JPH06167385A - Sound diagnosing method - Google Patents

Abstract

PURPOSE:To establish a position estimating method for the waveform to be diagnosed, when one or a plurality of diagnostic points are present in one data, by detecting rough position of the waveform to be diagnosed basing on an arbitrary trigger outputted immediately before function of machine at the time of data sampling. CONSTITUTION:Sound produced from a machine 100 to be diagnosed is sampled by a microphone 211 and diagnosed by means of a sound diagnosis unit comprising a measurement data sampling section 210, a waveform feature extracting section 230, a waveform polyline approximating section 240, a waveform collating section 250, a waveform diagnostic section 260, a diagnostic result display section 270, etc. At the time of data sampling, an arbitrary trigger is outputted immediately before operation of the internal mechanism of the machine and appearing time of waveform to be diagnosed is estimated basing on the sampled data with reference to the appearing time of normal waveform at the point to be diagnosed. When easily collatable and not easily collatable patterns are present subsequently, easily collatable pattern is diagnosed first and the appearing time of not easily collatable pattern is estimated from the collating time of the easily collatable pattern.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for diagnosing abnormality and deterioration using sound generated by equipment and devices.

[0002]

2. Description of the Related Art When diagnosing a device abnormality using sound,
A method has been adopted in which the sound emitted by a device is sampled as data, subjected to Fourier transform, and the power spectrum at each frequency is investigated. This method is effective for a stationary sound such as a rotating sound (a state in which sounds of constant sound pressure and timbre are continuously produced). However, it is difficult to diagnose a non-steady sound such as a single sound (sound pressure / timbre changes with time). Since the sound pressure and timbre of a steady sound is constant, it is possible to detect abnormalities by analyzing the sound at any time, but for a non-steady sound, the sound pressure and timbre of that sound changes with time, It is impossible to detect anomalies by the same method as for stationary sounds. Therefore, for abnormality diagnosis of equipment that emits non-stationary sound, it is effective to use not the frequency analysis of the sound but the change of the sound with respect to the time axis direction at a certain frequency, that is, the waveform of the sound. No. 4-115803.

[0003]

When dealing with non-stationary sounds,
The appearance position of the waveform to be diagnosed in the collected data is
Depending on the data, it will be mixed up. Therefore,
When an arbitrary section is extracted from the sampled data, it may not contain the waveform to be diagnosed, and in this case, proper diagnosis cannot be performed. Furthermore, when there are multiple diagnosis target waveforms in the sampled data, the greater the number of diagnosis target waveforms, the more difficult it is to detect an appropriate diagnosis target waveform from among them, but it is difficult. It was
Therefore, an object of the present invention is to estimate a rough position of a waveform to be diagnosed and to determine the estimated position of the waveform to be diagnosed when one or more diagnostic points are present in one data. And a method of detecting a portion to be diagnosed (a portion where the device has operated) further than the range.

[0004]

In order to solve the above problems, the rough position of the waveform to be diagnosed is detected by outputting an arbitrary signal (trigger) immediately before the device is operated at the time of data collection, or the diagnosis is performed. The method of estimating from the appearance time of the normal waveform (reference pattern) at the target location is used.

For the above problems, a matching method using dynamic programming is adopted.

[0006]

By operating the trigger before the operation of the device to be diagnosed, the start of operation of the device can be accurately known.
Further, even when the device or the internal mechanism forming the device is abnormal, the operating order of the mechanism does not change, and therefore the appearance time of the diagnostic target waveform is near the appearance time of the reference pattern. Therefore, an appropriate measurement data series (section estimated to include the waveform to be diagnosed) can be extracted from the measurement data by the above method. Also, by extracting the diagnostic target part from the measurement vector series that approximates the measurement data series with a broken line by collating with the reference pattern using dynamic programming, all possible correspondences between both vector series can be obtained. From among the above, it is possible to efficiently perform the association that best satisfies the given evaluation formula,
Accurate diagnosis of abnormality is possible.

