JP2008153255A - Model registration method for mounting component inspection, mounting component inspection method, and board inspection apparatus using this method - Google Patents

Abstract

It is possible to easily and accurately perform registration of a defective model necessary for correctly distinguishing between correct and incorrect parts.
A component classification processing unit 103 uses a CAD data of a board input by a CAD data input unit 101 to classify each component on the board into parts having the same component type and size. For each product number, the model registration unit 104 registers a good model of an image of a part with the product number in the model storage unit 111 and sets a defective model using a good model of another part belonging to the same group. The inspection execution unit 107 calculates the similarity of the inspection target image with respect to each model using the good model and the defective model corresponding to the part to be inspected. As a result, when the similarity to the good model is higher than the similarity to any defective model, it is determined that the part to be inspected is a correct part.
[Selection] Figure 5

Description

  The present invention relates to an automatic visual inspection for determining whether or not a correct component is mounted on a substrate after a component mounting process is completed.

  In recent board production lines, parts can be automatically mounted by installing a mounter, but parts are replenished or replaced manually in the parts feeder, so wrong parts are mounted due to human error. Sometimes. For this reason, conventionally, it is possible to detect component mounting errors by extracting a character string printed on the main body of the component from the board image after mounting the component and determining whether this is the correct character string. (See Patent Document 1). In addition, even for a component without a character string, a mounting error is detected based on the color and size of the component.

JP2004-235582

  In conventional inspection of mounted components including Patent Document 1, an image of a component to be inspected (hereinafter referred to as “inspection target image”) is referred to as an image of a correct component registered in advance (hereinafter referred to as “good model”). ) To determine whether the part is correct or not. For example, for a part on which a character string is printed, the similarity between the character string cut out from the image to be inspected and the character string in the good model is obtained, and when this similarity exceeds a predetermined criterion value, This part is determined to be a correct part.

  By the way, a component mounting mistake often occurs due to a mistake with a component having a similar appearance. In particular, for a part printed with a character string, there are a plurality of parts whose appearances are almost the same except that only a part of the character string is different.

  Therefore, when discriminating whether a part is correct or not by collating character strings, it is necessary to set the determination reference value high so that even a part of the character strings can be detected. However, if the criterion value is increased, even if it is a correct part, if the similarity to the good model decreases due to blurring or blurring of characters, it may be determined that the part is wrong. It becomes impossible to ensure the accuracy of the.

  As a method for solving such a problem, in addition to the good model, an image of a part that may be erroneously mounted (hereinafter referred to as “defective model”) is registered, and each of these good / defective items is registered. A method of matching the model individually with the character string to be examined can be considered. In this case, for the inspection target image, if the similarity to the good model is higher than the similarity to the defective model, the part to be inspected is determined to be a correct part, and if the similarity to the defective model is higher, It is determined that the part to be inspected is an incorrect part.

  However, in the above method, since it is necessary for the user to determine what kind of defective model should be registered, it takes time to register. Even if a large number of models are registered, there is a problem that the accuracy of the inspection cannot be ensured if there is an error in the model selection by the user.

  The present invention has been made paying attention to the above-mentioned problem, and images a board that has undergone a component mounting process, and uses the characteristics (for example, character strings or colors) related to the parts in the generated image to correct the parts on the board. It is an object of the present invention to enable easy and accurate registration of a defect model necessary for correctly identifying whether a component is correct or not when inspecting whether or not a component is mounted.

  In order to solve the above-described problems, in the present invention, prior to the inspection, the design data (for example, CAD data) of the board is used to group the mounted parts based on the component type and size, and each step on the board. Pay attention to the good model registered in Step B, Step B, where the good model of the part image is registered, and select the good model of the other parts classified into the same group as the part to which the good model under consideration is applied. Step C is set as a defective model corresponding to the good model under consideration. Then, in the inspection, for each part to be inspected, the similarity of the image of the part to the good model and the defective model is obtained for the image of the part, and the similarity to the good model is higher than the similarity to any defective model When is higher, it is determined that the part to be inspected is the correct part, and when the similarity to one of the defective models is higher than the similarity to the good model, the part to be inspected is incorrect. It is determined that it is a part.

  The majority of component mounting mistakes are thought to be caused by misreading similar parts when refilling or replacing parts in the parts feeder. Therefore, it is considered that there is a high possibility that a component that exists as a component of the same type and the same size as the correct component and that is prepared in the field for mounting on the same substrate is erroneously mounted.

  In the model registration method according to the present invention, in consideration of the above points, each mounted component on the board is grouped into components having the same component type and the same size, and each component has the same group as that component. Since a good model of a part belonging to the item is set as a defective model, it is possible to automatically set a defective model suitable for detecting a mounting error. In the defect model setting, various good models may be used as the defective models of other parts as they are, but not limited to this, for example, for the model to be a defective model, the region most similar to the good models of other parts is selected. The image in this area may be specified as a defective model.

  In step A, in addition to the component type and size, whether a character string is printed on the component may be set as a grouping condition. This is because the inspection algorithm differs between a part with a character string and a part without a character string.

