Generating Estimates of Failure Risk for a Vehicular Component in Situations of High-Dimensional and Low Sample Size Data

• US 20160093118A1
• Filed: 09/26/2014
• Published: 03/31/2016
• Est. Priority Date: 09/26/2014
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Ahn, Hyung-il, Denesuk, Matthew, Hochstein, Axel, Leung, Ying Tat
• Granted Patent: US 10,540,828 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: B60K 35/00 Instruments specially adapted for vehicles; Arrangement of instruments in or on vehicles
• CPC Class Codes: G07C 5/006 Indicating maintenance
• CPC Class Codes: G07C 5/0808 Diagnosing performance data testing of vehicles G01M17/00; testing of electrical installation on vehicles G01R31/005

Abstract

Methods, systems, and computer program products for generating estimates of failure risk for a vehicular component in situations of high-dimensional and low sample size data are provided herein. A method includes splitting a first input time series comprising multiple data points derived from a vehicular component across a fleet of multiple vehicles into multiple sub-time series; generating a first failure status predicting function of a first selected sub-time series; deleting, from the first input time series, the portion of the data points that corresponds to the first selected sub-time series; repeating the preceding two steps for a second selected sub-time series; generating a second failure status predicting function of each selected sub-time series; applying each second failure status predicting function to a second input time series to calculate prediction of failure values; and identifying the largest prediction of failure value as an estimate of failure risk for the vehicular component.

Claims

1. A method comprising the following steps:

(a) splitting a first input time series comprising multiple data points derived from a vehicular component across a fleet of multiple vehicles into multiple sub-time series, wherein each of the multiple sub-time series comprises a portion of the multiple data points in the first input time series;

(b) generating, based on a full likelihood model fitting across the multiple data points in the first input time series, a first failure status predicting function of a first selected sub-time series from the multiple sub-time series that has the best fit to the multiple data points;

(c) deleting, from the first input time series, the portion of the multiple data points that corresponds to the first selected sub-time series, thereby generating a modified first input time series;

(d) generating, based on a full likelihood model fitting across the multiple data points in the modified first input time series, a first failure status predicting function of a second selected sub-time series from the multiple sub-time series that has the best fit to the multiple data points excluding the deleted portion;

(e) deleting, from the modified first input time series, the portion of the multiple data points that corresponds to the second selected sub-time series, thereby generating a further modified first input time series;

(f) generating, based on a partial likelihood model fitting across a given sub-set of the multiple data points in the first input time series, a second failure status predicting function of each selected sub-time series that has the best fit to the given sub-set of the multiple data points;

(g) applying the second failure status predicting function of each selected sub-time series to a second input time series derived from the vehicular component to calculate multiple prediction of failure values for the second input time series; and

(h) identifying the largest of the multiple prediction of failure values as an estimate of failure risk for the vehicular component;

wherein at least one of the steps is carried out by a computing device.

2. The method of claim 1, wherein the multiple data points correspond to multiple measurements derived from the vehicular component.

3. The method of claim 2, wherein the portion of the multiple data points associated with each sub-time series corresponds to one of the multiple measurements derived from the vehicular component.

4. The method of claim 1, comprising:

storing each selected sub-time series in a database.

5. The method of claim 1, wherein each of said first input time series and said second input time series is derived from a set of sensor readings corresponding to the vehicular component.

6. The method of claim 1, comprising generating the first failure status predicting function via logistic regression.

7. The method of claim 1, comprising generating the second failure status predicting function via logistic regression.

8. The method of claim 1, comprising:

repeating, for a given number of additional selected sub-time series from the multiple sub-time series;

generating, based on a full likelihood model fitting across the multiple data points in the further modified first input time series, a first failure status predicting function of an additional selected sub-time series from the multiple sub-time series that has the best fit to the multiple data points excluding the deleted portions; and

deleting, from the further modified first input time series, the portion of the multiple data points that corresponds to the additional selected sub-time series.

9. The method of claim 8, wherein said given number of additional selected sub-time series comprises all of the multiple sub-time series.

10. The method of claim 8, wherein said given number of additional selected sub-time series comprises a number of additional selected sub-time series prior to generating a failure status prediction function that indicates no improvement compared to an empty set.

11. The method of claim 1, wherein the given sub-set of the multiple data points in the first input time series comprises measurements from each time point corresponding to a failure of the vehicular component.