[0007]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a diagram for explaining the processing steps of the acoustic diagnosis method according to the present invention. FIG. 2 is a system configuration diagram for implementing the present invention. In FIG. 2, reference numeral 100 is a device to be diagnosed, and 200 is an acoustic diagnostic device. The sound emitted by the diagnosis target device 100 is collected by the microphone 211 and diagnosed by the sound diagnostic apparatus 200. The acoustic diagnostic device 200 is
A measurement data collection unit 210, a measurement data processing unit 220, a waveform feature extraction unit 230, a waveform broken line approximation unit 240, and a waveform matching unit 2 to which a microphone 211 for collecting sound is connected.
50, a waveform diagnostic unit 260, a diagnostic result display unit 270 including a display device 271, a feature extraction filter storage file 284, a reference pattern storage file 285, a diagnostic knowledge storage file 286, and a process for changing the contents of these files and measurement data. The system comprises a system control unit 280 for displaying the contents of the file 283, the diagnostic result storage file 287, controlling the entire system inside the apparatus, and a common memory 288 for temporarily saving the calculation results in each unit. It is connected by bus 290. The system control unit 280 controls program execution and requests the user from an external input device (keyboard 281, mouse 2).
82).

Each processing step of this embodiment will be described with reference to FIGS.

Measurement data collection process 310 The measurement data collection process 310 includes a measurement data collection unit 210.
Will be enforced at. When a data read command is sent from the system controller 280, the measurement data collection process 330 is performed. In the measurement data collection process 310, the sound emitted by the diagnostic target device 100 is collected using the microphone 211, and the collected data is a power spectrum of a certain constant frequency component with a bandpass filter at a predetermined constant sampling period. Is stored in the measurement data storage file 283.

Reference pattern reading process 320 The reference pattern reading process 320 is performed by the system controller 28.
When a reference pattern read command is sent from 0, the reference pattern is read from the reference pattern storage file and stored in the common memory 288.

Operation of Measurement Data Processing Step 330 The measurement data processing step 330 is performed by the measurement data processing section 220.
Will be enforced in. The measurement data processing step 330 is used for diagnosis from the measurement data storage file 283 based on the appearance time of the reference pattern stored in the common memory 288 when the system control unit 280 sends an instruction to execute this step. A partial data string (measurement data series) is extracted, converted into a decibel value using (Equation 1), the data values are averaged by (Equation 2), and stored in the common memory 288.

[0012]

[Equation 1]

[0013]

[Equation 2]

This process smoothes the data and facilitates the extraction of waveform features in subsequent processes. The measurement data series extracted from the measurement data storage file is the whole or a part of the measurement data. Retrieving an appropriate measurement data series is very important in the reliability of diagnostic results. This is because it is indispensable that the measurement data sequence taken out contains the waveform to be diagnosed. The method of determining the time range vs _i -ve _i of the measurement data series is shown below.

When the appearance time of the reference pattern read in the reference pattern reading step 320 is st _i -et _i ,

[0016]

[Equation 3]

It is assumed that

In the measurement data collection process 310, a signal (trigger) is output immediately before the internal device to be diagnosed is activated, and subsequent data is collected and stored in the measurement data storage file 283. In the measurement data processing step 330, data in an arbitrary section is used as a measurement data series. The method for determining an arbitrary section is the same as. By attaching a trigger, the operating part of the device becomes clear, and there is an advantage that the time lag of the waveform to be diagnosed is prevented.

The number of internal diagnostic target mechanisms differs depending on the diagnostic target device. As a method of coping with this, a case where a plurality of diagnosis target mechanisms exist inside the diagnosis target device will be described. When there are n diagnosis target mechanisms, one sound data has n diagnosis target waveforms. In order to deal with this, it is necessary to have n reference patterns corresponding to each target diagnostic waveform, and to collate n times during measurement data diagnosis. The problem at that time is that the time at which the pattern appears on the measurement data is before or after the time at which the reference pattern appears. Further, it is conceivable that the later the pattern appears, the greater the deviation from the reference pattern appearance time. However, retrieving an appropriate measurement data series is very important to bring more accurate diagnostic results. In the measurement data processing unit 220,
The following is a method of extracting data in an appropriate section from the measurement data as a measurement data series.

In the measurement data collection process 310, a signal (trigger) is output immediately before the internal device to be diagnosed is activated, and subsequent data is collected and stored in the measurement data storage file 283. FIG. 3 shows a flow chart for explaining a data collection processing procedure with a trigger.

Step 1: 510 i is initialized to 1.

[Step 2: 520] When all the data have been read, the process ends.

Step 3: 530 Read the data.

Step 4: 540 When a trigger is output, go to Step 5. Otherwise,
Go to step 3.

