitor into two distinct sets, a training set and a validation set. The system also includes a training module for training a model using the sensor values belonging to the training set and applying the model to each sensor value belonging to the validation set so as to determine a range of acceptable sensor values. The system further includes an estimating module for obtaining an estimated sensor value for each sensor using the model, and a fault-determining module for testing at least one sensor combination if a sensor value is not within its range of acceptable sensor values. A sensor combination includes at least one sensor whose estimated sensor value is not within the range of acceptable values.

Claims

1. A system for detecting a faulty sensor in a multi-sensor monitor, the system comprising:

a partitioning module for partitioning a plurality of sensor values generated by the multi-sensor monitor into two sets, the sets defining a training set comprising a portion of the plurality of sensor values and a validation set comprising another portion of the plurality of sensor values;

a training module for training a model using the sensor values belonging to the training set and applying the model to each sensor value belonging to the validation set so as to determine a range of acceptable sensor values;

an estimating module for obtaining an estimated sensor value corresponding to each sensor of the multi-sensor monitor using the model; and

a fault-determining module for testing at least one sensor combination if at least one sensor value is not within the range of acceptable sensor values, the at least one sensor combination including at least one sensor whose estimated sensor value is not within the range of acceptable values.

2. The system of claim 1, wherein the training module, for each at least one sensor combination, obtains recomputed estimated sensor values corresponding to each sensor not included in the sensor combination.

3. The system of claim 1, wherein the fault-determining module determines at least one sensor is faulty if recomputed estimated sensor values corresponding to each other sensor are each within the range of acceptable sensor values when the at least one faulty sensor is excluded.

4. The system of claim 1, further comprising determining an out-of-range event.

5. The system of claim 1, wherein the estimating module determines a revised range of acceptable sensor values for each at least one sensor combination.

6. The system of claim 1, wherein the partitioning module randomly allocates a portion of the plurality of sensor values to the training set and allocates a different portion of the plurality of sensor values to the validation set.

7. The system of claim 1, wherein each range of acceptable sensor values comprises a minimum residual and a maximum residual.

8. A method of detecting a faulty sensor in a multi-sensor monitor, the method comprising the steps of:

partitioning a plurality of sensor values generated by the multi-sensor monitor into two sets, the sets defining a training set comprising a portion of the plurality of sensor values and a validation set comprising another portion of the plurality of sensor values;

training a model using the sensor values belonging to the training set and applying the model to each sensor value belonging to the validation set so as to determine a range of acceptable sensor values;

obtaining an estimated sensor value corresponding to each sensor of the multi-sensor monitor using the model; and

if at least one estimated sensor value corresponding to one of the sensors is not within the range of acceptable sensor values, testing at least one sensor combination comprising at least one sensor whose estimated sensor value is not within the range of acceptable values so as to determine whether the combination includes a faulty sensor.

9. The method of claim 8, wherein testing comprises, for each at least one sensor combination, obtaining recomputed estimated sensor values corresponding to each sensor not included in the sensor combination.

10. The method of claim 8, wherein the at least one sensor is faulty if recomputed estimated sensor values corresponding to each other sensor are each within the range of acceptable sensor values when the at least one faulty sensor is excluded.

11. The method of claim 8, further comprising determining an out-of-range event.

12. The method of claim 8, further comprising determining a revised range of acceptable sensor values for each at least one sensor combination.

13. The method of claim 8, wherein partitioning comprises randomly allocating a portion of the plurality of sensor values to the training set and allocating a different portion of the plurality of sensor values to the validation set.

14. The method of claim 8, wherein each range of acceptable sensor values comprises a minimum residual and a maximum residual.

15. A computer readable storage medium for use with multi-sensor monitor, the storage medium comprising computer instructions for:

partitioning a plurality of sensor values generated by the multi-sensor monitor into two sets, the sets defining a training set comprising a portion of the plurality of sensor values and a validation set comprising another portion of the plurality of sensor values;

training a model using the sensor values belonging to the training set and applying the model to each sensor value belonging to the validation set so as to determine a range of acceptable sensor values;

obtaining an estimated sensor value corresponding to each sensor of the multi-sensor monitor using the model; and

if at least one estimated sensor value corresponding to one of the sensors is not within the range of acceptable sensor values, testing at least one sensor combination comprising at least one sensor whose estimated sensor value is not within the range of acceptable values so as to determine whether the combination includes a faulty sensor.

16. The computer readable storage of claim 15, wherein the instruction for testing comprises an instruction to obtain, for each at least one sensor combination, recomputed estimated sensor values corresponding to each sensor not included in the sensor combination.

17. The computer readable storage of claim 15, wherein the at least one sensor is faulty if recomputed estimated sensor values corresponding to each other sensor are each within the range of acceptable sensor values when the at least one faulty sensor is excluded.

18. The computer readable storage of claim 15, further comprising an instruction for determining an out-of-range event.

19. The computer readable storage of claim 15, further comprising an instruction for determining a revised range of acceptable sensor values for each at least one sensor combination.

20. The computer readable storage of claim 15, wherein partitioning comprises randomly allocating a portion of the plurality of sensor values to the training set and allocating a different portion of the plurality of sensor values to the validation set.

21. The method of claim 15, wherein each range of acceptable sensor values comprises a minimum residual and a maximum residual.