Providing early warning and assessment of vehicle design problems with potential operational impact

• US 10,372,872 B2
• Filed: 04/22/2016
• Issued: 08/06/2019
• Est. Priority Date: 04/22/2016
• Status: Active Grant

• Current Assignee: The Boeing Co.
• Original Assignee: The Boeing Co.
• Inventors: Brelje, Benjamin J., Bouvier, Kenneth D.
• Primary Examiner(s): Roni, Syed A
• Application Number:
  US15/136,170
• Publication Number:
  US 20170308640A1
• Time in Patent Office: 1,201 Days
• Field of Search: None
• CPC Class Codes: G05B 2219/45071 Aircraft, airplane, ship cleaning manipulator, paint stripping
• CPC Class Codes: G05B 23/0227 Qualitative history assessment, whereby the type of data acted upon, e.g. waveforms, images or patterns, is not relevant, e.g. rule based assessment; if-then decisions
• CPC Class Codes: G06F 30/15 Vehicle, aircraft or watercraft design

Abstract

Method and apparatus for unsupervised aircraft design. A plurality of design problem data and service event data for an aircraft is received from an electronic data repository. Embodiments communicate with sensors on the aircraft during flight operations and capturing service data and sensor data. A high order vector is generated for each received problem report and service event data and each high order vector is concatenated into a high order vector matrix. Embodiments generate a reduced order symptom-normalized matrix by factorization of the concatenated high order vector matrix and generate a similarity matrix from the symptom-normalized matrix. An impact score is computed for each in-service event data as a function of similar problem reports using the similarity matrix. Embodiments generate a priority matrix configured to identify service event data having high impact scores and communicate a real-time alert of the high impact scored service event.

Claims

1. An unsupervised machine learning aircraft design method, comprising:

receiving, by operation of at least one processor coupled to an electronic data repository and a communication network, a plurality of design problem data and in-service event data for an aircraft from the electronic data repository, wherein the plurality of design problem data and in-service event data includes heterogeneous natural language data, free text data, numeric data, and scalar and vector sensor data;

generating a high order vector for each of the plurality of design problem data and in-service event data, wherein a first high order vector is generated for a first design problem data and in-service event data of the plurality of design problem data and in-service event data, based on natural language text contained within the first design problem data and in-service event data;

concatenating each high order vector, including the first high order vector, into a high order vector matrix;

generating a reduced order symptom-normalized matrix by dimensionality reduction of the high order vector matrix;

generating a similarity matrix from the reduced order symptom-normalized matrix by computing a similarity metric between the reduced order symptom-normalized matrix and each of the plurality of problem data and in-service event data;

computing an impact score for each design problem represented in the similarity matrix as a function of a corresponding portion of the plurality of design problem data and in-service event data;

generating a priority matrix configured to identify design problems having high impact scores;

communicating an alert of the high impact scored design problems; and

prioritizing corresponding maintenance schedule and design updates for at least one aircraft during development, based on the generated similarity matrix.

2. The method of claim 1, wherein the plurality of design problem data and in-service event data includes one or more of in-service event fault codes and sensor data.

3. The method of claim 1, wherein computing the impact score for each design problem is based on analyzing a database of event reports comprising maintenance or other operational consequences including historical dispatch delays, maintenance time and maintenance cost associated with a plurality of historical in in-service problem reports on vehicles, to determine a respective cost value.

4. The method of claim 3, further comprising:

prioritizing a plurality of design updates on at least one aircraft during development, based on the generated similarity matrix, in order to minimize an overall severity of the maintenance or other operational consequences for a delivered vehicle.

5. The method of claim 1, wherein receiving the plurality of design problem data and in-service event data for the aircraft from the electronic data repository further comprises:

data mining a database comprising a plurality of lab tests problem reports, a plurality of vehicle test problem reports, a plurality of vehicle system health management fault reports, and a plurality of vehicle sensor inputs.

6. The method of claim 1, further comprising:

providing a plurality of predefined trigger conditions, each specifying conditional logic for one or more types of sensor data collected from the sensors on the aircraft; and

upon determining that one or more sensor data values from the sensors on the aircraft satisfy one of plurality of predefined trigger conditions, recording a sensor event in the plurality of design problem data and in-service event data.

7. The method of claim 1, wherein in- service events in the in-service event data are automatically recorded by control logic within the aircraft.

8. The method of claim 1, wherein generating the high order vector for each of the plurality of design problem data and in-service event data, further comprises:

performing a vector space analysis of the plurality of design problem data and in-service event data, such that a respective vectorial representation is calculated based on the heterogeneous natural language data, free text data, numeric data, and scalar and vector sensor data.