  In a more preferable aspect, prior to the inspection, weights corresponding to differences with respect to corresponding pixels of the good model are set for each pixel of the defective model set in Step C, and these weights are associated with the defective model. Step D of registration is further executed. In the inspection in this case, the larger the difference between the images, using a value obtained by weighting the difference between corresponding pixels by the weight registered for the model in Step D between the inspection target image and each defective model. The similarity of the image to be inspected with respect to each defective model is calculated by calculating the similarity that is set to increase the value. On the other hand, for the good model, the degree of similarity of the image to be inspected with respect to the good model is calculated in a plurality of ways by executing the calculation a plurality of times using the same weight set for each defective model. Then, the similarity to the good model and the similarity to the bad model are combined and compared for each of those calculated using the same weight, and in all the combinations, the similarity to the good model is higher than the similarity to the bad model. When it becomes higher, the part to be inspected is determined to be a correct part, and in any combination, when the similarity to the defective model is higher than the similarity to the good model, the part to be inspected Is determined to be an incorrect part.

  According to the above aspect, when there is a difference between the inspection target image and the model, the degree of similarity changes depending on the weight value for weighting the difference. Since the weight for each pixel is set according to the difference between the defective model and the good model, a large weight is set for a pixel that actually has a large difference between the good and bad models. In the image to be inspected when the part to be inspected is a correct part, it is considered that the same difference as that obtained for the good model is generated at the same position for any defective model. Emphasized by weighting, the calculated similarity is low (value is high). On the other hand, for a good model, it is considered that no large difference is calculated for any pixel, so the calculated similarity is high (value is low). Therefore, in any combination of similarities, it is considered that the similarity to the good model is higher than the similarity to the defective model.

  If the part to be inspected is the wrong part, the difference between the image to be inspected and the model is the smallest when the defective model corresponding to the wrong part is targeted. It is thought to be higher. On the other hand, between the image to be inspected and the good model, a difference similar to that obtained between the bad model corresponding to the wrong part and the good model occurs at the same position. When the similarity to the good model is calculated using, the similarity is considered to be lower than the similarity to the defective model.

  Therefore, when comparing the similarity calculated using the same weight for the good model and the defective model, if the part to be inspected is correct, the similarity to the good model is higher than the similarity to the defective model. Conceivable. Further, when the part to be inspected is an incorrect part, the similarity to the defective model corresponding to the part is considered to be higher than the similarity to the good model calculated using the same weight. Therefore, according to the above aspect, since the similarity to the model corresponding to the actual part is larger than the similarity to the other model, the success or failure of the part can be determined with high accuracy.

  The board inspection apparatus using the above method accepts input of design data of a memory on which a good model of an image of each component on the board to be inspected is registered and design data of the board to be inspected. Using the classification means for grouping each mounted component based on component type and size, and focusing on the good model registered in the memory in order, it was classified into the same group as the component to which the good model under consideration is applied When a defective model setting unit that sets a good model of another part as a defective model corresponding to the good model under consideration and the board to be inspected are imaged, for each part to be inspected in the generated image, A determination unit is provided for determining the similarity of the image of the part to the corresponding good model and the defective model, and determining whether each part is a correct part based on the similarity. Further, the discrimination means discriminates that the part to be inspected is a correct part when the similarity to the good model is higher than the similarity to any defective model, and any of the similarities to the good model When the similarity to the defective model becomes higher, it is determined that the part to be inspected is an incorrect part.

  According to the board inspection apparatus having the above configuration, by inputting design data of a board to be inspected, a defect model suitable for determining whether components are correctly mounted is automatically set, and high-precision inspection is executed. It becomes possible to do.

  According to the above method and the board inspection apparatus, a defect corresponding to a component that can be mounted by mistake using an image registered as a board design data or a good model for checking whether a correct component is mounted. The model can be set easily and accurately. Therefore, it is possible to ensure the accuracy of the inspection for determining the success or failure of the mounted component.

FIG. 1 shows the configuration of an automatic visual inspection apparatus for printed circuit boards to which the present invention is applied.
This automatic visual inspection apparatus (hereinafter simply referred to as “inspection apparatus”) inspects a board that has undergone a component mounting process on a board production line, and includes a controller 1, a camera 2, an illumination apparatus 3, a board stage 4, The input unit 5 and the monitor 6 are configured.

  The substrate stage 4 includes a table unit 41 for supporting the substrate 8 and a moving mechanism 42 including an X-axis stage and a Y-axis stage (both not shown).

  The camera 2 and the illumination unit 3 constitute an optical system called “color highlight method”. The camera 2 generates a color still image, and is provided above the substrate stage 41 with the imaging surface directed downward and the optical axis aligned with the vertical direction. The illumination unit 3 includes three annular light sources 3R, 3G, and 3B disposed between the substrate stage 4 and the camera 2. These light sources 3R, 3G, and 3B emit red, green, and blue color lights, respectively, and are arranged in a state in which each central portion is aligned with the optical axis of the camera 4. The light sources 3R, 3G, and 3B are set to have different diameters so that the substrate 8 can be irradiated with light from different directions.

  The controller 1 includes a computer control unit 10, an image input unit 11, an imaging control unit 12, an illumination control unit 13, an XY stage control unit 14, a memory 15, a CD-ROM drive 16, and a communication interface 17. It is done. The image input unit 11 includes an interface circuit for the camera 2 and the like. The imaging control unit 12 is for outputting a timing signal instructing imaging to the camera 2.

