JP5464986B2 - Pass / fail judgment device, pass / fail judgment method, and pass / fail judgment program - Google Patents

Description

  The present invention relates to a quality determination device, a quality determination method, and a quality determination program for determining quality of a product.

  2. Description of the Related Art Conventionally, a quality determination device that uses multivariate linear discriminant analysis (hereinafter referred to as linear discriminant analysis) is known as a quality determination device that determines the quality of a product (see, for example, Patent Document 1 and Patent Document 2). . In such a device, multiple parameters are acquired from a large number of samples with known pass / fail attributes and used as variables (explanatory variables) for linear discriminant analysis, and the frequency distribution of pass / fail categories is separated from these multiple parameters. A linear discriminant function (hereinafter referred to as a discriminant function) that gives a variable to be obtained is obtained, and pass / fail is determined by comparing the calculation result of the discriminant function obtained by substituting a plurality of parameters measured from the pass / fail judgment target with a threshold value. judge.

  As another quality determination device, a quality determination device using multivariate distance discriminant analysis (hereinafter referred to as distance discriminant analysis) is known (for example, see Patent Document 3). In such an apparatus, a plurality of parameters are obtained from a large number of samples with known pass / fail attributes and used as variables (explanatory variables) for distance discriminant analysis, and approximation of distances between clusters in a parameter space in which these multiple parameters are projected. Based on the Euclidean distance between the center position of each cluster and the position of the element corresponding to multiple parameters measured from the pass / fail judgment target, which element corresponds to the pass / fail judgment target Pass / fail is determined by determining whether it belongs to a cluster.

Japanese Patent Laid-Open No. 8-254501 JP 2004-085216 A JP 2006-226708 A

However, in the pass / fail judgment device based on the linear discriminant analysis as described in Patent Documents 1 and 2, the pass / fail judgment with excellent discrimination accuracy can be realized by using the discriminant function that best separates the pass / fail category. On the other hand, in order to obtain a discriminant function having a high degree of separation, it is necessary to collect a large amount of sample data consisting of parameters representing the characteristics of each category. In addition, since the discriminant function is calculated based on the collected sample data, for example, a defective product with a large number of samples is more reliable than a defective product with different causes such as floating, missing products, and non-wetting. However, a defective product with a small number of samples causing a defective cause may be erroneously determined as a non-defective product. There was a problem of requiring sample data.
Furthermore, in the case of linear discriminant analysis, since the two categories are divided into two by the boundary line (plane) serving as a threshold, the discriminant result corresponds to an outlier that is far from the average value of the frequency distribution of the category. However, there is a problem that pass / fail is determined depending on which side of the threshold value belongs. Therefore, there is a problem that it is difficult to accurately determine an unknown defect having a low correlation with a parameter selected for obtaining a discriminant function.

  On the other hand, in the pass / fail determination device by the distance discriminant analysis as in Patent Document 3, since pass / fail can be determined based on the Euclidean distance from the center of the cluster, even when there are few defective products such as at the start of inspection, There is the convenience that the thing similar to the non-defective product can be identified using only the non-defective sample data. However, in order to prevent an increase in the so-called outflow rate, which may result in incorrect determination of a non-defective product as a non-defective product in some cases, the pass / fail categories may not be sufficiently separated. Necessary. For this reason, there is a problem that a so-called oversight rate in which a non-defective product is erroneously determined as a defective product becomes high due to a high safety factor threshold.

The present invention is intended to solve the above problem, and by reflecting the determination result of the highly reliable linear discriminant analysis in the distance discriminant analysis, the convergence of the threshold determination of the distance discriminant analysis is improved, and the distance discriminant analysis is performed. The pass / fail judgment device and the pass / fail judgment method that can improve the simple judgment performance of, and can share the linear discriminant analysis and the distance discriminant analysis, and make use of the characteristics of each other's good / bad category discriminator and realize excellent judgment accuracy And it aims at providing a pass / fail judgment program.