Step 5: 550 Measured data i is data of a predetermined arbitrary section.
Is stored in the measurement data storage file 283. When the data in the section is collected, i is incremented by s
Go to step 2.

The measurement data storage file 283 stores n pieces of measurement data according to the internal equipment. When the appearance time of the reference pattern i read in the reference pattern reading step 320 is st _i −et _i , the time range of the measurement data series i is

[0027]

[Equation 4]

It is assumed that A flow chart for explaining the processing procedure is shown in FIG.

Step 1: 610 i is initialized to 1.

Step 2: 620: Read the reference pattern i.

Step 3: 630 The time range vs _i -ve _i to be a measurement data series is determined from the time range st _i -et _{i of the} reference pattern i by (Equation 4).

Step 4: 640 From the measurement data storage file 283, the time range vs _i −
ve _i is taken out as the measurement data series i.

Step 5: 650 The reference pattern i and the measurement data series i are diagnosed.

Step 6: 660 If there is an undiagnosed reference pattern, i is incremented and the process goes to Step 2. Otherwise it ends.

When a plurality of diagnostic waveforms appear in one data, the later pattern may have a larger time lag, and it is difficult to extract an appropriate diagnostic portion from the measurement data. By attaching a trigger, the operating part of the device becomes clear, the problem of time lag of the waveform to be diagnosed can be overcome, and there is an advantage that an appropriate diagnostic part can be extracted from the measurement data.

The reference pattern i-1 is the measurement vector sequence i.
−1 and (FIG. 5), the time range vs _i −v of the measurement data series i with respect to the reference pattern i
e _i

[0037]

[Equation 5]

[0038] FIG. 6 shows a flowchart for explaining the processing procedure.

Step 1: 710 i is initialized to 1.

Step 2: 720: Read the reference pattern i.

Step 3: 730 A time range vs _i -ve _i to be a measurement data series is determined from the time range st _i -et _{i of the} reference pattern i by (Equation 5).

Step 4: 740 From the measurement data storage file 283, the time range vs _i −
ve _i is taken out as the measurement data series i.

Step 5: 750 Diagnosis of the reference pattern i and the measurement data series i is performed. Let the matching position be ms _i −me _i of the measurement data series.

Step 6: 760 If there is an undiagnosed reference pattern, i is incremented. If not, end.

Step 7: 770 The time range vs _i -ve _i to be the measurement data series is determined from the time range st _i -et _{i of the} reference pattern i and the matching position ms _i-1 -me _{i-1 of the} measurement data series. To do.

Step 8: 780 Read the reference pattern i, and go to Step 3.

When the deviation between the appearance time of the reference pattern and the estimated appearance time of these patterns in the measurement data is approximated to some function f (t) on the time axis, the measurement for the reference pattern i is performed. Time range of data series vs _i -ve
_i ,

[0048]

[Equation 6]

It is assumed that FIG. 7 shows a flowchart for explaining the processing procedure.

Step 1: Initialize 810 i to 1.

Step 2: 820 The reference pattern i is read.

Step 3: 830 The time range vs _i -ve _i to be a measurement data series is determined by (Equation 5) from the time range st _i -et _{i of the} reference pattern i and the function f (t).

Step 4: 840 From the measurement data storage file 283, the time range vs _i −
ve _i is taken out as the measurement data series i.

Step 5: 850 The standard pattern i and the measurement data series i are diagnosed.

Step 6: 860 If there is an undiagnosed reference pattern, i is incremented to go to Step 2. Otherwise it ends.

There are some reference patterns that are easy to collate and some that are difficult to collate. When the difficult pattern, the easy pattern, and the reference pattern are lined up on the time axis, the pattern that is easy to collate is collated first by (Equation 6), and the time of the measurement data series is calculated by (Equation 7) from the collation time. presume. further,
Next, obtain the time range of the waveform to be diagnosed by (Equation 8)

[0057]

[Equation 7]

[0058]

[Equation 8]

[0059]

[Equation 9]

A flowchart of the processing is shown in FIG.

Step 1: 905 The reference pattern 1 is read and the time range st of the reference pattern 1 is read.
_1- et ₁ indicates the time range vs that should be the measurement data series.
The ₁ -ve ₁ was determined by equation (3), from the measurement data storage file 283, retrieves the time range vs ₁ -ve ₁ as measured data series 1, the diagnosis.

Step 2: 910 i is incremented.

Step 3: 915 Read the reference pattern i.