  The illumination control unit 13 controls the operation of turning on / off the light sources 3R, 3G, and 3B of the illumination unit 3 and adjusts the amount of light. The XY stage control unit 14 controls the movement timing and movement amount of the substrate stage 4.

  The memory 15 includes an inspection program (including a program for a component inspection system described later) and various inspection data (for example, component information of a component to be inspected, setting data for an inspection region, and a part to be inspected). Are stored as binarization thresholds. Furthermore, the memory 15 of this embodiment is also provided with areas corresponding to the storage units of the model storage unit 111, the group data storage unit 112, and the weight data storage unit 113 described later.

  The control unit 10 controls the movement of the substrate stage 42 via the XY stage control unit 14, thereby aligning the camera 2 and the substrate 8 and taking an image. The color image generated by this imaging is input to the control unit 10 via the image input unit 11 and stored in its internal memory (RAM). The control unit 10 sequentially executes inspections for each component using the color image stored in the RAM. Further, the control unit 10 transmits the measurement result and determination result for each component and the image used for the inspection to an external device (not shown) using the communication interface 17.

  The inspection executed by the above-described inspection apparatus includes an inspection of whether or not the mounted component is correct (hereinafter referred to as “component inspection”) and an inspection of whether or not the position and orientation of the component are correct. In the following, the description will be limited to component inspection only.

  The component inspection in this embodiment is further classified into two types. One is to determine whether a part is correct or not based on whether or not the character string on the image is correct for a part on which a predetermined character string (model number, resistance value, etc.) is printed on the part main body. The other is to determine whether a part is correct or not by checking whether the size or color of the part on the image is correct for a part on which no character string is printed. In any case, for each part, a correct sample image of the part and a sample image of the wrong part are registered as a good model and a defective model, respectively.

  The control unit 10 calculates the degree of similarity between the good model and the defective model for the actual image of the part according to the procedure described later, and compares the degree of similarity between the models to determine whether the part is correct. Determine.

  Further, in this inspection apparatus, a defect model suitable for detecting an erroneous mounting of a component can be automatically set in the control unit 10 using the CAD data of the board. Hereinafter, a method for setting a defect model will be described using a case where a character string is inspected as an example.

  FIG. 2 shows a configuration example of a substrate to be inspected, and FIG. 3 shows the contents of CAD data of the substrate. In this example, in order to simplify the explanation, it is assumed that the number of mounting parts on the board is limited to nine, and that each part is assigned a part number of 1 to 9. Also in the following description, when referring to individual parts, they are indicated by using part numbers such as “No. parts”.

  The CAD data in the example of FIG. 3 includes the above part number, the part number, the part type, the size, the x and y coordinates of the part center, and the mounting angle (based on the part in a predetermined direction). (The actual rotation angle of the part with respect to the state).) In FIG. 3, information other than the part number and the mounting angle is indicated by characters such as alphabets, but data having the same content is expressed by the same characters.

  According to the above CAD data, each part except for the part No. 6 belongs to the same part type I (in this example, a square chip). Of the parts belonging to the part type I, each part except for No. 4 (in FIG. 3, the part with a halftone dot added to the part number) has the same size. Further, among the parts of the same size of the part type I, the parts Nos. 2, 5, 8, and 9 have the same part number, and the parts Nos. 3 and 7 have the same part number.

  Further, according to the configuration diagram of the board of FIG. 2, all the parts belonging to the part type I, except for the fourth part, are printed with character strings on the part main body. The character strings (101, 102, 103) in this example indicate resistance values, and the same character string is printed on all parts having the same product number.

  By the way, mis-mounting of parts is caused by a mistake of an operator, but the most likely mistake is considered to be a mistake with a part having a similar appearance. In addition, when a component with an apparently different appearance is mistaken, it is possible to detect an erroneous mounting by matching with a good model. However, when a component with a very close appearance to a correct component is mistaken (for example, FIG. 2). , 3, when the part number A is mistaken for the part number B or C part), there is a possibility that erroneous mounting cannot be detected only by the good model. In addition, in order to improve the detection accuracy of such erroneous mounting, if the similarity criterion value is increased, the wrong part will be the part that is a correct part but has a defect such as a missing or faint character string. It is determined that

  In view of such a problem, in this embodiment, using the CAD data, each mounted component on the board is grouped into components having the same component type and the same size. Furthermore, when parts of the same type and the same size contain parts with and without a character string in the parts body, the group may be divided depending on the presence or absence of the character string. For each part, a good model of the part is registered, and a good model of another part classified into the same group is set and registered as a defective model. Furthermore, the accuracy of part correctness determination is improved by matching processing using weight data described later.

  FIG. 4 shows the result of grouping the parts shown in FIG. 2 into parts having the same part type and size. According to this example, among the four types of parts belonging to the part type I, three types of parts (part numbers A, B, C) having the same size are classified into the same group (group 1), and the other part numbers D, E is classified into separate groups 2 and 3, respectively. Therefore, for the parts of part number A in group 1, the good models of parts B and C are set as defective models. Similarly, for the part number B, the good models of the part numbers A and C are set as defective models, and for the part number C, the good models of the part numbers A and B are set as defective models.

  According to the model registration method described above, when a component that is clearly different in appearance from the component to be inspected is erroneously mounted, the image of the component to be inspected is compared with the good model as in the past. Incorrect mounting can be detected. In addition, when a component whose appearance is similar to that of the component to be inspected is erroneously mounted, the erroneous mounting can be detected by collation processing using a failure model set from a good model of the component.

