JPH09243503A - Structure non-defect/defect inspecting apparatus - Google Patents

Abstract

PROBLEM TO BE SOLVED: To rapidly and effectively inspect the non-defect/defect of structural rigidity of an object to be measured. SOLUTION: The structure non-defect/defect inspecting apparatus comprises a feature part extractor 12 for extracting the feature part of a transfer function generated by a transfer function generator 10, and a structure non-defect/defect discriminator 14 for discriminating the non-defect/defect of the structure of the object based on the feature part of the function. The extractor 12 has a resonance point quarrying function 20 for quarrying the waveform of a plurality of the resonance point parts of the function based on the position of the point of the a predetermined standard transfer function. Further, the discriminator 14 has non-defect/defect according-tofrequency discriminating function 26 for discriminating that the structure has defect if the frequency of the point is low when the value of the point is smaller than the standard value by the standard function based on the amplitude and frequency of the point of the quarried waveform by the function 20 and discriminating that the structure has non-defect if the frequency of the point is high when the value of the point is larger than the standard value.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a structural quality inspection device, and more particularly to a structural quality inspection device for determining the quality of welded parts.

The object of the quality judgment according to the present invention is
It is a structure that requires certain rigidity such as welded parts, cast parts, forged parts, and pressed parts. The present invention belongs to a non-destructive inspection for inspecting the quality of a structure without destroying it.

[0003]

2. Description of the Related Art Conventionally, as shown in FIG. 10, determination is made by using all waveforms of transfer function of vibration analysis, or as shown in FIG. 11, determination is made only by height of a part of peak value. .

In the example shown in FIG. 10, an OK range 51 is defined in advance, and if the measured data 50 is outside this OK range, it is determined to be NG (defective). On the other hand, FIG.
In the example shown in FIG. 1, the NG range 52 is included in the peak part of the transfer function.
When the measurement data 50 reaches this NG range, it is determined that the defect is defective.

[0005]

However, in the above-mentioned conventional example, since the entire transfer function is used for the determination, the calculation time is long, and the S / N ratio becomes poor, so that the determination becomes uncertain. Also, if you use only the height of some peaks for judgment,
Although the calculation time is short, there is an inconvenience that it is not possible to cope with the deviation of the resonance frequency and sufficient quality judgment cannot be performed.

Further, in the conventional example, there was a disadvantage that the experiment had to be redone in order to change the judgment standard.

[0007]

SUMMARY OF THE INVENTION It is an object of the present invention to provide a structural quality inspection device capable of improving the disadvantages of the conventional example and, in particular, capable of inspecting the quality of the object to be measured at high speed and reliably. And

[0008]

Therefore, in the present invention, as a first means (Claim 1), a transfer function generating section for generating a transfer function based on free vibration of an object to be measured, and the transfer function generating section. And a structure quality determination unit that determines the quality of the structure of the DUT based on the characteristic part of the transfer function extracted by the characteristic part extraction unit. It has and. Moreover, the characteristic portion extraction unit has a resonance point cutting function that cuts out the waveforms of a plurality of resonance point portions of the transfer function based on the positions of the resonance points of the predetermined standard transfer function. Further, the structure quality determination unit, when the value of the resonance point is smaller than the standard value by the standard transfer function based on the size and frequency of the resonance point of the cutout waveform cut out by the resonance point cutting function, When the frequency is low, it is judged as defective, and when the value of the resonance point is larger than the standard value by the standard transfer function, if the frequency of the resonance point is high, it is judged as good product by frequency. , Is adopted.

In the first means, the transfer function generating section generates a transfer function from the force applied to the object to be measured and the vibration of the object to be measured due to the force applied to the object to be measured. The transfer function can have various configurations as long as it exhibits the characteristic of vibration, and is generated by, for example, the input force, the acceleration of the measured object, the position of the measured object, or the speed. Here, it includes all of the transfer functions generally used for vibration analysis.