9. The method of claim 8, wherein generating the reduced order symptom-normalized matrix by dimensionality reduction of the high order vector matrix further comprises:

performing a truncated singular value decomposition factorization of the high order vector matrix, wherein the reduced order symptom-normalized matrix approximates the high order vector matrix.

10. The method of claim 9, wherein computing the similarity metric between the reduced order symptom-normalized matrix and each of the plurality of design problem data and in-service event data further comprises:

calculating a measure of similarity between each vectorial representation within the reduced order symptom-normalized matrix.

11. The method of claim 10, further comprising:

performing a clustering algorithm on a plurality of vectors within the reduced order symptom-normalized matrix.

12. The method of claim 1, wherein computing the impact score for each design problem is further based on an operational impact estimation algorithm.

13. The method of claim 12, wherein the operational impact estimation algorithm includes a measure of delay time, a measure of repair cost, and a measure of whether a flight was cancelled.

14. A system, comprising:

one or more computer processors; and

a memory containing computer program code that, when executed by operation of the one or more computer processors, performs an operation comprising;

retrieving problem report data for at least a first class of vehicle, comprising;

retrieving a plurality of user-submitted problem reports, each comprising a natural language description of a respective problem occurrence for a respective in-service vehicle of the first class of vehicle;

retrieving a plurality of service events, each specifying a fault code; and

retrieving a plurality of sensor events, each corresponding to a respective occurrence of a pattern of data being received from one or more sensor devices;

calculating a vector representation for each of the plurality of user-submitted problem reports, the plurality of service events and the plurality of sensor events, wherein a first vector representation is generated for a first user-submitted problem report containing natural language text describing the respective problem occurrence for the respective in-service vehicle of the first class of vehicle;

reducing a dimensionality of the vector representations;

calculating similarity values between the vector representations;

categorizing each of the vector representations into one of a plurality of problem categories, based on the similarity values;

calculating a respective measure of impact for each of the plurality of problem categories; and

prioritizing corresponding maintenance schedule and design updates for at least one aircraft during development, based on the calculated measures of impact.

15. The system of claim 14, the operation further comprising:

providing a plurality of predefined trigger conditions, each specifying conditional logic for one or more types of sensor data received from the one or more sensor devices; and

upon determining that one or more sensor data values from the one or more sensor devices satisfy one of plurality of predefined trigger conditions, recording one of the plurality of sensor events.

16. The system of claim 14, wherein calculating the vector representation for each of the plurality of user-submitted problem reports, the plurality of service events and the plurality of sensor events, comprises:

performing a vectorial analysis of the problem report data, such that a respective vectorial representation is calculated for each of a plurality of problem reports of the problem report data, based on a corresponding portion of the plurality of user-submitted problem reports, the plurality of service events and the plurality of sensor events; and

generating a matrix that includes all of the vectorial representations for the plurality of user-submitted problem reports, the plurality of service events and the plurality of sensor events.

17. The system of claim 16, wherein reducing the dimensionality of the vectorial representations further comprises:

performing a truncated singular value decomposition factorization of the matrix, to generate a reduced matrix, wherein reduced matrix approximates the matrix.

18. The system of claim 17, wherein calculating the similarity values between the vector representations further comprises calculating a respective similarity metric between each vectorial representation within the reduced matrix and each other vectorial representation within the reduced matrix, and

wherein categorizing each of the vector representations within the reduced matrix into one of a plurality of problem categories further comprises performing a clustering algorithm on each of a plurality of vectorial representations within the reduced matrix.

19. A non-transitory computer-readable medium containing computer program code that, when executed by operation of one or more computer processors, performs an operation comprising:

retrieving problem report data for at least a first class of vehicle, comprising;

retrieving a plurality of user-submitted problem reports, each comprising a natural language description of a respective problem occurrence for a respective in-service vehicle of the first class of vehicle;

retrieving a plurality of service events, each specifying a fault code; and

retrieving a plurality of sensor events, each corresponding to a respective occurrence of a pattern of data being received from one or more sensor devices;

calculating a vector representation for each of the plurality of user-submitted problem reports, the plurality of service events and the plurality of sensor events, wherein a first vector representation is generated for a first user-submitted problem report containing natural language text describing the respective problem occurrence for the respective in-service vehicle of the first class of vehicle;

reducing a dimensionality of the vector representations;

calculating similarity values between the vector representations;

categorizing each of the vector representations into one of a plurality of problem categories, based on the similarity values;

calculating a respective measure of impact for each of the plurality of problem categories; and

prioritizing corresponding maintenance schedule and design updates for at least one aircraft during development, based on the calculated measures of impact.