EVENT-DRIVEN FAULT DIAGNOSIS FRAMEWORK FOR AUTOMOTIVE SYSTEMS

• US 20110238258A1
• Filed: 03/24/2010
• Published: 09/29/2011
• Est. Priority Date: 03/24/2010
• Status: Active Grant

• Current Assignee: GM Global Technology Operations LLC (General Motors Company)
• Original Assignee: GM Global Technology Operations Incorporated (Motors Liquidation Co. GUC Trust)
• Inventors: Singh, Satnam, Bandyopadhyay, Pulak, Chougule, Rahul
• Granted Patent: US 8,301,333 B2
• Time in Patent Office: Days
• US Class Current: 701/33
• CPC Class Codes: G07C 5/0808 Diagnosing performance data testing of vehicles G01M17/00; testing of electrical installation on vehicles G01R31/005

Abstract

Systems and methods for capturing and analyzing significant parameter data from vehicle systems whenever a diagnostic trouble code (DTC) is triggered. A multi-dimensional matrix is constructed, with vehicles, DTCs, and parameter data comprising three dimensions of the matrix. The data matrix is populated with DTC and parameter data from many different vehicles, either when vehicles are taken to a dealer for service, or via wireless data download. Time can be added as a fourth dimension of the matrix, providing an indication of whether a particular system or component is temporally degrading. When sufficient data is accumulated, the data matrix is pre-processed, features are extracted from the data, and the features are classified, using a variety of mathematical techniques. Trained classifiers are then used to diagnose the root cause of any particular fault signal, and also to provide a prognosis of system health and remaining useful life.

Claims

1. A method for diagnosing causes of faults in vehicle systems, said method comprising:

defining a plurality of diagnostic trouble codes which are applicable to various fault conditions in a vehicle'"'"'s systems, criteria for triggering each diagnostic trouble code, and a list of parameter identification data which should be captured for each diagnostic trouble code;

providing a controller onboard a vehicle which is programmed with the diagnostic trouble codes and the criteria for triggering each diagnostic trouble code, and which is also programmed with the list of parameter identification data which should be captured for each diagnostic trouble code;

providing sensors onboard the vehicle as needed to capture the parameter identification data;

capturing the diagnostic trouble code and parameter identification data by the controller when a fault condition occurs in a vehicle system, where each fault condition occurrence is known as a fault event;

downloading the diagnostic trouble code and parameter identification data from the controller on the vehicle to a central computer;

storing diagnostic trouble code and parameter identification data for fault events from multiple vehicles on the central computer, where each fault event has an actual failure mode attributed to it;

analyzing the diagnostic trouble code and parameter identification data from the fault events stored on the central computer using mathematical models which can learn to correlate features in the data with the actual failure mode for each fault event; and

using the mathematical models to diagnose a failure mode for additional fault events.

2. The method of claim 1 wherein analyzing the diagnostic trouble code and parameter identification data includes a training phase and a testing phase.

3. The method of claim 2 wherein the training phase includes:

providing input data comprising a matrix of data for each of a plurality of failure modes, where each matrix of data identifies the failure mode to which it pertains, and includes diagnostic trouble code and parameter identification data for fault events from a plurality of vehicles;

using a pre-processing module to normalize and reduce variation in the input data;

using a feature extraction module to extract characteristic features of the input data;

using a classifier module to identify patterns in the features of the input data and associate the patterns with the actual failure mode; and

storing techniques and settings used in the feature extraction module and the classifier module which most effectively correlate the input data with the actual failure modes.

4. The method of claim 3 wherein the feature extraction module includes at least one of the techniques comprising key statistics, principal component analysis, and partial least squares.

5. The method of claim 3 wherein the classifier module includes at least one of the techniques comprising support vector machine, probabilistic neural networks, k-nearest neighbor, decision trees, linear discriminant analysis, and quadratic discriminant analysis.

6. The method of claim 3 wherein the testing phase includes:

providing a testing data matrix which includes diagnostic trouble code and parameter identification data for fault events from a plurality of vehicles, where each fault event has the actual failure mode attributed to it;

using the pre-processing module, the feature extraction module, the classifier module, and the stored techniques and settings from the training phase to produce a diagnosed failure mode for each fault event; and

verifying that the stored techniques and settings from the training phase are effective with the testing data matrix by comparing the diagnosed failure mode with the actual failure mode for each fault event.

7. The method of claim 6 further comprising downloading the mathematical models after the training and testing phases to a computer at a vehicle repair service center, and using the mathematical models to diagnose a failure mode for fault events on vehicles which are brought to the vehicle repair service center for diagnosis.

8. The method of claim 6 further comprising downloading the mathematical models after the training and testing phases to a computer onboard a vehicle, and using the mathematical models to diagnose a failure mode for fault events on said vehicle.

9. The method of claim 1 wherein the diagnostic trouble code and parameter identification data also includes a time stamp for each fault event.

