JP2006208186A - Shape acceptable/unacceptable determining device and shape acceptable/unacceptable determining method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a shape acceptable/unacceptable determining device and shape acceptable/unacceptable determining method which determine the shape acceptance/unacceptance without preparing a sample for teaching in advance. <P>SOLUTION: The feature value F of a defective of feature values F and G is calculated from an estimated image 33b produced by estimation instead of an actually picked up image 33a, so that it is not required that many defectives are produced intentionally and actually picked up. Labor hour and cost for collecting the feature values F of the defectives are eliminated. The shape data 34d showing the shapes of the defectives can be produced infinitely within the range of the defectives. The feature values F calculated on the basis of the shape data 34d can be infinitely prepared thereby. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a shape pass / fail determination device and a shape pass / fail determination method for determining pass / fail of a shape to be examined.

2. Description of the Related Art Conventionally, as this kind of shape pass / fail judgment device, a device for photographing a teaching board on which a defect is formed in advance and calculating a plurality of feature values from the image data is known (for example, Patent Document 1). ,reference.).
According to this configuration, the feature value of the image data obtained by imaging the substrate having a defect and the feature value obtained from the image data obtained by imaging the product substrate can be compared using the discriminant function. It was possible to determine the similarity between the captured image data and the image data captured of the substrate having a defect. That is, when the similarity between the image data obtained by imaging a product substrate and the image data obtained by imaging a substrate having a defect is strong, it can be determined that the product substrate has a defect, and the product substrate is imaged. When the similarity between the obtained image data and the image data obtained by imaging the substrate having a defect is weak, it is possible to determine that the product substrate is a non-defective product.
JP 11-344450 A

In the conventional shape pass / fail judgment device described above, a teaching board on which a defect is formed must be photographed in advance, and a teaching board having a defect is prepared before the pass / fail judgment is performed. There was a need. However, if the yield of the product substrate is good, a substrate having a defect is rarely produced, and a substrate having a defect must be created intentionally. Therefore, the effort and cost for creating a substrate having a defect intentionally has been wasted. In addition, in order to calculate a statistically reliable discriminant function, a large number of teaching boards must be prepared, which increases the labor and cost. Furthermore, if the cause of the defect is not clear, there is a problem that even a defective substrate cannot be created intentionally.
The present invention has been made in view of the above problems, and an object of the present invention is to provide a shape pass / fail judgment apparatus capable of executing pass / fail judgment without preparing a teaching sample in advance.

  In order to achieve the above object, according to the first aspect of the present invention, an imaging means for imaging an inspection object and generating a captured image, an actual feature value calculating means for calculating a feature value of the captured image as an actual feature value, and , Shape data acquisition means for acquiring three-dimensional data representing the shape of a non-defective product or a defective product, and estimation for generating an estimated image obtained by estimating the captured image when the good product or defective product is imaged based on the three-dimensional data Means for calculating the feature value of the estimated image estimated as the estimated feature value; and comparing the actual feature value with the estimated feature value to determine pass / fail of the inspection object. And a pass / fail judgment means for performing the above.

  In the invention of claim 1 configured as described above, the imaging means generates a captured image by imaging the inspection object. The actual feature value calculation means calculates a feature value for the captured image from the captured image as an actual feature value. The shape data acquisition means acquires three-dimensional data representing the shape of a non-defective product or a defective product. Then, based on the three-dimensional data, the estimation unit estimates the captured image of the non-defective product or the defective product, thereby generating an estimated image.

  Further, the estimated feature value calculating means calculates the estimated image feature value for the non-defective product or the defective product as an estimated feature value. The pass / fail judgment means judges pass / fail of the inspection object by comparing the actual feature value with the estimated feature value. That is, if the actual feature value obtained by imaging the inspection object approximates the estimated feature value based on the three-dimensional data representing the shape of the non-defective product, the inspection object is a non-defective product. Can do. Conversely, if the actual feature value obtained by imaging the inspection object approximates the estimated feature value based on the three-dimensional data representing the shape of the defective product, the inspection object is a defective product. It can be said that there is.

  According to such a configuration, the estimated feature value can be obtained without actually imaging the non-defective product or the defective product, and the quality determination can be performed using the estimated feature value. Therefore, since it is not necessary to actually create the non-defective product or the defective product, it is possible to eliminate the labor and cost for creating the non-defective product or the defective product.

  Furthermore, in the invention according to claim 2, the imaging unit generates the captured image by acquiring the state of the electromagnetic wave incident via the inspection object for each position, and the estimating unit includes: Based on the incident angle acquisition means for acquiring the incident angle of the electromagnetic wave incident on the inspection object when the imaging means images the inspection object, the electromagnetic wave is based on the three-dimensional data and the incident angle. Behavior estimating means for estimating behavior when reaching a non-defective product or a defective product, arrival state estimating means for estimating the state of the electromagnetic wave reaching the imaging means based on the behavior, and the electromagnetic wave reaching the imaging means And an estimated image generating means for generating the estimated image by arranging values indicating the state of each position for each position.

  In the invention of claim 2 configured as described above, the imaging unit generates a captured image by acquiring the state of the electromagnetic wave incident via the inspection object for each position. In other words, since the characteristics of the inspection object are reflected in the electromagnetic wave incident through the inspection object, the characteristics of the inspection object are expressed by acquiring the state of the electromagnetic wave for each position. The captured image can be generated. For example, by inputting visible light reflected by the inspection object with a camera as the imaging means, a reflected image reflecting the characteristics of the inspection object can be generated. Of course, even when the electromagnetic wave passes through the inspection object and enters the imaging means, it is needless to say that a reflected image reflecting the characteristics of the inspection object can be generated.

  On the other hand, the incident angle acquisition means acquires the incident angle of the electromagnetic wave incident on the inspection object when the imaging means images the inspection object. The behavior estimation means estimates a behavior when the electromagnetic wave reaches the non-defective product or the defective product. That is, assuming the case where the electromagnetic wave reaches the non-defective product or the defective product, the behavior at that time is estimated. For example, reflection, transmission, refraction, scattering, diffraction, and the like are estimated as the behavior when the electromagnetic wave reaches the non-defective product or the defective product. Since the behavior estimation means acquires the incident angle and the shape of the non-defective product or defective product from the three-dimensional data, the behavior when the electromagnetic wave reaches the good product or defective product can be estimated from these. it can.

