JP5013058B2 - Defect detection method and apparatus - Google Patents

Description

  The present invention relates to a defect detection method and apparatus.

  Products such as automobile bodies that are manufactured by press working or industrial products are prone to defects such as cracks and cracks where stress is concentrated. For this reason, in the past, visual inspections by a plurality of inspectors were performed to detect this defect, but the environment caused by noise was not good, and human-specific inspection errors occurred. It was sought after.

  As an example of an apparatus for detecting a defect, there is a quality detection apparatus disclosed in Patent Document 1. This quality detection device is used when press working is performed using a mold, and a vibration sensor is embedded in the mold. Note that the vibration sensor is embedded in the vicinity of a portion where a processing defect such as a crack is likely to occur in the product. The quality detection device compares the normal vibration resonance frequency monitored by the vibration sensor with the vibration resonance frequency detected by the vibration sensor when performing press processing, and determines whether or not a processing defect has occurred. Yes.

Further, as a product inspection method, there is a method disclosed in Patent Document 2, which has the following configuration. First, the subject is hammered with an impulse hammer, and the vibration waveform of the subject obtained by the hammering is collected. Next, the collected vibration waveform is taken out. The extracted vibration waveform is subjected to a fast Fourier transform process to obtain a frequency spectrum of the subject. Next, the peak frequency of the obtained frequency spectrum of the subject is compared with the peak frequency of the frequency spectrum of a plurality of non-defective products that have been measured in advance, and the peak frequency of the subject is the peak frequency of the non-defective product. The quality of the subject is determined based on whether or not it exists in the distribution.
JP-A-6-328156 JP-A-9-171008

  By the way, the quality detection apparatus disclosed by patent document 1 may generate | occur | produce a large vibration in a short time at the time of a press, and there exists a possibility that the vibration which generate | occur | produces with the processing defect smaller than it cannot be measured correctly.

  Further, according to the inspection method disclosed in Patent Document 2, it is possible to perform an inspection without human intervention. However, the subject may vary depending on the shape and production accuracy of the subject. In such a case, the statistical distribution for each peak frequency of the non-defective product that becomes the reference is complicated, and there is a risk of causing a determination error. In addition, since the fast Fourier transform process is not suitable for analyzing short-time phenomena or transient phenomena, it may not be possible to deal with defects that are different from normal defects in the subject. In addition, in the hammering using the impulse hammer, the range in which the subject can be vibrated is limited, and it is considered difficult to excite high-frequency vibrations that can be used when finding the presence of minute flaws. Further, hammering by an impulse hammer may give a secondary defect to the subject depending on the shape and structure of the subject.

  There is still no defect detection device in the industry that can perform a defect inspection in a short time without a human hand while the handling robot is holding (sucking and holding) a detection target.

  An object of this invention is to provide the defect detection method and apparatus which test | inspect a defect in the middle of conveying a detection target object.

In the defect detection method according to the present invention, the object to be detected is sucked and held by the holding unit that is moved by the conveying unit, and the detection object that is sucked and held by the holding unit is being conveyed by the conveying unit, or The conveyance means is stopped halfway, the detection target is vibrated with a vibration nozzle provided in the holding unit, the vibration of the detection target is measured, and a feature amount is extracted from the measurement result in advance. It is characterized in that a correlation with the obtained reference feature value is obtained and the presence or absence of a defect is determined. The timing for vibrating the detection object described above, the transfer means starts vibrating when the way carrying the detection object, or that once stopped in the middle, leaving the determination of the presence or absence of defects It is characterized by sometimes ending the vibration.

Further, the defect detection apparatus according to the present invention includes a holding unit that holds the detection target by suction , and is provided in the holding unit that transfers the detection target by moving the holding unit. A feature quantity from a vibration nozzle that vibrates a detection target, a vibration sensor in which at least a part of a movement range of the detection target held by the transport unit is a measurement range, and a vibration waveform obtained by the vibration sensor And a determination means for determining the presence / absence of a defect by obtaining a correlation between the feature quantity and a reference feature quantity obtained in advance. The holding unit described above is provided with a vacuum cup that comes into contact with the object to be detected, and suction means for sucking air inside the vacuum cup is connected to the vacuum cup.

