Data processing: artificial intelligence – Neural network – Learning task
Reexamination Certificate
2001-03-20
2004-01-20
Davis, George B. (Department: 2121)
Data processing: artificial intelligence
Neural network
Learning task
C706S016000
Reexamination Certificate
active
06681215
ABSTRACT:
BACKGROUND OF THE INVENTION
The present invention relates to a learning method and apparatus for improving causal networks, and particularly to a learning method and apparatus for Bayesian belief networks.
Complex electromechanical systems such as locomotives are composed of several complex sub-systems. Each of these sub-systems is built from components that may fail over time. When a component does fail, it is difficult to identify the failed component. This is because the effects or problems that the failure has on the sub-system are often neither obvious in terms of their source nor unique. The ability to automatically diagnose problems that have occurred or will occur in the locomotive sub-systems has a positive impact on minimizing down-time of the electromechanical systems.
Computer-based systems are used to automatically diagnose problems in a locomotive in order to overcome some of the disadvantages associated with completely relying on experienced personnel. Typically, a computer-based system utilizes a mapping between the observed symptoms of the failures and the equipment problems using techniques such as a table look-up, a symptom-problem matrix, and production rules. These techniques work well for simplified systems having simple mapping between symptoms and problems. However, complex equipment and process diagnostics seldom have simple correspondences between the symptoms and the problems. In addition, not all symptoms are necessarily present if a problem has occurred, thus making other approaches more cumbersome.
Morjaria et al. U.S. Pat. No. 5,845,272 teaches a system and method for isolating failures in a locomotive. A locomotive comprising several complex sub-systems is detailed. A method and system is set forth for isolating causes of failure, generally including supplying incident information occurring in each of the several sub-systems during operation of the locomotive; mapping some of the incidents to indicators, wherein each indicator is representative of an observable symptom detected in a sub-system; determining causes for any failures associated with the incidents with a fault isolator; wherein the fault isolator comprises a diagnostic knowledge base having diagnostic information about failures occurring in each of the plurality of sub-systems and the indicators, and a diagnostic engine for processing the mapped indicators with the diagnostic information in the diagnostic knowledge base; and providing a course of action to be performed for correcting the failures.
A particularly useful tool for determining probabilities of certain isolated failures in a locomotive is a causal network, as detailed in Morjaria et al. One type of a causal network is a Bayesian Belief Network (BBN). BBNs are conventionally used to determine the conditional probability of the occurrence of a given event. For a detailed description of BBNs, reference is made to certain useful texts, including Neopolitan, Richard E.,
Probabilistic Reasoning in Expert Systems,
pp. 251-316, John Wiley and Sons, 1990.
The ability to automatically improve the performance of a BBN is important for improving its performance and eliminating the time-consuming and complicated task of physically modifying the BBN. In application to locomotive fault diagnosis, present BBNs do not have the ability to automatically improve their performance, or learn, when they make errors in diagnosis. To improve their performance, an expert usually examines the current BBN, and makes modifications to it based on his/her expertise and the type of misdiagnoses produced by the BBN. This task is time-consuming and involved, and does not provide the ability to adapt the BBNs performance based on the locomotive”s operational characteristics.
There is a need to improve the performance of a BBN so as minimize or eliminate the time consuming and complicated task of physically modifying the BBN.
SUMMARY OF THE INVENTION
There is provided a system and method for improving a causal network. A new apriori probability is determined for a repair or a configuration factor within the causal network. The new apriori probability is compared to an old apriori probability for the repair or the configuration factor. If the new apriori probability differs from the old apriori probability by more than a predetermined amount, the causal network is updated.
In another aspect, a causal network result is stored for a causal network. The causal network includes a plurality of root causes with a symptom being associated with each of the root causes. The causal network further includes an existing link probability related to the symptom and root cause. An expert result or an actual data result related to each of the symptoms is stored. A new link probability is computed based on the stored causal network result, and expert result or the actual data result.
These and other features and advantages of the present invention will be apparent from the following brief description of the drawings, detailed description, and appended claims and drawings.
REFERENCES:
patent: 5802256 (1998-09-01), Heckerman et al.
patent: 5845272 (1998-12-01), Morjaria et al.
patent: 6353679 (2002-03-01), Cham et al.
Agosti Ann M.
Davis George B.
General Electric Company
Patnode Patrick K.
LandOfFree
Learning method and apparatus for a causal network does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Learning method and apparatus for a causal network, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Learning method and apparatus for a causal network will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3194654