Fault diagnosis in power electronics using adaptive PCA

• US 10,354,462 B1
• Filed: 04/06/2018
• Issued: 07/16/2019
• Est. Priority Date: 04/06/2018
• Status: Active Grant

• Current Assignee: University of Connecticut (State of Connecticut), Toyota Jidosha Kabushiki Kaisha (Toyota Motor Corporation)
• Original Assignee: Toyota Motor Engineering & Manufacturing North America Incorporated (Toyota Motor Corporation)
• Inventors: Zhang, Lingyi, Chen, Weiqiang, Pattipatti, Krishna, Bazzi, Ali M., Joshi, Shailesh N., Dede, Ercan M.
• Primary Examiner(s): Soofi, Yazan A
• Application Number:
  US15/947,702
• Time in Patent Office: 466 Days
• Field of Search: 701 329
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G07C 5/008 communicating information to a remotely located station transmission systems for measured values G08C
• CPC Class Codes: G07C 5/0808 Diagnosing performance data testing of vehicles G01M17/00; testing of electrical installation on vehicles G01R31/005
• CPC Class Codes: G07C 5/0816 Indicating performance data, e.g. occurrence of a malfunction
• CPC Class Codes: G07C 5/085 using electronic data carriers

Abstract

A system includes an electronic device and a sensor to detect sensor data corresponding to the electronic device. The system also includes a machine learning processor that receives the sensor data and generates a model of the electronic device to determine a T squared threshold and a Q threshold using a machine learning algorithm. The machine learning processor also performs a T squared analysis of the electronic device by comparing a T squared value to the T squared threshold, and a Q analysis of the electronic device by comparing a Q value to the Q threshold. The machine learning processor also determines that the model is faulty when the T squared value is less than the T squared threshold and the Q value is greater than or equal to the Q threshold, and generates a new model or updates the model when the model is determined to be faulty.

Claims

1. A system for analyzing a health status of a component of vehicles, comprising:

an electronic device;

a sensor configured to detect sensor data corresponding to at least one performance characteristic of the electronic device;

a memory configured to store a machine learning algorithm; and

a machine learning processor coupled to the sensor and configured to;

receive the sensor data;

generate, using the machine learning algorithm, a model of the electronic device to determine a T squared threshold and a Q threshold;

perform a T squared analysis of the electronic device by comparing a T squared value to the T squared threshold;

perform a Q analysis of the electronic device by comparing a Q value to the Q threshold;

determine that the model of the electronic device is faulty when the T squared value is less than the T squared threshold and the Q value is greater than or equal to the Q threshold; and

generate a new model of the electronic device or update the model of the electronic device when the model of the electronic device is determined to be faulty.

2. The system of claim 1 wherein the machine learning processor is further configured to perform an error detection algorithm that is different than the model of the electronic device to determine whether the electronic device has failed or is likely to fail when the model of the electronic device is faulty.

3. The system of claim 2 wherein the machine learning processor is further configured to predict a residual life of the electronic device based on at least one of the error detection algorithm or the T squared analysis when the electronic device is determined to be likely to fail.

4. The system of claim 2 wherein the machine learning processor is further configured to generate the new model of the electronic device or update the model of the electronic device when the error detection algorithm indicates that the electronic device has failed or is likely to fail.

5. The system of claim 1 wherein the machine learning processor is further configured to identify that the electronic device has failed or is likely to fail when the T squared value is equal to or greater than the T squared threshold.

6. The system of claim 1 wherein the machine learning processor is further configured to determine that a failure or likelihood of failure is less severe when the T squared value reaches or exceeds the T squared threshold before the Q value reaches or exceeds the Q threshold relative to when the T squared value reaches or exceeds the T squared threshold at the same time that the Q value reaches or exceeds the Q threshold.

7. The system of claim 1 wherein the model of the electronic device is a principal component analysis (PCA) model of the electronic device.

8. The system of claim 7 wherein the machine learning processor is further configured to generate eigenvalues and corresponding eigenvectors, and to determine the T squared threshold and the Q threshold based on the eigenvalues and the corresponding eigenvectors.