Moreover, the defect detection apparatus according to the present invention is characterized in that the vibration nozzle is provided in the holding portion. Further, in the defect detection apparatus according to the present invention, the excitation nozzle jets gas to vibrate the detection target, and a gas supply source is provided in a factory in which the transport unit is disposed. The gas supply source and the vibration nozzle are connected by a hose, and a valve is provided on the hose. The detection object described above is a press-processed specimen.

  According to such a defect detection method and apparatus, it is possible to detect a defect in the middle of transporting the detection target object for the presence or absence of a defect such as a crack, acne-like protrusion, or constriction included in the detection target object. For this reason, an inspection process can be easily added to an existing production line, and there is no need to newly establish an inspection process. In addition, since the handling robot can detect the defect while the object to be detected is sucked and held, the inspection can be completed in a short time, and the inspection can be performed without the intervention of the inspector.

In addition, since the vacuum cup is disposed in the holding portion, the handling robot can be sucked and held regardless of the shape of the object to be detected, and can detect a defect. In addition, since the handling robot uses vacuum cups to attract and hold the detection object, the vibration of the detection object is transmitted to the handling robot and not to the vibration sensor. Presence or absence can be accurately determined.
Moreover, since the detection target is vibrated by jetting gas from the vibration nozzle, it is possible to prevent the detection target from being given secondary defects that may occur due to hammering.

  Further, the defect detection method and apparatus according to the present invention does not require driving (advancing / retreating) of the support part of the detection object, the rocket, the excitation nozzle / vibration sensor as in the prior art apparatus, and thus the structure is simplified. At the same time, the inspection time can be shortened.

  The best embodiment of the defect detection method and apparatus according to the present invention will be described below. FIG. 1 is an explanatory diagram of a defect detection apparatus. The defect detection apparatus 10 includes a handling robot 12 (conveyance means), a gas supply source 32, a suction means 42, a vibration sensor 44, a control determination unit 50, and the like. The handling robot 12 is installed inside and outside the press machine, and is used when the detection object 48 processed by the press machine is transported to the next process. The handling robot 12 has a plurality of joints, and these operations are controlled by the robot control means 16 connected to the handling robot 12.

  A holding unit 18 equipped with a vacuum cup 34 and a vibration nozzle 26 is provided at the tip of the arm 14 in the handling robot 12. The holding portion 18 includes first holding means 20, second holding means 22, and third holding means 24 each having a bar shape. And the 1st holding means 20 has faced the vertical direction, the base end side adheres to the front-end | tip part of the arm 14, and the front end side is extended toward the front. Further, the second holding means 22 faces in the lateral direction, and the central portion thereof is fixed to the front end side of the first holding means 20. Furthermore, the 3rd holding means 24 has faced the vertical direction, The center part and each edge part of the 2nd holding means 22 have adhered.

  The vibration nozzle 26 is provided in the vicinity where the first holding means 20 and the second holding means 22 are connected. The vibration nozzle 26 ejects gas to vibrate the detection target object 48, and is arranged in the holding unit 18 with the gas outlet facing the detection target object 48. That is, the direction of the vibration nozzle 26 can be set freely. Further, a hose for supplying gas is connected to the vibration nozzle 26. The other end of this hose is connected to a gas supply source 32 via a valve 30. This valve 30 is provided in the handling robot 12, and opening / closing control is performed by the robot control means 16. Further, the valve 30 may be any one that opens and closes when a signal is input from the robot control means 16. As a specific example, the valve 30 may be the electromagnetic valve 30. A part of the hose shown in FIG. 1 is omitted.