  Moreover, the said electromagnetic waves can apply the thing of various wavelengths, such as an electromagnetic wave, light (infrared light, ultraviolet light, visible light), an X-ray, and a gamma ray. Depending on the wavelength of the electromagnetic wave, the behavior when the non-defective product or the defective product is reached varies greatly. Furthermore, depending on the physical properties of the non-defective product or the defective product, the behavior when the non-defective product or the defective product is reached varies greatly. Therefore, the behavior estimation means acquires not only the incident angle and the shape of the non-defective product or defective product, but also other behavior fluctuation factors as described above, and when the electromagnetic wave reaches the non-defective product or defective product. Estimate the behavior of

  The arrival state estimation unit estimates the state of the electromagnetic wave reaching the imaging unit based on the estimated behavior. That is, since the behavior of the electromagnetic wave when it reaches the non-defective product or the defective product is obtained, it is possible to estimate the state where the electromagnetic wave that has passed through the non-defective product or the defective product reaches the imaging means. For example, if the behavior of the electromagnetic wave when reaching the non-defective product or the defective product is reflection in a direction different from that of the imaging unit, it is possible to estimate a reaching state in which the reflected electromagnetic wave does not reach the imaging unit. it can. On the other hand, if the behavior of the electromagnetic wave when it reaches the non-defective product or the defective product is a reflection in the direction toward the imaging means, the estimated state of how much the reflected electromagnetic wave reaches the imaging means is estimated. can do. Further, the estimated image generation means generates the estimated image by arranging values indicating the state of the electromagnetic wave reaching the imaging means for each position. The estimated image composed of a plurality of pixel data each having a value indicating the state of the electromagnetic wave can be generated.

  Furthermore, in the invention concerning Claim 3, the said behavior estimation means is set as the structure which estimates the reflective behavior at the time of the said electromagnetic wave reaching | attaining the said non-defective product or a defective product. In the invention of claim 3 configured as described above, a reflection behavior when the electromagnetic wave reaches the non-defective product or the defective product is estimated by the behavior estimating means. That is, when the imaging reflection image of the inspection object is imaged by the imaging means, it is possible to generate an estimated reflection image of the good product or the defective product that can be compared with this.

Further, the invention according to claim 4 is configured such that specular reflection and diffuse reflection are taken into account when the behavior estimation means estimates the reflection behavior.
In the invention of claim 4 configured as described above, it is possible to estimate the reflection behavior in consideration of both specular reflection and diffuse reflection.

In the invention according to claim 5, the shape data obtaining means obtains a plurality of three-dimensional data in which the shapes of the non-defective product or the defective product are varied.
In the invention of claim 5 configured as described above, since the shape data acquisition means acquires a plurality of three-dimensional data in which the shape of the non-defective product or the defective product has a variation, the estimation feature considering the variation of the shape A value can be calculated. Therefore, reliability can be improved when the pass / fail determination means performs pass / fail determination by comparing the actual feature value and the estimated feature value by a statistical method. Of course, since the frequency of the estimated feature value can be easily increased, the statistical reliability can be further improved.

In the invention according to claim 6, the actual feature value calculating means and the estimated feature value calculating means calculate various actual feature values and estimated feature values from the captured image and the estimated image, and The pass / fail determination means is configured to perform pass / fail determination by comparing discriminant values obtained by linearly combining the various actual feature values and the estimated feature values.
In the invention of claim 6 configured as described above, various actual feature values and estimated feature values are calculated from the captured image and the estimated image by the actual feature value calculating means and the estimated feature value calculating means. . Then, the pass / fail determination means performs pass / fail determination by comparing a discriminant value obtained by linearly combining various actual feature values with a discriminant value obtained by linearly combining various estimated feature values. When it is not possible to make an accurate pass / fail judgment using only a single type of the actual feature value and the estimated feature value, it is possible to make an accurate pass / fail judgment using the various actual feature values and the estimated feature value. . The coefficients for linearly combining the various actual feature values and the estimated feature values can be optimized using a known multivariate analysis technique.

    The above-described device can also be applied as a method for realizing such a device, and the invention according to claim 7 has basically the same operation. Of course, it is also possible to make the structure described in claims 2 to 6 correspond to the method of claim 7.

As described above, according to the first and seventh aspects of the invention, it is possible to provide a shape quality determination device and a shape quality determination method that can execute a quality determination without preparing a teaching sample in advance. .
According to the second aspect of the present invention, the estimated image can be estimated from the behavior of the electromagnetic wave.
According to the invention of claim 3, it is possible to estimate a reflected image of the inspection object.
According to the invention of claim 4, the reflected image can be estimated accurately.
According to the fifth aspect of the present invention, the quality determination can be performed accurately.
According to the sixth aspect of the invention, erroneous determination can be prevented.

Here, embodiments of the present invention will be described in the following order.
(1) Configuration of shape pass / fail judgment device:
(2) Defect data accumulation processing:
(3) Good data storage processing:
(4) About discriminant function:
(5) Pass / fail judgment processing:
(6) Modification:

(1) Configuration of shape pass / fail judgment device:
FIG. 1 shows the configuration of a shape pass / fail judgment apparatus according to the present invention. In the figure, the shape pass / fail judgment apparatus 10 is composed of an imaging unit 20 and a computer 30 which are connected to each other. The imaging unit 20 images the mounting board 50 of the inspection object to generate a captured image, and outputs the captured image to the computer 30. The computer 30 inputs the captured image and analyzes the captured image to determine whether the mounted substrate 50 that has captured the image is acceptable. In FIG. 1, the imaging unit 20 includes an XY stage 23 on which a mounting board 50 is placed at a fixed position and movable in the XY (horizontal) direction based on a command from the controller 21. The camera 22 has an optical system 22a composed of a predetermined optical lens oriented vertically downward so that an image vertically below can be input. A CCD image pickup plate 22b is provided inside the camera 22, and the optical system 22a can form an image vertically below the CCD image pickup plate 22b.

  The CCD imaging plate 22b is composed of a plurality of CCD imaging elements arranged in a dot matrix. The CCD image sensor is a photoelectric element that generates a charge according to each input light, and temporarily stores the generated charge. The electric charges generated by the CCD image pickup device are sequentially transferred to the controller 21 while being converted into digital signals. The controller 21 stores the transferred digital signal in an image memory (VRAM) 21a while associating it with each address of the CCD image pickup device on the CCD image pickup plate 22b. That is, image data having the gradation of the digital signal is generated for each pixel corresponding to the CCD image sensor in the VRAM 21a. In the present embodiment, the digital signal represents the luminance of light input to the CCD image sensor. That is, image data having a luminance value for each pixel is stored in the VRAM 21a.