The present invention is as follows.
1. Parameter acquisition means for acquiring a plurality of parameters that are pass / fail judgment factors from the inspection target;
Obtain a discriminant function consisting of a plurality of parameters selected to give a variable that separates the frequency distribution of the good category and the bad category from the plurality of parameters corresponding to a plurality of samples with known pass / fail attributes. The linear discrimination threshold of the discriminant function as a judgment criterion set based on this and the calculation result obtained by substituting the parameter corresponding to the test object into the discriminant function, the pass / fail category of the test object Linear discriminant analysis means for discriminating
In the parameter space in which the plurality of parameters are projected, a distance determination threshold as a determination reference set as a distance to a center position of a good cluster configured by elements corresponding to a plurality of samples having good attributes, and the center position A distance discriminating and analyzing means for comparing the distance of the element corresponding to the inspection object with respect to and determining the pass / fail category of the inspection object;
With respect to the inspection object discriminated as a good category by the linear discriminant analysis means, a good / bad category is discriminated by the distance discriminating / analyzing means, and a good category is discriminated by both the linear discriminant analysis means and the distance discriminant analysis means. A quality determination device comprising: quality determination means for determining quality only when the
2. Above 1. In visual identification means for visually determining the quality of the inspection target,
When the visual discrimination means discriminates the pass / fail category for the inspection object determined by the distance discrimination analysis means as the negative category, and when the distance discrimination analysis means and the visual discrimination means determine the negative category Preliminary pass / fail judgment means for judging only whether or not
Sample data storage means for storing the plurality of parameters corresponding to the inspection object determined by the preliminary quality determination means and the quality determination results as sample data,
The pass / fail judgment means
The linear discriminant analysis means discriminates a pass / fail category by the linear discriminant threshold set based on the sample data,
The distance discriminating / analyzing means discriminates a pass / fail category by the distance discriminating threshold set based on data determined to be a good category among the sample data, and the pass / fail of the inspection object is determined based on the discrimination results. It is characterized by determining.
3. A parameter acquisition step for acquiring a plurality of parameters that are pass / fail judgment factors from the inspection target;
Obtain a discriminant function consisting of a plurality of parameters selected to give a variable that separates the frequency distribution of the good category and the bad category from the plurality of parameters corresponding to a plurality of samples with known pass / fail attributes. The linear discrimination threshold of the discriminant function as a judgment criterion set based on this and the calculation result obtained by substituting the parameter corresponding to the test object into the discriminant function, the pass / fail category of the test object A linear discriminant analysis process for discriminating
In the parameter space in which the plurality of parameters are projected, a distance determination threshold as a determination reference set as a distance to a center position of a good cluster configured by elements corresponding to a plurality of samples having good attributes, and the center position A distance discriminating and analyzing step of comparing the distance of the element corresponding to the inspection object with respect to and determining the pass / fail category of the inspection object;
For the inspection object discriminated as a good category by the linear discriminant analysis step, a pass / fail category is discriminated by the distance discriminant analysis step, and it is discriminated as a good category in both the linear discriminant analysis step and the distance discriminant analysis step. A quality determination method comprising: a quality determination step that determines quality only when the
4). 3. above. In the visual discrimination step of discriminating the quality category of the inspection target by visual inspection,
When the visual discrimination step determines the pass / fail category for the inspection object determined by the distance discriminant analysis step, and when it is determined as the negative category in both the distance discriminant analysis step and the visual discrimination step Preliminary pass / fail judgment process for judging only
A sample data accumulating step for accumulating the plurality of parameters corresponding to the inspection target determined by the preliminary pass / fail determination step and the pass / fail determination result as sample data;
The pass / fail judgment step includes
The linear discriminant analysis step discriminates pass / fail categories by the linear discriminant threshold set based on the sample data,
The distance discrimination analysis step discriminates a pass / fail category by the distance discrimination threshold set based on data determined to be a good category among the sample data, and the pass / fail of the inspection object is determined based on the discrimination results It is characterized by determining.
5. A parameter acquisition function for acquiring a plurality of parameters that are pass / fail judgment factors from the inspection target;
Obtain a discriminant function consisting of a plurality of parameters selected to give a variable that separates the frequency distribution of the good category and the bad category from the plurality of parameters corresponding to a plurality of samples with known pass / fail attributes. The linear discrimination threshold of the discriminant function as a judgment criterion set based on this and the calculation result obtained by substituting the parameter corresponding to the test object into the discriminant function, the pass / fail category of the test object Linear discriminant analysis function to discriminate
In the parameter space in which the plurality of parameters are projected, a distance determination threshold as a determination reference set as a distance to a center position of a good cluster configured by elements corresponding to a plurality of samples having good attributes, and the center position A distance discriminant analysis function that compares the distance of the element corresponding to the inspection object with respect to and determines the pass / fail category of the inspection object;
For the inspection object discriminated as a good category by the linear discriminant analysis function, a pass / fail category is discriminated by the distance discriminant analysis function, and it is discriminated as a good category by both the linear discriminant analysis function and the distance discriminant analysis function. A non-defective / non-defective program for causing a computer to realize a non-defective / unsatisfactory function only when
6). 5. above. In the visual discrimination function for visually determining the quality of the inspection target,
When the visual discrimination function determines the pass / fail category for the inspection object determined by the distance discriminant analysis function as a negative category, and the distance discriminant analysis function and the visual discrimination function determine the pass category. Preliminary pass / fail judgment function for judging only
A sample data accumulation function for accumulating the plurality of parameters corresponding to the inspection object determined by the preliminary quality determination function and the quality determination result as sample data;
The pass / fail judgment function is
The linear discriminant analysis function discriminates a pass / fail category by the linear discriminant threshold set based on the sample data,
The distance discriminant analysis function discriminates a pass / fail category by the distance discriminating threshold set based on the data determined to be a good category among the sample data, and the pass / fail of the inspection object is determined based on these discrimination results It is characterized by determining.

  According to the pass / fail determination apparatus of the present invention, the pass / fail determination means discriminates the pass / fail category by the distance discriminant analysis means for the inspection object determined by the linear discriminant analysis means, and the linear discriminant analysis means and the distance discriminant analysis. A good judgment is made only when a good category is determined in any of the means. By sharing the linear discriminant analysis and the distance discriminant analysis in this way, even if the good / bad categories are not sufficiently separated in the distance discriminant analysis, it is possible to eliminate specific negative categories in advance by the linear discriminant analysis. As a result, it is not necessary to set a distance discrimination threshold value with a high safety factor, and the convergence of the threshold value for distance discrimination analysis can be improved. Thereby, the quality determination apparatus excellent in determination accuracy is realizable.

  Further, the visual discrimination means for visually determining the pass / fail category of the inspection object, and the visual discrimination means for determining the pass / fail category for the inspection object determined by the distance discrimination / analysis means, and the distance discrimination analysis A preliminary pass / fail judgment means that judges whether or not only in both cases and the visual discrimination means are judged as a negative category, the plurality of parameters corresponding to the inspection object judged pass / fail by the preliminary pass / fail judgment means, and Sample data accumulating means for accumulating the pass / fail judgment result as sample data, wherein the pass / fail judgment means uses the linear discrimination threshold set based on the sample data by the linear discriminant analysis means. And the distance discriminating / analyzing means is set based on the sample data. When determining the pass / fail category based on the distance determination threshold and determining the pass / fail of the inspection target based on the determination results, the pass / fail determination can be executed even in a state where the number of sample data at the initial stage of operation is small. At the same time, it is possible to execute the pass / fail determination for actual operation based on a sufficient number of sample data acquired by the pass / fail determination by the preliminary pass / fail determination means. That is, in a normal production process, the yield is necessarily high, and it is difficult to obtain a defective sample. On the other hand, since the non-defective product sample can be easily acquired, if the pass / fail determination is executed based on these, the pass / fail determination device can be installed from a relatively early stage without requiring a lot of time to acquire the defective product sample. It can be operated. Further, for the inspection object determined to be the negative category by the distance discriminant analysis, a visual inspection is executed to determine true quality. As a result, it is possible to reliably prevent a non-defective product from being erroneously determined as a defective product.

Although the case where the present invention is realized as an apparatus has been described above, the present invention can be realized as a method and program for realizing the apparatus and a medium recording the program. In addition, the above-described pass / fail judgment device may be realized alone, applied to a certain method, or used in a state where the method is incorporated in another device. The present invention is not limited to a pass / fail determination device, a pass / fail determination method, and a pass / fail determination program, and includes various aspects. Therefore, the idea of the present invention can be changed as appropriate, such as a program, software, or hardware.
In the case of software for controlling the above apparatus as an embodiment of the idea of the invention, it exists and is used also on such a program, software, or a recording medium on which the software is recorded.

Further, the program and software recording media may be magnetic recording media, magneto-optical recording media, or any recording media that will be developed in the future. The same is true for the replication stage of primary replicas and secondary replicas. In addition, even when the communication apparatus is used as the supply device, the present invention is not used. Further, even when a part is software and a part is realized by hardware, the idea of the invention is not completely different, and a part is stored on a recording medium and is appropriately changed as necessary. It may be in a form that is read.