  In the above method, the CAD data of the board is used to group each component for each component having the same component type and size, so that the components that are likely to be mistaken by the operator are actually the same. Can be classified into groups. In many cases, components mounted on the board are classified and stocked for each component type and size in the vicinity of the mounter, so that the components having similar appearance are likely to be mixed up. However, in this embodiment, since it is possible to register a defective model for detecting each part, which is likely to be mistaken, it is possible to register the determination reference value for parts inspection to a high value. It becomes possible to ensure the accuracy of the inspection.

  FIG. 5 is a functional block diagram of a component inspection system introduced into the inspection apparatus of FIG. The component inspection system 100 executes the above-described component classification processing and model registration processing, and then performs inspection by fetching an image of a substrate to be inspected. The system 100 includes a model storage unit 111, a group data storage unit 112, and a weight data storage unit 113, a CAD data input unit 101, a non-defective image acquisition unit 102, a component classification processing unit 103, and a model registration unit 104. , A weight data setting unit 105, a board image acquisition unit 106, an inspection execution unit 107, an inspection result output unit 108, and the like.

  The CAD data input unit 101 takes in CAD data of a substrate to be inspected from a storage medium such as a CD-ROM or an external device. The non-defective product image acquisition unit 102 captures an image generated when the model of the substrate 8 is captured by the camera 2 as an image of a non-defective substrate (hereinafter referred to as “non-defective product image”). The acquired CAD data and non-defective product images are stored in a working memory (not shown).

  The component classification processing unit 103 groups each mounted component for each component having the same component type and size based on the product number, component type, and size information in the CAD data. In addition, a unique identification number (hereinafter referred to as “group number”) is set for each set group, and is stored in the group data storage unit 112. In addition, for each group, the part number and part number of the part belonging to the group, and group attribute information (part type and size common to each belonging part, and presence / absence of character string) are associated with the above group number. , Saving in the group data storage unit 112 (hereinafter, the process of saving the group number in the group data storage unit 112 is referred to as “group registration process”, which is a process of saving data in association with the group number. Is referred to as “processing for registering data in a group”).

  Based on the CAD data, the model registration unit 104 cuts out an image of the first focused component from the non-defective image of the board for each product number, and registers it in the model storage unit 111 as a good model common to the product number. Also, the model registration unit 104 recognizes other product numbers belonging to the same group as the product number for each product number based on the grouping result of the part classification processing unit 103, and sets a good model of the recognized product number as a defective model. .

  The weight data setting unit 105 sets weight data reflecting the difference from the corresponding good model for each defective model, and registers this in the weight data storage unit 113.

  The board image acquisition unit 106 captures the generated image when the board 2 to be inspected is imaged by the camera 2 and supplies it to the inspection execution unit 107. The inspection execution unit 107 performs component inspection by processing the image of the target component using registered data in the model storage unit 111 and the weight data storage unit 113 while paying attention to each component included in the supplied image in order. Execute. The inspection result output unit 108 integrates the inspection results and outputs them to the monitor 6 or an external device (not shown) when the component inspection for all the components on the board is completed.

  Next, the good / bad model and weight data registered in the component inspection system 100 will be described.

  FIGS. 6A and 6B show examples in which good models are registered for two parts belonging to the same group. FIG. 6A is an enlarged view of a component (the above-described product numbers A and B) in which a character string is printed on the component main body. FIG. 6B is an example of a component that has no character string in the component main body but has a common color.

  In this embodiment, for each component, a rectangular area 50 having a predetermined size is set in the part main body in the non-defective image, and the image in this area 50 is cut out and registered as a good model. In the case of a component having a character string, it is necessary to adjust the position and size of the region so that all the character strings are included.

  As in this embodiment, when a good model set for each of a plurality of parts belonging to the same group is used as a defective model for other parts, registration processing as a defective model is not particularly necessary. However, in this embodiment, in order to set a defect model that accurately reflects a substantial difference from the good model, the matching process using the good model being focused is used in the good model of other parts. An area most similar to the good model is identified, and an image in the area is registered in the model storage unit 111 as a defective model.

  The weight data is set based on a substantial difference between the set defective model and the good model under attention. Specifically, after creating a difference image between these models, the difference image is smoothed, and the smoothed difference image is set as weight data corresponding to the defective model.

  FIG. 7 shows the density (gradation) of each pixel for a difference image (upper stage) and an image after smoothing (lower stage) as an example of setting weight data. The smoothing process of this example replaces the gradation of the pixel of interest with the average value of the gradations of the pixel and the surrounding eight pixels while paying attention to the constituent pixels of the difference image in order. The pixel data of each component pixel of the difference image after the smoothing process is used to calculate the degree of difference with respect to the model of the inspection target image as a “weight” for the corresponding pixel (pixel having the same coordinate) of the defective model.

  Since color images are processed in this embodiment, each of the good / bad models is a combination of three types of image data of R, G, and B. Therefore, the weight data is also set for each of R, G, and B.

