Incremental learning of nonlinear regression networks for machine condition monitoring

• US 7,844,558 B2
• Filed: 10/03/2007
• Issued: 11/30/2010
• Est. Priority Date: 10/05/2006
• Status: Active Grant

• Current Assignee: Siemens Corp. (Siemens AG)
• Original Assignee: Siemens Corp. (Siemens AG)
• Inventors: Neubauer, Claus, Yuan, Chao
• Primary Examiner(s): Holmes; Michael B.
• Assistant Examiner(s): GONZALES, VINCENT
• Application Number:
  US11/866,554
• Publication Number:
  US 20080086437A1
• Time in Patent Office: 1,154 Days
• Field of Search: 706/45, 706/62
• US Class Current: 706/45
• 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: G06F 11/0706 the processing taking place on a specific hardware platform or in a specific software environment
• CPC Class Codes: G06F 11/0751 Error or fault detection not based on redundancy power supply failures G06F1/30; network fault management H04L41/06

Abstract

A method for identifying a potential fault in a system includes obtaining a set of training data. A first kernel is selected from a library of two or more kernels and the first kernel is added to a regression network. A next kernel is selected from the library of two or more kernels and the next kernel is added to the regression network. The regression network is refined. A potential fault is identified in the system using the refined regression network.

Claims

1. A method for identifying a potential fault in a system, comprising:

obtaining a set of training data;

selecting a first kernel from a library of two or more kernels and adding the first kernel to a regression network;

selecting a next kernel from the library of two or more kernels and adding the next kernel to the regression network;

refining the regression network using a leave-one-out method in which the regression network is iteratively improved by removing a single kernel from the regression network and replacing the removed kernel with a replacement kernel from the library of kernels and then repeating the removing and replacing steps for the kernels of the regression network until a desired level of convergence between the training data and the regression network is achieved; and

identifying a potential fault in the system using the refined regression network.

2. The method of claim 1, wherein the step of selecting a next kernel from the library of kernels and adding the next kernel to the regression network is repeated until a regression error has been reduced to a predetermined level.

3. The method of claim 1, wherein the step of selecting a next kernel from the library of kernels and adding the next kernel to the regression network is repeated until a regression error has been minimized.

4. The method of claim 1, wherein the first kernel is selected from among the library of kernels by calculating a cost function for each of the kernels of the library and selecting the kernel that results in a lowest cost function.

5. The method of claim 1, wherein the next kernel is selected from among the library of kernels by calculating a cost function for each of the kernels of the library and selecting the kernel that results in a lowest cost function.

6. The method of claim 1, wherein refining the regression network additionally comprising repeating the steps of removing and replacing a kernel from the regression network until a cost function is minimized.

7. The method of claim 1, wherein the library of kernels includes a linear kernel, a Gaussian kernel, a polynomial kernel, quadratic kernel, or a sigmoid kernel.

8. The method of claim 1, wherein the step of identifying a potential fault in the system using the refined regression network comprises:

obtaining operational sensor data;

estimating an expected monitored sensor value based on the operational sensor data and the refined regression network;

obtaining an actual value from the monitored sensor; and

identifying a potential fault when the actual value of the monitored sensor deviates from the expected monitored sensor value by more than a predetermined tolerance level.

9. A system for identifying potential faults in a machine, comprising:

a training data database including set of training data;

a kernel database including two or more kernels;

a selection unit for selecting a kernel from a library of two or more kernels and adding the kernel to a regression network;

a refining unit for refining the regression network using a leave-one-out method in which the regression network is iteratively improved by removing a single kernel from the regression network and replacing the removed kernel with a replacement kernel from the library of kernels and then repeating the removing and replacing steps for the kernels of the regression network until a desired level of convergence between the training data and the regression network is achieved;

a set of sensors for monitoring attributes of the machine; and

an identification unit for identifying a potential fault in the machine using the refined regression network and data obtained from the set of sensors.

10. The system of claim 9, wherein the selection unit repeats the selecting of a kernel and adding the selected kernel to the regression network until a regression error has been minimized or reduced to a predetermined level.

11. The system of claim 9, wherein the selection unit selects a kernel from among the library of kernels by calculating a cost function for each of the kernels of the library and selecting the kernel that results in a lowest cost function.

12. The system of claim 9, wherein the refining unit repeats the refining process until a regression error has been reduced to a predetermined level or a cost function is minimized.

13. A computer system comprising:

a processor; and

a program storage device readable by the computer system, embodying a program of instructions executable by the processor to perform method steps for identifying a potential fault in a system, the method comprising;

obtaining a set of training data;

selecting a first kernel from a library of two or more kernels and adding the first kernel to a regression network;

selecting a next kernel from the library of two or more kernels and adding the next kernel to the regression network;

refining the regression network using a leave-one-out method in which the regression network is iteratively improved by removing a single kernel from the regression network and replacing the removed kernel with a replacement kernel from the library of kernels and then repeating the removing and replacing steps for the kernels of the regression network until a desired level of convergence between the training data and the regression network is achieved; and

identifying a potential fault in the system using the refined regression network.

14. The computer system of claim 13, wherein the step of selecting a next kernel from the library of kernels and adding the next kernel to the regression network is repeated until a regression error has been minimized or reduced to a predetermined level.

15. The computer system of claim 13, wherein the first kernel and the next kernel are each selected from among the library of kernels by calculating a cost function for each of the kernels of the library and selecting the kernel that results in a lowest cost function.