  The vacuum cup 34 is provided for sucking and holding the detection object 48 and is provided in the vicinity of each end of each third holding means 24. FIG. 2 is a side view of the vacuum cup. The vacuum cup 34 has a body portion 36. A cup portion 38 is provided on the lower side of the body portion 36 so as to expand sideward as it goes downward. A recessed portion that opens downward is provided inside the body portion 36 and the cup portion 38. A suction hose 40 is connected to the upper side of the body portion 36, and the other end of the suction hose 40 is connected to the suction means 42 as shown in FIG. By using the suction means 42, the air in the cavity formed by the recessed portion can be sucked through the suction hose 40. The suction means 42 is connected to the robot control means 16. The robot control means 16 is connected to the control determination unit 50.

  Further, the defect detection apparatus 10 has a vibration sensor 44. Although the vibration sensor 44 may be either a contact type or a non-contact type, FIG. 1 shows the non-contact type vibration sensor 44. An example of the non-contact vibration sensor 44 may be a laser Doppler vibration sensor. And the vibration sensor 44 should just be provided in the path | route along which the detection target object 48 which the handling robot 12 adsorbs and holds is conveyed. That is, the measurement range of the vibration sensor 44 overlaps at least a part of the range in which the detection object 48 moves when the detection object 48 is sucked and held by the handling robot 12. The vibration sensor 44 is connected to the control determination unit 50.

  FIG. 3 is a block diagram of the control determination unit. The control determination unit 50 includes control display means 52 for controlling the amount of air ejected from the vibration nozzle 26. A signal output from the control display means 52 is input to the valve 30 via the robot control means 16 to control opening and closing of the valve 30. Although illustration of the robot control means 16 and the suction means 42 is omitted in FIG. 3, the suction means 42 and the robot control means 16 can be controlled by the control display means 52.

  The control determination unit 50 includes a first filter 54, an analog / digital (A / D) converter 56, a waveform cutout unit 58, a second filter 60, a linear prediction coefficient calculation unit 62, and a determination unit 64. The first filter 54 is connected to the vibration sensor 44. The control display unit 52 is connected to the A / D converter 56, the second filter 60, the linear prediction coefficient calculation unit 62, and the determination unit 64.

  Next, the defect detection method will be described. The handling robot 12 drives the arm 14 to place the vacuum cup 34 on the detection object 48 processed by the press. Then, the suction means 42 is operated to suck the air in the cavity of the vacuum cup 34. Then, the detection object 48 is attracted and held by the vacuum cup 34 (handling robot 12). Thereafter, the handling robot 12 conveys the detection object 48 to the next process, but temporarily stops during the conveyance. This stop position is a position where the vibration sensor 44 is disposed. That is, the handling robot 12 is controlled so that the detection object 48 stops within the range that the vibration sensor 44 can measure. In the case shown in FIG. 1, the detection object 48 is stopped below the vibration sensor 44.

  Thereafter, the robot control means 16 outputs an inspection preparation completion signal to the control determination unit 50. Upon receiving this inspection preparation completion signal, the control determination unit 50 outputs an excitation start signal to the valve 30 via the robot control means 16. The valve 30 opens upon receiving a vibration start signal and supplies gas from the gas supply source 32 to the vibration nozzle 26. Then, gas is ejected from the jet nozzle of the vibration nozzle 26 toward the detection target object 48 to vibrate the detection target object 48 with air. At this time, the handling robot 12 holds the detection object 48, but the detection object 48 is sucked and held using the vacuum cup 34, so that the vibration of the detection object 48 is not transmitted to the handling robot 12.

  The vibration sensor 44 measures the vibration waveform of the detection object 48. FIG. 4 shows an output waveform of the vibration sensor when a detection target having no defect is measured. FIG. 5 shows an output waveform of the vibration sensor when a detection target having a defect is measured. As shown in FIGS. 4 and 5, there is almost no difference in the output waveform of the vibration sensor 44 regardless of the presence or absence of defects. Such a measurement result is input to the control determination unit 50.