  The image data stored in the VRAM 21a as described above is output to the computer 30. Further, new image data is stored in the VRAM 21a every time the camera 22 performs imaging. The controller 21 outputs a drive signal to the XY stage 23, and the XY stage 23 is driven in the horizontal direction according to the drive signal. Of course, the mounting board 50 placed at a fixed position on the XY stage 23 also moves horizontally. In this way, the field of view of the camera 22 can be moved to any position on the mounting board 50 without moving the camera 22. Note that when the camera 22 captures an image, the XY stage 23 is stopped, so that a still image is captured.

  A ring light 24 is held above the XY stage 23 by a predetermined amount. The ring light 24 includes an LED as a light emitting element. The LED is formed in an annular shape, and the center position thereof is matched with the center position of the optical system 22a of the camera 22 when viewed from above. That is, the ring light 24 can irradiate the visual field of the camera 22 from the outside at a certain angle. Since the field of view of the camera 22 is sufficiently smaller than the size of the ring light 24 and the distance from the mounting substrate 50, errors in the incident angle and light quantity of the ring light 24 in the field of view can be ignored. By doing so, it is possible to generate two-dimensional image data obtained by capturing the reflection image of the mounting board 50 by the ring light 24 with the camera 22. The image data captured by the camera 22 is referred to as a captured image.

  FIG. 2 is a block diagram showing the internal configuration of the computer 30. In FIG. 1, a computer 30 includes a CPU 31, a RAM 32, a video memory (VRAM) 33, and a hard disk (HDD) 34 as main components. The CPU 31 is an arithmetic unit that executes processing based on an operating system (O / S) 34a, a pass / fail determination program 34b, and the like stored in the HDD 34, and uses the RAM 32 as a work area when executing the processing. The VRAM 33 is a memory specialized for storing image data, and stores a captured image 33a and an estimated image 33b. The captured image 33a is image data transferred from the VRAM 21a of the controller 21.

  As another configuration, the computer 30 includes a video interface (I / F) 35, an input interface (I / F) 37, and an I / O 39. A display 36 is connected to the video I / F 35, a keyboard 38 a and a mouse 38 b are connected to the input I / F 37, and the controller 21 of the imaging unit 20 is connected to the I / O 39. The I / O 39 outputs a signal for driving the XY stage 23 to the controller 21, a signal for executing imaging by the camera 22, and the like, and receives an input of a captured image from the controller 21. Yes.

  FIG. 3 shows the configuration of software implemented by the computer 30 and the data flow. In the CPU 31 and the RAM 32, an O / S 34a and a pass / fail determination program 34b are executed, and its module configuration is shown. The pass / fail determination program M includes, as main components, an imaging execution unit M1, a shape data acquisition unit M2, an estimation unit M3, a feature value calculation unit M4, a discriminant function calculation unit M5, and a pass / fail determination unit M6. Furthermore, the estimation unit M3 includes an incident angle acquisition unit M3a, a reflection behavior estimation unit M3b, a reaching light amount estimation unit M3c, and an estimated image generation unit M3d.

  The imaging execution unit M1 acquires the board data 34c stored in the HDD 34, and causes the imaging unit 20 to execute imaging based on the board data 34c. Specifically, the board data 34c stores information such as the size of the mounting board 50, the type, position, size, shape, and number of mounted components, and the XY stage 23 is determined based on these pieces of information. Drive. Therefore, the visual field of the camera 22 can be moved to a desired position on the mounting board 50, and pass / fail judgment can be performed on the desired position on the mounting board 50.

  The shape data acquisition unit M2 acquires the shape data 34d from the HDD 34. The shape data 34d is three-dimensional data representing the solder shape on the mounting substrate 50. For example, the shape of the solder is specified by a plurality of points represented by three components (X, Y, Z). Of course, a data format in which the three-dimensional shape of the solder is expressed by an approximate function or a vector may be used. The shape data acquisition unit M2 outputs a plurality of solder shape images based on the shape data 34d to the display 36. Then, the user selects one of the solder shapes displayed on the display 36 with the mouse 38b or the keyboard 38a. The shape data 34d corresponding to the selected solder shape is output to the reflection behavior estimation unit M3b of the estimation unit M3.

  The incident angle acquisition unit M3a of the estimation unit M3 inputs the size (diameter) of the ring light 24 and the distance between the ring light 24 and the mounting substrate 50 from the imaging unit 20, and the camera 22 uses the values based on these values. The angle at which the light emitted from the ring light 24 enters the visual field is calculated. The incident angle calculated by the incident angle acquisition unit M3a is output to the reflection behavior estimation unit M3b. The reflection behavior estimation unit M3b receives the shape data 34d and the incident angle, and estimates the reflection behavior when the light emitted from the ring light 24 enters the solder shape expressed by the shape data 34d based on these. Since the incident angle of the light emitted from the ring light 24 and the angle of the solder with which the light collides are known, the light reflection behavior can be estimated. Further, the reflection characteristics of the solder formed on the mounting board 50 are also acquired from the board data 34c, and the reflection characteristics are also taken into account when estimating the reflection behavior. The reflection behavior estimated by the reflection behavior estimation unit M3b is input to the reaching light amount estimation unit M3c.

  The arrival light amount estimation unit M3c receives the reflection behavior, and estimates the amount of light reaching each CCD image sensor of the CCD image pickup plate 22b based on the reflection behavior. Since the reflection behavior is estimated by the reflection behavior estimation unit M3b, the amount of light that the CCD image pickup plate 22b reaches each CCD image pickup device of the CCD image pickup plate 22b is estimated by tracing the reflected light to the CCD image pickup plate 22b. be able to. The light amount of each CCD image sensor estimated by the reaching light amount estimation unit M3c is input to the estimated image generation unit M3d, where an estimated image 33b as image data is generated. In other words, a luminance gradation corresponding to the amount of light of each CCD image sensor is calculated, and a two-dimensional estimated image 33b in which this gradation is arranged according to the position of each CCD image sensor is generated. The estimated image 33b is stored in the VRAM 33 in the same manner as the captured image 33a.

  The feature value calculation unit M4 receives the captured image 33a and the estimated image 33b from the VRAM 33, and calculates an actual feature value 34e1 and an estimated feature value 34e2 based on each of them. The calculated actual feature value 34e1 and estimated feature value 34e2 are stored in the HDD 34. The discriminant function calculation unit M5 acquires the actual feature value 34e1 and the estimated feature value 34e2 stored in the HDD 34, and performs a multivariate analysis on these to calculate the discriminant function 34f. The discriminant function 34f is stored in the HDD 34, and is used when the pass / fail judgment unit M6 performs pass / fail judgment. Specifically, the quality determination unit M6 determines whether the quality is good or not based on the value obtained by substituting the actual feature value 34e1 into the discriminant function 34f. The result of pass / fail judgment may be output to an output device such as the display 36 or a printer, or may be stored in the HDD 34. In the present embodiment, all processes are performed by a single computer 30. However, a plurality of computers may be communicably connected, and the processes may be distributed among the computers.