The feature extraction unit cuts out a portion at a predetermined position from the transfer function. This cut-out position is the peak portion (resonance frequency portion) of the transfer function of the structure that is considered good for inspection purposes. The peak portion of the transfer function appears in the resonance frequency portion that specifies the vibration mode, and the point of the transfer function at this resonance frequency is called the resonance point (see FIG. 3). The feature extraction unit cuts out the transfer function of the portion including the resonance point to obtain a cut-out waveform.

Further, the structure quality determination unit uses the frequency-based quality determination function to determine, based on the size of the resonance point and its resonance frequency, when the value of the resonance point is smaller than the standard value of the standard transfer function. If the frequency of the point is low, it is determined to be defective (first frequency-specific determination function). This standard value is obtained in advance from an object to be measured having normal rigidity. When the resonance frequency is lower than the standard frequency,
Since the rigidity of the object to be measured is considered to be low, it is determined to be defective even if the resonance point is small.

On the other hand, the frequency-specific pass / fail judgment function determines that the product is non-defective if the frequency of the resonance point is high when the value of the resonance point is larger than the standard value (second frequency-specific judgment function). This is because even if the amplitude (magnitude) becomes large, the resonance frequency becomes high as the rigidity of the object to be measured becomes high, so that the resonance frequency becomes a good product. Is determined. Therefore, in the present invention, it is precisely measured whether or not it has a certain rigidity, which is the purpose of inspection.

In the second means (claim 2), in addition to the matter for specifying the first means, the characteristic portion extracting section has a standardization function for standardizing a plurality of cutout waveforms cut out by the resonance point cutout function. ing. In addition, the structure quality determination unit has a quality determination function by resonance point number that determines a defect when the number of resonance points of the waveform shape based on the standardized data is 2 or more or 0 based on the standardized data standardized by the standardization function, Is adopted.

In the second means, when a part of the resonance frequencies is changed, or when the peak at the resonance frequency is out of the cut-out position, the quality judgment function according to the number of peaks is used.
Judge as defective.

In the third means (Claim 3), in addition to the matter for specifying the second means, the structural quality judging unit normalizes the cut-out standardized data standardized by the standardizing function, and The neural network which outputs the quality level of the rigidity of the measured object using the normalized data normalized by the normalization function as input data.
Moreover, the neural network uses the standard transfer function as a defective area in a region with a higher frequency as the frequency becomes higher, and defines the region larger than the standard value as a defective region in the low frequency region as a standard transfer function as the frequency becomes lower. Based on the learning data in which a region smaller than the standard value is a defective region, and the learning data that is defective when the number of peaks in the region determined to be good by the learning data is 2 or more and when the number of peaks is 0, It is self-organized.

In the third means, the frequency-based quality determination function of the first means and the peak-number-based quality determination function of the second means are realized by a neural network. Therefore, the quality of the structure is determined only by generating learning data from the transfer function of the standard structure or creating learning data in a pseudo manner and inputting the learning data to the network.

The present invention aims to achieve the above-mentioned object by each of these means.

[0018]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing the structure of a structural quality inspection apparatus according to the present invention. The structure quality inspection apparatus includes a transfer function generation unit 10 that generates a transfer function based on free vibration of the DUT 1, and a characteristic portion extraction unit that extracts a characteristic portion of the transfer function generated by the transfer function generation unit 10. 12 and a structural quality determination unit 14 that determines the quality of the structure of the DUT based on the characteristic portion of the transfer function extracted by the characteristic portion extraction unit 12.

In addition, the characteristic portion extracting section 12 has a resonance point cutting function 20 for cutting out the waveforms of a plurality of resonance point portions of the transfer function based on the positions of the resonance points of the predetermined standard transfer function.

Further, when the structure quality judgment unit 14 determines that the value of the resonance point is smaller than the standard value by the standard transfer function based on the size and frequency of the resonance point of the cut-out waveform cut out by the resonance point cutting function 20. When the frequency of the resonance point is low, it is determined as defective, and when the value of the resonance point is higher than the standard value by the standard transfer function, it is determined as good if the frequency of the resonance point is high. The judgment function 26 is provided.

This will be described in detail.

