Identifying failures in an aeroengine

• US 8,682,616 B2
• Filed: 12/14/2009
• Issued: 03/25/2014
• Est. Priority Date: 12/15/2008
• Status: Active Grant

• Current Assignee: Safran Aircraft Engine Company
• Original Assignee: Snecma (Safran SA)
• Inventors: Lacaille, Jrme
• Primary Examiner(s): TSAI, CAROL S W
• Application Number:
  US13/139,774
• Publication Number:
  US 20110307220A1
• Time in Patent Office: 1,562 Days
• Field of Search: 702/56, 702/85, 702/116, 702/185, 702/193, 702/113, 702/121, 700/100, 700/114, 700/30.9, 324/537, 607/73, 60/392.81, 60/390.91, 602/04, 602/23, 703/2, 734/89, 73/18.4, 731/120.1, 73/13.7
• US Class Current: 702/185
• CPC Class Codes: G05B 2219/32191 Real time statistical process monitoring
• CPC Class Codes: G05B 2219/32201 Build statistical model of past normal proces, compare with actual process
• CPC Class Codes: G05B 23/024 Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks
• CPC Class Codes: G05B 23/0281 Quantitative, e.g. mathematical distance; Clustering; Neural networks; Statistical analysis
• CPC Class Codes: Y02P 90/02 Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]

Abstract

A method and a system for identifying failures in an aeroengine. The system includes: a mechanism defining a set of standardized indicators representative of operation of the aeroengine; a mechanism constructing an anomaly vector representative of a behavior of the engine as a function of the set of standardized indicators; a mechanism selecting in an event of an anomaly being revealed by the anomaly vector a subset of reference vectors having directions belonging to a determined neighborhood of a direction of the anomaly vector, the subset of reference vectors being selected from a set of reference vectors associated with failures of the aeroengine and determined using criteria established by experts; and a mechanism identifying failures associated with the subset of reference vectors.

Claims

1. A method of identifying failures in an aeroengine, the method comprising:

using sensors to collect time-series measurements from said aeroengine and its environment;

using a processor means to calculate from said time-series measurements indicators that are specific to elements of said aeroengine;

using the processor means to define from said specific indicators a set of standardized indicators that are representative of an operation of said aeroengine;

using the processor means to construct an anomaly vector representative of a behavior of said engine as a function of said set of standardized indicators;

using the processor means, in an event of an abnormality being revealed by said anomaly vector, to select a subset of reference vectors having directions belonging to a determined neighborhood of a direction of said anomaly vector, said subset of reference vectors being selected from a set of reference vectors associated with failures of said aeroengine and determined using criteria established by experts; and

using the processor means to identify the failures associated with said subset of reference vectors;

and wherein the selecting said subset of reference vectors comprises;

using the processor means to calculate geodesic distances between a projection of said anomaly vector and projections of said reference vectors on a sphere in a space of dimension equal to a number of indicators in said set of standardized indicators minus a number of linear relationships between the indicators;

using the processor means to compare said geodesic distances in pairs;

using the processor means to classify the reference vectors in increasing order of their geodesic distances relative to said anomaly vector; and

using the processor means to form said subset of reference vectors from first reference vectors having a classification order less than a determined rank.

2. A method according to claim 1, wherein said sphere is of radius 1.

3. A method according to claim 1, further comprising:

using the processor means to define, for each reference vector, an a priori probability of occurrence on the basis of criteria established by experts; and

using the processor means to calculate, for each reference vector, an a posteriori probability of occurrence as a function of said a priori probability of occurrence and of said geodesic distances.

4. A method according to claim 1, wherein said set of standardized indicators {tilde over ({tilde over (y)}₁, . . . {tilde over ({tilde over (y)}_m comprises indicators {tilde over (y)}₁, . . . {tilde over (y)}_n identified by the processor means using criteria established by experts.

