Measuring and testing – Vibration
Reexamination Certificate
1998-03-13
2001-01-09
Moller, Richard A. (Department: 2856)
Measuring and testing
Vibration
Reexamination Certificate
active
06170333
ABSTRACT:
FIELD OF THE INVENTION
This invention relates to an apparatus for detecting an abnormal sound and a method for judging something wrong in a machine. More specifically, this invention relates to a method for detecting something wrong with regard to a sound or an oscillation generated by an equipment in motion, a method for judging something wrong in a machine based on a detected value of the abnormal sound or the abnormal oscillation, a method for detecting a similarity between a standard oscillation wave and an arbitrary oscillation wave to be monitored and a method for recognizing a speech by using a detected value of the similarity.
BACKGROUND OF THE INVENTION
In a nuclear plant, a high pressure steam flows inside an equipment such as a heat exchanger and a pipe. In such a plant, for dealing with something wrong with the equipment, such as a leak of a steam, there is equipped a monitoring apparatus which detects abnormal sound generated by the equipment, then monitors whether there is something wrong with the equipment based on the detected value of the sound.
In a prior art, a standard pattern vector having, as an element, a feature value, such as a power spectrum of a normal sound, is previously registered, then an Euclid distance or an angle between the standard pattern vector and an input pattern vector is calculated. The input pattern vector is produced as a vector having, as an element, a feature value of a sound to be monitored. Further, in a prior art, something wrong in a machine is judged by comparing an allowed value with a calculated value of the Euclid distance or the angle.
Namely, in an N-dimensional space wherein N is equal to a number of kinds of the feature, a similarity of two pattern vectors is numerically evaluated, and then something wrong in the machine is judged based on the evaluated value. The similarity represents the angle or the Euclid distance which is a linear distance between a point of the standard pattern vector and a point of the input pattern vector.
For example, as shown in
FIG. 17
, a seven-dimensional standard pattern vector
10
A is previously registered corresponding to a normal sound
10
of which a shape of a power spectrum is flat. The vector
10
A has, an element or component, the power spectrum of the normal sound
10
.
Then, corresponding to sounds
11
,
12
,
13
respectively to be monitored, seven-dimensional input pattern vectors
11
A,
12
A,
13
A are produced. Each energy of monitored sounds
11
,
12
,
13
is equal to the energy of the normal sound
10
, but a shape of a power spectrum of each monitored sounds
11
,
12
,
13
is different from the normal sound
10
. The vector
11
A has, an element or component, the power spectrum of the monitored sound
11
. The vector
12
A has, an element or component, the power spectrum of the monitored sound
12
. The vector
13
A has, an element or component, the power spectrum of the monitored sound
13
. As a measure of a similarity between the standard pattern vector
10
A and each of the input pattern vectors
11
A,
12
A,
13
A, the Euclid distance or a cosine of the angle indicated by d
11
, d
12
, d
13
is calculated.
It is assumed there is a relation between a parameter a and each of monitored sounds
11
,
12
,
13
as shown in FIG.
17
.
Namely, as the relationship shown in
FIG. 17
, the parameter &agr; prescribes a change of the each power spectrum shape of sounds
11
,
12
,
13
from the power spectrum shape of the normal sound
10
.
The Euclid distance is obtained as a square root of a value which is a sum of a square of each difference between an element of the standard pattern vector and corresponding element of the input pattern vector. A cosine of the angle is obtained by dividing an inner product of two pattern vectors by a magnitude of two pattern vectors.
By the way, in case of using the Euclid distance or the angle as the measure of the similarity to the normal sound, it happens that the same value is obtained in plural sounds, while a shape of a power spectrum is different from each other. In such a case, it is impossible to distinguish sounds having different feature from each other, thus it is impossible to precisely detect an abnormal sound.
The following is a detailed description.
FIG. 18
shows a change of calculated value d
11
, d
12
, d
13
of the Euclid distance when the parameter a in
FIG. 17
is increased from 0 to 1.
FIG. 19
shows a change of calculated value d
11
, d
12
, d
13
of the cosine of the angle when the parameter a in
FIG. 17
is increased from 0 to 1.
As shown in
FIGS. 18 and 19
, the calculated value d
11
, d
12
, d
13
of the Euclid distance or the cosine of the angle are always equal each other (d
11
=d
12
=d
13
). According to an increase of the parameter a, the Euclid distance value d
11
, d
12
, d
13
are increased and the cosine value d
11
, d
12
, d
13
of the angle are decreased. The angle itself is increased. By the way, generally a power spectrum shape of a white noise is flat and a power spectrum shape of a normal sound generated by the equipment in normal motion is almost the same as a white noise. A power spectrum shape of a noise is slightly changed according to time. Hereinafter, such slight change is called a “sway.”
In
FIG. 17
, in a case that the parameter a is small, it is assumed that the sounds
11
,
12
are “sway” sounds which slightly swayed from the normal sound
10
and that the sounds
13
is an abnormal sound based on a small leak of a steam etc.
As shown in
FIGS. 18 and 19
, when the parameter a is the same, the Euclid distance or the angle from the normal sound
10
is the same in each monitored sounds
11
,
12
,
13
. Therefore, by comparing the value with an arbitrary determined allowed value, it is judged that all of three sounds
11
,
12
,
13
are normal, or conversely, it is judged that all of three sounds
11
,
12
,
13
are abnormal, then it is impossible to distinguish three sounds
11
,
12
,
13
.
On the other hand, it may be considered to register many standard pattern vectors which correspond to the “sway” sounds from the normal sound
10
. However, since there is a limitation in available number of registration of the standard pattern vector because of a memory capacity or processing speed of a computer, it is practically limited to distinguish the “sway” sounds generated by the equipment in normal motion from the abnormal generated by a small steam leak.
As mentioned-above, because prior art uses, as a measure of the similarity, the Euclid distance or the angle among two vectors, it is impossible to exactly detect the abnormal sound and it is impossible to judge something wrong in the machine with a greatly sufficient accuracy.
Therefore, an object of the present invention is to provide an apparatus and a method for detecting an abnormal sound, each of which can exactly obtain a geometric distance between the standard pattern vector and the input pattern vector from two vectors.
Another object of the present invention is to provide a method for exactly judging something wrong in a machine by using a value detected by the above-mentioned abnormal sound detection apparatus or method.
Another object of the present invention is to provide a method for detecting a similarity between a standard oscillation such wave as a voice and an arbitrary oscillation wave as a voice to be monitored, which can exactly obtain a geometric distance between the standard pattern vector and the input pattern vector from two vectors.
Another object of the present invention is to provide a method for recognizing a speech by using a similarity of the oscillation wave detected by the above-mentioned similarity detection method.
SUMMARY OF THE INVENTION
An apparatus for detecting an abnormal sound, according to the present invention comprises: means for producing a standard pattern vector having a feature value of a normal sound as an element, an input pattern vector having a feature value of a sound to be monitored as an element, a positive vector of a reference pattern having a value of arbi
Arakawa Masahiro
Ishihara Yoshinao
Jinnai Michihiro
Kidu Yoshikazu
Ohshima Jun
Entropy Software Laboratory, Inc.
Moller Richard A.
LandOfFree
Apparatus for detecting abnormal sound and method for... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Apparatus for detecting abnormal sound and method for..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Apparatus for detecting abnormal sound and method for... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2519128