It is a general view of the quality determination apparatus concerning this embodiment. It is an internal block diagram of the computer concerning this embodiment. It is a software block diagram for performing quality determination according to the present embodiment. It is a figure which shows the example of a sample data table. It is a figure which shows the flowchart of the sample data registration process concerning this embodiment. 4 is a plan view showing an example of a field of view S. FIG. 5 is a plan view showing an example of an evaluation region E. FIG. It is a figure which shows the flowchart of the distance discrimination | determination analysis determination reference | standard preparation process concerning this embodiment. It is a figure which shows the example of the graph which shows parameter space. It is a figure which shows the example of the frequency distribution diagram with respect to distance. It is a figure which shows the flowchart of the preliminary | backup quality determination process concerning this embodiment. It is a figure which shows the flowchart of the linear discriminant-analysis determination criteria preparation process concerning this embodiment. It is a figure which shows the example of the frequency distribution figure with respect to the variable of a discriminant function. It is a figure which shows the flowchart of the quality determination process at the time of the actual operation concerning this embodiment.

Here, embodiments of the present invention will be described in the following order.
(1) Configuration of pass / fail judgment device:
(2) Sample data registration process:
(3) Pass / fail judgment processing:
(3-1) Preliminary pass / fail judgment processing:
(3-2) Pass / fail judgment processing during actual operation:

(1) Configuration of pass / fail judgment device:
FIG. 1 shows the configuration of a quality determination apparatus according to the present invention. In the same figure, the pass / fail judgment device 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. As a result, two-dimensional image data obtained by capturing the reflection image of the mounting board 50 by the ring light 24 with the camera 22 can be generated. As shown in the figure, the ring light 24 is not limited to a single stage, and a plurality of ring lights having different diameters may be arranged above and below to generate image data having different illumination angles. 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 (OS) 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. 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 unit 36 such as a display is connected to the video I / F 35, an input unit 38 such as a keyboard and a mouse is 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. An OS 34a and a pass / fail judgment program 34b are executed by the CPU 31 and the RAM 32, and their module configurations are shown. The pass / fail judgment program MP includes an imaging data input unit SP1, a distance discriminant analysis unit SP2, a linear discriminant analysis unit SP3, a visual data input unit SP4, and a display control unit SP5.

  The imaging data input unit SP1 includes an imaging execution unit SP11, a captured image acquisition unit SP12, and a parameter calculation unit SP13. The imaging execution unit SP11 acquires the board data 34c stored in the HDD 34, and causes the imaging unit 20 to perform imaging based on the board data 34c. More specifically, the board data 34c stores information such as the size of the mounting board 50 and the 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 substrate 50, and pass / fail determination can be performed for the desired position on the mounting substrate 50.

The captured image acquisition unit SP12 inputs a captured image from the controller 21 and stores the captured image 33a in the VRAM 33. The parameter calculation unit SP13 receives the captured image 33a and calculates a plurality of parameters from the captured image 33a. Here, the parameter is a characteristic value obtained from the pixel area where the solder is imaged in the captured image 33a, and is a value such as an average brightness, a maximum brightness, a minimum brightness, or the number of edges. Of course, it is not limited to these. In the present embodiment, h (h is a positive integer) types of parameters are prepared. Note that the value of each parameter is expressed as a parameter value p _k (k is an integer from 0 to h).

Further, the parameter calculation unit SP13 extracts the pixel data for the pixel region corresponding to the solder to obtain the position information indicating the position of the solder in the field of view of the camera 22 from the board data 34c, and the parameter value p _k can be calculated. Parameter value _{p k} parameter calculation unit SP13 has been calculated is stored in the sample data table 34d assigned to the HDD 34. FIG. 4 shows a sample data table 34d, and good / bad is identified for each sample data. This pass / fail attribute is input by the user via the input unit 38. In actual mass-produced products, since the yield is high, the number of defective products is often significantly smaller than the number of non-defective products. Therefore, when such an actual mass-produced product is used as a sample, the number of negative attribute samples is significantly smaller than that of a good attribute sample.

The distance discrimination analysis unit SP2 includes a center value calculation unit SP21, a distance calculation unit SP22, a distance discrimination threshold setting unit SP23, and a distance discrimination pass / fail judgment unit SP24. Center value calculating section SP21 obtains the parameter value p _k which corresponds from the sample data table 34d in the element of the group of good quality attributes, to calculate the central value of the parameter value p _k. Then, the distance calculation unit SP22 is calculated and the center value, and the parameter value p _k which corresponds to the parameter value p _k and the inspected sample data having a good attributes used for calculating the center value, the Euclidean distance To do.

Distance determination threshold setting unit SP23 has a central value of the above, and the parameter value p _k of sample data with good attributes used to calculate the central value, the Euclidean distance, and the frequency average and the standard deviation for this distance Based on these, a value that gives a predetermined oversight rate is set as the distance determination threshold.
The distance determination pass / fail determination unit SP24 compares the distance calculated by the distance calculation unit SP22 and the distance determination threshold set by the distance determination threshold setting unit SP23 to determine the pass / fail category of the inspection target in the distance discrimination analysis.

The linear discriminant analyzer SP3 includes a discriminant function calculator SP31, a linear discriminant threshold calculator SP32, and a linear discriminant pass / fail determiner SP33. The discriminant function calculation unit SP31 performs a process of calculating a discriminant function that gives a certain variable, and when the frequency distribution of the good category and the bad category is created for the variable, the two distributions are well separated. This variable is a function of the parameter value p _k. In the present embodiment, discriminant function, and the parameter value p _k stored in the sample data table 34d and their quality determination result is calculated using. Further, the parameter value p _k in this embodiment is present h type, determined in this embodiment by selecting an effective variable in separating the distribution of the quality category in a predetermined number of parameter values p _k Specifies the function.

  The linear discrimination threshold calculation unit SP32 calculates a frequency average and a standard deviation for each frequency distribution of the good category and the bad category for the variable Z corresponding to each sample data given by the discrimination function calculated by the discrimination function calculation unit SP31. Then, based on these, a linear discrimination threshold value that satisfies a predetermined outflow rate and oversight rate is calculated.

  The linear discriminant pass / fail judgment unit SP33 compares the variable Z corresponding to the test target given by the discriminant function with the linear discriminant threshold calculated by the linear discriminant threshold calculation unit SP32 to determine the pass / fail category of the test target in the linear discriminant analysis. Is determined.

Visual data input unit SP4, and accepts the input of the sample data registration processing quality determination result is associated to the parameter value p _k during the later, the inspected result of the distance discriminant analysis in the preliminary quality judgment processing is negative The input of the true determination result when the visual re-inspection is performed is accepted. It also accepts input of numerical values for setting a distance discrimination threshold and a linear discrimination threshold, which will be described later.
The display control unit SP5 outputs the captured image 33a and various data generated in the course of the pass / fail determination process to the display unit 36, thereby displaying those images on the display.

(2) Sample data registration process:
In the sample data registration process according to the present embodiment, a plurality of parameters for creating a determination criterion used in a quality determination process to be described later are calculated from a sample with a known quality attribute prepared in advance, and the plurality of parameters and the quality attribute are determined. Are registered as sample data.
FIG. 5 shows the flow of the sample data registration process. In the figure, in step S100, the mounting substrate 50 prepared as a sample is imaged. For the mounting board 50 prepared as a sample, the quality of the solder is determined in advance by other methods such as visual inspection, and the quality of the solder formed on the mounting board 50 is known.