(2) Defect data accumulation processing:
FIG. 4 shows the flow of defective data storage processing executed in the present invention. In step S100, the shape data acquisition unit M2 outputs a plurality of solder shape images based on the shape data 34d to the display 36. Here, a procedure for creating defect data for determining a small solder as a defect will be described as an example. FIG. 5 shows an example of an image displayed on the display 36 in step S100. In the figure, based on the shape data 34d prepared in advance, six central cross-sectional shapes of the solder 52 when the chip component 51 is mounted on the mounting substrate 50 are shown as A to F. The shape data 34d is data representing a three-dimensional shape. Here, the shape data 34d is represented by a two-dimensional image for easy comparison of the shapes. In addition, the amount of solder decreases in order from A to F, and the height of the highest point of solder (hereinafter referred to as solder height h) tends to decrease in order.

In step S105, selection of a boundary shape is accepted. That is, the upper limit of the cross-sectional shapes A to F that the user wants to select as a small solder is selected. For example, the following description will be made assuming that the user selects the cross-sectional shape D. In this case, it is understood that the user wants to sort the cross-sectional shapes A, B, and C as good products and the cross-sectional shapes D, E, and F as defective products. In step S110, variation is set for the shape of the solder that becomes a defective product. According to the intention of the user, all of the solders whose height h is lower than the cross-sectional shape D are defective. Therefore, the variation is set for the solder height h lower than the solder height h ₁ of the cross-sectional shape D.

Specifically, as shown in FIG. 6, the solder height h ₁ to the solder height h = 0 are equally divided into ₁₀₀ , and the shape data 34d is acquired for each of the solder heights h ₁ to ₁₀₀ . In acquiring the shape data 34d corresponding to the respective solder heights h ₁ to ₁₀₀ , the shape data 34d may be prepared for all solder shapes in advance, or may be predicted from the shape data 34d of the cross-sectional shape D. . For example, new shape data 34d satisfying the solder height h _n is obtained by simply multiplying all coordinates (X, Y, Z) constituting the shape data 34d by (h _n / h ₁ ). be able to. Further, the present invention is not limited to setting 100 variations.

In step S115, 1 is selected as n. That is, the shape data 34d corresponding to the solder height h ₁ is selected, and the shape data 34d is acquired in step S120. When the shape data acquisition unit M2 acquires the shape data 34d, the pixel is defined for the shape data 34d in step S125. FIG. 7 virtually shows a state in which the solder 52 represented by the shape data 34d is viewed from above. In the figure, solder 52 is partitioned in a grid pattern by broken lines in the X and Y directions, and a region surrounded by the broken lines is generated as _a pixel Pa _{, b} . Note that a (1 to a _max ) indicates the position in the X direction, and b (1 to b _max ) indicates the position in the Y direction. In the figure, for the sake of simplification, the pixel Pa _{, b} is shown large, but actually, the pixel Pa _{, b} is very small.

In step S130, it calculates an average tilt angle theta _{a, b} of each pixel P _{a, b.} The lower part of FIG. 7 illustrates how the curved surface in the pixel Pa _{, b} is approximated by a plane, and the inclination angle of the plane is calculated as the average inclination angle θa _{, b} .

In step S135, the incident angle acquisition unit M3a calculates the incident angle. FIG. 8 illustrates how the incident angle acquisition unit M3a calculates the incident angle. In the figure, the incident angle acquisition unit M3a shows the diameter r of the ring light 24 and the distance d between the ring light 24 and the mounting substrate 50. The diameter r and the distance d are acquired from the imaging unit 20. Of course, the user may directly input the diameter r and the distance d using the keyboard 38a. An angle at which the light emitted from the ring light 24 enters the field of view S is an incident angle γ. From the relationship shown in the figure, the incident angle γ can be expressed as the following formula (1) by the diameter r and the distance d.
γ = arctan (d / r) (1)

  The field of view S is 2 cm × 2 cm, which is sufficiently smaller than the diameter r and the distance d, so that the incident angle γ can be considered to be uniform in the field of view S. In addition, the solder protrudes upward in the field of view S, but the height h of the solder is also sufficiently smaller than the diameter r and the distance d, so that the incident angle does not depend on the solder height h. γ can be considered uniform.

In step S140, the reflection behavior estimation unit M3b estimates the reflection behavior. FIG. 9 virtually shows how the light emitted from the ring light 24 enters the solder. Of course, the solder 52 and light shown in the figure are estimated by calculation, and light is not actually incident on the solder 52. In the figure, it is assumed that the CCD image pickup device of the CCD image pickup plate 22b picks up the solder 52 from above. On the other hand, the pixel P _a, the _b are incident incident light C _i emitted from the ring light 24, the incident light C _i it is found that reflective pixel P _a, the _b in the predetermined direction.

It is estimated that specular reflection light C _s and diffuse reflection light C _d are generated from the incident light C _{i at the} pixel Pa _{, b} . The specular reflected light C _s is reflected light emitted only in a direction symmetric with the incident light C _i . Then, 'When, arrives component C _s' components reaching the CCD imaging device directly out of the specular reflection light C _s reach component C _s quantity l _s of' can be represented by the following formula (2).
l _s ′ = l _i · (1−c) · cos ^m β (2)

In the above formula (2), β represents the angle formed between the specular reflected light C _s and the reaching component C _s ′, c (0 to 1) represents the roughness of the solder surface, and m represents the local specular reflection. L _i represents the amount of light per unit area of the incident light C _i (area of the pixels Pa _{, b} ). In the above formula (2), 'quantity l _s of' arrival component C _s is proportional to a power of the cosine of the specular reflection light C _s reach component C _s' and the angle beta. The degree of localization m is a coefficient representing the clarity of the specular reflection contour, and depends on the surface state of the solder. For example, in the case of eutectic solder, the contour of specular reflection is clear, and it can be said that the degree of localization m is high. On the other hand, in the case of lead-free solder, it can be said that the degree of localization m is low because the contour of specular reflection is blurred. In the reflection behavior estimation unit M3b, the roughness c and the localization degree m are acquired as the reflection characteristics of the solder formed on the mounting board 50 from the information on the solder material included in the board data 34c. since the light quantity l _i of the incident light C _i from the specification is known, it is possible to calculate the l _s' based on the equation (2).