The resonance point (peak value) is the frequency response amplitude at the resonance frequency, and the resonance frequency and the size of the resonance point change depending on the shape, material and rigidity of the object. In the present embodiment, the rigidity of the object to be measured is precisely inspected by utilizing the fact that the resonance frequency increases as the rigidity of the object increases and the frequency response amplitude decreases.

Further, in the present embodiment, the characteristic portion extraction unit 1
2 has a standardization function 22 for standardizing a plurality of cutout waveforms cut out by the resonance point cutout function. In addition, the structure quality determination unit 14 includes a quality determination function by resonance point number 28 that determines a defect when the number of resonance points of the waveform shape based on the standardized data is 2 or more or 0 based on the standardized data standardized by the standardization function. ing.

[First Embodiment] In order to determine the quality of a structure using vibration analysis, a frequency transfer function is used in this embodiment. FIG. 2 is an explanatory diagram showing an example of data input to the transfer function generation unit 10. In the present embodiment, the transfer function is calculated from the force applied to the measured object and the acceleration due to the force applied by the measured object.

As shown in FIG. 2A, when an object to be measured is hit with an impact hammer having a force sensor 2, the illustrated signal is input to the transfer function generator 10.
Further, when the object to be measured vibrates due to this impact, the acceleration pickup 3 captures the acceleration of the object to be measured and outputs a signal as shown to the transfer function generating unit 10.

The transfer function generating section performs FFT processing (fast Fourier transform) on the input waveform to generate the transfer function shown in FIG. In the example shown in FIG. 3, five resonance points 5 appear. Of the measured transfer functions, some resonance frequency peaks (resonance points) are selected and cut out.

The position of the clipping window 6 for cutting out the frequency peak is one in which a characteristic peak is selected by its transfer function. The characteristic peak is a peak that has a high gain and can be easily discriminated, and is a peak in which the height or the frequency greatly changes between the non-defective product and the defective product.

The width of the cutout is ± 5 [H] centered on the peak.
z], but change depending on the memory capacity, speed, number of peaks, etc. The actual cutting peak position and width selection is
The optimum one is selected from the transfer function. Therefore, the cutout peak position and width are predetermined and fixed. Also, the width of the gain is a fixed value that is determined in advance.

FIG. 5 is an explanatory diagram showing a cutout waveform cut out by the resonance point cutout function 20. In the present embodiment, the quality of the rigidity of the structure is determined by evaluating the cutout waveform based on the criteria shown in FIGS. 6 and 7.
Further, instead of evaluating all the cut-out waveforms respectively, the cut-out waveform of the transfer function obtained by one measurement may be standardized, and the pass / fail judgment may be performed based on the standardized data.

FIG. 6 is a table showing the criteria for quality judgment according to this embodiment. In this embodiment, the quality is judged by the relationship between the size of the resonance point and the frequency. Specifically, when the frequency is low, the amplitude is large, or when the amplitude is normal, NG
If the amplitude is small, the O
Judge as K.

FIG. 7 is a diagram for comparing the criteria shown in FIG. 6 with the conventional criteria. As shown in FIG. 7A, the conventional NG region 7 has a constant size regardless of the frequency height. On the other hand, as shown in FIG. 7B, in the present embodiment, when the resonance frequency is high, the amplitude is large even if OK, and when the resonance frequency is low, the amplitude is small but NG.

As described above, in the first embodiment, regarding each of the cut out data or the standardized data,
The quality of the rigidity of the object to be measured is determined based on the criteria shown in FIG. Therefore, it is possible to change the judgment criterion of the quality judgment in accordance with the change of the resonance frequency due to the change of the rigidity of the object to be measured. can do.

[Second Embodiment] In the second embodiment, the quality of the structure is determined by a neural network.

Here, the cut-out waveform shown in FIG. 5 is standardized, and is further normalized by setting the upper limit of the gain (amplitude) width to 1.0 and the lower limit to 0.0 (normalization function 24). The gain width is set so that the peak value is about 0.5 and the quantization error does not occur in order to magnify a minute change in the transfer function and use it for the determination. The number of quantized data is the number of windows × frequency width. The width of this frequency depends on the frequency resolution of the system.