  Specifically, the R, G, and B values of the pixel at the coordinates (x, y) of the good or bad model are Mr (x, y), Mg (x, y), and Mb (x, y), respectively. The R, G, B values of the corresponding pixels of the inspection target image are Ir (x, y), Ig (x, y), Ib (x, y), and the weight data are Hr (x, y), Hg (x , Y) and Hb (x, y), the difference S between the images to be inspected with respect to this model is obtained by the following equation (1) (X and Y in the equation are the maximum values of the x and y coordinates). is there.).

  The degree of difference S according to the above equation (1) substantially represents the degree of similarity of the inspection target image with respect to the model. That is, the degree of similarity is expressed with a parameter whose value decreases as the inspection target image is closer to the model. Further, according to the expression (1), when a large weight is associated with a pixel having a large difference value, the difference is further strengthened and the similarity is lowered.

  The values of the weights Hr (x, y), Hg (x, y), and Hb (x, y) are determined according to the difference between corresponding pixels of the defective model and the good model (a large weight is applied to a pixel having a large difference). Is set.) If the part to be inspected is a part corresponding to the defective model, when the difference between the defective model and the inspection target image is obtained, the value of the difference between the corresponding pixels is considered to be close to 0 at any position. Therefore, the similarity is considered to be relatively high.

  On the other hand, when the difference between the good model and the image to be inspected is obtained, the difference value corresponding to the pixel to which the large weight is associated is comparable to the value calculated between the good model and the defective model. The difference is strengthened by the weight. Therefore, the similarity in this case is considered to be smaller than the similarity to the defective model.

  When the part to be inspected is correct, if the difference is found between the defective model and the inspection target image, the difference between the pixels associated with a large weight becomes large, and the difference in this pixel becomes stronger. Therefore, the similarity is low. On the other hand, when the difference between the good model and the inspection target image is obtained, the difference between the corresponding pixels is close to 0 at any position. Therefore, for any defective model, it is considered that the similarity to the good model is greater than the similarity to the defective model.

  In the component inspection system of this embodiment, in view of the above points, for each defective model, the difference degree S (hereinafter referred to as the inspection target image with respect to the defective model) is calculated using equation (1) using the weight data obtained for the defective model. , The difference degree S (n) is calculated. Further, for the good model, the difference degree S (hereinafter referred to as difference degree S (o)) of the inspection target image is calculated by the equation (1) using the same weight data, and S (o), S ( Compare the values of n).

  According to the above processing, for each defective model, while calculating the similarity to the image to be inspected using the weight data corresponding to each model, for the good model, a plurality of weights corresponding to each defective model are calculated. A plurality of similarities are calculated using the data, and each similarity is combined and compared for each of those calculated by the same weight data.

  Here, for any weight data, if the similarity to the good model is greater than the similarity to the defective model, it is determined that the part to be inspected is the correct part (however, each similarity is (Assuming that it exceeds the predetermined reference value.)

  On the other hand, when the similarity is calculated using any of the weight data, the similarity to the defective model is greater than the similarity to the good model, and the similarity is greater than the reference value. In such a case, there is a high possibility that the part to be inspected is a part indicated by the defective model. Therefore, in this case, it is determined that the part to be inspected is not a correct part.

  According to such processing, the similarity value for each model varies depending on which model the image corresponding to the difference between the good and bad models in the inspection target image is close to. Therefore, by comparing these similarities, it is possible to determine the presence or absence of incorrect mounting with high accuracy.

  In the example of FIG. 7, the weight corresponding to the magnitude of the difference with respect to the good model is set for each constituent pixel of the defective model. However, the present invention is not limited to this. For example, the magnitude of the difference with respect to the good model is predetermined. A constant value weight may be set only for pixels that are greater than or equal to the value.

Hereinafter, the flow of processing executed in the component inspection system will be described for each type.
First, FIG. 8 shows a processing procedure related to component classification and registration of a good model. In ST101, which is the first step in this procedure, the initial value “1” is set to the counter i for specifying the part number, and the group number j _max is set to “0”.

  In ST102, the CAD data of the i-th component Bi is read, and it is checked whether the product number in the data has already been classified into any group (ST103). At the stage of i = 1, since no group has been set yet, this determination is “NO”, the process proceeds to ST104, and the initial value 0 is set to the counter j indicating the group of interest.

In the next ST105, the counter j is updated to one larger value (that is, 1), and then the updated j is compared with _{jmax in} ST106. Immediately after the process is started, j _max = 0, so j> j _max and the determination in ST 106 is “NO”. In this case, the process proceeds to ST111, and j _max is updated with the value of j (ie, j _max = 1).

In the next ST 112, it registers the j-th group _{G j} on a group data storage unit 112 (specifically, will register a group number j.). In subsequent ST113, the attribute information of the part _{B i} (whether component types and sizes as well as a string), as attribute information of the j-th group Gj, is registered in the group data storage unit 112. Furthermore Then, the process proceeds to ST114, and registers the part number and part number of the component _{B i} to the group Gj.

  In ST115, an image of the component Bi is cut out from the non-defective image on the board. Further, when the mounting angle of the component Bi is an angle other than 0 degrees, the cut-out image is rotationally corrected so that an image with a mounting angle of 0 degrees is obtained. In ST116, the image acquired in ST115 is registered in the model storage unit 111 as a good model with a part number corresponding to the component Bi.