  In step S110, a captured image 33a is generated, and the captured image acquisition unit SP12 stores the captured image 33a in the VRAM 33. FIG. 6 shows the field of view S of the camera 22. In the figure, the visual field S imaged by the camera 22 includes a plurality of mounting components and solder on the mounting substrate 50. In the present embodiment, the quality determination for the shape of the solder 52 of the chip component 51 indicated by a solid line will be described. In step S110, a captured image 33a is generated as image data representing the entire visual field S.

  In step S120, only pixel data including the solder 52 is extracted from the captured image 33a. That is, only pixel data inside the evaluation area E in which the periphery of the solder 52 is surrounded by a solid line in FIG. 6 is extracted. In step S <b> 120, the CPU 31 specifies where in the field of view S the solder 52 subject to pass / fail determination is arranged based on the board data 34 c stored in the HDD 34. Then, an evaluation area E having a size registered in advance is set in the pixel in which the solder 52 is imaged, and pixel data belonging to the evaluation area E is extracted.

  FIG. 7 schematically shows the state of the evaluation region E. In the figure, the evaluation area E is composed of a large number of pixels Pa and b divided by broken lines. Note that the luminance values Ba, b of the pixels Pa, b mean the average luminance of the area where each CCD image sensor on the mounting substrate 50 is in charge of image input. Note that the evaluation region E is not limited to the illustrated shape, and the shape and number are determined according to the parameters to be calculated.

In step S130, a parameter value _pk is calculated for the pixel data extracted in step S120. For example, in the case of calculating the average luminance as one of the parameter values p _k is calculated divided by the total number of pixels the luminance value Ba, the integrated value of b of all the pixels included in the evaluation region E as a parameter value p _k Will be. Similarly, the luminance value Ba of all the pixels included in the evaluation area E, by the parameter calculation unit SP13 performs a predetermined operation on b, to h kind calculated in all other types of parameter values p _k. Calculated parameter values p _k is stored in the sample data table 34d as described above with a unique serial number given for soldering corresponding to the evaluation region E.

In step S140, and it registers the quality attributes in association with the parameter value p _k stored in the sample data table 34d. The user acquires the quality of the solder on the mounting substrate 50 in advance by visual inspection or the like, and inputs this information via the input unit 38. Then, information is added to the column of “good / bad attribute” in the sample data table 34d. Thereby, the parameter value _pk and the known pass / fail judgment result can be registered for one solder 52 in association with each other.

In the above description, the process of registering the parameter value _pk and the known pass / fail judgment result in association with one solder 52 has been described. Normally, the image 33a includes “no solder”, “part floating”, “part missing part”. Since a plurality of inspection items such as “component displacement” are included, the processing of steps S120 to S140 is repeated for the evaluation area E of each solder in the captured image 33a. Furthermore, since many components are mounted on the single mounting substrate 50 and there are many evaluation regions E, steps S100 to S140 are repeated while moving the visual field S by the XY stage 23. In this case, the parameter values _pk may be calculated for all the inspection items in advance, and the pass / fail attributes may be registered in a lump without impairing the correspondence between each solder and the pass / fail attributes.

(3) Pass / fail judgment processing:
Next, a quality determination process performed when actually inspecting a product will be described. In the present embodiment, the preliminary pass / fail judgment process by the distance discriminant analysis and the visual discrimination is executed before the pass / fail judgment process by the linear discriminant analysis and the distance discriminant analysis, which are the pass / fail judgment processes in actual operation. In other words, in this embodiment, after the preliminary pass / fail determination process is performed, the actual operation pass / fail determination process is performed.

  As described above, in linear discriminant analysis, it is necessary to collect a large amount of sample data consisting of parameters representing the characteristics of each category in order to obtain a discriminant function with high resolution for good and bad categories. In products, since the yield is high, the number of defective products is often much smaller than the number of non-defective products, and it is extremely difficult to collect data on defective samples. On the other hand, in the distance discriminant analysis, it is possible to discriminate pass / fail using only data of non-defective samples that are relatively easily available. Therefore, in the present embodiment, sample data for pass / fail determination processing during actual operation is accumulated while executing pass / fail determination by the preliminary pass / fail determination processing. Then, when a sufficient number of sample data is collected, the process shifts to an actual operation pass / fail judgment process.

(3-1) Preliminary Pass / Fail Judgment Processing The preliminary pass / fail judgment processing in this embodiment is a judgment criterion used for distance discriminant analysis based on a relatively small number of good sample data whose pass / fail attributes are known in advance for a plurality of parameters. Is created, and discrimination by distance discrimination analysis using this and discrimination by visual observation are performed. In other words, the preliminary pass / fail judgment process is a pass / fail judgment process that can start operation without collecting defective sample data when the apparatus is started up, and creates a judgment criterion used for the pass / fail judgment process during actual operation described later. This is a temporary operation pass / fail judgment process for accumulating a large amount of sample data.

  First, a process for creating a criterion used for distance discriminant analysis will be described. In the distance discriminant analysis in the present embodiment, a distance discriminating threshold set as the Euclidean distance from the center position of the cluster constituted by elements corresponding to parameters having good attributes in the parameter space is used as a criterion. Therefore, in the determination criterion creation process of distance discriminant analysis, using the sample data having good attributes among the sample data, the distance that defines the range of the good cluster is calculated from the center value in those parameter spaces, and the determination criterion is created. . The determination criterion creation processing for distance discrimination analysis is executed according to the flowchart of FIG.

なお、図９において、各サンプルデータに対応する要素の座標Ｐ（ｓ）は×で示されている。 First, the center value calculation unit SP21 refers to the sample data table 34d, and extracts sample data having good attributes from all the registered sample data (step S200). In step S210, it calculates the center value of the configured good clusters by each of these parameters values p _k and the coordinate in the h dimension of parameter space elements. Figure 9 shows a parameter value p _k described in the sample data table 34d as coordinates in h-dimensional parameter space of the parameters. In the figure, for the sake of simplification, the coordinates are shown in the plane of h = 2, but in actuality, this is an h-dimensional space in which an axis is formed for each parameter type h. Of course, in some cases the h = 1, 2, in which case it is possible to plot the parameter value p _k on the line and plane, respectively. In the above, the coordinates P (s) of the sth sample data can be expressed as the following formula (1).

In FIG. 9, the coordinates P (s) of the elements corresponding to each sample data are indicated by x.

