System and method for estimating a measure of confidence in a match generated from a case-based reasoning system

• US 5,799,148 A
• Filed: 12/23/1996
• Issued: 08/25/1998
• Est. Priority Date: 12/23/1996
• Status: Expired due to Fees

• Current Assignee: General Electric Company
• Original Assignee: General Electric Company
• Inventors: Cuddihy, Paul Edward, Cheetham, William Estel
• Primary Examiner(s): PALYS, JOSEPH E
• Application Number:
  US08/772,451
• Time in Patent Office: 610 Days
• Field of Search: 395/183.02, 395/183.01, 395/183.13, 395/183.22, 395/68, 395/77, 395/50, 395/185.01, 395/185.1, 395/183.16
• US Class Current: 714/26
• CPC Class Codes: G06F 11/2257 using expert systems
• CPC Class Codes: G06N 5/04 Inference or reasoning models

Abstract

The present invention discloses a system and a method for estimating a measure of confidence in a match generated from a case-based reasoning system. The measure of confidence is determined by using a confidence function that is derived for all existing cases in the case base. The confidence function is then used to estimate the confidence of any matches between a new case and an existing case. In the present invention, a retrieval unit uses a similarity index and retrieves any existing cases from the case base that has a measurement of similarity to the new case that is within a predetermined similarity threshold. The confidence function then maps the similarity measurement for each retrieved case to a corresponding measure of how many different outcomes are likely given the level of similarity. A report of the existing cases that have the best measures of confidence are then provided in a list. In another embodiment, the confidence estimation is used to analyze an error log generated from a malfunctioning machine.

Claims

1. A system for estimating a measure of confidence in a match generated from a case-based reasoning system, comprising:

a case base storing a plurality of existing cases, each of the plurality of existing cases having similarity information on the similarity to other existing cases in the case base;

means for retrieving existing cases from the case base that are similar to a new case, the retrieving means retrieving existing cases from the case base according to a predetermined similarity threshold, each retrieved existing case having a measure of similarity to the new case;

means for mapping the similarity measurement in each of the retrieved existing cases to a measure of confidence on how closely the existing case matches the new case, the mapping means comprising means for generating a confidence function for each of the retrieved cases, wherein the confidence function is formed by plotting confidence function points determined from solutions of the similar existing cases, the confidence function providing a relationship between confidence and similarity with other existing cases; and

means for providing a list of the retrieved existing cases having the best measures of confidence.

2. The system according to claim 1, wherein the confidence function points are connected to form the confidence function.

3. The system according to claim 2, wherein the similarity measurement for each of the retrieved existing cases are mapped onto its confidence function to generate the measure of confidence for the case.

4. A method for estimating a measure of confidence in a match generated from a case-based reasoning system, the method comprising the steps of:

storing a plurality of existing cases, each of the plurality of existing cases having similarity information on the similarity to other existing cases in the case base;

retrieving existing cases from the case base that are similar to a new case, the retrieved existing cases being retrieved according to a predetermined similarity threshold, each retrieved existing case having a measure of similarity to the new case;

mapping the similarity measurement in each of the retrieved existing cases to a measure of confidence on how closely the existing case matches the new case, the mapping comprising generating a confidence function for each of the retrieved cases, wherein the confidence function is formed by plotting confidence function points determined from solutions of the similar cases, the confidence function providing a relationship between confidence and similarity with other existing cases; and

providing a list of the existing cases having the best measures of confidence.

5. The method according to claim 4, wherein the confidence function points are connected to form the confidence function.

6. The method according to claim 5, wherein the similarity measurement for each of the retrieved existing cases are mapped onto its confidence function to generate the measure of confidence for the case.

7. A system for analyzing an error log generated from a malfunctioning machine, comprising:

a case base storing a plurality of historical cases generated from a plurality of malfunctioning machines, each of the plurality of historical cases containing data representative of events occurring at the plurality of malfunctioning machines during operation and solutions for correcting the malfunctioning machines in accordance with the events, and each of the plurality of historical cases containing information on the similarity to other historical cases in the case base;

means for receiving a new error log from a malfunctioning machine, the new error log containing data representative of events occurring at the malfunctioning machine during operation;

means for retrieving historical cases from the case base having similar solutions to the new error log, the retrieving means retrieving historical cases from the case base according to a predetermined similarity threshold, each retrieved historical case having a measure of similarity to the new error log;

means for estimating a measure of confidence on how closely each historical case matches the new error log, the estimating means comprising means for mapping the similarity measurement in each of the retrieved historical cases to a measure of confidence on how closely the historical case matches the new error log, the mapping means comprising means for generating a confidence function for each of the retrieved historical cases, wherein the confidence function is formed by plotting confidence function points determined from solutions of the similar historical cases, the confidence function providing a relationship between confidence and similarity with other historical cases; and

means for providing a list of the historical error logs having the best measures of confidence.

8. The system according to claim 7, wherein the confidence function points are connected to form the confidence function.

9. The system according to claim 8, wherein the similarity measurement for each of the retrieved historical cases are mapped onto its confidence function to generate the measure of confidence.

10. A method for analyzing an error log generated from a malfunctioning machine, comprising the steps of:

storing a plurality of historical cases generated from a plurality of malfunctioning machines, each of the plurality of historical cases containing data representative of events occurring at the plurality of malfunctioning machines during operation and solutions for correcting the malfunctioning machines in accordance with the events, and each of the plurality of historical cases containing information on the similarity to other historical cases in the case base;

receiving a new error log from a malfunctioning machine, the new error log containing data representative of events occurring at the malfunctioning machine during operation;

retrieving historical cases from the case base having similar solutions to the new error log, the retrieved historical cases being retrieved according to a predetermined similarity threshold, each retrieved historical case having a measure of similarity to the new error log;

estimating a measure of confidence on how closely each historical case matches the new error log, the estimating comprising mapping the similarity measurement in each of the retrieved historical cases to a measure of confidence on how closely the historical case matches the new error log, the mapping comprising generating a confidence function for each of the retrieved historical error logs, wherein the confidence function is formed by plotting confidence function points determined from solutions of the similar historical cases, the confidence function providing a relationship between the similarity of the retrieved historical error log and the measure of confidence on how closely the historical error log matches the new error log; and

providing a list of the historical error logs having the best measures of confidence.

11. The method according to claim 10, wherein the confidence function points are connected to form the confidence function.

12. The method according to claim 11, wherein the similarity measurement for each of the retrieved historical cases are mapped onto its confidence function to generate the measure of confidence.