Patent
1995-12-20
1998-06-30
Hafiz, Tariq R.
395 60, 395 61, G06F 1518
Patent
active
057746290
DESCRIPTION:
BRIEF SUMMARY
TECHNICAL FIELD
The present invention relates to an inference apparatus for inferring the causes of a failure of a drive machine or the like based on the results of so-called normal diagnosis or reverse diagnosis.
BACKGROUND ART
The inference of the cause of failure, etc., of a drive machine in construction machinery or the like by means of so-called normal diagnosis has been common practice for some time.
A normal diagnosis treatment involves inferring the causes of a failure from the causal relationship between various phenomena and various causes, that is, from the current incidence of various phenomena and normal diagnosis knowledge indicating the degree of effect, incidence, etc., between a phenomenon and a cause. This normal diagnosis tends to infer at a higher degree of certainty the higher is the incidence of the cause.
Normal diagnosis is performed by the means (a) and (b) shown below, for example.
(a) This is based on Bayesian theory. When a great deal of past diagnostic data are available, these data are analyzed and the occurrence probability of the cause with respect to the occurring phenomenon (the conditional probability) is calculated from the following formula (1). {.SIGMA.F(Pi.linevert split.Cn).multidot.F(Cn)} (1)
Here, F(Cj.linevert split.Pi) is the conditional probability of cause Cj at the time of occurrence of phenomenon Pi, F(Pi.linevert split.Cj) is the conditional probability of phenomenon Pi at the time of occurrence of cause Cj, F(Cj) is the occurrence probability of cause Cj, and n is the number of causes.
The drawback to a method based on Bayesian theory is that although when numerous failure cause data are available the terms on the right side of the above-mentioned formula (1) are determined subjectively and the F(Cj.linevert split.Pi) on the left side can be determined effectively, in actual practice it is rare for numerous failure diagnosis data to be available, and this method cannot be applied when there are few data available. Also, this method involves computing the average occurrence probability for all data, and does not allow trends in causes that have occurred recently to be considered.
(b) Another method is to extract the occurrence probability of a cause with respect to the occurring phenomenon (the conditional probability) based on the past experience of experts, and to compile this in a matrix and use it as normal diagnosis knowledge to perform normal diagnosis. The advantage of this method is that it is easy to obtain knowledge because the knowledge is expressed in a matrix form, and system construction is easier.
Normal diagnosis knowledge can be expressed as in FIG. 2(a), for example. In the figure, Wij can be treated as corresponding to the conditional probability F(Cj.linevert split.Pi) in the Bayesian theory as indicated by the following formula (2).
This normal diagnosis knowledge is based on the past experience of experts, the advantage of which is that knowledge can be easily extracted, without a need for a large quantity of failure data as in the case of the Bayesian theory in (a). This is considered to be the most salient advantage to an expert system.
However, since the knowledge of an expert is not obtained from the analysis of actual failure diagnosis data, the reliability thereof is not considered to be that high.
Also, incidents continue to occur even after system construction. At this point, an expert can learn this and modify his or her knowledge, but the system itself cannot update its information, so the knowledge becomes obsolete, and before long the system can no longer make the correct evaluation. This is considered to be the major drawback to an expert system.
If one were to attempt to construct a system that performs diagnosis based on continuously up-to-date information, then one would have to keep modifying the knowledge whenever needed, entailing a tremendous amount of labor and driving up the cost markedly.
Meanwhile, an error back propagation method (BP method) that utilizes a neural network is generally used for learning. This
REFERENCES:
International Search Report for International Application No. PCT/JP94/01091, mailed Oct. 4, 1994.
Kitamura Tatsuya
Suzuki Kunitoshi
Yamaguchi Hiroyoshi
Hafiz Tariq R.
Komatsu Ltd.
LandOfFree
Inference apparatus using occurence rate data and degree of rele does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Inference apparatus using occurence rate data and degree of rele, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Inference apparatus using occurence rate data and degree of rele will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-1869123