Fault diagnosis

• US 6,606,580 B1
• Filed: 06/12/2000
• Issued: 08/12/2003
• Est. Priority Date: 05/09/2000
• Status: Expired due to Term

• Current Assignee: Rolls-Royce PLC (Rolls-Royce Holdings Plc)
• Original Assignee: Rolls-Royce PLC (Rolls-Royce Holdings Plc)
• Inventors: Zedda, Marco, Gulati, Ankush, Singh, Ritindar
• Primary Examiner(s): Hoff, Marc S.
• Assistant Examiner(s): SUAREZ, FELIX E
• Application Number:
  US09/592,296
• Time in Patent Office: 1,156 Days
• Field of Search: 702/123, 702/179, 702/182, 702/183, 702/185, 702/186, 702/188, 705/8, 712/227
• US Class Current: 702/185
• CPC Class Codes: G05B 13/0265 the criterion being a learning criterion
• 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 11/008 Reliability or availability analysis
• CPC Class Codes: G06Q 10/04 Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem" market predictions or forecasting for commercial activities G06Q30/0202
• CPC Class Codes: G07C 3/00 Registering or indicating the condition or the working of machines or other apparatus, other than vehicles

Abstract

Methods and systems for fault diagnosis are described. The methods are in particular, although not necessarily exclusively, for the detection of faults in multi-component systems, typically power plant, whose performance is characterised by a series of indirectly measurable performance parameters. The methods proposed are based on “optimisation techniques”, involving the optimisation of an objective function to obtain an estimate of performance parameters and operating parameters. In one aspect, a method is proposed that takes account of measurement biases. In another aspect, a method for optimising an objective function is proposed, which adopts a unique Evolution Program approach in which a population of possible solutions is sub-divided into a number of groups.

Claims

1. A method for estimating the performance of and/or detecting faults in components of a multi-component system,

the performance of each component in the system being defined by one or more performance parameters x, and

the performance parameters x being related to measurement parameters z which can be expressed as a function h( ) of the performance parameters x and operating parameters w defining an operating condition of the system,

the method comprising:

selecting a maximum number M_bias of fault effected measurements, and

optimising an objective function J(x,w) that gives a measure of the consistency between measured values of the measurement parameters z collected from the system and calculated values of the measurement parameters calculated using said function of the performance parameters and the operating parameters h(x,w), said optimisation including multiple evaluations of the objective function J(x,w),

each time the objective function J(x,w)is evaluated during said optimisation, the method comprising calculating a plurality of possible values for the objective function, each possible value being calculated by determining the objective function with a different M_bias of the measurement parameters eliminated, the value of the minimum one of said plurality of calculated possible values being assigned as the value of the objective function.

2. A method according to claim 1, wherein said function relating the measurement parameters to the performance and operating parameters is embodied in a performance simulation model designed to simulate the performance of the multi-component system to which the method is applied.

3. A method according to claim 1, wherein an Evolution Program is used for said optimisation of the objective function.

4. A method according to claim 3, wherein said Evolution Program is a Genetic Algorithm.

5. A method according to claim 3, wherein said optimisation of the objective function comprises the steps of:

(a) creating an initial population of strings, each string comprising a plurality of elements corresponding to the performance parameters x and operating parameters w, values being assigned to the string elements representing estimated values of said parameters;

(b) evaluating the objective function for each of the strings in the population and, on the basis of this evaluation, assigning a fitness value to each string; and

(c) creating a next generation population of strings using a selection operator and at least one of a cross-over operator and a mutation operator, the selection operator employing the assigned fitness values to determine the strings that survive to the next generation;

wherein at the outset of the method an assumed maximum number of fault affected components is set and, based on this, a series of fault classes defined corresponding to possible outcomes in terms of faulty components;

each of the strings of the initially created population being assigned to one of the fault classes, the values assigned to the elements of strings in a fault class representing performance parameters being constrained to only indicate fault affected values for performance parameters defining the performance of the respective fault affected component or components of the class; and

during creation of subsequent generation populations, the cross-over operator, if used, being constrained to operate on pairs of strings from the same fault class, and the mutation operator, if used, being constrained so that it does not operate on string elements representing performance parameters other than those performance parameters associated with the fault affected components within a class.

6. A method according to claim 5, wherein said selection operator is applied across the complete population of strings in all of the fault classes.

7. A method for estimating the performance of and/or detecting faults in components of a multi-component system,

the performance of each component in the system being defined by one or more performance parameters x, and

the performance parameters x being related to measurement parameters z which can be expressed as a function h( ) of the performance parameters x and operating parameters w defining an operating condition of the system,

the method comprising:

optimising an objective function J(x,w) which gives a measure of the consistency between measured values of the measurement parameters z collected from the system and calculated values of the measurement parameters calculated using said function of the performance parameters and the operating parameters h(x,w),

said optimisation being an Evolutionary Program based method comprising the steps of;

(a) creating an initial population of strings, each string comprising a plurality of elements corresponding to the performance parameters x and operating parameters w, values being assigned to the string elements representing estimated values of said parameters;

(b) evaluating the objective function for each of the strings in the population and, on the basis of this evaluation, assigning a fitness value to each string; and

(c) creating a next generation population of strings using a selection operator and at least one of a cross-over operator and a mutation operator, the selection operator employing the assigned fitness values to determine the strings that survive to the next generation;

wherein at the outset of the method an assumed maximum number of fault affected components is set and, based on this, a series of fault classes defined corresponding to possible outcomes in terms of faulty components;

each of the strings of the initially created population being assigned to one of the fault classes, the values assigned to the elements of strings in a fault class representing performance parameters being constrained to only indicate fault affected values for performance parameters defining the performance of the respective fault affected component or components of the class; and

during creation of subsequent generation populations, the cross-over operator, if used, being constrained to operate on pairs of strings from the same fault class, and the mutation operator, if used, being constrained so that it does not operate on string elements representing performance parameters other than those performance parameters associated with the fault affected components within a class.