If all the transfer functions are quantized and put into the neural network, the gain is low (noisy).
Since the data is also used for the determination and a long calculation time is required, the cutout by the cutout window 6 is used.

FIG. 8 is an explanatory diagram showing the structure of the neural network 15. As shown in FIG. 8, the neural network 15 has an input layer 15A having a number of units equal to the number of data in the frequency direction determined by the frequency width of the clipping window and the unit frequency according to the frequency resolution, and the input layer 15A. An intermediate layer 15B having a plurality of units connected to each unit and an output layer 15C having a unit that outputs a value from 1.0 to 0.0 are provided.

A value normalized to 1.0 to 0.0 by the normalization function 24 is input to each unit of the input layer 15A. The unit of the output layer 15C is OK.
Learn so that the quality is 1.0 and the quality is 0.0.
Thus, as the neural network 15, a general three-layered one is used. 0 at the time of actual judgment
Although a value between 1 is output, the larger the value, the higher the probability of being an OK product. Since the output value and the degree of failure have a correlative relationship, the threshold value for the output value is determined by the user.

The range of learning OK and NG is shown in FIG. 7 (B). Therefore, the learning data shown in FIG.
Cutout data having peak values as shown in (A) to (C) are prepared, and O is set so that the output becomes 0 in the case of NG.
Learning so that the output is 1 when K (self-organization)
Let it. As a result, the neural network 15 operates as the frequency pass / fail judgment function 26.

The criteria shown in FIGS. 9A to 9C take into consideration that the resonance frequency increases as the rigidity of the object to be measured increases, and according to this, the amplitude (magnitude) becomes large. However, it can be regarded as an OK product. Further, when the resonance frequency is lowered, the rigidity is lowered even if the size is the same, so that it becomes NG.

Further, a waveform having a significantly different waveform is determined to be NG. That is, the data shown in FIGS. 9D and 9E is pseudo-created and the neural network is made to learn. As a result, the neural network 15 operates as the pass / fail judgment function 28 according to the number of resonance points.

As described above, in the second embodiment, by using the neural network, it is possible to feed back the actual data and make a judgment with higher accuracy. Further, since the learning of the neural network can be performed off-line, the judgment standard can be updated without stopping the line. Further, since the learned neural net is used for the determination, the determination time is short.

Further, according to the present embodiment, when a single component whose vibration modes are relatively clearly separated is used as an object to be measured, for example, when the quality of a welded part is judged, highly accurate judgment can be performed. It can be performed and the determination time can be shortened. Further, when the determination is uncertain, it can be updated by feeding back.

[0044]

Since the present invention is constructed and functions as described above, according to the present invention, the structure quality determining unit firstly detects the resonance when the value of the resonance point is smaller than the standard value of the standard transfer function. If the frequency of the point is low, it is determined to be defective,
Even if it is determined as a non-defective product when it is determined based on only the size of the resonance point, it can be accurately determined that the rigidity is poor when the rigidity is low and the resonance frequency is low. Secondly, when the value of the resonance point is larger than the standard value by the standard transfer function, the product is judged to be a good product if the frequency of the resonance point is high. Therefore, if the judgment is based only on the value of the resonance point, it is defective. Even if it is determined that the rigidity is high and the resonance frequency is high, it is possible to determine that the rigidity of the object to be inspected is equal to or higher than a certain level, and determine that the object is a good product. For this reason, for example, when inspecting the quality of welding of a welded part, it is possible to accurately inspect a measured object that does not have the necessary rigidity due to defective welding as a defective product, and at the resonance point portion. Since only the part is cut out and inspected by comparison with the standard transfer function, it is possible to inspect at higher speed than the conventional example in which all transfer functions are inspected. As described above, it is possible to provide an unprecedented excellent structural quality inspection device capable of inspecting the quality of the object to be measured at high speed and reliably.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an embodiment of the present invention.

