Monitoring and diagnosing vehicle system problems using machine learning classifiers

Patent metadata

Field	Value
Publication number	US-10922906-B2
Application number	US-201916367827-A
Country	US
Kind code	B2
Filing date	Mar 28, 2019
Priority date	Mar 28, 2019
Publication date	Feb 16, 2021
Grant date	Feb 16, 2021

Abstract

A system for monitoring operation of a vehicle includes a processing device including an interface configured to receive measurement data from sensing devices configured to measure parameters of a vehicle system. The processing device is configured to receive measurement data from each of the plurality of sensing devices, and in response to detection of a malfunction in the vehicle, input at least a subset of the measurement data to a machine learning classifier associated with a vehicle subsystem, the classifier configured to define a class associated with normal operation of the vehicle subsystem. The processing device is also configured to determine whether the subset of the measurement data belongs to the class, and based on at least a selected amount of the subset of the measurement data being outside of the class, output a fault indication, the fault indication identifying the vehicle subsystem as having a contribution to the malfunction.

First claim

What is claimed is:

1. A system for monitoring operation of a vehicle, comprising:
a processing device including an interface configured to receive measurement data from a plurality of sensing devices, each sensing device of the plurality of sensing devices configured to measure a parameter of a vehicle system, the processing device including a plurality of machine learning classifiers, each classifier of the plurality of machine learning classifiers associated with a different vehicle subsystem of a plurality of vehicle subsystems, the processing device configured to perform:
receiving measurement data from each of the plurality of sensing devices, the measurement data having a plurality of subsets;
in response to detection of a malfunction in the vehicle, inputting a respective subset of the plurality of subsets to each classifier, wherein each classifier is configured to define a class associated with normal operation of a respective vehicle subsystem;
determining by each classifier whether the respective subset of the measurement data belongs to the class; and
based on one or more classifiers determining that at least a selected amount of the respective subset of the measurement data is outside of the class, outputting a fault indication, the fault indication identifying which of the plurality of vehicle subsystems has a contribution to the malfunction.

2. The system of claim 1 , wherein each classifier is configured to plot the respective subset of the measurement data as position vectors in a feature space having axes representing parameters related to operation of the vehicle subsystem, the feature space including a boundary that defines a region of the feature space associated with a healthy condition of the vehicle subsystem.

3. The system of claim 2 , wherein each classifier is configured to output the fault indication based on a selected proportion of the position vectors being outside of the region.

4. The system of claim 1 , wherein each classifier is a one-class classifier, and the class is a healthy class representing normal operation of the respective vehicle subsystem.

5. The system of claim 4 , wherein the processing device is configured to:
determine, for each one-class classifier, whether the respective subset belongs to the healthy class; and
based on at least one classifier determining that the respective subset does not belong to the healthy class, identify which of the plurality of vehicle subsystems has a contribution to the malfunction.

6. The system of claim 5 , wherein the plurality of one-class classifiers includes an air subsystem classifier, a fuel subsystem classifier and an ignition subsystem classifier.

7. The system of claim 1 , wherein the plurality of machine learning classifiers includes a plurality of two-dimensional one-class classifiers, each one-class classifier configured to receive a subset of the measurement data for a pair of parameters, and the processing device is configured to identify one or more individual components or vehicle subsystem as having a contribution to the malfunction by comparing outputs from the plurality of one-class classifiers to a knowledge base.

8. The system of claim 1 , wherein the plurality of machine learning classifiers includes a two-class classifier having a healthy class associated with normal operation of the vehicle subsystem, and a faulty class.

9. The system of claim 1 , wherein the plurality of machine learning classifiers includes a multi-class classifier having at least three classes associated with a plurality of vehicle subsystems and/or components.

10. The system of claim 9 , wherein the at least three classes include a healthy class associated with normal operation of all of the plurality of vehicle subsystems or components, and a faulty class for each vehicle subsystem and/or component.

11. A method of monitoring operation of a vehicle, comprising:
receiving measurement data from a plurality of sensing devices by a processing device, each sensing device of the plurality of sensing devices configured to measure a parameter of a vehicle system, the measurement data having a plurality of subsets, the processing device including a plurality of machine learning classifiers, each classifier of the plurality of machine learning classifiers associated with a different vehicle subsystem of a plurality of vehicle subsystems;
in response to detection of a malfunction in the vehicle, inputting at least a respective subset of the plurality of subsets to each classifier, wherein each classifier is configured to define a class associated with normal operation of the a respective vehicle subsystem;
determining by each classifier whether the respective subset of the measurement data belongs to the class; and
based on one or more classifiers determining that at least a selected amount of the subset of the measurement data is outside of the class, outputting a fault indication, the fault indication identifying which of the plurality of vehicle subsystems has a contribution to the malfunction.

12. The method of claim 11 , wherein each classifier is configured to plot the respective subset of the measurement data as position vectors in a feature space having axes representing parameters related to operation of the vehicle subsystem, the feature space including a boundary that defines a region of the feature space associated with a healthy condition of the vehicle subsystem.

13. The method of claim 12 , further comprising outputting the fault indication from the classifier based on a selected proportion of the position vectors being outside of the region.

14. The method of claim 11 , wherein each classifier is a one-class classifier, and the class is a healthy class representing normal operation of the respective vehicle subsystem.

15. The method of claim 14 , further comprising:
determining, for each one-class classifier, whether the respective subset belongs to the healthy class; and
based on at least one classifier determining that the respective subset does not belong to the healthy class, identifying which of the plurality of vehicle subsystems has a contribution to the malfunction.

16. The method of claim 15 , wherein the plurality of one-class classifiers includes an air subsystem classifier, a fuel subsystem classifier and an ignition subsystem classifier.

17. The method of claim 11 , wherein the plurality of machine learning classifiers includes a plurality of two-dimensional one-class classifiers, each one-class classifier configured to receive a subset of the measurement data for a pair of parameters, the method further comprising identifying one or more individual components or vehicle subsystem as having a contribution to the malfunction by comparing outputs from the plurality of one-class classifiers to a knowledge base.

18. The method of claim 11 , wherein the plurality of machine learning classifiers includes a two-class classifier having a healthy class associated with normal operation of the vehicle subsystem, and a faulty class.

19. The method of claim 11 , wherein the plurality of machine learning classifiers includes a multi-class classifier having at least three classes associated with a plurality of vehicle subsystems and/or components.

20. The method of claim 19 , wherein the at least three classes include a healthy class associated with normal operation of all of the plurality of vehicle subsystems or components, and a faulty class for each vehicle subsystem and/or component.

Assignees

• Gm Global Tech Operations Llc

Inventors

• Wang Yue-Yun
• Haskara Ibrahim
• Sun David
• Zou Yusheng
• Duan Shiming
• Kao Chi-Kuan
• Lu Xiangxing

Classifications

• G06F18/2431

  Multiple classes · CPC title

• G06F18/2411 Primary

  based on the proximity to a decision surface, e.g. support vector machines · CPC title

• G06N20/10

  using kernel methods, e.g. support vector machines [SVM] · CPC title

• G06N5/02

  Knowledge representation; Symbolic representation · CPC title

• G06N3/02

  Neural networks · CPC title