のように表すことができる。
次に、ステップＳ２２０において、距離算出部ＳＰ２２は、中心値Ｃと各要素の座標Ｐ（ｓ）とのユークリッド距離を算出する。中心値Ｃとｓ番目の要素の座標Ｐ（ｓ）との距離ＣＰ（ｓ）は以下の式で表すことができる。

The center value C of the cluster consisting of t pieces of elements, the h types of parameters p _k integrating all elements content, calculated the integrated value as an arithmetic mean of the parameter p _k divided by the total number of elements t Is done. That is, the center value C is

It can be expressed as
Next, in step S220, the distance calculation unit SP22 calculates the Euclidean distance between the center value C and the coordinates P (s) of each element. The distance CP (s) between the center value C and the coordinate P (s) of the sth element can be expressed by the following equation.

  In step S230, a frequency distribution diagram for the Euclidean distance CP (s) acquired as described above is created. The created frequency distribution diagram is, for example, as shown in FIG. In step S240, the frequency average and the standard deviation are calculated in the frequency distribution diagram. When the frequency average and the standard deviation are calculated, the distance determination threshold value setting unit SP23 receives an oversight rate input from the input unit 38 in step S250. That is, in the determination criterion creation process of this distance discrimination analysis, the oversight rate can be input by the input unit 38, and the distance discrimination threshold is calculated based on this oversight rate. That is, the overlook rate is a probability that a test object to be determined to be good or bad is determined to be non-defective, but can be calculated by the standard deviation σ in the above-described frequency distribution diagram.

  According to the standard deviation, in the normal distribution, the probability that the variable is included between the frequency average and the standard deviation or the probability that the variable is included in the tail part outside the standard deviation can be easily specified. By multiplying by a constant, the input overlook rate can be obtained. That is, by inputting the oversight rate in step S250, the coefficient α by which the standard deviation σ is multiplied is determined. For example, when the overlook rate 0.00031 is input, the coefficient α = 4. In step S250, an input of coefficient α itself for multiplying the standard deviation may be accepted, or the oversight rate may be specified based on pre-stored oversight rate data (or standard deviation data). good.

In any case, when the oversight rate is received in step S250 and its value is specified, in step S260, a threshold T _ds = A _ds + ασ that gives the oversight rate is calculated. The calculated threshold T _ds is stored in the HDD 34 (step S270). In this way, the threshold value T serving as a reference for determining the quality category in the distance discrimination analysis is set.

The determination criterion for the distance discrimination analysis is created as described above, and the determination criterion for the distance discrimination analysis used in the preliminary pass / fail determination processing uses a numerical value with a high overlook rate coefficient less than α = 4. In the distance discriminant analysis, it is possible to discriminate pass / fail using only good samples that are relatively easy to obtain, but it still takes a lot of labor to visually judge the quality of a large number of mounted boards. For this reason, in the preliminary acceptance / rejection determination process according to the present embodiment, a determination criterion for distance discrimination analysis is created based on a very small number of samples. And what is clearly a good product is judged as a good product without visual judgment, and only a product that is suspected of being defective is visually judged. Then, by setting the overwatch rate relatively high as described above, it is possible to suppress the outflow of defective products even with the criteria for distance discriminant analysis created based on a very small number of good sample data.
Note that the criteria for distance discriminant analysis used for the pass / fail judgment process in actual operation, which will be described later, are created based on a sufficient number of sample data accumulated in the preliminary pass / fail judgment process, so the overlook rate coefficient α is set to a numerical value of 4 or more.

Next, the preliminary quality determination process will be described. FIG. 11 shows the flow of preliminary acceptance / rejection determination processing. Incidentally, omitted in the figure, the step of inspecting the mounting board 50 of the object in steps S330 and imaging for calculating a parameter value p _k, the description is the same as to the step S130 of the sample data registration processing. However, the mounting board 50 that performs imaging here is a board that is a target of quality determination, and the result of quality determination is unknown. In step S330, the parameter value _p 1 for the solder with the upper mounting board _{50, p 2 ·· p k ··} p (h-1), p h is calculated. In step S340, a distance CX to be compared with a distance determination threshold value that is a determination criterion for distance determination analysis is calculated.

At step S340 of FIG. 11, the distance calculation unit SP22, the parameter values _{p 1, p 2 ·· p k} ·· p (h-1), the coordinates X = _{{p 1} in the parameter space defined by _{p h} , P ₂ · p _k ·· p _(h−1) , p _h } and the center value C of the good category calculated when creating the determination criterion of the distance discriminant analysis. That is, the Euclidean distance between the coordinate X and the center value C is calculated by the same method as the above equation (3). In FIG. 9, the coordinates X indicated by ◯ are shown as the coordinates of the element corresponding to the inspection object.

In step S350, the distance determination pass / fail determination unit SP24 compares the distance CX with respect to the center value C of the good category of the coordinate X calculated in step S340 with the distance determination threshold T _ds calculated in advance as a determination criterion. If it is determined that the distance CX is smaller than the threshold value T _ds , it is determined that the inspection target is a good category, and the determination result for the inspection target is stored (step S355). Then, the same processing is repeated for the next inspection target. In addition to the determination result, the parameter value to be inspected may be stored. These determination results and parameter values may be output as data, or may be additionally registered as sample data in the sample data table 34d, as in the visual discrimination described later.

In step S350, if the distance CX is not judged to be smaller than the threshold value _{T ds,} implementing the determination by visual inspection at step S360. In the visual discrimination, it is discriminated whether or not the inspection object determined to be the negative category by the distance discriminant analysis is truly negative. Specifically, the image of the evaluation area E to be inspected is displayed on the display unit 36, and based on this, the quality category is discriminated visually. If it is determined that it is a good category by visual determination, a good determination is made in step S370, and if it is determined that it is a negative category, a negative determination is made in step S380. Finally, the pass / fail judgment result and parameter value for this inspection object are additionally registered in the sample data table 34d (step S390). At this time, the cause of the defect is also registered for the inspection object determined to be negative by visual determination. Then, the same processing is repeated for the next inspection target.

  In this way, in the preliminary pass / fail judgment process, the pass / fail judgment by distance discriminant analysis and visual discrimination is performed, and sample data of pass / fail categories used to create the judgment criteria of the linear discriminant analysis in the pass / fail judgment during actual operation is collected. To do. Then, when these processes are repeatedly executed and a predetermined number (eg, 300) of non-category sample data sufficient to create a criterion for linear discriminant analysis is collected, the pass / fail judgment process during actual operation is collected. Migrate to

(3-2) Pass / Fail Judgment Process at Actual Operation In the pass / fail judgment process at the actual operation in the present embodiment, a judgment criterion for linear discriminant analysis and a judgment criterion for distance discriminant analysis based on a necessary number of sample data by visual judgment. Are created in advance, and linear discriminant analysis and distance discriminant analysis are performed using these. In the present embodiment, the necessary number of sample data for creating the criterion for linear discriminant analysis and the criterion for distance discriminant analysis is the sample data accumulated in the above-described preliminary pass / fail determination process.