FIG. 2 is an explanatory diagram showing an example of input data to a transfer function generation unit shown in FIG. 1, FIG. 2 (A) is a diagram showing input of force applied to an object to be measured, and FIG. FIG. 4 is a diagram showing an output from the object to be measured.

3 is a waveform diagram showing an example of a transfer function generated by a transfer function generating unit shown in FIG.

FIG. 4 is an explanatory diagram showing a clipping window when clipping a part of the transfer function shown in FIG.

5 is an explanatory diagram showing an example of a cutout waveform cut out by the cutout window shown in FIG.

FIG. 6 is a table showing an example of a judgment standard of quality judgment by the frequency-specific quality judgment function shown in FIG. 1.

FIG. 7 is a diagram comparing the determination criteria shown in FIG. 6 with a conventional determination criteria, FIG. 7 (A) is an explanatory view showing the conventional determination criteria, and FIG. 7 (B) is a determination according to the present embodiment. It is explanatory drawing which shows a reference | standard.

8 is an explanatory diagram showing an example of a neural network that functions as the structural quality determination unit shown in FIG.

9 is an explanatory diagram showing an example of learning data for self-organizing the neural network shown in FIG.
(A) to (C) are diagrams showing learning data for learning the frequency-based quality determination function, and FIGS. 9 (D) and (E) show learning data for learning the resonance point-based quality determination function. It is a figure.

FIG. 10 is an explanatory diagram showing a conventional example in which the quality of rigidity of a structure is determined based on a predetermined OK range.

FIG. 11 is an explanatory diagram showing a conventional example in which the quality of the rigidity of a structure is determined based on a predetermined NG range.

[Explanation of symbols]

 10 Transfer Function Generating Section 12 Characteristic Extracting Section 14 Structural Quality Judgment Section 20 Resonance Point Extraction Function 22 Standardization Function 24 Normalization Function 26 Frequency-Specific Quality Judgment Function 28 Resonance Point-Specific Quality Judgment Function

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI Technical display location G06F 15/18 560 G06F 15/18 560Z

Claims (3)

[Claims] 1. A transfer function generating section for generating a transfer function based on free vibration of an object to be measured, a characteristic part extracting section for extracting a characteristic part of the transfer function generated by the transfer function generating section, and this characteristic. In a structure quality inspection device comprising a structure quality determination unit that determines the quality of the structure of the object to be measured based on the characteristic portion of the transfer function extracted by the portion extraction unit, the characteristic portion extraction unit is predetermined A resonance point cutout function for cutting out waveforms of a plurality of resonance point portions of the transfer function based on the position of the resonance point of the standard transfer function is provided, and the structural quality determination unit is a cutout waveform cut out by the resonance point cutout function. When the value of the resonance point is smaller than the standard value by the standard transfer function based on the size and frequency of the resonance point, the frequency is determined to be defective if the frequency of the resonance point is low. In addition, when the value of the resonance point is larger than the standard value by the standard transfer function, if the frequency of the resonance point is high, the structure is characterized by having a quality-by-frequency determination function for determining good products. Inspection device. 2. The feature extraction unit includes a standardization function for standardizing a plurality of cutout waveforms cut out by the resonance point cutout function, and the structural quality determination unit converts the standardized data standardized by the standardization function. The structure quality inspection device according to claim 1, further comprising a quality determination function according to the number of resonance points that determines a defect when the number of resonance points of the waveform shape based on the standardized data is 2 or more or 0. 3. The structure quality determining unit normalizes a cut-out standardized data standardized by the standardizing function, and the measured data using the normalized data normalized by the normalizing function as input data. A neural network for outputting the quality level of the rigidity of the object, wherein the neural network defines a region having a higher frequency than the standard value by the standard transfer function as the frequency becomes higher and the frequency of the frequency becomes higher. As for the low region, the learning data in which the region smaller than the standard value by the standard transfer function becomes a defective region as the frequency becomes lower, and the number of peaks in the region determined to be good by the learning data is 2 or more If the number is 0, it is self-organized by the learning data that is considered to be defective. Structure quality inspection apparatus of claim 2 wherein.