In ST117, the value of i is updated to one larger value. If i after the update does not become larger than the total number i _{max of} mounted parts, the next ST118 is “YES” and the process returns to ST102. Similarly, the same processing is executed while changing the component of interest by updating i. However, since there is at least one group for the second transition component, ST102 to ST105 After the processing, ST106 becomes “YES”, and the process proceeds to ST107.

  In ST107, the attribute information of the component Bi (component type and size and presence / absence of character string) is compared with the attribute information of the group Gj. If it is determined that the component type is the same, the size is the same, and the presence / absence state of the character string is the same (when all of ST108 to 110 are “YES”), the process proceeds to ST114 and the component Bi The product number and the part number are registered in the group Gj of interest. Thereafter, by executing ST115 and 116, a good model having a product number corresponding to the component Bi is set and registered.

On the other hand, when the attribute information match cannot be confirmed for any of the set groups, the loop of ST105 to 110 is repeated until j = _jmax , and then ST106 becomes “NO”. . In this case, ST111 to 113 are executed, and thereafter ST114 to 116 are executed to set a new group and register the component Bi in the group.

  When the part number of the part Bi is already classified into any group, ST103 becomes “YES” and the process proceeds to ST119, and the part number of the part Bi is registered in the group in which the part number is classified. Correspond with.

By executing the above processing until i reaches the total number i _{max of the} mounted components, the components are grouped for each component having the same attribute information. A good model is set for each product number and registered in the model storage unit 111.

  FIG. 9 shows a processing procedure for registering a defect model and weight data in units of product numbers for one group registered by the processing of FIG. 8 (when there are a plurality of groups, the processing of FIG. Is executed.)

  First, in ST201, an initial value “1” is set in a counter m for specifying the product number and the good model. In ST202, the good model Om indicated by m is read.

  In the next ST203, an initial value “1” is set in a counter k (k ≦ M−1 (M is the number of registered product numbers in the group)) for storing the number of readings of a good model other than the good model under consideration. set. In ST204, a good model Ok other than the good model being noticed (Ok means the good model read k-th) is read out.

  In ST205, matching processing is performed between the two read models Om and Ok, and an area most similar to Om in the model Ok is specified. In ST206, the image in the specified area is registered in the model storage unit 111 as the kth defective model Nk of the mth product number.

  Further, in ST207, a difference image between the models Om and Nk is generated. In ST208, a process of smoothing the generated difference image is executed. Further, in ST209, each pixel data of the difference image after the smoothing process is registered in the weight data storage unit 113 as the weight data Hk corresponding to the defective model Nk.

  Thereafter, in ST210, the value of k is updated to one larger value. Thereafter, the loop of ST203 to 211 is repeated until k> (M−1) after this update. As a result, for each part number registered in the same group as the part number corresponding to the good model Om of interest, the failure model Nk and weight data Hk for detecting the case where the part is mounted by mistake are registered. Is done.

  Similarly, the counter m is updated until m> M, and the loop of ST202 to ST211 is repeated for the part specified by the hourly m. As a result, for all product numbers in the group, a defective model is registered in order to cope with a case where components of other product numbers belonging to the same group are erroneously mounted.

FIG. 10 shows an inspection procedure using each model and weight data.
First, in ST301, an image generated by imaging a substrate to be inspected is input. Thereafter, based on the component information registered in the memory 15, the following processing is executed while paying attention to the mounted components in order.

  The component information includes data having the same contents as the CAD data in FIG. In ST302, the group data storage unit 112 is searched with the part number in this data, and the group to which the part under consideration belongs and the number M of the part numbers registered in this group are recognized. In ST303, the model storage unit 111 is searched by the part number of the component under consideration, and the corresponding good model is read out.

In ST304, “1” is set to the counter k indicating the number of readings of the defective model, and the maximum value S _max of the dissimilarity with respect to the good model is set to 0 as an initial value. In the next ST305, the k-th defective model Nk and the corresponding weight data Hk are read out. In ST306, the degree of difference S (o) k of the processing target image with respect to the good model is calculated by executing the above-described equation (1) using each weight included in the weight data Hk.

  In ST307, the dissimilarity S (n) k of the processing target image with respect to the defective model Nk is similarly calculated by the equation (1) using the weight data Hk.

In ST308, the dissimilarity S (o) k obtained in ST306 is compared with the maximum value _Smax . If S (o) k> _Smax , the process proceeds to ST309, and _Smax is rewritten to the current value of S (o) k. If S (o) k ≦ _Smax , ST309 is skipped.

  In ST310, the differences S (o) k and S (n) k obtained in ST306 and 307 are compared. If S (o) k <S (n) k, that is, if the similarity to the good model is higher, the process proceeds to ST311 to update the value of k, and then returns to ST305 via ST312.

Thereafter, the processes of ST305 to 312 are executed for all the defective models set, and when the similarity to the good model is higher than the similarity to any defective model, ST312 is “NO”. Then, the process proceeds to ST313, and the maximum value S _{max of the} degree of difference with respect to the good model at that time is compared with a predetermined threshold value Sth. Here, if S _max <Sth (that is, if the smallest of the similarities to the good model obtained using various weight data exceeds a certain reference value), the process proceeds to ST314, and the focus is on This part is determined to be a correct part.

On the other hand, in the above comparison processing, when S _max ≧ Sth, the process proceeds to ST 316 and determines that the component under attention is an incorrect component. Such a case may occur when a group of parts different from the correct part is erroneously mounted.