  First, a description will be given of a process for creating a determination criterion used for a quality determination process during actual operation. As described above, the linear discriminant analysis and the distance discriminant analysis are performed in the pass / fail judgment process during actual operation. Among these, the distance discriminant analysis determination criterion is created in the same manner as the distance discriminant analysis determination criterion used in the preliminary quality determination process described above, and therefore the creation method is omitted, and the determination criterion generation process used in the linear discriminant analysis is omitted. Will be described.

  In addition, about the distance discrimination | determination analysis determination criteria used for this quality determination process, the sample data used for the preparation are sample data belonging to the good category among the sample data accumulated in the preliminary quality determination process as described above. . That is, the distance discriminant analysis determination criterion used for the pass / fail determination process at the time of actual operation is different from the distance discriminant analysis determination criterion used for the preliminary pass / fail determination process described above. The oversight rate coefficient is set to less than α = 4, which is lower than the numerical value of the oversight rate coefficient of the distance discrimination analysis determination criterion used in the preliminary pass / fail determination processing.

  In the linear discriminant analysis in the present embodiment, a discriminant function that gives a variable that separates the frequency distribution of the good category and the bad category from the parameter value having a known pass / fail attribute and the pass / fail attribute is calculated, and the variable given from the discriminant function A linear discriminating threshold calculated by performing statistical estimation from the distribution tendency is used as a judgment criterion. Therefore, in this determination criterion creation process, a discriminant function for separating the pass / fail categories in the frequency distribution is calculated using the sample data, and a threshold value as a determination criterion is calculated based on the frequency distribution. The criterion creation processing for linear discriminant analysis is executed according to the flowchart of FIG.

First, the discriminant function calculator SP31 refers to the sample data table 34d and acquires all registered sample data (step S400). In step S410, it selects a parameter value p _k incorporated as explanatory variables of the discriminant function from among the parameter values p _k. In other words, all parameters were used when creating the criteria used for the above-mentioned distance discriminant analysis, but in creating criteria used for the linear discriminant analysis, the frequency distribution of non-defective products and defective products out of all parameters is good. The parameter values _pk to be separated are selected and used as explanatory variables of the discriminant function.

In the present embodiment, when selecting parameters for use in the discriminant function from among the parameter values p _k, the parameter value indicating the strength of the correlation is greater than or equal to a predetermined size with any other non-parameter- The parameters to be used are excluded in advance, and are selected only from parameters having a relatively low correlation with other parameters. Thereby, the multicollinearity of the discriminant function is removed. Specifically, a correlation coefficient that quantifies the strength of correlation between each parameter is calculated, and the number of parameters for which the correlation coefficient is equal to or greater than a predetermined value is aggregated for each of the good category and the bad category. Repeat the process of making the parameter with the large total value of the non-use parameter. As a result, parameters whose correlation coefficient is equal to or greater than a certain value are sequentially eliminated, and only parameters that ultimately have a correlation coefficient less than a certain value remain.

で表現される。なお、ここで、ａ_ｉは係数、ｘ_ｉは各パラメータが代入される変数、ｎは選択されたパラメータの数である。上記判別関数は各パラメータ値を変数として有しており、この判別関数により変数"Ｚ"が与えられる。なお、選択されるパラメータ値ｐ_ｋは、「はんだ無」、「部品浮き」、「部品欠品」、「部品位置ずれ」などの検査項目毎に異なるものを選択し、それぞれについての判別関数が算出される。 The parameter used for the discriminant function is selected from the parameters selected as described above. In this parameter selection, for example, a brute force method, a forward selection method, a backward erasure method, a sequential method, or the like is adopted to select a specific parameter that well separates the pass / fail frequency distribution. In the present embodiment, combinations of all parameters are obtained by the brute force method, and a combination that best separates the frequency distribution of pass / fail is selected. Next, a discriminant function using these parameters as variables is calculated in step S420. The discriminant function is

It is expressed by Here, a _i is a coefficient, x _i is a variable to which each parameter is substituted, and n is the number of selected parameters. The discriminant function has each parameter value as a variable, and the variable “Z” is given by this discriminant function. The parameter values p _k to be selected, "solder Mu", "parts lifting", "parts shortage", and select the different each inspection item such as "component positional displacement", the discriminant function for each Calculated.

なお、上記式（５）でｌは良カテゴリーあるいは否カテゴリーを識別する識別符号であり、ｎ_lはカテゴリー毎のサンプル数であり、上部に直線が付されたｘは総平均である。 The discriminant function is calculated by determining the coefficient so as to give Z that most separates the good category and the bad category when the frequency distribution diagram of the good category and the bad category is created. A coefficient that maximizes the correlation ratio η ² in (5) can be determined and calculated.

In the above formula (5), l is an identification code for identifying a good category or a bad category, n _l is the number of samples for each category, and x with a straight line on the top is the total average.

In the above formula (5), the ratio of the total sum of squares S _T and between groups sum of squares S _B and correlation ratio eta ² is maximized is maximized. In this state, the variance between groups of both categories is as large as possible (the average of each category is as far as possible), and the variance of each category is as small as possible. Therefore, both categories are very well separated in the frequency distribution for the discriminant function. The calculation of the discriminant function as described above is described in detail in “Multivariate Statistical Analysis Method” by Yutaka Tanaka and Kazumasa Wakimoto, published on May 1, 1983, Hyundai Mathematics. When the discriminant function calculating unit SP31 calculates each coefficient a _i as described above, data indicating the coefficient and the parameter is stored in the HDD 34 as discriminant function data.

  When the discriminant function that gives Z that most separates the good category and the bad category is calculated as described above, the linear discriminant threshold value calculation unit SP32 creates a frequency distribution chart of each category with respect to the discriminant function (step) S430). That is, the parameter value data and the pass / fail judgment result data are acquired, and the value of Z is calculated by substituting each parameter value of the sample into the parameter defined as the variable of the discriminant function, and is registered in the sample data table 34d. A frequency distribution diagram is created by measuring the frequency according to pass / fail based on the pass / fail judgment by visual inspection.

In step S440, in the frequency distribution diagram created as shown in FIG. 13, the frequency average and standard deviation for the samples are calculated for each pass / fail category. Located in discriminant function Z = 0 is the center of the power average of the applicable category in the figure, "A _dc" the power average of not categories, shows the the frequency average "-A _dc" goodness category. The standard deviation of the negative category is represented by σ _n , and the standard deviation of the good category is represented by σ _o .