9. A system for analyzing a health status of a component of vehicles, comprising:

an electronic device;

a sensor configured to detect sensor data corresponding to at least one performance characteristic of the electronic device;

a memory configured to store a machine learning algorithm; and

a machine learning processor coupled to the sensor and configured to;

receive the sensor data;

generate, using the machine learning algorithm, a principal component analysis (PCA) model of the electronic device to determine a reduced data matrix;

determine a T squared threshold and a Q threshold based on the reduced data matrix;

perform a T squared analysis of the electronic device by comparing a T squared value to the T squared threshold;

perform a Q analysis of the electronic device by comparing a Q value to the Q threshold;

determine that the PCA model of the electronic device is faulty when the T squared value is less than the T squared threshold and the Q value is greater than or equal to the Q threshold; and

generate a new PCA model of the electronic device or update the PCA model of the electronic device when the PCA model of the electronic device is determined to be faulty.

10. The system of claim 9 wherein the machine learning processor is further configured to perform an error detection algorithm that is different than the PCA model of the electronic device to determine whether the electronic device has failed or is likely to fail when the PCA model of the electronic device is faulty.

11. The system of claim 10 wherein the machine learning processor is further configured to predict a residual life of the electronic device based on at least one of the error detection algorithm or the T squared analysis when the electronic device is determined to be likely to fail.

12. The system of claim 10 wherein the machine learning processor is further configured to generate the new PCA model of the electronic device or update the PCA model of the electronic device when the error detection algorithm indicates that the electronic device has failed or is likely to fail.

13. The system of claim 9 wherein the machine learning processor is further configured to identify that the electronic device has failed or is likely to fail when the T squared value is equal to or greater than the T squared threshold.

14. The system of claim 9 wherein the machine learning processor is further configured to determine that a failure or likelihood of failure is less severe when the T squared value reaches or exceeds the T squared threshold before the Q value reaches or exceeds the Q threshold relative to when the T squared value reaches or exceeds the T squared threshold at the same time that the Q value reaches or exceeds the Q threshold.

15. The system of claim 9 wherein the machine learning processor is further configured to generate the reduced data matrix by generating eigenvalues and corresponding eigenvectors based on the PCA model of the electronic device and the sensor data.

16. A method for analyzing a health status of a component of vehicles, comprising:

detecting, by a sensor, sensor data corresponding to at least one performance characteristic of an electronic device of a vehicle;

storing, in a memory, a machine learning algorithm;

receiving, by a machine learning processor, the sensor data;

generating, by the machine learning processor using the machine learning algorithm, a model of the electronic device to determine a T squared threshold and a Q threshold;

performing, by the machine learning processor, a T squared analysis of the electronic device by comparing a T squared value to the T squared threshold;

performing, by the machine learning processor, a Q analysis of the electronic device by comparing a Q value to the Q threshold;

determining, by the machine learning processor, that the model of the electronic device is faulty when the T squared value is less than the T squared threshold and the Q value is greater than or equal to the Q threshold; and

at least one of generating, by the machine learning processor, a new model of the electronic device, or updating, by the machine learning processor, the model of the electronic device when the model of the electronic device is determined to be faulty.

17. The method of claim 16 further comprising determining, by the machine learning processor, whether the electronic device has failed or is likely to fail using an error detection algorithm that is different than the model of the electronic device when the model of the electronic device is faulty.

18. The method of claim 17 further comprising predicting, by the machine learning algorithm, a residual life of the electronic device based on at least one of the error detection algorithm or the T squared analysis when the electronic device determined to be likely to fail.

19. The method of claim 17 further comprising the at least one of generating the new model of the electronic device or updating the model of the electronic device when the error detection algorithm indicates that the electronic device has failed or is likely to fail.

20. The method of claim 16 further comprising determining, by the machine learning processor, that a failure or likelihood of failure is less severe when the T squared value reaches or exceeds the T squared threshold before the Q value reaches or exceeds the Q threshold relative to when the T squared value reaches or exceeds the T squared threshold at the same time that the Q value reaches or exceeds the Q threshold.