  If a component that belongs to the same group as the correct component is erroneously mounted, the determination when the comparison process of ST310 is performed for the defective model corresponding to the component is “NO”. In this case, the process proceeds to ST315, and it is checked whether or not the difference degree S (n) k is smaller than the threshold value Sth. If S (n) k <Sth, the process proceeds to ST316, and it is determined that the component of interest is an incorrect component.

  By executing the processing of ST302 to 316 for all the mounted components on the board, it is determined for each component whether the component is a correct component or an incorrect component. When the determination process for all the mounted components is completed, ST317 is “YES”, the process proceeds to ST318, and the inspection result for the entire board is output. This output destination can be sent to an external device (not shown) in addition to the monitor 6.

  In the above-described embodiment, the weight data is set for each of the R, G, and B color data for the defective model. However, the weight data is not limited to this, and the weight data is limited to the color data that is different from the non-defective model. Data may be set. For example, when a red character string is printed on a part, weight data is set for R, but weight data is not set for G and B. When weight data is set only for specific color data in this way, the similarity may be calculated using only the image data of the color for which the weight data is set. However, the present invention is not limited to this. For specific color data, weight data is set and the similarity is calculated. For other color data, the similarity is calculated without setting the weight data, and all similarities are calculated. You may ask for the sum of.

  The calculation of the similarity according to the above equation (1) is to obtain the sum of the differences between the corresponding pixels, but instead of this, a normalized correlation calculation may be executed. In this case as well, by setting an expression such that the higher the similarity is, the lower the value of the operation result is, and if the part to be inspected corresponds to one of the defective models, the similarity to the corresponding defective model is the highest. Since a result that is high can be obtained, it is possible to determine whether the part is correct.

When the above description is arranged, in parts inspection, the similarity to each model can be obtained by the following three methods.
(1) A method of obtaining a difference between corresponding pixels for each color data of R, G, and B using the above-described equation (1) and using the sum as a similarity.
The similarity obtained by this method is hereinafter referred to as “similarity by comprehensive difference calculation”.
(2) A method of obtaining a difference between corresponding pixels only for color data (for example, R) for which weight data is set, and using the difference as a similarity.
The similarity obtained by this method is hereinafter referred to as “similarity based on a specific color difference calculation”.
(3) A method of obtaining similarity by normalized correlation calculation.
The similarity obtained by this method is hereinafter referred to as “similarity by normalized correlation calculation”.

  In actual component inspection, among the similarities obtained by the above three arithmetic processes, the difference between the similarity to the model that matches the image to be inspected and the similarity to other models becomes as significant as possible. It is desirable to select. Various components are mounted on the substrate, and which of the above three similarities is suitable depends on the type of component. Therefore, if the similarity calculation method is selected in advance for each group, the accuracy of the inspection can be further improved.

  As for the method of (2), as described above, for color data other than the specific color data, the similarity without setting the weight data is obtained, and these are used as the similarity using the weight data of the specific color data. The method of adding to can be replaced. In the method (3), as in the case of similarity by difference calculation, a correlation value is obtained using R, G, and B color data, and a correlation value is obtained using only specific color data. May be classified.

  Hereinafter, the process of automatically selecting the similarity calculation method will be specifically described with reference to FIGS. In this example, attention is paid to a specific product number for which registration of the good model and the defective model and the weight data is completed by the processing of FIGS. 8 and 9, and any of the similarities by the methods (1) to (3) above Is automatically determined using a non-defective image of the substrate.

  In FIG. 11, OK <b> 1 to OK <b> 3 are names of sample images (hereinafter referred to as “good samples”) extracted from the non-defective images for the part of the product number under attention. Further, NG1 to NG5 are names of sample images (hereinafter referred to as “defective samples”) extracted from non-defective products for parts of other product numbers that belong to the same group as the product number of interest. Each sample is individually cut out from a non-defective image. Therefore, even the samples with the same part number have slightly different colors. As for defective samples, when there are a plurality of product numbers other than the part under consideration, it is desirable that defective samples are set for all of these product numbers.

  FIG. 11 shows the result of obtaining the similarity of each sample with respect to the non-defective product model of the part of the product number of interest by the above three calculation methods. Note that “OK1” of the good samples is used to create the good model, and thus all the calculation results are “0” (that is, the similarity is 100%).

FIG. 12 shows parameters used for discrimination processing on these similarity graphs.
In the figure, the mark ● indicates the similarity of the non-defective sample shown in FIG. 11, and the mark ■ indicates the similarity of the defective sample. The thick alternate long and short dash line is the intermediate value of the similarity distribution range of the non-defective samples (average value of the maximum and minimum values), and the thick two-dotted line is the intermediate value of the similarity distribution range of the defective samples. It is.

  In each graph, D is the difference between the intermediate value of the good sample and the intermediate value of the defective sample. R is a minimum similarity value (a numerical value is the maximum value) of a good sample to a maximum similarity value (a difference from the minimum value is a numerical value) of a defective sample.

  D is considered to indicate the magnitude of the average difference in similarity between the good sample and the bad sample. R is considered to indicate the size of the gap between the distribution range of good samples and the distribution range of defective samples. It is considered that the larger R is, the easier it is to distinguish between correct parts and wrong parts. Therefore, in this embodiment, for each calculation of (1) to (3), the ratio of R to D (R / D) is obtained, and the calculation process that maximizes the ratio is selected. .

