Method for detection and isolation of faulty sensors
Technical effects of the invention include use of a data-driven multivariate statistical method for the detection and isolation of sensor faults applied in a virtual flow metering context. In one implementation, the data-driven multivariate statistical method employs principal components analysis, weighted squared prediction error, and partial decomposition contribution plots.
1. A virtual flow meter, comprising:
a processor-based controller configured to:
acquire training measurements from a plurality of sensors over time, wherein the plurality of sensors are positioned in a flow path of a fluid and at least one of the plurality of sensors measures a flow of a multiphase production fluid;
perform a principal components analysis of the training data to generate a principal components model defining a principal components component and a residual subspace;
acquire current measurements from the plurality of sensors;
define a number of operating points using the previously acquired training measurements of the plurality of sensors,
project the current measurements of each operating point into the residual subspace to detect one or more outliers wherein the defined operation points are used as inputs, if present;
if one or more outliers are detected, apply partial decomposition to generate one or more contribution plots; and
identify a faulty sensor based on the contribution plots.
2. The virtual flow meter of claim 1 , wherein the controller comprises a processor based-controller.
3. The virtual flow meter of claim 1 , wherein the controller comprises an application specific integrated circuit.
4. The virtual flow meter of claim 1 , wherein the plurality of sensors comprise temperature sensors or pressure sensors.
5. The virtual flow meter of claim 1 , wherein projecting the current measurements into residual space defines a residual vector.
6. The virtual flow meter of claim 1 , wherein detecting one or more outliers comprises generating a weighted square prediction error metric that is assessed for deviations from expected measurements.
7. The virtual flow meter of claim 1 , wherein detecting one or more outliers comprises computing a weighted square prediction error index for each measurement period.
8. The virtual flow meter of claim 1 , comprising indicating detection of a sensor fault if one or more outliers are detected.
9. The virtual flow meter of claim 1 , wherein the partial decomposition allocates the contribution of each sensor of the plurality of sensors to the one or more outliers.
10. The virtual flow meter of claim 1 , wherein the number of operating points are defined through a receding horizon filtering, comparison with filtered values, and clustering based on measurements of specified variables.
11. A processor-based method for identifying faulty sensors in a fluid production network, comprising:
acquiring training measurements from a plurality of sensors over time, wherein the plurality of sensors are positioned in a flow path of a fluid and at least one of the plurality of sensors measures a flow of a multiphase production fluid;
performing a principal components analysis of the training data to generate a principal components model defining a principal components component and a residual subspace;
acquiring current measurements from the plurality of sensors;
defining a number of operating points using on the previously acquired training measurements of the plurality of sensors;
projecting the current measurements of each operating point into the residual subspace to detect one or more outliers wherein the defined operating points are used as inputs, if present;
if one or more outliers are detected, applying partial decomposition to generate one or more contribution plots; and
identify a faulty sensor based on the contribution plots.
12. The method of claim 11 , wherein the plurality of sensors comprise temperature sensors or pressure sensors.
13. The method of claim 11 , wherein projecting the current measurements into residual space defines a residual vector.
14. The method of claim 11 , comprising indicating detection of a sensor fault if one or more outliers are detected.
15. The method of claim 11 , wherein the partial decomposition allocates the contribution of each sensor of the plurality of sensors to the one or more outliers.
16. One or more computer-readable media comprising executable routines, which when executed by a processor cause acts to be performed comprising:
acquiring training measurements from a plurality of sensors over time, wherein the plurality of sensors are positioned in a flow path of a fluid and at least one of the plurality of sensors measures a flow of a multiphase production fluid;
performing a principal components analysis of the training data to generate a principal components model defining a principal components component and a residual subspace;
acquiring current measurements from the plurality of sensors;
defining a number of operating points using the previously acquired training measurements of the plurality of sensors;
projecting the current measurements of each operating point into the residual subspace to detect one or more outliers wherein the operating points are used as inputs, if present;
if one or more outliers are detected, applying partial decomposition to generate one or more contribution plots; and
identify a faulty sensor based on the contribution plots.
17. The one or more computer-readable media of claim 16 , wherein projecting the current measurements into residual space defines a residual vector.
18. The one or more computer-readable media of claim 16 , wherein the routines, when executed by the processor causes the act to be performed of indicating detection of a sensor fault if one or more outliers are detected.
19. The one or more computer-readable media of claim 16 , wherein the partial decomposition allocates the contribution of each sensor of the plurality of sensors to the one or more outliers.