10. The method of claim 9 further comprising using the diagnostic trouble code and parameter identification data and the time stamp for each fault event to make a prognosis of system health and remaining useful life for any system involved in the fault event.

11. The method of claim 1 wherein the actual failure mode for each fault event is provided by a vehicle repair service center which has resolved the fault condition.

12. A method for diagnosing causes of faults and making a prognosis of system health in vehicle systems, said method comprising:

defining a plurality of diagnostic trouble codes which are applicable to various fault conditions in a vehicle'"'"'s systems, criteria for triggering each diagnostic trouble code, and a list of parameter identification data which should be captured for each diagnostic trouble code;

providing a controller onboard a vehicle which is programmed with the diagnostic trouble codes and the criteria for triggering each diagnostic trouble code, and which is also programmed with the list of parameter identification data which should be captured for each diagnostic trouble code;

providing sensors onboard the vehicle as needed to capture the parameter identification data;

capturing the diagnostic trouble code and parameter identification data by the controller when a fault condition occurs in a vehicle system, where each fault condition occurrence is known as a fault event, and data captured for each fault event also includes a time stamp;

downloading the diagnostic trouble code and parameter identification data and time stamps from the controller on the vehicle to a central computer;

storing diagnostic trouble code and parameter identification data and time stamps for fault events from multiple vehicles on the central computer, where each fault event has an actual failure mode attributed to it;

analyzing the diagnostic trouble code and parameter identification data from the fault events stored on the central computer using mathematical models which can learn to correlate features in the data with the actual failure mode for each fault event;

using the mathematical models to diagnose a failure mode for additional fault events; and

using the diagnostic trouble code and parameter identification data and the time stamp for each fault event to make a prognosis of system health and remaining useful life for any system involved in the fault event.

13. The method of claim 12 wherein analyzing the diagnostic trouble code and parameter identification data includes a training phase, where the training phase includes:

providing input data comprising a matrix of data for each of a plurality of failure modes, where each matrix of data identifies the failure mode to which it pertains, and includes diagnostic trouble code and parameter identification data for fault events from a plurality of vehicles;

using a pre-processing module to normalize and reduce variation in the input data;

using a feature extraction module to extract characteristic features of the input data, where the feature extraction module includes at least one of the techniques comprising key statistics, principal component analysis, and partial least squares;

using a classifier module to identify patterns in the features of the input data and associate the patterns with the actual failure mode, where the classifier module includes at least one of the techniques comprising support vector machine, probabilistic neural networks, k-nearest neighbor, decision trees, linear discriminant analysis, and quadratic discriminant analysis; and

storing techniques and settings used in the feature extraction module and the classifier module which most effectively correlate the input data with the actual failure modes.

14. The method of claim 13 wherein analyzing the diagnostic trouble code and parameter identification data includes a testing phase, where the testing phase includes:

providing a testing data matrix which includes diagnostic trouble code and parameter identification data for fault events from a plurality of vehicles, where each fault event has the actual failure mode attributed to it;

using the pre-processing module, the feature extraction module, the classifier module, and the stored techniques and settings from the training phase to produce a diagnosed failure mode for each fault event; and

verifying that the stored techniques and settings from the training phase are effective with the testing data matrix by comparing the diagnosed failure mode with the actual failure mode for each fault event.

15. A system for diagnosing causes of faults in vehicle systems, said system comprising:

a vehicle with a plurality of systems to be monitored;

a plurality of sensors onboard the vehicle for measuring various conditions in the systems;

a controller onboard the vehicle, where the controller is configured to receive parameter data from the sensors, monitor the parameter data to determine if any fault conditions exist, and capture and store a plurality of parameter data when a fault event occurs;

a central computer for collecting parameter data for fault events from a plurality of vehicles, where the computer uses mathematical models to analyze the parameter data to diagnose failure modes for groups of fault events and for individual fault events; and

a communications system onboard the vehicle for transmitting the parameter data for fault events to the central computer.

16. The system of claim 15 wherein the controller includes a pre-defined list of diagnostic trouble codes, criteria for triggering each diagnostic trouble code, and parameter data to capture and store for each diagnostic trouble code.

17. The system of claim 15 wherein the communications system is a wireless system which can transmit the parameter data to the central computer without the vehicle having to be taken to a vehicle repair service center.

18. The system of claim 15 wherein the mathematical models used by the central computer include data pre-processing, feature extraction, and trained classifier modules.

19. The system of claim 18 wherein:

the data pre-processing module includes normalization and variance reduction techniques;

the feature extraction module includes key statistics, principal component analysis, and partial least squares techniques; and

the trained classifier module includes support vector machine, probabilistic neural networks, k-nearest neighbor, decision trees, linear discriminant analysis, and quadratic discriminant analysis techniques.

20. The system of claim 15 further comprising a computer onboard the vehicle which is configured to use the mathematical models from the central computer to diagnose a failure mode for any new fault event which occurs on the vehicle'"'"'s systems.