  FIG. 13 shows the results of obtaining R, D, and R / D for each calculation result shown in FIG. According to this example, the R / D becomes the largest when the degree of similarity based on the correlation value of (3) is obtained. Therefore, the method of (3) is also selected as the operation to be performed at the time of inspection. Is done.

  The selection process of the similarity calculation method shown in FIGS. 11 to 13 is preferably performed for each type of substrate to be inspected, similarly to the model registration process.

It is a figure which shows the electrical structure and optical structure of a board | substrate external appearance automatic inspection apparatus. It is explanatory drawing which shows the example of arrangement | positioning of the components on a board | substrate. It is explanatory drawing which shows the content of the CAD data of each component of FIG. It is explanatory drawing which shows the result of having grouped the product number of each component of FIG. It is a functional block diagram of a component inspection system. It is explanatory drawing which shows the example of registration of a good model. It is explanatory drawing which shows the example of a setting of weight data. It is a flowchart which shows the process sequence regarding the classification of parts, and registration of an OK model. It is a flowchart which shows the process sequence regarding registration of a defect model and weight data. It is a flowchart which shows the flow of a test | inspection. It is a table which shows the example which calculated | required the similarity with respect to a sample image with three kinds of arithmetic processing methods. It is a graph which shows the method of calculating | requiring the parameter used for the selection process of the calculation method of similarity from the calculation result of FIG. 12 is a table showing R, D, and R / D values obtained from the calculation result of FIG. 11.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Controller 2 Camera 10 Control part 15 Memory 100 Component inspection system 101 CAD data input part 103 Part classification process part 104 Model registration part 105 Weight data setting part 107 Inspection execution part 111 Model storage part 112 Group data storage part 113 Weight data storage part

Claims (4)

A method of registering a model for inspecting whether or not a correct component is mounted on the substrate by using a feature related to the component in the generated image by imaging a substrate that has undergone a component mounting process,
Step A for grouping each mounted component based on component type and size using the board design data,
Step B for registering a good model of an image of each component on the board,
Pay attention to the good model registered in step B in order, and select a good model of another part classified into the same group as the part to which the good model being noticed is applied as a defective model corresponding to the good model being noticed. Step C to set and register as
A model registration method for inspecting a mounted component, characterized in that each of the steps is executed. A method of inspecting whether or not a correct component is mounted on the substrate by imaging a substrate that has undergone a component mounting process, and using features related to the component in the generated image,
Step A for grouping each mounted component based on component type and size using the board design data,
Step B for registering a good model of an image of each component on the board,
Pay attention to the good model registered in step B in order, and select a good model of another part classified into the same group as the part to which the good model being noticed is applied as a defective model corresponding to the good model being noticed. Step C to set and register as
Each step is performed before inspection,
In the inspection, for each part to be inspected, the similarity of the image of the part to the good model and the defective model is obtained for the image of the part, and the similarity to the good model is better than the similarity to any defective model When it becomes high, it is determined that the part to be inspected is a correct part, and when the similarity to any defective model is higher than the similarity to the good model, the part to be inspected is incorrect. A method for inspecting a mounted component that is identified as a component. The method of claim 2, wherein
Before the inspection, a step of setting a weight corresponding to a difference with respect to a corresponding pixel of the good model for each pixel of the defective model set in the step C, and registering these weights in association with the defective model D is executed further,
In the inspection,
A value obtained by weighting the difference between corresponding pixels by the weight registered for the model in Step D between the inspection target image and each defective model is used so that the value increases as the difference between the images increases. By calculating the similarity set for the defect model, the similarity of the image to be inspected with respect to each defective model is obtained, and for the good model, the calculation is executed a plurality of times using the same weights set for each defective model. By calculating a plurality of similarities of the inspection target image with respect to the good model,
The similarity for the good model and the similarity for the bad model are compared and combined for each one calculated using the same weight, and in all combinations, the similarity for the good model is better than the similarity for the bad model. When it is high, it is determined that the part to be inspected is a correct part, and in any combination, when the similarity to the defective model is higher than the similarity to the good model, the part to be inspected is erroneous. A method for inspecting a mounted component that determines that the component is a damaged component. An apparatus for imaging a board that has undergone a component mounting process and inspecting whether or not a correct component is mounted on the board by using a feature related to the component in the generated image,
A memory in which good models of images of each part on the board to be inspected are registered,
A classifying means that accepts input of design data of a board to be inspected, and uses the input design data to group each mounted component based on component type and size;
Pay attention to the good model registered in the memory in order, and select a good model of another part classified into the same group as the part to which the good model being noticed is applied, and a defective model corresponding to the good model being noticed Defective model setting means to set as,
When the board to be inspected is imaged, for each part to be inspected in the generated image, the similarity of the image of the part to the corresponding good model and defective model is obtained, and each part is based on these similarities And a determination means for determining whether or not is a correct part,
When the similarity to the good model is higher than the similarity to any defective model, the determination means determines that the part to be inspected is a correct part, and any of the similarities to the good model A board inspection apparatus that determines that the part to be inspected is an incorrect part when the degree of similarity with respect to the defective model is higher.