Step 4: 920 If the reference pattern has a shape that makes it easy to collate, Step 5
What. If not, go to step 8.

[0065] step5: 925 reference pattern i time range st _i -et _i and the time range vs _i -ve _i should measurement data series from interrogation position ms _i-1 -me _i-1 of the measurement data series ( It is determined by the formula 7).

[0066] step6: from 930 measured data storage file 283, retrieves the time range vs _i -ve _i should measurement data series as the measurement data series i, diagnose and reference pattern i.

Step 7: 935 If the undiagnosed reference pattern remains, go to Step 2. Otherwise it ends.

Step 8: 940 i is incremented.

Step 9: 945 The time range st _i of the reference pattern i having a shape that is easy to match.
-Et _i and collation position ms _{i + 2-} m of the diagnostic waveform two before
Time range vs _i −ve to be measured data series from e _{i + 2
i} is determined by (Equation 8).

[0070] step10: than 950 measurement data storage file 283, taken out of the time range vs _i -ve _i should be in the measurement data series as the measurement data series i, make a diagnosis and the reference pattern i.

Decrement step 11: 955 i.

Step 12: 960 The time range st _i of the reference pattern i having a shape that is difficult to collate.
−et _i and the reference range i + 1 of the already-matched reference pattern and the matching position ms _{i + 1} −me _{i + 1 of the} measurement data series _{i + 1} determine the time range vs _i −ve _i to be the measurement data series by (Equation 9). .

[0073] step13: than 965 measurement data storage file 283, taken out of the time range vs _i -ve _i should be in the measurement data series as the measurement data series i, make a diagnosis and the reference pattern i.

Step 14: Double increments 970 i, and goes to step 3.

When the device to be diagnosed has a plurality of mechanisms, extracting an appropriate measurement data series from the collected data is a condition that is essential for making an appropriate diagnosis, and the above method should be used. Makes it possible with.

Waveform Feature Extraction Process 330 The waveform feature extraction process 330 is performed by the waveform feature extraction unit 230. In the waveform feature extraction process 330, the measurement data stored in the common memory 288 is retrieved, the feature extraction filter retrieved from the feature extraction filter storage file 284 is acted on to obtain the variation amount of the unevenness of the waveform, and the shape of the data waveform. The points (feature points) that are indispensable for expressing the characteristics of the waveform are calculated. The calculated feature points are stored in the common memory 288. The waveform feature extraction process 330 is performed in the following 2 steps. The processing flow is shown in FIG.

Step 1: Feature extraction processing 331 The unevenness of the waveform of the measurement data series is calculated. The product-sum operation of the measurement data series f (t) and the feature extraction filter (w ₂ (x)) that calculates the variation y ₂ (t) of the unevenness of f (t) is performed. y ₂
The larger the absolute value of (t) is, the more rapidly the waveform changes. The change amount y ₂ (t) is calculated by the following formula.

[0078]

[Equation 10]

Step 2: Feature point extraction processing 332 From the variation amount y ₂ (t) of the unevenness of the measurement data series f (t), a point having a large change is calculated as a feature point. Figure 1 shows the calculation method
It will be described using 0. The y ₂ (t) calculated by the feature extraction process 332 is a curve as shown in FIG.
Since the feature point is a point having a large unevenness, it is a point having a large absolute value of the variation amount y ₂ (t). The center of gravity of the variation y ₂ (t) in the range where the positive and negative values continue in the portion where the variation y ₂ (t) of the unevenness has a positive value and the portion where the variation y ₂ (t) has a negative value. The position is set as a feature point t _j . In FIG. 10-b, the feature point t _j is a point at which the area of s (t ₁ , t ₂ ) is halved, that is,

[0080]

[Equation 11]

This is a point that satisfies the condition. Using the feature extraction filter (w ₀ (x)) that calculates the value y ₀ (t) obtained by smoothing f (t) from the data value g (t _j ) at the feature point t _j , Restore

[0082]

[Equation 12]

(T _j , g (t _j )) is the point at which the gradient of the measurement data series changes.

Waveform polygonal line approximation process 350 The waveform polygonal line approximation process 350 is performed by the waveform polygonal line approximation unit 240.
Will be enforced at. In the waveform polygonal line approximation process 350, the waveform feature extraction process 3 stored in the common memory 288 is extracted.
The feature points (t _j , g (y _j )) (j =
0, ..., N) are connected and the measurement data series is approximated by a polygonal line. This polygonal line is used as a measurement vector series (the time width between characteristic points and the amount of data change between them are the components of each vector) and stored in the common memory 288. By vectorization: Since the waveform pattern is stored in the state of being broken into lines, the amount of information can be reduced.