12. A computer program product, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computing device to cause the computing device to:

(a) split a first input time series comprising multiple data points derived from a vehicular component across a fleet of multiple vehicles into multiple sub-time series, wherein each of the multiple sub-time series comprises a portion of the multiple data points in the first input time series;

(b) generate, based on a full likelihood model fitting across the multiple data points in the first input time series, a first failure status predicting function of a first selected sub-time series from the multiple sub-time series that has the best fit to the multiple data points;

(c) delete, from the first input time series, the portion of the multiple data points that corresponds to the first selected sub-time series, thereby generating a modified first input time series;

(d) generate, based on a full likelihood model fitting across the multiple data points in the modified first input time series, a first failure status predicting function of a second selected sub-time series from the multiple sub-time series that has the best fit to the multiple data points excluding the deleted portion;

(e) delete, from the modified first input time series, the portion of the multiple data points that corresponds to the second selected sub-time series, thereby generating a further modified first input time series;

(f) generate, based on a partial likelihood model fitting across a given sub-set of the multiple data points in the first input time series, a second failure status predicting function of each selected sub-time series that has the best fit to the given sub-set of the multiple data points;

(g) apply the second failure status predicting function of each selected sub-time series to a second input time series derived from the vehicular component to calculate multiple prediction of failure values for the second input time series; and

(h) identify the largest of the multiple prediction of failure values as an estimate of failure risk for the vehicular component.

13. A system comprising:

a memory; and

at least one processor coupled to the memory and configured for;

(a) splitting a first input time series comprising multiple data points derived from a vehicular component across a fleet of multiple vehicles into multiple sub-time series, wherein each of the multiple sub-time series comprises a portion of the multiple data points in the first input time series;

(b) generating, based on a full likelihood model fitting across the multiple data points in the first input time series, a first failure status predicting function of a first selected sub-time series from the multiple sub-time series that has the best fit to the multiple data points;

(c) deleting, from the first input time series, the portion of the multiple data points that corresponds to the first selected sub-time series, thereby generating a modified first input time series;

(d) generating, based on a full likelihood model fitting across the multiple data points in the modified first input time series, a first failure status predicting function of a second selected sub-time series from the multiple sub-time series that has the best fit to the multiple data points excluding the deleted portion;

(e) deleting, from the modified first input time series, the portion of the multiple data points that corresponds to the second selected sub-time series, thereby generating a further modified first input time series;

(f) generating, based on a partial likelihood model fitting across a given sub-set of the multiple data points in the first input time series, a second failure status predicting function of each selected sub-time series that has the best fit to the given sub-set of the multiple data points;

(g) applying the second failure status predicting function of each selected sub-time series to a second input time series derived from the vehicular component to calculate multiple prediction of failure values for the second input time series; and

(h) identifying the largest of the multiple prediction of failure values as an estimate of failure risk for the vehicular component.

14. A method comprising the following steps:

(a) splitting a first input time series comprising multiple data points derived from a vehicular component across a fleet of multiple vehicles into multiple sub-time series, wherein each of the multiple sub-time series comprises a portion of the multiple data points in the first input time series;

(b) generating, based on a full likelihood model fitting across the multiple data points in the first input time series, a first failure status predicting function of a first selected sub-time series from the multiple sub-time series that has the best fit to the multiple data points;

(c) deleting, from the first input time series, the portion of the multiple data points that corresponds to the first selected sub-time series, thereby generating a modified first input time series;

(d) generating, based on a full likelihood model fitting across the multiple data points in the modified first input time series, a first failure status predicting function of a second selected sub-time series from the multiple sub-time series that has the best fit to the multiple data points excluding the deleted portion;

(e) deleting, from the modified first input time series, the portion of the multiple data points that corresponds to the second selected sub-time series, thereby generating a further modified first input time series;

(f) repeating, for a given number of additional selected sub-time series from the multiple sub-time series;

generating, based on a full likelihood model fitting across the multiple data points in the further modified first input time series, a first failure status predicting function of an additional selected sub-time series from the multiple sub-time series that has the best fit to the multiple data points excluding the deleted portions; and

deleting, from the further modified first input time series, the portion of the multiple data points that corresponds to the additional selected sub-time series;

(g) generating, based on a partial likelihood model fitting across a given sub-set of the multiple data points in the first input time series, a second failure status predicting function of multiple combinations of two or more of the selected sub-time series that has the best fit to the given sub-set of the multiple data points;

(h) applying the second failure status predicting function of each of the multiple combinations of two or more selected sub-time series to a second input time series derived from the vehicular component to calculate multiple prediction of failure values for the second input time series; and

(i) identifying the largest of the multiple prediction of failure values as an estimate of failure risk for the vehicular component;

wherein at least one of the steps is carried out by a computing device.

15. The method of claim 14, wherein the multiple data points correspond to multiple measurements derived from the vehicular component.

16. The method of claim 14, comprising:

storing each selected sub-time series in a database.

17. The method of claim 14, comprising generating the first failure status predicting function via logistic regression.

18. The method of claim 14, comprising generating the second failure status predicting function via logistic regression.

19. The method of claim 14, wherein said given number of additional selected sub-time series comprises all of the multiple sub-time series.

20. The method of claim 14, wherein said given number of additional selected sub-time series comprises a number of additional selected sub-time series prior to generating a failure status prediction function that indicates no improvement compared to an empty set.