  The measurement result input to the control determination unit 50 is first input to the first filter 54. The first filter 54 removes unnecessary frequency bands from the vibration waveform input from the vibration sensor 44 to increase the signal-to-noise ratio (SN ratio). The signal from which noise has been removed is input to the A / D converter 56. The A / D converter 56 converts the input analog signal into a digital signal and outputs it to the waveform cutout means 58. The waveform cutout unit 58 cuts out a signal in a certain time region from the input signal and outputs it to the second filter 60. The second filter 60 passes only the frequency band of vibration caused by the defect in the input signal and outputs it to the linear prediction coefficient calculation means 62.

ここでｘ（ｎ）は時刻ｎΔｔにおける信号の値、Δｔはサンプリング間隔、ｍはモデル次数である。数式１は、任意の時刻の信号はそれ以前の信号の加重和として表すことを示しており、自己回帰モデルと呼ばれている。 The linear prediction coefficient calculation means 62 calculates a linear prediction coefficient that is a feature amount of the input signal based on the maximum entropy method. This linear prediction coefficient is a coefficient represented by a _k in Equation 1.

Here, x (n) is the value of the signal at time nΔt, Δt is the sampling interval, and m is the model order. Formula 1 indicates that a signal at an arbitrary time is expressed as a weighted sum of signals before that, and is called an autoregressive model.

ここでＰ（ｆ）は周波数ｆのスペクトル値、Ｐｍは予測誤差の分散である。 Based on the linear prediction coefficient, the signal power spectrum can be calculated as shown in Equation 2.

Here, P (f) is the spectrum value of the frequency f, and Pm is the variance of the prediction error.

  As shown in Formula 2, there is a correspondence relationship between the spectrum and the linear prediction coefficient, and if the spectrum is different, the linear prediction coefficient is different, so that the linear prediction coefficient can be used as the signal feature amount. In addition, since the spectrum can be usually calculated with a small number of linear prediction coefficients, the number of feature amounts is reduced. Usually, in order to identify the characteristics of a signal, the number of linear prediction coefficients is about 20 and reaches a level where self-diagnosis is possible, which is an effective method in automatic diagnosis based on a waveform.

  FIG. 6 shows linear prediction coefficients of a detection target having no defect. FIG. 7 shows linear prediction coefficients of a detection target having a defect. Here, FIG. 6 and FIG. 7 are linear prediction coefficients for 1 second from the time when 0.5 second has elapsed from the head of the vibration waveform measured by the vibration sensor 44. FIG. 6 and FIG. 7 show the linear prediction coefficients calculated every 0.1 second calculated over 1 second. As can be seen from these figures, the linear prediction coefficient differs between the detection object 48 having no defect and the detection object 48 having a defect.

  The linear prediction coefficient is input to the determination means 64. The determination unit 64 determines the presence or absence of a defect by obtaining a correlation between the linear prediction coefficient (feature amount) obtained by the linear prediction coefficient calculation unit 62 and a reference linear prediction coefficient (reference feature amount) obtained in advance. The judgment result is output. This determination means 64 is constructed by a boundary learning type neural network. In the determination method, a pattern to be identified is input so as to obtain a control output corresponding to the signal, a learning operation for correcting the corresponding output by comparing it with a teacher pattern is repeated, and thereafter an arbitrary identification pattern is input. It will be automatically determined.

  Specifically, first, a plurality of learning input patterns are input to the input unit group, and the output pattern of the output unit group corresponding to the input learning input pattern is compared with a predetermined teacher pattern to obtain an error between them. Next, based on this error, an operation for correcting the threshold values of the units constituting each unit group and the weighting factors of the coupling elements connecting the unit groups is repeated, and the learning input pattern is changed. Let them learn.

  Next, an arbitrary pattern is input to the input unit group, and the output pattern of the output unit group is compared with the output pattern corresponding to the learned input pattern. Then, an average value m and a standard deviation σ of feature quantities are calculated from the learning data, and data obtained by changing part or all of the learning data to m ± aσ (a is a real value) is used as boundary data. The neural network is learned by adding the boundary data created in this way to the learning data. In addition, how to determine the real value a may be obtained with reference to knowledge of normal distribution. Thereby, a neural network having a desirable pattern recognition capability is constructed.

