METHOD AND SYSTEM FOR DIAGNOSTICS OF APPARATUS

• US 20100121609A1
• Filed: 11/10/2008
• Published: 05/13/2010
• Est. Priority Date: 11/10/2008
• Status: Active Grant

• Current Assignee: Mitek Analytics LLC
• Original Assignee: Mitek Analytics LLC
• Inventors: GORINEVSKY, Dimitry
• Granted Patent: US 8,121,818 B2
• Time in Patent Office: Days
• US Class Current: 702/183
• CPC Class Codes: G05B 23/0281 Quantitative, e.g. mathematical distance; Clustering; Neural networks; Statistical analysis

Abstract

Proposed is a method, implemented in software, for estimating fault state of an apparatus outfitted with sensors. At each execution period the method processes sensor data from the apparatus to obtain a set of parity parameters, which are further used for estimating fault state. The estimation method formulates a convex optimization problem for each fault hypothesis and employs a convex solver to compute fault parameter estimates and fault likelihoods for each fault hypothesis. The highest likelihoods and corresponding parameter estimates are transmitted to a display device or an automated decision and control system. The obtained accurate estimate of fault state can be used to improve safety, performance, or maintenance processes for the apparatus.

Claims

1. A method for computing diagnostic estimates for faults of an apparatus with condition sensors connected to a computer;

the method comprising;

processing data from the condition sensors to obtain a set of parity parameters y reflecting apparatus condition deviation from normality at time period t,

collecting the parity parameters y over a moving horizon interval of time of a fixed maximal duration and ending at time period t in a data vector Y(t),

computing estimates of at least one fault condition and likelihood parameters for each of the at least one fault condition, and

transmitting the computed estimates of the fault conditions to a display device or to an automated decision and control system or storing the estimates in memory;

wherein a fault condition k at time period t is characterized by fault intensity parameter x_k(t);

computing estimates of the fault intensity parameter x_k(t) over the moving horizon interval of time and likelihood parameters p_k for each fault condition k, said computation being done for one fault condition k at a time, said computation further being performed in two steps, the first step being a formulator step and the second step being an optimizer step,

wherein the formulator step formulates a convex optimization program for a fault condition using the data vector Y(t), and the fault signature corresponding to the fault condition k,

wherein the optimizer step numerically finds a solution of the convex optimization program encoded by the formulator step, the solution being computed with a pre-defined accuracy for fault condition k;

and whereby the computed estimates for faults comprises

estimates of fault condition intensity parameters x_k(t) over the moving horizon interval of time computed as an optimal solution or as a transformation of the solution, and

likelihood parameter p_k computed as an optimum value of the program or as a transformation of the optimum value.

2. The method of claim 1 wherein the formulator step further comprises formulating and the optimizer step further comprises solving a convex optimization program of either:

a) an isotonic or monotonic regression program,

b) a univariate convex program,

c) a Quadratic Program,

d) a Linear Program,

e) a Second-order Cone Program, or

f) a constrained convex optimization program.

3. A method of claim 1 wherein the convex optimization program further comprises additional decision variables in addition to the fault intensity parameters x_k(t).

4. The method of claim 1 wherein the parity parameters y further comprise prediction residuals obtained as a difference of obtained readings from the sensor and readings predicted for an apparatus model which receives the same inputs as the apparatus;

the apparatus model comprising either a dynamic model, a nonlinear map, a set of static values corresponding to a chosen steady state regime, or another computer simulation model of the apparatus.

5. The method of claim 1 wherein fault signatures represent responses observed in the data y when a fault occurs.

6. The method of claim 1 wherein the method is implemented on-line in a computer or computers connected to the sensors of the apparatus or implemented off-line by collecting data from the apparatus, transmitting it by electronic means to a computer implementing the method, and performing the method computations at a later time.

7. The method of claim 1 wherein the fault condition parameters and likelihood parameters computed by the optimizer are used for improving safety of the apparatus operation, or for improving apparatus performance, or for scheduling a maintenance action.

8. The method of claim 1 where the formulator step further comprises formulating and the optimizer step further comprises solving the convex problem when one or more of the components of vector Y(t) is missing or unavailable.

9. A system for computing diagnostic estimates for faults of an apparatus with condition sensors connected to a computer;

the system comprising;

a processor processing data from the sensors to obtain a set of parity parameters y reflecting apparatus condition deviation from normality at time period t,

a collector collecting the parity parameters y over a moving horizon interval of time of a fixed maximal duration and ending at time period t in a data vector Y(t),

a computing circuit computing estimates of fault conditions and likelihood parameters for each of the fault conditions, and

a transmitter transmitting the computed estimates of the fault conditions to a display device or to an automated decision and control system or storing the estimates in memory;

wherein a fault condition k at time period t is characterized by fault intensity parameter x_k(t),

a computing circuit computing fault intensity parameters x_k(t) over the moving horizon interval of time and likelihood parameters p_k for each fault condition k, said computing is done for one fault condition k at a time, said computing is performed in two steps, the first step being a formulator step and the second step being an optimizer step,

the formulator step formulates a convex optimization program for fault condition using the moving horizon data vector Y(t), and the fault signature corresponding to the fault condition k,

the optimizer step numerically finds the solution of the convex optimization program encoded by the formulator, the solution is computed with a pre-defined accuracy for fault condition k;

whereby the diagnostic estimates for faults comprises

estimates of fault condition intensity parameters x_k(t) over the moving horizon interval of time computed as the optimal solution or as a transformation of the said solution, and

likelihood parameter p_k computed as the optimum value of the program or as a transformation of the said optimum value.