Furthermore, geometrically, the following expressions (3) and (4) are established for the angle β formed between the specular reflection light C _s and the arrival component C _s ′.
α = π / 2−γ−θ _{a, b} (3)
β = θ _{a, b} −α = −π / 2 + γ + 2θ _{a, b} (4)
In the above formulas (3) and (4), α represents an angle formed by the normal of the solder surface and incident light C _i in the pixel Pa _{, b} . By substituting the above equation (4) into the above equation (2), the following equation (5) can be obtained.
l _s ′ = l _i · (1−c) · cos ^m (−π / 2 + γ + 2θ _{a, b} ) (5)
As described above, the light quantity l _s ′ of the component C _s ′ reaching the CCD image sensor derived from the specular reflection, the light quantity l _{i of the} incident light, the roughness c of the solder surface, the localization degree m of the specular reflection, and the incident angle γ It can be calculated from the average inclination angle θ _{a, b} .

On the other hand, the light quantity l _d of the diffuse reflected light C _d can be expressed as the following formula (6).
l _d = l _i · c · cos α (6)
That is, the light quantity l _d of the diffusely reflected light C _d is proportional to the cosine of the angle α formed by the normal of the solder surface in the pixel Pa _{, b} and the incident light C _i . Further, the diffuse reflected light C _d is a reflected light generated by the irregular reflection of the incident light C _{i on} the surface of the solder 52, and can be considered to be reflected with a uniform light amount in all directions. Accordingly, the light amount l _d ′ of the component C _d ′ reaching the CCD image sensor due to diffuse reflection is the same as the light amount l _d . Furthermore, the following formula (7) can be obtained by substituting the above formula (3) into the above formula (6).
l _d = l _i · c · cos (π / 2−γ−θ _{a, b} ) (7)

In the above equation (7), the light quantity l _d ′ of the component C _d ′ reaching the CCD image sensor derived from diffuse reflection is also equal to the light quantity l _{i of the} incident light, the roughness c of the solder surface, the incident angle γ, and the average tilt angle. It can be calculated from θ _{a, b} . Furthermore, the total amount of light reaching l _o reaching the CCD image sensor is calculated by superimposing the amount of light l _s ′ derived from specular reflection and the amount of light l _d ′ derived from diffuse reflection by the following equation (8). can do.
l _o = l _s '+ l _d ' (8)

Furthermore, the following formula (9) can be obtained by substituting the above formula (5) and the above formula (7) into the above formula (8).
l _o = l _i · (1−c) · sin ^m (γ + 2θ _{a, b} )
+ L _i · c · sin (γ + θ _{a, b} ) (9)

In the above formula (9), the first term is a specular reflection term, and the second term is a diffuse reflection term. According to the above formula (9), the amount of light reaching l _o reaching the CCD image sensor can be accurately calculated from the known parameters l _i , c, m, γ, θ _{a, b} . In step S145, the light amount l _o reaching the CCD image pickup plate 22b is calculated for each pixel Pa _{, b} based on the above equation (9). At step S150, the all pixel P _{a, b} can reach the light quantity l _o is calculated is checked for all the pixels P _a, if calculated for _b is completed step S155 is executed.

In step S155, the estimated image generation unit M3d generates an estimated image. That is, since the reaching light quantity l _o is calculated for all the pixels P _{a, b} , the addresses according to the positions a, b of the pixels P _{a, b} are given to them, so that the number of pixels is a _max × An estimated image serving as b _max pieces of image data can be generated. In addition, since the reaching light quantity l _o and the luminance value are in a proportional relationship, it is possible to generate an estimated image having luminance for each pixel Pa _{, b} .

By the way, when the camera 22 actually captures an image, a captured image 33a having the same number of pixels as the number of effective CCD image pickup elements of the CCD image pickup plate 22b is generated unless resolution conversion is performed halfway. In step S155, resolution conversion is performed so that the resolution on the mounting board 50 in the estimated image generated as described above matches the resolution on the mounting board 50 in the captured image 33a, and the converted estimated image 33b is converted into the VRAM 33. To remember. In this resolution conversion, methods such as thinning and interpolation can be applied. Of course, if the size of the pixel Pa _{, b} on the mounting board 50 is set in advance so that the resolution on the mounting board 50 in the estimated image matches the resolution on the mounting board 50 in the captured image 33a, There is no need to perform resolution conversion.

  In step S160, the feature value calculation unit M4 calculates an estimated feature value. As the feature value, an estimated image 33b stored in the VRAM 33 is input, and a large number of estimated feature values F are calculated from the estimated image 33b. The estimated feature value F is a large number of parameters calculated from the estimated image 33b, and it is desirable to prepare about several tens to several hundreds of types. In the present embodiment, 300 types of estimated feature values F are calculated. Specifically, as the estimated feature value F, an average luminance, a luminance range, a luminance edge number, a luminance dispersion value, or the like of the estimated image 33b can be applied. The calculated estimated feature value F is stored in HDD 34 as estimated feature value 34e2.

In step S165, it is confirmed whether or not n is 100. If it is not 100, (n + 1) is substituted for n, and step S120 and subsequent steps are repeated. That is, it is confirmed whether or not the estimated feature value F has been calculated for all the solder shapes (solder height h _n ) set in step S110, and if not all have been calculated, the next solder shape (solder height h _{n). The} process of calculating the estimated feature value F for ₊₁ ) is repeated. When the process is repeated until n reaches 100, the estimated feature value F is calculated for all solder shapes (solder height h _n ), and the process is terminated.

As a result, the estimated feature value F can be calculated and stored for each of 100 types of solder shapes to be selected as small solder defects. FIG. 10 shows a list of estimated feature values F calculated by the defective data accumulation process. In the figure, estimated feature values F _{1 to} F ₃₀₀ are calculated for each solder height h _n . Further, in the right column, the estimated feature values F _{1 to} F ₃₀₀ are identified so as to be understood as values indicating the characteristics of defective products.

(3) Good data storage processing:
Next, the good data accumulation process will be described. The good data accumulation process is executed separately from the defective data accumulation process. FIG. 11 shows the flow of good data storage processing. In the figure, in step S <b> 200, the camera 22 captures an image of the actually created mounting board 50. When imaging is performed, illumination is performed from the ring light 24 with a light amount l _i per unit area. In addition, since the mounting substrate 50 which is non-defective can be created by applying the manufacturing conditions when mass-producing the mounting substrate 50, the creation of the non-defective mounting substrate 50 is not troublesome.

  Further, since the non-defective product can be shipped as a product, the mounting substrate 50 itself is not wasted. As described above, by taking an image with the camera 22, a captured image 33 a as image data is generated, transferred to the VRAM 33, and stored. As shown in FIG. 12, the field of view S when the image is taken by the camera 22 includes other mounting parts and other solder, but is indicated by a solid line so as to have the same angle of view as the estimated image 33b. It is assumed that the masked area other than the periphery of the solder 52 is stored as the captured image 33a. In addition, when another inspection location exists in the visual field S, another captured image 33a masking other than the inspection location is also generated.