8. A method according to claim 7, wherein said selection operator is applied across the complete population of strings in all of the fault classes.

9. A method according to claim 7, wherein steps (b) and (c) of the optimisation are sequentially repeated a plurality of times.

10. A method according to claim 9, further comprising employing a local search algorithm to periodically refine the better strings in the population.

11. A method according to claim 7, further comprising:

selecting a maximum number M_bias of fault effected measurements,

each time the objective function J(x,w)is evaluated during said optimisation, the method comprising calculating a plurality of possible values for the objective function, each possible value being calculated by determining the objective function with a different M_bias of the measurement parameters eliminated, the value of the minimum one of said plurality of calculated possible values being assigned as the value of the objective function.

12. A method according to claim 7, wherein:

said measurement parameters z are provided for two or more different operating points of said system,

step (b) of said optimisation comprising evaluating the objective function for each of said strings for each of said two or more operating points.

13. A method according to claim 12, wherein for each said string the calculated values of the objective function are combined, the fitness value assigned to the string being assigned on the basis of the combined objective function values.

14. A method according to claim 12, wherein said optimisation comprises a pareto-based approach.

15. An evolution Program based search or optimisation method for solving a problem, comprising the steps of:

(a) creating an initial population of strings, each string representing a possible solution to the problem;

(b) evaluating the problem for each of the strings in the population and, on the basis of this evaluation, assigning a fitness value to each string; and

(c) creating a next generation population of strings using a selection operator and at least one of a cross-over operator and a mutation operator, the selection operator employing the assigned fitness values to determine the strings that survive to the next generation;

wherein said strings are categorised at the outset of the method, in accordance with a pre-selected characteristic of said strings, to form a plurality of sub-populations comprising strings of the having the same value of said pre-selected characteristic; and

during creation of subsequent generation populations, the cross-over operator, if used, being constrained to operate on pairs of strings from the same sub-population, and the mutation operator, if used, being constrained to ensure that the strings of each sub-population retain their initial value of said pre-selected characteristic.

16. A diagnostic system for estimating the performance of and/or detecting faults in components of a multi-component system,

the performance of each component in the multi-component system being defined by one or more performance parameters x, and

the performance parameters x being related to measurement parameters z which can be expressed as a function h( ) of the performance parameters x and operating parameters w defining an operating condition of the multi-component system,

the diagnostic system comprising:

means for inputting and storing measured values of the measurement parameters z collected from the system,

means for storing a maximum number M_bias of fault affected measurements,

means for calculating calculated values of the measurement parameters using said function of the performance parameters and the operating parameters h(x,w), and

an optimiser for optimising an objective function J(x,w) that gives a measure of the consistency between said measured values of the measurement parameters z and said calculated values of the measurement parameters,

said optimiser comprising means for calculating a plurality of possible values for the objective function, each possible value being calculated by determining the objective function with a different M_bias of the measurement parameters eliminated, and means for assigning the value of the minimum one of said plurality of calculated possible values as the value of the objective function.

17. A diagnostic system according to claim 16, wherein said means for calculating calculated values of the measurement parameters comprises a performance simulation model designed to simulate the performance of the multi-component system.

18. A diagnostic system for estimating the performance of and/or detecting faults in components of a multi-component system,

the performance of each component in the multi-component system being defined by one or more performance parameters x, and

the performance parameters x being related to measurement parameters z which can be expressed as a function h( ) of the performance parameters x and operating parameters w defining an operating condition of the multi-component system,

the diagnostic system comprising:

means for inputting and storing measured values of the measurement parameters z collected from the system,

means for calculating calculated values of the measurement parameters using said function of the performance parameters and the operating parameters h(x,w),

means for setting an assumed maximum number of fault affected components is set,

means for defining a series of fault classes corresponding to possible outcomes in terms of faulty components; and

an optimiser for optimising an objective function J(x,w) that gives a measure of the consistency between said measured values of the measurement parameters z and said calculated values of the measurement parameters,

said optimiser comprising Evolutionary Program means including;

(a) means for creating an initial population of strings, each string comprising a plurality of elements corresponding to the performance parameters x and operating parameters w, values being assigned to the string elements representing estimated values of said parameters and means for assigning each of said strings to a related fault class;

(b) means for evaluating the objective function for each of the strings in the population and, on the basis of this evaluation, assigning a fitness value to each string; and

(c) means for creating a next generation population of strings using a selection operator and at least one of a cross-over operator and a mutation operator, the selection operator employing the assigned fitness values to determine the strings that survive to the next generation;

said means for creating the next generation comprising constraints for constraining the cross-over operator, if used, to operate on pairs of strings from the same fault class, and constraining the mutation operator, if used, so that it does not operate on string elements representing performance parameters other than those performance parameters associated with the fault affected components within a class.

19. A diagnostic system according to claim 18, wherein said means for calculating calculated values of the measurement parameters comprises a performance simulation model designed to simulate the performance of the multi-component system.

20. A diagnostic system according to claim 19, wherein said optimiser further comprises a local search algorithm for periodically refining selected strings in the population.