Since the calculation is performed using vectorized data, the amount of calculation can be reduced.

There is an advantage in that the shape of the waveform can be clearly expressed, and in the waveform matching process 350, appropriate matching can be performed by capturing the characteristics of the waveform.

Waveform Matching Step 360 The waveform matching step 360 is performed by the waveform matching section 250. In the waveform matching step 360, the reference pattern storage file 28
3 and the reference pattern stored in the common memory 288 and the measurement vector series created in the waveform polygonal line approximation process 350 are collated, and the portion which is evaluated to be the closest to the reference pattern (diagnosis Target waveform) is extracted from the measurement vector series. The diagnostic target waveform extracted as the collation result is stored in the common memory 288.

A method of collating the measurement vector series with the reference pattern will be described. This matching method is based on dynamic programming. Dynamic programming is the first to satisfy the property that the decision at each stage depends only on the state of the system at that time and is independent of how it was decided at the previous stage. The multi-step decision process is decomposed into a one-step decision process by utilizing the fact that the subsequent results are optimized in the state of this result regardless of the state / decision. FIG. 11 shows the state of the matching method using the dynamic programming method. At each node x _pq , there is a value v _pq, and when there is a problem of minimizing this system through all the systems x _p (p = 1, ..., N), the system x _p-1 (... x _{p-1q -1} , x
_p-1q , x _{p-1q + 1} ...), the optimal route in x _pq is the system x _p-1 to the system x _p
It depends only on the cost. Using this principle, two vector sequences are matched. The characters enclosed in [] below are regarded as vector notation. In FIG. 12, the horizontal axis is the reference pattern Σ.
The right side of [s _i ] is the time direction, and the vertical axis is the measurement vector sequence Σ.
[A _i ] is arranged in the time direction. Each vector sequence consists of m and n vectors,
Since adjacent feature points of each vector constituting the reference pattern are as shown in FIG. 13,

[0089]

[Equation 13]

Similarly, each vector constituting the measurement vector series is

[0091]

[Equation 14]

Each node represents the correspondence between [s _j ] and [a _j ]. The correspondence between these vectors is determined by
In FIG. 11, the same problem occurs as finding the optimum route from x ₁ to x _n that satisfies a certain evaluation formula. However, in this case, since there is no correspondence between the vectors as shown in FIG. 14, the optimum route is obtained from the route shown in FIG. That is, if the node representing the correspondence between [s _j ] and [a _i ] is f (s _j , a _i ), then f (s ₁ , a _k ) (i ≦ k ≦ n)
To f (s _n , a _l ) (i ≦ l ≦ m), the optimum route is obtained. The waveform matching process consists of six steps as shown in the flowchart of FIG.

Step 1: 351 The vector information of the reference pattern is read from the reference pattern storage file 283. The measurement vector series obtained in the previous step 350 is received. Further, a large number that cannot be taken is substituted for the variable s _g that stores the evaluation value of the approximation of both vector sequences = the minimum value of the error. Set i and j to 1.
Go to step 2.

Step 2: 352 left f (s _j-1 , a _i ), lower left f (s _j-1 , a _i-1 ), lower f (s _j ,
a _i-1 ), of the connectable routes, the optimum route is selected. The optimum route minimizes the error between vectors. Go to step 3.

Step 3: Store the optimum route leading to 353 f (s _j , a _i ) and the error g _ji at the time of correspondence between both vectors by the route. If j has reached n, st (if the optimum correspondence between both vector sequences when [s _n ] corresponds to [a _i ] is determined) st
Go to ep4. If not, go to step 1 (d2).

Step 4: Calculate the error g _n when the two vector sequences are associated with each other on the optimum route up to 354 f (s _n , a _i ). g _n is to step5 smaller than s _g. Otherwise, go to d3.

Step 5: Update 355 s _g to g _n . The route to f (s _n , a _i ) is updated to the optimum one for the correspondence between the two vector sequences. i
Reaches m (when all matching is finished)
Go to step 6. Otherwise, go to step 1 (d1).

Step 6: 356 The matching result is sent to the waveform diagnosis process 370 and is sent to the bus 290.
Send through.