  In step S210, the feature value calculation unit M4 calculates the actual feature value G based on the captured image 33a. The actual feature value G is a parameter similar to the estimated feature value F, and is calculated by the same method. Further, since the resolution on the mounting substrate 50 is matched between the captured image 33a and the estimated image 33b as described above, the actual feature value G and the estimated feature value F can be calculated under the same conditions. In step S220, the actual feature value G is stored in the HDD 34. When step S220 is completed, the process returns to step S200 again, and processing for calculating the actual feature value G for the next good product is performed. Thereby, the actual feature value G can be accumulated for many good products. Then, when the actual feature value G is accumulated by a predetermined amount (for example, up to 100 non-defective products), the good data accumulation process is terminated.

FIG. 13 shows a list of actual feature values G calculated by the good data storage process. In the figure, actual feature values G _{1 to} G ₃₀₀ are calculated for each good product. Further, in the right column, the actual feature values G _{1 to} G ₃₀₀ are identified so as to be understood to be values indicating the features of the non-defective products.

(3) About discriminant function:
As described above, when the estimated feature values F _{1 to} F ₃₀₀ and the actual feature values G _{1 to} G ₃₀₀ can be calculated, the discriminant function can be calculated based on them. FIG. 14 conceptually shows the discriminant function. In the figure, the horizontal axis and the actual feature values G ₁ to estimate characteristic values F _1, the vertical axis is graph plotting the actual characteristic value G ₂ the estimated feature value F ₂ are shown, the actual feature values G ₁ , G ₂ and estimated feature values F ₁ and F ₂ are shown. Since the distribution of (G ₁ , G ₂ ) is obtained by imaging good products, it is shown as a set of good products. On the other hand, since the distribution of (F ₁ , F ₂ ) is estimated from the defective shape data 34d, it is shown as a set of defective products. (G ₁ , G ₂ ) is distributed in the range of variation in the manufacturing process, and (F ₁ , F ₂ ) is distributed in the range of variation set in step S110.

As described above, since the non-defective product and the defective product have some shape characteristics, the feature values F and G have different distribution tendencies. Therefore, the quality determination can be performed by analyzing the distribution tendency of the feature values F and G and setting an appropriate threshold value. However, if it is determined to pass or fail only with the frequency distribution of the single kind of feature values G ₁ and F ₁ , the non-defective product distribution and the defective product distribution may overlap as shown in the lower part of FIG. Similarly, when it is determined that the quality is determined only by the frequency distribution of the single type of characteristic values G ₂ and F ₂ , the non-defective product distribution and the defective product distribution overlap as shown in the left side of the figure. In such a state, it is impossible to set a threshold value that can distinguish good products from defective products without erroneous determination.

On the other hand, the discriminant function is defined by the following equation (10).
H _G = s ₁ G ₁ + s ₂ G ₂ + s ₃
H _F = s ₁ F ₁ + s ₂ F ₂ + s ₃ (10)
In the above equation (10), H _G and H _F indicate discriminant functions, and s ₁ , s ₂ , and s ₃ indicate coefficients. That is, in the above equation (10), the discriminant functions H _G and H _F are given by a linear combination of the actual feature values G ₁ and G ₂ and the estimated feature values F ₁ and F ₂ .

The coefficient s _1, s ₂ is the distribution of the discriminant function H _G of the good, and the distribution of the discriminant function H _F of defective products are set as much as possible away. According to the coefficient s _1, s _2, feature values _{(G 1, G 2),} (F 1, F 2) it is possible to provide an inclination to, discriminant function H discriminant function distribution and defective _G of good it is possible to set the inclination of separating the distribution of the H _F. Furthermore, the constant term s ₃ is set so that the center value of the distribution of the non-defective product discrimination function H _{G and} the distribution of the non-defective product discrimination function H _F becomes zero. Therefore, the discriminant functions H _G and H _F can be separated using 0 as a threshold value. Thus, if the discrimination function H _G positive obtained by substituting the actual characteristic values calculated from the image pickup image 33a obtained by capturing the solder 52 is good or bad do not know the (G _1, G _2), the solder 52 You can see that it is a good product.

In FIG. 14, two-dimensional discriminant functions H _G and H _F using two types of feature values G ₁ , G ₂ , F ₁ , and F ₂ as variables are illustrated for ease of explanation and illustration. Since 300 types of feature values F and G are calculated, several feature values F and G that can most isolate the quality distribution are selected from these, and the selected feature values F and G are selected. High-dimensional discriminant functions H _G and H _F are calculated by linearly combining. In calculating the discriminant functions H _G and H _F , a known multivariate analysis method can be applied.

(5) Pass / fail judgment processing:
FIG. 15 shows the flow of pass / fail judgment processing. In the figure, in step S300, the mounting substrate 50 to be inspected is imaged. Here, as in step S200 of the good data accumulation process, the captured image 33a masked except for the periphery of the solder 52 is stored in the VRAM 33. In step S310, the feature value calculation unit M4 calculates the actual feature value G as in step S210. In step S320, it reads out the discrimination function H _G from HDD 34, substituting the actual characteristic value G in the discrimination function H _G, to calculate the value of the discriminant function H _G.

Then, in step S330, and when the discrimination function H _G is positive, a solder 52 of the mounting board 50 is good in step S340. On the other hand, when the discrimination function H _G is less than or equal to zero, it is determined that the solder 52 is defective the mounting board 50 at step S350. According to the discrimination function H _G, because the a good distribution and defective distribution are separated across the 0, it is possible to perform accurate quality determination.

  As described above, in general, in pass / fail determination, pass / fail is determined using a statistical method like the discriminant function H. In that case, it is required to collect specimen data in advance for good or defective standards. In the statistical method, as the number of samples increases, the reliability of the estimation increases, and the accuracy of the quality determination performed based on the reliability increases. In the present embodiment, collecting as many feature values F and G as possible is a condition for improving the accuracy of the pass / fail determination. Since the feature value F of the defective product among the feature values F and G is calculated from the estimated image 33b generated by the estimation instead of the captured image 33a actually captured, a large number of defective products are created intentionally, There is no need to actually take an image.

  Therefore, the labor and cost for collecting the feature value F of defective products can be eliminated. Further, the shape data 34d representing the shape of the defective product can be created infinitely within the range of the defective product. Therefore, the feature value F calculated based on the shape data 34d can be prepared infinitely. Furthermore, since the shape data 34d representing the shape of the defective product can be created arbitrarily, even if it is a shape of a defective mode that occurs very rarely, the shape data 34d can be easily created and its feature value F Can be collected.