  FIG. 8 shows the determination result of the determination means. The horizontal axis of FIG. 8 is the number of the detection object 48, and the vertical axis is the output value (NN output value) of the neural network. As the output value of the boundary learning type neural network, a large value is output if the detection object 48 is a non-defective product, and a small value is output if the detection object 48 has a defect. Yes. By comparing the magnitudes of the output values shown in FIG. 8, the presence / absence of a defect occurring in the detection object 48 is determined.

  When the quality determination of the detection object 48 is completed in the control determination unit 50, a quality determination signal is output from the control display means 52 to the robot control means 16. Accordingly, the robot control means outputs a control signal to close the valve 30 and outputs a movement signal for resuming the conveyance of the detection object 48 to the handling robot 12. Then, the handling robot 12 stops the excitation of the detection object 48 and conveys the detection object 48 to the open position. Note that the position where the detection object 48 is opened differs depending on the quality determination of the detection object 48. For this reason, the handling robot 12 conveys the detection object 48 to a position corresponding to the difference in quality. When the movement of the handling robot 12 stops, a suction stop signal is output from the robot control means 16 to the suction means 42 to stop the suction. As a result, the detection object 48 is released from the handling robot 12, and the defect detection ends.

  According to such a defect detection method and apparatus 10, since the inspection of the defect is performed in the middle of conveying the detection object 48, it is not necessary to newly establish an inspection process in the production line, and it is related to the presence or absence of the factory space. Defect detection. That is, the entire length of the production line can be shortened. In addition, since the handling robot 12 can detect a defect while the object to be detected 48 is sucked and held, the inspection can be completed in a short time, and the inspection can be performed without an inspector's hand. large. Further, if a vibration nozzle 26 or the like is added to the existing handling robot 12, defects can be detected, so that it is easy to deal with existing manufacturing equipment.

  Further, since the vibration nozzle 26 can be attached to an arbitrary position of the holding unit 18, vibration can be performed regardless of the shape of the detection target object 48. In addition, since the direction of the gas outlet provided in the vibration nozzle 26 can be arbitrarily adjusted, the vibration direction of the detection object 48 can be freely set. In addition, the vacuum cup 34 can be disposed not only in the third holding means 24 but also in an arbitrary position of the holding portion 18, and the distance between the vacuum cup 34 and the holding portion 18 can be freely set. The handling robot 12 can suck and hold regardless of the shape of the detection object 48 or the like. Further, since the handling robot 12 sucks and holds the detection target object 48 using the vacuum cup 34, vibration of the detection target object 48 is transmitted to the handling robot 12 and is not transmitted to the vibration sensor 44. The presence or absence of a defect in the detection object 48 can be accurately determined.

  Further, the defect detection apparatus 10 according to the present embodiment does not require driving (advancing / retreating) of the detection object support section, rocket, excitation nozzle / vibration sensor as in the prior art apparatus, and thus the structure is simplified. .

  In addition, the defect detection method and apparatus 10 injects gas toward the detection object 48 and vibrates. For this reason, by adjusting the diameter, gas pressure, and vibration time of the jet nozzle of the vibration nozzle 26, stable and high-frequency vibration can be applied to the detection object 48. In addition, since the vibration sensor 44 captures vibrations up to a high frequency, the information amount of the output of the vibration sensor 44 increases, and the presence or absence of a defect can be determined more accurately. Further, it is possible to prevent the detection object 48 from being given secondary defects that may occur due to hammering.

