Data processing: measuring – calibrating – or testing – Measurement system – Performance or efficiency evaluation
Reexamination Certificate
2000-03-10
2004-08-31
Wachsman, Hal (Department: 2857)
Data processing: measuring, calibrating, or testing
Measurement system
Performance or efficiency evaluation
C702S181000, C702S183000
Reexamination Certificate
active
06785636
ABSTRACT:
TECHNICAL FIELD
This invention relates to fault diagnosis in a complex system, such as a nuclear plant, using probabilistic reasoning techniques. In a preferred embodiment, a Bayesian network is utilized. The technique provides a way to capture knowledge and reach rational decisions in uncertain domains by casting the decision-making process as computation with a discrete probability distribution represented by a causal network. This technique also allow multiple hypotheses and/or diagnoses to be presented simultaneously to the operator.
BACKGROUND OF THE INVENTION
Modern complex systems such as nuclear power plants create challenges to the operating staff for understanding and trouble-shooting system problems. One important issue is posed by the large amount of information that the operator must absorb before making operational decisions. Although this issue is not conspicuous during normal operation, it becomes very important during abnormal or accident condition due to the limited decision time. Moreover, it is known that the reliability of human action is adversely affected at the time of crisis due to psychological factors. Decision support tools for helping the operator to perform diagnosis and initiate recovery action may greatly improve the system safety and reliability. For the purposes of explanation, the following description is based upon a nuclear power plant, with the understanding that the invention is applicable to other systems. “Intelligent” approaches to various diagnostic problems in the nuclear plant domain have been developed. One approach uses Probabilistic Risk Assessment that focus mainly on modeling the potential response of a reactor to an abnormal condition and its consequences. Other work explores AI and learning techniques such as artificial neural network and expert systems. With these approaches, it is difficult to explain how the system arrived at its decision to the user. Some expert systems are easier to explain, but do not correspond well to the probabilistic relationships between causes and effects that occur in complex systems due to the large number of degrees of freedom which are not observed.
Bayesien networks may also be used. Exemplary Bayesien networks are shown in
FIGS. 1 and 2
. The network of
FIG. 1
is constructed using an automated system, many of which are known to those of skill in the art. Although efficient, these systems often do not account for dependence between variables that only an expert would know.
Bayesien networks such as that of
FIG. 2
may be constructed by hand. Although such networks may more accurately reflect underlying interrelationships, the manual method requires highly skilled engineers and is extremely time consuming.
In view of the above, there exists a need in the art for a system to diagnose faults in complex environments and which supports a comprehensible explanation of the results, as well as a strong correspondence to probabilistic relationships between causes and effects.
SUMMARY OF THE INVENTION
In accordance with the present invention, a graph of a casual network is constructed by hand based on the knowledge of an expert, and conditional probability tables are generated through automation. Additionally, a user interface is provided to conveniently display results. The graph of the network is a relatively simple object, well suited for human design. The conditional probability tables, on the other hand, may contain very many numbers. Thus automation is appropriate. A further advantage of designing the graph by hand, is that when adjustments to the probability tables by hand are unavoidable, due to inadequacies of the available data set, it is easier to adjust the tables in a manually created graph than one produced automatically. If the graph is constructed automatically, correcting a data problem may require the adjustment of certain conditional probabilities that seem unnatural and cannot be readily calculated. Another aspect of the invention includes using the probabilistic graph which is trained to provide a diagnosis based upon input of the extracted features from real line measurement values and outputting simultaneously multiple hypotheses through the graphical user interface.
REFERENCES:
patent: 5101337 (1992-03-01), Ebizuka
patent: 5133046 (1992-07-01), Kaplan
patent: 5293323 (1994-03-01), Doskocil et al.
patent: 5528516 (1996-06-01), Yemini et al.
patent: 5581459 (1996-12-01), Enbutsu et al.
patent: 6076083 (2000-06-01), Baker
patent: 6345265 (2002-02-01), Thiesson et al.
Darken Christian
Erdmann Jochen
Santoso Nugroho I.
Paschburg Donald B.
Siemens Corporate Research Inc.
Wachsman Hal
LandOfFree
Fault diagnosis in a complex system, such as a nuclear... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Fault diagnosis in a complex system, such as a nuclear..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Fault diagnosis in a complex system, such as a nuclear... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3271716