(6) Modification:
In the embodiment described above, the example of calculating the estimated feature value F of the defective product based on the shape data 34d representing the shape of the defective product has been exemplified. However, the estimated feature value F may be calculated for the non-defective product. For example, the estimated feature value F may be calculated based on the shape data 34d corresponding to the non-defective section shapes A, B, and C shown in FIG. In that case, it is identified in the right column shown in FIG. 10 that the estimated feature value F is non-defective, and when calculating the discriminant function, the estimated feature value F is referred to by referring to the right column shown in FIG. The value F may be attributed to the non-defective product distribution. In this way, it is possible to eliminate the need to actually perform imaging for non-defective products.

  Moreover, although what judged the quality of small solder as a failure mode was illustrated, naturally this invention is applicable also in the quality determination of another failure mode. For example, it is possible to make a pass / fail judgment by the same apparatus configuration for a defective mode such as a large solder, a missing part, a wet defect, or a solder flow. That is, by preparing the shape data 34d corresponding to each failure mode on the computer, it is possible to acquire a large number of estimated feature values F and perform a quality determination with high accuracy. Each failure mode such as solder size, missing item, wetting failure, solder flow, etc. has a unique shape feature, so that an estimation image reflecting the same shape feature can be estimated. Of course, the subject of inspection is not limited to solder, and the present invention can be applied as long as it determines the quality of the shape. For example, the present invention can also be applied to pass / fail judgments such as metal and plastic molded products and printed circuit board wiring.

  Furthermore, in the above-described embodiment, the camera 22 is used to image the reflected light and determine pass / fail, but the present invention can be applied to other imaging methods. That is, the present invention is not limited to the case where the inspection object is irradiated with visible light as an electromagnetic wave as in the above embodiment, but the inspection object is irradiated with X-rays, infrared light, or the like, and the state of the transmitted light is imaged. However, the present invention can be applied. According to X-rays, the transmittance with respect to a metal such as solder varies depending on the thickness. Therefore, an estimated image as a transmission image can be generated by preparing three-dimensional shape data in advance. In addition, since infrared light is transmitted through plastic, an estimated image as a transmission image can be generated in the same manner.

  That is, the present invention can be applied if the estimated image can be estimated by estimating the behavior when the electromagnetic wave passes through the inspection object. Therefore, it is possible to prepare in advance the estimated feature value of a good product or a defective product even in an imaging system that captures a refraction image, a scattered image, a diffraction image, or the like of an electromagnetic wave. In particular, when an estimated image of a transmission image is generated, the estimated image can be easily generated from three-dimensional shape data because the thickness of the inspection object corresponds to the light amount (luminance) in the estimated image. Furthermore, in the transmission image, an estimated image of the inspection object in which the void interposes can be estimated by designating the diameter and position of the void that can be generated inside the inspection object. Specifically, an estimated image having a feature that a light amount (luminance) is locally large is created for a pixel group through which electromagnetic waves penetrate a void. Then, if the captured image when the inspection object is actually imaged is similar to the estimated image having the above characteristics and the surface shape is flat, it is specified that the inspection object has a void. be able to.

  Furthermore, in the above-described embodiment, an example in which an estimated image is generated for a single ring light is illustrated. However, for example, a plurality of estimated images may be generated assuming two or three ring lights. In other words, estimated images for a plurality of incident angles may be generated, and these estimated images may be compared with captured images actually captured at the respective incident angles. By doing in this way, the comparison result about several sets of presumed images and picked-up images can be obtained, and quality can be determined comprehensively from these comparison results. Therefore, even under conditions where the reliability of the determination result is low at a certain incident angle, it is possible to perform the pass / fail determination while considering the determination result of other incident angles with high reliability. Can be realized.

  Moreover, although what performed the quality determination using a discriminant function in the said embodiment was illustrated, it does not necessarily need to use a discriminant function. That is, if the actual feature value obtained by actually imaging the inspection object is a value far from the estimated feature value of the defective product, it can be determined that the inspection object is a non-defective product, If the value is close to the estimated feature value of the non-defective product, it can be determined that the inspection object is a defective product. That is, if an estimated feature value of a non-defective product or a defective product has been calculated in advance, pass / fail judgment is performed by comparing the estimated feature value with an actual feature value actually obtained by imaging the inspection object. Can do.

  Moreover, in the said embodiment, although the dispersion | variation which sets the solder height h uniformly from the upper limit shape D of a solder small defect was set, you may set dispersion | variation in another aspect. That is, when the distribution tendency of the solder height h in mass production can be predicted, weighting may be performed according to the distribution tendency. For example, when the distribution frequency of the solder height h in the vicinity of the upper limit shape D is larger than the other solder heights h, the deviation of the solder height h is fine in the vicinity of the solder height h in the vicinity of the upper limit shape D. A larger number of estimated feature values than other solder heights h may be calculated. Thereby, the estimated feature value can be weighted. Further, the range of variation may be set based on the process capability of the previous process.

  Furthermore, various techniques for obtaining the shape data 34d can also be applied. In the embodiment described above, the user specifies the shape data 34d that is visually defective. As another method, for example, in the case of a solder shape, the shape data 34d may be created from physical characteristics at the time of melting the solder. In this case, the three-dimensional shape data 34d can be predicted by inputting conditions such as the wettability of the mounted component or pad, the surface tension or viscosity of the molten solder, the melting temperature profile, and the amount of solder.

  FIGS. 16 and 17 illustrate how the three-dimensional shape data 34d is created. FIG. 16 shows a list of parameters that affect the solder shape. Specifically, the amount of solder, the shape of the mounted component, the shape of the pad, the specific gravity of the solder, the solder contact angle, the gravity, the surface tension of the molten solder, the position of the mounted component with respect to the pad, etc. are listed as parameters that affect the solder shape. . Then, by inputting specific numerical values of these parameters into the fluid simulation program and executing the fluid simulation, the three-dimensional shape data 34d can be created.

  FIG. 17 illustrates some of the three-dimensional shape data 34d created by the fluid simulation program. In the fluid simulation, the shape of the molten solder surface is calculated along a time series in consideration of the above parameters. Further, during the calculation, polygons (triangular unit figures shown in the figure) may be appropriately divided or corrected. For example, when the amount of solder input as a parameter is reduced, the “small solder” shape data 34d shown in the upper left of the figure can be created. On the contrary, when the amount of solder input as a parameter is increased, the shape data 34d of “solder large” shown in the upper right of the figure can be created.