  Further, the defect detection method and apparatus 10 uses a linear prediction coefficient as a feature value of the output of the vibration sensor 44. For this reason, a defect such as a finer crack can be detected as compared with a method using the amplitude value as an evaluation criterion and a method using the frequency characteristic of the original signal calculated by fast Fourier transform analysis to obtain the feature value. In addition, since the calculation time of the linear prediction coefficient is fast, the feature amount of the vibration sensor 44 can be calculated in a short time. Further, since the linear prediction coefficient reflects the frequency characteristics of the signal, the change can be detected even with a minute defect that does not occur as a change in amplitude, and the feature quantity extraction capability is improved.

  In addition, the defect detection method and apparatus 10 uses a boundary learning type neural network to determine the presence or absence of a defect that has occurred in the detection object 48. For this reason, the boundary learning type neural network is most suitable as a method for determining whether or not there is a defect feature amount in the output of the vibration sensor 44 based on the feature amounts of a plurality of linear prediction coefficients, and the determination accuracy is improved. Further, since the output of the boundary learning type neural network is normalized and takes a value from 0 to 1, the degree of the defect of the detection object 48 can be determined based on the magnitude of the output value.

  Since the valve 30 connected to the vibration nozzle 26 is controlled to be opened and closed by the robot control means 16 and the control determination unit 50, it is not necessary to control when the defect detection of the detection object 48 is not required. . That is, when it is not necessary to detect the defect of the detection target object 48, the handling robot 12 can transfer the detection target object 48 to the next process without stopping once while the detection target object 48 is being transferred.

It is explanatory drawing of a defect detection apparatus. It is a side view of a vacuum cup. It is a block diagram of a control determination part. It is an output waveform of a vibration sensor when a detection target without a defect is measured. It is an output waveform of a vibration sensor when a detection object having a defect is measured. It is a linear prediction coefficient of the detection target without a defect. It is a linear prediction coefficient of a detection target with a defect. It is the determination result of a determination means.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 ......... Defect detection apparatus, 12 ......... Handling robot, 14 ......... Arm, 18 ......... Holding part, 26 ......... Excitation nozzle, 30 ......... Valve, 32 ......... Gas supply source, 34 ......... Vacuum cup, 42 ....... Suction means, 44 .... Vibration sensor, 48 .... Detection object, 50 .... Control determination unit, 62 .... Linear prediction coefficient calculation means, 64 .... ... determination means.

Claims (6)

The object to be detected is adsorbed and held in a holding unit that is moved by a conveying means,
In the middle of conveying the detection object attracted and held by the holding unit by the conveying means, or stopping the conveying means in the middle,
Exciting the detection object with an excitation nozzle provided in the holding unit, measuring the vibration of the detection object,
Extract the feature value from this measurement result to obtain the correlation with the reference feature value obtained in advance, and determine the presence or absence of defects,
A defect detection method characterized by the above. The timing for vibrating the detection object is as follows:
When vibration is temporarily stopped while the conveying means is conveying the detection target, vibration is started,
To end the cuff when the determination of the presence or absence of defects came out,
The defect detection method according to claim 1. A holding unit that holds the detection target by suction , and a transport unit that transports the detection target by moving the holding unit ;
A vibration nozzle that is provided in the holding unit and vibrates the detection target ;
A vibration sensor in which at least a part of the movement range of the detection object held by the transport means is a measurement range;
Linear prediction coefficient calculation means for extracting a feature value from the vibration waveform obtained by the vibration sensor;
A determination means for examining the presence or absence of a defect by obtaining a correlation between the feature amount and a reference feature amount obtained in advance;
A defect detection apparatus comprising:   The defect detection apparatus according to claim 3, wherein the holding unit is provided with a vacuum cup that contacts the detection object, and suction means for sucking air inside the vacuum cup is connected to the vacuum cup. The vibration nozzle is for jetting gas to vibrate the detection target. A gas supply source is provided in a factory in which the conveying means is disposed, and the gas supply source and the vibration nozzle are provided. connect a hose, the defect detection apparatus according to claim 3 or 4, characterized in that a valve to the hose. The detection object, the defect detection apparatus according to any one of claims 3 to 5, characterized in that a subject is pressing.

Images