5. A method according to claim 4, wherein said set of standardized indicators {tilde over ({tilde over (y)}₁, . . . {tilde over ({tilde over (y)}_m further comprises dynamic indicators constructed by the processor means as a function of the indicators at present and past instants {tilde over ({tilde over (y)}(t)=f({tilde over (y)}(s);

s≦

t) and representative of the behavior of said aeroengine over time.

6. A method according to claim 1, wherein said constructing said anomaly vector comprises:

using the processor means to form an indicator vector {tilde over ({tilde over (y)} from said set of indicators; and

using the processor means to construct said anomaly vector z by renormalizing said indicator vector {tilde over ({tilde over (y)} using formula;

z=Σ

<sup>−

1/2</sup>({tilde over ({tilde over (y)}−

μ

)

in which μ

is the mean of the indicator vectors and Σ

<sup>−

1/2</sup> is the root of a pseudo-inverse signal Σ

<sup>−

1</sup> of a covariance matrix Σ

.

7. A method according to claim 6, further comprising:

using the processor means to calculate the norm of said anomaly vector using a Mahalanobis distance;

d²=∥

z∥

²=({tilde over ({tilde over (y)}−

μ

)^TΣ

<sup>−

1</sup>({tilde over ({tilde over (y)}−

μ

); and

using the processor means to detect an abnormality of said aeroengine using a trigger threshold defined as a function of statistical distribution of said norm of the anomaly vector.

8. A method according to claim 1, wherein said set of reference vectors is constructed in accordance with caricatural behaviors of the indicators in the event of anomalies.

9. A method according to claim 3, further comprising:

using the processor means to establish a decision grid in application of criteria established by experts;

using the processor means to apply Bayesian rules to deduce per component probabilities of failures from said a posteriori probabilities of occurrence and from said decision grid; and

using the processor means to detect faulty physical components that are responsible for said failures in application of said per component failure probabilities.

10. A method according to claim 9, wherein said decision grid is formed of a matrix of conditional probabilities that a component is faulty, knowing that a failure has been observed and of a series of coefficients corresponding to a priori probabilities of failure of each component.

11. A method according to claim 9, wherein said decision grid is corroborated by machine learning.

12. A non-transitory computer readable medium including computer executable instructions for implementing the method of identifying failures according to claim 1, when executed by a processor.

13. A system for identifying failures in an aeroengine, the system comprising:

sensors for collecting time-series measurements from said aeroengine and its environment;

means for calculating from said time-series measurements indicators that are specific to elements of said aeroengine;

means for using said specific indicators to define a set of standardized indicators representative of an operation of said aeroengine;

means for constructing an anomaly vector representative of a behavior of said engine as a function of said set of standardized indicators;

means for selecting in an event of an anomaly being revealed by said anomaly vector a subset of reference vectors having directions belonging to a determined neighborhood of a direction of said anomaly vector, said subset of reference vectors being selected from a set of reference vectors associated with failures of said aeroengine and determined using criteria established by experts;

means for identifying the failures associated with said subset of reference vectors;

and wherein the means for selecting a subset of reference vectors comprises;

means for calculating geodesic distances between a projection of said anomaly vector and projections of said reference vectors on a sphere in a space of dimension equal to a number of indicators of said set of standardized indicators minus a number of linear relationships between said indicators;

means for comparing said geodesic distances in pairs;

means for classifying the reference vectors in an increasing order of their geodesic distances relative to said anomaly vector; and

means for forming said subset of reference vectors from first reference vectors having a classification order less than a determined rank.

14. A system according to claim 13, further comprising:

means for defining, for each reference vector, an a priori probability of occurrence in application of criteria established by experts; and

means for calculating, for each reference vector, an a posteriori probability of occurrence as a function of said a priori probability of occurrence and of said geodesic distances.

15. A system according to claim 13, further comprising:

means for establishing a decision grid in application of criteria established by experts;

means for using Bayesian rules to deduce per component failure probabilities from said a posteriori probabilities of occurrence and from said decision grid; and

means for detecting faulty physical components that are responsible for said failures according to said per component failure probabilities.