  Furthermore, when the contact angle input as a parameter is increased, the “wetting defect” shape data 34d shown in the lower left of the figure can be created. Further, when the position of the mounted component with respect to the pad is largely shifted, the “out of stock” shape data 34d shown in the lower right of the figure can be created. As described above, by acquiring the shape data 34d using the fluid simulation, it is possible to acquire a large amount of the shape data 34d without creating any defective sample or the like. Note that a melting temperature profile, a reflow conveyance direction, a solder component, a flux component, and the like may be specified as parameters specified for the fluid simulation program.

  Further, in the case of using the fluid simulation, it is possible to obtain a large number of shape data 34d in consideration of the variation in the solder formation conditions by setting the variation for the above-described parameters. For example, if the variation in the amount of solder printed on the pad is known in advance, a large number of shape data 34d having variations in the solder height h can be obtained by setting the variation in the amount of solder. . Therefore, it is possible to obtain the variation of the shape data 34d in accordance with the physical laws and the process capability, rather than setting the uniform variation as in the above embodiment.

  The shape data 34d created by the fluid simulation program is acquired in step S120 shown in FIG. 4, and the subsequent processing is executed as in the previous embodiment. That is, when the shape data acquisition unit M2 and the estimation unit M3 perform processing on the shape data 34d created by the fluid simulation program, an estimated image 33b corresponding to the shape data 34d can be obtained. Therefore, the shape data 34d created by the fluid simulation program needs to be in a format that can be handled by the shape data acquisition unit M2 and the estimation unit M3. By the way, various commercially available software for creating CG can be applied as the fluid simulation program. On the other hand, as the shape data acquisition unit M2 and the estimation unit M3 that process the shape data 34d, general ray tracing (ray tracing) software can be applied. Estimated images can be created by setting solder surface information.

  In this way, when each software is individually provided and the shape data 34d in each software is not compatible, the shape data 34d created by the CG creation software can be handled by the ray tracing software. You may make it provide the cooperation software which adjusts to a format or implement | achieves exchange of the data, command, etc. between both software.

1 is an overall view of a quality determination apparatus according to the present invention. It is an internal block diagram of a computer. It is a software block diagram for performing quality determination. It is a flowchart of defective data accumulation processing. It is a display screen of a display. It is a figure explaining dispersion | variation setting. It is a figure which shows the pixel in shape data. It is a figure explaining calculation of incident angle γ. It is a figure explaining reflection behavior. It is a table in which estimated feature values are accumulated. It is a flowchart of a good data storage process. 3 is a plan view showing a field of view S. FIG. It is a table in which actual feature values are accumulated. It is a graph explaining a discriminant function. It is a flowchart of a quality determination process. It is a list of parameters used for creating three-dimensional data. It is explanatory drawing of three-dimensional data.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 ... Shape quality determination apparatus 20 ... Imaging unit 21 ... Controller 21a ... Image memory (VRAM)
22 ... Camera 22a ... Optical system 22b ... CCD imaging plate 23 ... XY stage 24 ... Ring light 30 ... Computer 31 ... CPU
32 ... RAM
33 ... Video memory (VRAM)
33a ... Captured image 33b ... Estimated image 34 ... Hard disk (HDD)
34b, M ... pass / fail judgment program 34c ... board data 34d ... shape data 34e1, G ... actual feature value 34e2, F ... estimated feature value 34f, H ... discriminant function 36 ... display 38a ... keyboard 38b ... mouse 50 ... mounting board 51 ... Chip part 52 ... Solder A to F ... Cross-sectional shape M1 ... Imaging execution part M2 ... Shape data acquisition part M3 ... Estimation part M3a ... Incident angle acquisition part M3b ... Reflection behavior estimation part M3c ... Amount of arrival light quantity estimation part M3d ... Estimated image generation part M4: Feature value calculation unit M5 ... Discrimination function calculation unit M6 ... Pass / fail determination unit P ... Pixel S ... Field of view γ ... Incident angle θ ... Average tilt angle

Claims (7)

Imaging means for imaging an inspection object and generating a captured image;
Actual feature value calculating means for calculating the feature value of the captured image as an actual feature value;
Shape data acquisition means for acquiring three-dimensional data representing the shape of a non-defective product or a defective product;
Estimating means for generating an estimated image obtained by estimating the captured image when the good product or defective product is imaged based on the three-dimensional data;
Estimated feature value calculating means for calculating the estimated feature value of the estimated image as the estimated feature value;
A shape pass / fail judgment device comprising: pass / fail judgment means for judging pass / fail of the inspection object by comparing the actual feature value and the estimated feature value. The imaging means generates the captured image by acquiring the state of the electromagnetic wave incident via the inspection object for each position, and
The estimation means is
Incident angle acquisition means for acquiring the incident angle of the electromagnetic wave incident on the inspection object when the imaging means images the inspection object;
Based on the three-dimensional data and the incident angle, behavior estimation means for estimating behavior when the electromagnetic wave reaches the non-defective product or the defective product,
Arrival state estimation means for estimating the state of the electromagnetic wave reaching the imaging means based on the behavior;
The shape quality determination device according to claim 1, further comprising estimated image generation means for generating the estimated image by arranging values indicating the state of the electromagnetic wave reaching the imaging means for each position. .   3. The shape pass / fail determination apparatus according to claim 2, wherein the behavior estimation means estimates a reflection behavior when the electromagnetic wave reaches the non-defective product or the defective product.   4. The shape pass / fail judgment apparatus according to claim 3, wherein specular reflection and diffuse reflection are taken into account when the behavior estimation means estimates the reflection behavior.   5. The shape pass / fail judgment device according to claim 1, wherein the shape data acquisition unit acquires a plurality of three-dimensional data having variations in the shape of the non-defective product or the defective product. The actual feature value calculating means and the estimated feature value calculating means calculate various actual feature values and estimated feature values from the captured image and the estimated image,
6. The quality determination unit according to claim 1, wherein the quality determination unit makes a quality determination by comparing a discrimination value obtained by linearly combining the actual feature value and the estimated feature value. The shape pass / fail judgment device described. An imaging process for imaging an inspection object and generating a captured image;
An actual feature value calculating step of calculating the feature value of the captured image as an actual feature value;
A shape data acquisition step for acquiring three-dimensional data representing the shape of a non-defective product or a defective product;
An estimation step for generating an estimated image obtained by estimating the captured image when the non-defective product or the defective product is imaged based on the three-dimensional data;
An estimated feature value calculating step of calculating the estimated feature value of the captured image as the estimated feature value;
A shape pass / fail determination method comprising: a pass / fail determination step for determining pass / fail of the inspection object by comparing the actual feature value and the estimated feature value.