The error evaluation formulas used in step 2 and step 4 are:

[0100]

[Equation 15]

When considering a plurality of corresponding vectors as one vector, it is as shown in FIG. An example of the matching result is shown in FIG.
Shown in. The upper waveform is the reference pattern 1010, and the lower waveform is the measurement vector series 1020.

Waveform Diagnosis Process 370 The waveform diagnosis process 370 is carried out by the waveform diagnosis unit 260. In the waveform diagnosis process 370, the result of the waveform matching process 360 stored in the common memory 288 and the diagnostic knowledge information in the diagnostic knowledge storage file 286 are used to determine an abnormality,
The degree of deterioration at the diagnosis point and the type of abnormality are specified. The diagnosis result is stored in the diagnosis result storage file 287.

The information stored in the diagnostic knowledge storage file 286 is as follows.

Error range that can be regarded as normal with respect to the reference pattern at the diagnosis location Various abnormal patterns at the diagnosis location The following diagnosis can be performed using these pieces of information.

(1) Diagnosis of Abnormality / Normality In the error distribution for each vector corresponding to the reference pattern and the measurement vector series obtained from the result of the waveform matching process 360, the portion outside the allowable range is indicated by the information in the diagnostic knowledge storage file 286. If it exists, it is diagnosed as abnormal. Let g _k be the error for each k-th corresponding vector in the reference pattern and the measurement vector series, and check whether g _k (1 ≦ k ≦ a) is within the reference value. If there is something outside the standard value, it is judged as abnormal, and if all are within the standard value, it is diagnosed as normal. The flowchart is shown in FIG. FIG. 19 shows an example of comparison results between the reference pattern and the waveform to be diagnosed. In FIG. 19, the upper vector series is the reference pattern 1010, and the lower vector series is the measurement vector series 1020. The bar graph displays the error 1030 for each corresponding vector. There is a portion where the error exceeds a certain value, that is, there is a portion where the waveform shape is different.Therefore, an abnormality or a sign of abnormality is detected in the mechanism outputting this waveform. Become.

(2) Diagnosis of Cause of Abnormality After the diagnosis of (1) is completed, if the abnormality is judged, the abnormality is specified using the information in the diagnostic knowledge storage file 40.
FIG. 20 shows an example of information used for identifying an abnormality. This diagnosis is
It consists of the processing of four steps shown in FIG. Each ste
p will be described.

Step 1: Error distribution evaluation process 371 The state of this process is the same as that in the diagnosis of abnormal normality. Let g _k be the error for each k-th corresponding vector in the reference pattern and the measurement vector series, and check whether g _k (1 ≦ k ≦ a) is within the reference value. If there is something out of the standard value, it is determined to be abnormal and go to step 2. If all are within the reference value, the waveform diagnosis process is terminated.

Step 2: Waveform change portion extraction processing 372 In the error distribution evaluation processing 510, both vectors of the portion exceeding the reference value A are cut out.

Step 3: Waveform change pattern confirmation processing 37
(3) The state of the change closest to the part extracted in the waveform change part extraction processing 520 is selected from the change pattern as shown in (a) of FIG.

Step 4: Abnormal event determination processing 374 The position between both vector sequences cut out by the waveform change portion extraction processing 520 and the change pattern determined by the waveform change pattern confirmation processing 530 are linked using the if ... Then rule. The abnormal event name is specified by. At that time, a rule map as shown in (FIG. 20-b) is used. Further, the degree of deterioration of the identified abnormal event is diagnosed by calculating the error between the cut out vector sequences.

Diagnosis Result Display Process 380 The diagnosis result display process is performed by the 380 diagnosis result display unit 270. In the diagnostic result display step 380, the result obtained by the waveform diagnostic step 370 is displayed. The displayed contents are detailed information (judgment result of normality / abnormality) and abnormal information (probable location of abnormality, estimated cause of abnormality and degree of deterioration). FIG. 22 shows an example in which an abnormality is detected as the diagnosis result and the estimated location of the abnormality is the “TM belt”. Displays a simplified diagram of the inside of the device to be diagnosed, warns of the abnormal part, and further, "belt loosening" that seems to be the cause of the abnormality
To present.

The above is the whole process in the embodiment.