When the frequency average and standard deviation for the sample are calculated in step S440, the outflow rate input from the input unit 38 is accepted in step S450. In the linear discriminant analysis determination criterion creation processing in the present embodiment, the outflow rate can be input by the input unit 38, and a threshold is determined based on this outflow rate. That is, the outflow rate is a probability of outflow without detection of a defective product by making a good determination even though the pass / fail determination target is defective, and can be calculated by the standard deviation σ _n of the negative category. it can.

As described above, according to the standard deviation, in the normal distribution, it is possible to easily specify the probability that the variable is included between the frequency average and the standard deviation or the probability that the variable is included in the tail part outside the standard deviation. it can. In this case, the input outflow rate can be obtained by multiplying the standard deviation σ _n calculated based on the frequency distribution of the non-category by a constant multiple (β times). In step S450, an input of the outflow rate itself may be received, an input of a coefficient by which the standard deviation is multiplied, or an outflow rate data (or standard deviation data) stored in advance. The spill rate may be specified.

In any case, when the outflow rate is received in step S450 and its value is specified, in step S460, a threshold T _dc for linear discriminant analysis that gives the outflow rate is calculated and provisionally determined. In the example shown in FIG. 13, the threshold value T _dc is provisionally determined by T _dc = Adc−βσ _n . In this embodiment, in addition to managing the outflow rate, the oversight rate is also managed. In this sense, the threshold value T _dc in step S460 is a provisionally determined value. The overlook rate is the probability of occurrence of a situation in which the threshold value is excessively tightened (overlook) due to the failure determination even though the pass / fail judgment target is a non-defective product, and the standard deviation σ of the good category It can be calculated by _o .

In the present embodiment, oversight rate data is acquired in step S470, it is determined whether (T _dc + A _dc ) / σ _o is greater than a predetermined coefficient γ, and the provisionally determined threshold value T _dc is determined. Whether or not is far from the frequency average of the good category is more than γ times the standard deviation σ _o . If it is determined in step S470 that (T _dc + A _dc ) / σ _o is larger than γ, the tentatively determined threshold T _dc is determined to be a sufficient value from both the outflow rate and the oversight rate. T _dc is stored as threshold data h in the HDD 34 (step S480). If it is not determined in step S470 that (T _dc + A _dc ) / σ _o is greater than γ, the parameter is redetermined because the oversight rate is high.

  In other words, parameter selection is performed on the assumption that a parameter that can properly determine pass / fail for the shape of the pass / fail determination target has not been obtained, and parameters that satisfy the above-described outflow rate and oversight rate are determined. By performing this re-determination, it is possible to prevent the operation of the pass / fail judgment apparatus from being started in a state where a sufficient outflow rate and oversight rate cannot be secured. In addition, the operation of the apparatus can be started with high performance without accumulating know-how. In this embodiment, the oversight rate data is stored in the HDD 34 in advance, but it is possible to adopt a configuration in which a value for specifying the oversight rate is input via the input unit 38. Further, in the present embodiment, the frequency average and standard deviation are calculated as in the example of Patent Document 2, and the linear discrimination threshold is set so as to obtain a predetermined outflow rate based on these. When the frequency distribution can be sufficiently separated and a parameter whose threshold value can be set to a value of 4σ or more from the frequency average of each category can be selected, the original discriminant function in the linear discriminant analysis as in the embodiment of Patent Document 1. You may employ | adopt as a threshold value the intermediate value of each frequency average which is a threshold value.

Next, quality determination processing during actual operation will be described. FIG. 14 shows the flow of pass / fail judgment processing during actual operation. Note that the steps up to step S530 are the same as those up to step S130 of the sample data registration process, as in the preliminary pass / fail judgment, and the description thereof is omitted.
In step S540 of FIG. 14, the variable Z for each discriminant function is calculated by substituting the value of the parameter selected above among the parameter values acquired in step S530 for each discriminant function calculated in advance. In step S550, the linear discrimination quality determination unit SP33 compares the calculated value of the variable Z for each discriminant function with the linear discrimination threshold T _dc corresponding to each discriminant function. If it is determined that each variable Z is greater than each threshold value T _dc , the inspection target of each variable Z is determined to be negative in step S585. If it is determined that each variable Z is smaller than the corresponding threshold value T _dc , the process proceeds to step S560.

In step S560, the distance calculation unit parameter value _p 1 by _{SP22, p 2 ·· p k ··} p (h-1), the coordinates of the parameter space defined by _{p h X = {p 1,} p 2 ·· The distance CX between p _k ·· p _(h−1) , p _h } and the center value C of the good category is calculated. In step S570, the distance determination pass / fail determination unit SP24 compares the distance CX with respect to the center value C of the good category of the coordinate X calculated in step S560 with the distance determination threshold T _ds calculated in advance as a determination criterion. To do. If the distance CX is judged to be smaller than the threshold value _{T ds,} good determines in step S580. If the distance CX is not judged to be smaller than the threshold value _{T ds} it is not determined at step S585.

  Finally, the quality determination result for the solder 52 to be inspected is stored (step S590). In addition, the parameter value of the inspection target may be stored together with the pass / fail determination result, and the cause of the defect may be stored together for the inspection target determined to be negative. These may be output as data or may be additionally registered in the sample data table 34d as sample data. When storing the cause of failure, it may be automatically stored based on the determination criteria when it is determined to be defective, or the failure cause identified by visual inspection is input as in the above-described preliminary pass / fail determination processing. The data may be input via the unit 38 and stored. In addition, when additionally registered as sample data, it can be used when correcting each determination criterion.

  Thus, in this pass / fail judgment process, the discriminant function and threshold used for the linear discriminant analysis are calculated for each cause of failure. Therefore, when the solder 52 to be inspected is a defective product having a specific cause of failure, linear discrimination is performed. In the analysis, it can be determined that the category is negative. On the other hand, in the distance discriminant analysis, pass / fail is discriminated based on the position of the cluster composed of elements having good attributes and the position of the element corresponding to the inspection target. It is possible to correctly determine whether or not a defective product has a cause of failure. That is, by sharing linear discriminant analysis and distance discriminant analysis, it is possible to realize pass / fail determination with excellent determination accuracy that can make use of the characteristics of both discriminant analysis methods and complement each other's defects.

  In the present embodiment, sample data is accumulated in the preliminary pass / fail judgment process, and the pass / fail judgment is performed based on the judgment criterion of linear discriminant analysis and the criteria of distance discriminant analysis created using them. It is not limited to. In other words, if the required number of samples with known pass / fail attributes can be prepared in advance, the sample data is accumulated by the sample data registration process, and the pass / fail by linear discriminant analysis and distance discriminant analysis without going through the preliminary pass / fail judgment process. The determination process may be performed.