10. A system of claim 9 wherein the formulator step formulates and the optimizer step solves one of the following optimization programs

a) an isotonic or monotonic regression program

b) a univariate convex program

c) a Quadratic Program

d) a Linear Program

e) a Second-order Cone Program, or

f) a constrained convex optimization program

11. The system of claim 9 wherein the convex program for the fault condition includes additional decision variables in addition to the fault intensity parameters x_k(t).

12. The system of claim 9 wherein the parity parameters y further comprise prediction residuals obtained as a difference of the obtained sensor readings and the readings predicted for an apparatus model which receives the same inputs as the apparatus;

the apparatus model comprising either a dynamic model, a nonlinear map, a set of static values corresponding to a chosen steady state regime, or another computer simulation model of the apparatus.

13. The system of claim 9 wherein fault signatures represent responses observed in the parity parameters y when a fault occurs.

14. The system of claim 9 wherein the system is implemented on-line in a computer or computers connected to the sensors of the apparatus or implemented off-line by collecting data from the apparatus, transmitting it by electronic means to a computer implementing the method, and performing the method computations at a later time.

15. The system of claim 9 wherein the fault condition parameters and likelihood parameters computed by the optimizer are used for improving safety of the apparatus operation, or for improving apparatus performance, or for scheduling a maintenance action.

16. The system of claim 9 wherein the formulator step formulates and the optimizer step solves the convex problem when one or more of the components of vector Y(t) is missing or unavailable.

17. A computer readable media, which, when executed on a computer, implements a method for computing diagnostic estimates for faults of an apparatus with condition sensors connected to a computer;

the method comprising;

processing data from the condition sensors to obtain a set of parity parameters y reflecting apparatus condition deviation from normality at time period t,

collecting the parity parameters y over a moving horizon interval of time of a fixed maximal duration and ending at time period t in a data vector Y(t),

computing estimates of at least one fault condition and likelihood parameters for each of the at least one fault condition, and

transmitting the computed estimates of the fault conditions to a display device or to an automated decision and control system or storing the estimates in memory;

wherein a fault condition k at time period t is characterized by fault intensity parameter x_k(t);

computing estimates of the fault intensity parameter x_k(t) over the moving horizon interval of time and likelihood parameters p_k for each fault condition k, said computation being done for one fault condition k at a time, said computation further being performed in two steps, the first step being a formulator step and the second step being an optimizer step,

wherein the formulator step formulates a convex optimization program for a fault condition using the data vector Y(t), and the fault signature corresponding to the fault condition k,

wherein the optimizer step numerically finds a solution of the convex optimization program encoded by the formulator step, the solution being computed with a pre-defined accuracy for fault condition k;

and wherein the computed estimates for faults comprises;

estimates of fault condition intensity parameters x_k(t) over the moving horizon interval of time computed as an optimal solution or as a transformation of the solution, and

likelihood parameter p_k computed as an optimum value of the program or as a transformation of the optimum value.

18. The computer readable media of claim 17 wherein the formulator step further comprises formulating and the optimizer step further comprises solving a convex optimization program of either:

a) an isotonic or monotonic regression program,

b) a univariate convex program,

c) a Quadratic Program,

d) a Linear Program,

e) a Second-order Cone Program, or

f) a constrained convex optimization program.

19. The computer readable media of claim 17 wherein the convex optimization program further comprises additional decision variables in addition to the fault intensity parameters x_k(t).

20. The computer readable media of claim 17 wherein the parity parameters y further comprise prediction residuals obtained as a difference of obtained readings from the sensor and readings predicted for an apparatus model which receives the same inputs as the apparatus;

the apparatus model comprising either a dynamic model, a nonlinear map, a set of static values corresponding to a chosen steady state regime, or another computer simulation model of the apparatus.

21. The computer readable media of claim 17 wherein fault signatures represent responses observed in the data y when a fault occurs.

22. The computer readable media of claim 17 wherein the method is implemented on-line in a computer or computers connected to the sensors of the apparatus or implemented off-line by collecting data from the apparatus, transmitting it by electronic means to a computer implementing the method, and performing the method computations at a later time.

23. The computer readable media of claim 17 where the fault condition parameters and likelihood parameters computed by the optimizer are used for improving safety of the apparatus operation, or for improving apparatus performance, or for scheduling a maintenance action.

24. The computer readable media of claim 17 where the formulator step further comprises formulating and the optimizer step further comprises solving the convex problem when one or more of the components of vector Y(t) is missing or unavailable.