By taking out an appropriate diagnostic target waveform from the measurement vector series and making an appropriate correspondence with the reference pattern in vector unit, it is possible to clearly calculate the error indicating the difference in waveform shape, and Various diagnoses are possible.
Also, if there are multiple mechanisms in the device to be diagnosed,
Extracting an appropriate measurement data series from the collected data is an indispensable condition for making an appropriate diagnosis, and the method in the example makes it possible.

[0114]

The present invention has the following effects.

It has become possible to diagnose equipment abnormalities by using acoustic data. Furthermore, it is possible to diagnose an apparatus in which a plurality of devices operate continuously, and an accurate diagnosis position can be determined, so the reliability of the diagnosis result is high.

Since the acoustic data is handled in the form of a vector,
The amount of calculation is greatly reduced and the processing speed is improved.

The unskilled person can make a diagnosis which was conventionally possible only by a skilled inspector.

It is possible to greatly reduce the time and effort required for the inspection to maintain the normal state of the equipment.

The time required for inspection can be shortened and efficiency can be improved.

The diagnosis is unified, and individual differences due to differences in the inspectors' achievements and experiences are eliminated.

[Brief description of drawings]

FIG. 1 is a diagram illustrating an internal processing process of an acoustic diagnostic device according to an embodiment.

FIG. 2 is a diagram showing a system configuration for implementing the present invention.

FIG. 3 is a diagram illustrating a procedure for collecting measurement data with a trigger.

FIG. 4 is a method for determining a time range of a measurement data series.

FIG. 5 shows an example of collation of a reference pattern and a measurement vector series.

FIG. 6 is a method for determining a time range of a measurement data series.

FIG. 7 is a method for determining a time range of a measurement data series.

FIG. 8 is a method for determining a time range of a measurement data series.

FIG. 9 is a diagram illustrating a process of a waveform feature extraction process.

FIG. 10A is an example in which a filter for extracting unevenness of a waveform is applied to the measurement data series. (B) It is a figure explaining the calculation method of a feature point.

FIG. 11 is a diagram illustrating dynamic programming.

FIG. 12 is a diagram for explaining connection of nodes.

FIG. 13 is a diagram illustrating a vector of each vector series.

[Fig. 14] Fig. 14 is a diagram for describing an example of correspondence of impossible vectors.

FIG. 15 is a diagram illustrating a waveform matching process.

FIG. 16 is an example in which a plurality of corresponding vectors are regarded as one vector.

FIG. 17 is a collation result example.

FIG. 18 is a diagram illustrating diagnosis of abnormal normality.

FIG. 19 is a diagram illustrating an error for each corresponding vector in the reference pattern and the measurement vector series.

20A is a diagram showing an example of diagnostic knowledge storage file information. FIG. (B) It is a figure which shows the diagnostic knowledge storage file information example.

FIG. 21 is a diagram illustrating a method of identifying and diagnosing a cause of abnormality.

FIG. 22 is a diagram showing a specific example of a diagnosis result display.

[Explanation of symbols]

100 ... Device to be diagnosed, 200 ... Acoustic diagnostic device, 210
… Measurement data collection unit, 211… Sound data input means, 2
20 ... Measurement data processing unit, 230 ... Waveform feature extraction unit, 2
40 ... Waveform polygonal line approximation unit, 250 ... Waveform matching unit, 260
... Waveform diagnostic unit, 270 ... Diagnostic result display unit, 271 ... Diagnostic result display device, 280 ... System control unit, 281 ... Keyboard, 282 ... Mouse, 283 ... Measurement data storage file, 284 ... Feature extraction storage file, 285 ... Standard Pattern storage file, 286 ... Diagnostic knowledge storage file, 2
87 ... Diagnosis result storage file, 290 ... System bus,
310 ... Measurement data collection process, 320 ... Reference pattern reading process, 330 ... Measurement data processing process, 340 ... Waveform feature extraction process, 350 ... Waveform polygonal line approximation process, 360 ...
Waveform matching process, 370 ... Waveform diagnosis process, 380 ... Diagnosis result display process.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Teruji Sekozawa, inventor Teruji Sekozawa 1099, Ozenji, Aso-ku, Kawasaki City, Kanagawa Prefecture, Ltd. System Development Laboratory, Hitachi, Ltd. (72) Koichi Kawaguchi, 5-2 Omika-cho, Hitachi City, Ibaraki Prefecture No. 1 Incorporated company Hitachi Ltd. Omika factory (72) Inventor Naoyoshi Machida 504-2 Shinanomachi, Totsuka-ku, Yokohama, Kanagawa Pref.