  10; Pass / fail judgment device, 20; Imaging unit, 21; Controller, 21a; Image memory (VRAM), 22; Camera, 22a; Optical system, 22b: CCD imaging plate, 23; XY stage, 24; Computer 31; CPU 32; RAM 33; video memory (VRAM) 33a; captured image 34; hard disk (HDD) 34a; operating system (OS) 34b; pass / fail judgment program 34c; 34d; Sample data table, 35; Video interface, 36; Display unit, 37; Input interface, 38; Input unit, 39; I / O, 50; Mounting substrate, 51; Chip component, 52; Solder, SP1; Input unit, SP11; imaging execution unit, SP12; captured image acquisition unit SP13; parameter calculation unit, SP2; distance discrimination analysis unit, SP21; center value calculation unit, SP22; distance calculation unit, SP23; distance discrimination threshold setting unit, SP24; distance discrimination pass / fail judgment unit, SP3: linear discrimination analysis unit, SP31 Discriminant function calculation unit, SP32; linear discrimination threshold value calculation unit, SP33; linear discrimination pass / fail judgment unit, SP4; visual data input unit, SP5; display control unit.

Claims (6)

Parameter acquisition means for acquiring a plurality of parameters that are pass / fail judgment factors from the inspection target;
Obtain a discriminant function consisting of a plurality of parameters selected to give a variable that separates the frequency distribution of the good category and the bad category from the plurality of parameters corresponding to a plurality of samples with known pass / fail attributes. The linear discrimination threshold of the discriminant function as a judgment criterion set based on this and the calculation result obtained by substituting the parameter corresponding to the test object into the discriminant function, the pass / fail category of the test object Linear discriminant analysis means for discriminating
In the parameter space in which the plurality of parameters are projected, a distance determination threshold as a determination reference set as a distance to a center position of a good cluster configured by elements corresponding to a plurality of samples having good attributes, and the center position A distance discriminating and analyzing means for comparing the distance of the element corresponding to the inspection object with respect to and determining the pass / fail category of the inspection object;
With respect to the inspection object discriminated as a good category by the linear discriminant analysis means, a good / bad category is discriminated by the distance discriminating / analyzing means, and a good category is discriminated by both the linear discriminant analysis means and the distance discriminant analysis means. A quality determination device comprising: quality determination means for determining quality only when the A visual discriminating means for visually discriminating a pass / fail category to be inspected;
When the visual discrimination means discriminates the pass / fail category for the inspection object determined by the distance discrimination analysis means as the negative category, and when the distance discrimination analysis means and the visual discrimination means determine the negative category Preliminary pass / fail judgment means for judging only whether or not
Sample data storage means for storing the plurality of parameters corresponding to the inspection object determined by the preliminary quality determination means and the quality determination results as sample data,
The pass / fail judgment means
The linear discriminant analysis means discriminates a pass / fail category by the linear discriminant threshold set based on the sample data,
The distance discriminating / analyzing means discriminates a pass / fail category by the distance discriminating threshold set based on data determined to be a good category among the sample data, and the pass / fail of the inspection object is determined based on the discrimination results. The pass / fail judgment device according to claim 1. A parameter acquisition step for acquiring a plurality of parameters that are pass / fail judgment factors from the inspection target;
Obtain a discriminant function consisting of a plurality of parameters selected to give a variable that separates the frequency distribution of the good category and the bad category from the plurality of parameters corresponding to a plurality of samples with known pass / fail attributes. The linear discrimination threshold of the discriminant function as a judgment criterion set based on this and the calculation result obtained by substituting the parameter corresponding to the test object into the discriminant function, the pass / fail category of the test object A linear discriminant analysis process for discriminating
In the parameter space in which the plurality of parameters are projected, a distance determination threshold as a determination reference set as a distance to a center position of a good cluster configured by elements corresponding to a plurality of samples having good attributes, and the center position A distance discriminating and analyzing step of comparing the distance of the element corresponding to the inspection object with respect to and determining the pass / fail category of the inspection object;
For the inspection object discriminated as a good category by the linear discriminant analysis step, a pass / fail category is discriminated by the distance discriminant analysis step, and it is discriminated as a good category in both the linear discriminant analysis step and the distance discriminant analysis step. A quality determination method comprising: a quality determination step that determines quality only when the
A visual discrimination process for visually determining the quality of the inspection target;
When the visual discrimination step determines the pass / fail category for the inspection object determined by the distance discriminant analysis step, and when it is determined as the negative category in both the distance discriminant analysis step and the visual discrimination step Preliminary pass / fail judgment process for judging only
A sample data accumulating step for accumulating the plurality of parameters corresponding to the inspection target determined by the preliminary pass / fail determination step and the pass / fail determination result as sample data;
The pass / fail judgment step includes
The linear discriminant analysis step discriminates pass / fail categories by the linear discriminant threshold set based on the sample data,
The distance discrimination analysis step discriminates a pass / fail category by the distance discrimination threshold set based on data determined to be a good category among the sample data, and the pass / fail of the inspection object is determined based on the discrimination results The pass / fail judgment method according to claim 3. A parameter acquisition function for acquiring a plurality of parameters that are pass / fail judgment factors from the inspection target;
Obtain a discriminant function consisting of a plurality of parameters selected to give a variable that separates the frequency distribution of the good category and the bad category from the plurality of parameters corresponding to a plurality of samples with known pass / fail attributes. The linear discrimination threshold of the discriminant function as a judgment criterion set based on this and the calculation result obtained by substituting the parameter corresponding to the test object into the discriminant function, the pass / fail category of the test object Linear discriminant analysis function to discriminate
In the parameter space in which the plurality of parameters are projected, a distance determination threshold as a determination reference set as a distance to a center position of a good cluster configured by elements corresponding to a plurality of samples having good attributes, and the center position A distance discriminant analysis function that compares the distance of the element corresponding to the inspection object with respect to and determines the pass / fail category of the inspection object;
For the inspection object discriminated as a good category by the linear discriminant analysis function, a pass / fail category is discriminated by the distance discriminant analysis function, and it is discriminated as a good category by both the linear discriminant analysis function and the distance discriminant analysis function. A non-defective / non-defective program for causing a computer to realize a non-defective / unsatisfactory function only when A visual discrimination function for visually determining the quality of the inspection target;
When the visual discrimination function determines the pass / fail category for the inspection object determined by the distance discriminant analysis function as a negative category, and the distance discriminant analysis function and the visual discrimination function determine the pass category. Preliminary pass / fail judgment function for judging only
A sample data accumulation function for accumulating the plurality of parameters corresponding to the inspection object determined by the preliminary quality determination function and the quality determination result as sample data;
The pass / fail judgment function is
The linear discriminant analysis function discriminates a pass / fail category by the linear discriminant threshold set based on the sample data,
The distance discriminant analysis function discriminates a pass / fail category by the distance discriminating threshold set based on the data determined to be a good category among the sample data, and the pass / fail of the inspection object is determined based on these discrimination results The quality determination program according to claim 5, wherein the determination is made as follows.

Images