Claims (9)

[Claims] 1. A sound generated when a device is operated is sampled as data, a pattern reflecting the characteristics of a data waveform is created, and it is compared with a previously stored normal waveform pattern to determine an abnormality. In the method, the device to be diagnosed is composed of one or more mechanisms when detecting the part where the device operates from the collected data and creating a waveform pattern, and these devices operate continuously or after a certain time. In that case, when the mechanism to be diagnosed operates in the sampled acoustic data and there is one or more output waveforms (diagnostic object waveform), first the data is used to detect the rough section where the mechanism is operating, A sound diagnostic method characterized by more strictly specifying the operating section of the mechanism. 2. The process of detecting the rough section is as follows: (1) When data is collected, a signal is output just before the internal mechanism of the device is activated. (2) The appearance time of the diagnosis target waveform is estimated from the collected data based on the appearance time of the normal waveform (reference pattern) at the diagnosis target location. (3) When there is a reference pattern that is difficult to collate and a pattern that is difficult to collate exists before and after, a pattern that is easy to collate is diagnosed first, and the appearance time of the pattern that is difficult to collate is estimated from the collation time. The acoustic diagnosis method according to claim 1, further comprising: 3. The process of estimating the appearance time of the waveform to be diagnosed is as follows: (1) Of the mechanisms that make up the device, the waveform of the mechanism that was operated immediately before and its reference time and the reference time The appearance time of the current waveform to be diagnosed is estimated in consideration of the time difference between the appearance times of the patterns. (2) The deviation between the appearance time of the reference pattern and the appearance time of the diagnosis target waveform appearing in the collected data is represented by a function, and the appearance time of the diagnosis target waveform is estimated. The acoustic diagnosis method according to claim 2, further comprising: 4. The process of estimating the appearance time of the waveform to be diagnosed, the matching time between the preceding waveform to be diagnosed and its reference pattern among the mechanisms constituting the device is ms _i-1 to me _{i-. 1} , when the appearance time of the reference pattern is from st _i to et _i , the estimated appearance time of the waveform to be diagnosed vs _i to ve
The _{i, vs i = st i +} (ms i-1 -st i-1) -Δt ve i = et i + (me i-1 -et i-1) + claim 3, characterized in that a Δt Acoustic diagnosis method. 5. The process of estimating the appearance time of the waveform to be diagnosed has a relationship expressed by a function f (t) in the matching time between the waveform to be diagnosed of the mechanism that constitutes the device and its reference pattern. When the appearance time of is from st _i to et _i , the estimated appearance times v s _i to ve _i of the waveform to be diagnosed are set as vs _i = st _i + f (t) ve _i = et _i + f (t) The acoustic diagnosis method according to claim 3. 6. The process of estimating the appearance time of the waveform to be diagnosed, the matching time between the diagnostic target waveform that is easy to collate and its reference pattern is ms _{i + 1} to me _{i + 1} When the appearance time is from st _i to et _i , the estimated appearance times v s _i to ve _i of the waveform to be diagnosed are vs _i = ms _{i + 1} − (st _{i + 1} −st _i ) −Δt ve _i = me _{i. +1 - (et i + 1 -et} i) + Δt acoustic diagnostic method according to claim 2, characterized in that. 7. The acoustic diagnosis method according to claim 2, wherein a dynamic programming method is used for the processing for strictly specifying the operating section of the mechanism. 8. The process of rigorously specifying the operation section of the mechanism using the dynamic programming method is a reference pattern that is obtained by approximating the waveform of acoustic data that is generated at all times by a polygonal line into a vector pattern, and the collected sound. A measurement vector series is obtained by approximating the waveform of data with a polygonal line, and the degree of coincidence between the i-th vector forming the reference pattern and the j-th vector forming the measurement vector series is determined by a dynamic programming method using a certain evaluation formula. I is stored as an evaluation value calculated by incrementing i within the number of vectors of the reference pattern based on
8. The acoustic diagnosis method according to claim 7, wherein is performed at any time while incrementing within the number of vectors of the measurement vector series, and a portion having a matching relationship having the best evaluation value is a diagnosis target portion. 9. The acoustic diagnosis method according to claim 8, wherein the following equation is used as the evaluation equation. (Error) = (difference in corresponding vector between reference pattern and measurement vector series)