JP2020030145A - Inspection apparatus and inspection system - Google Patents

Abstract

To provide an inspection apparatus and an inspection system that can quickly perform inspection of a plurality of kinds of objects.SOLUTION: An inspection apparatus according to an embodiment comprises: a plurality of recognition processing units that extract image data of a corresponding object from the image data; a plurality of inspection processing units that perform inspection corresponding to the object on the basis of the image data of the object; and a control unit that controls the plurality of recognition processing units and the plurality of inspection processing units.SELECTED DRAWING: Figure 5

Description

  The present invention relates to an inspection device and an inspection system.

  2. Description of the Related Art Conventionally, an inspection system has been used in which an object conveyed on a manufacturing line is photographed by a photographing device, and the object is inspected based on obtained image data. The inspection system can perform an inspection of the target object by recognizing the target object based on the image data and applying a prepared inspection model to the recognized target object.

JP-A-11-110560

  However, in the conventional inspection system, the recognition of the object and the application of the inspection model are performed by one software. Therefore, when a plurality of types of objects are photographed at the same time, it takes time to perform the processing, and it is difficult to perform the inspection. Became. For this reason, when using the conventional inspection system, it is necessary to align the objects so that a plurality of types of objects are not photographed simultaneously on the production line.

  The present invention has been made in view of the above problems, and has as its object to provide an inspection apparatus and an inspection system that can quickly execute inspection of a plurality of types of objects.

  An inspection device according to one embodiment performs a plurality of recognition processing units that extract image data of a corresponding object from image data, and performs an inspection corresponding to the object based on the image data of the object. A plurality of inspection processing units, and a control unit that controls the plurality of recognition processing units and the plurality of inspection processing units are provided.

  According to each embodiment of the present invention, it is possible to provide an inspection apparatus and an inspection system that can quickly execute inspection of a plurality of types of objects.

The figure which shows the schematic structure of an inspection system. The figure which shows an example of a test model management table. The figure which shows an example of an inspection device management table. The figure which shows an example of an inspection device management table. FIG. 3 is a diagram illustrating an example of a hardware configuration of the inspection device. FIG. 2 is a diagram illustrating an example of a hardware configuration of a learning device. 5 is a flowchart illustrating an example of an operation of the inspection device and the imaging device. 9 is a flowchart illustrating an example of the operation of the learning device. 9 is a flowchart illustrating an example of an operation of the management device.

  Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the description of the specification and the drawings according to the embodiments, components having substantially the same function and configuration are denoted by the same reference numerals, and overlapping descriptions will be omitted.

  An inspection system 100 according to one embodiment will be described with reference to FIGS. The inspection system 100 according to the present embodiment is a system that performs inspection of an object based on image data of the object conveyed on a manufacturing line. Object inspection includes defect inspection, scratch inspection, dirt inspection, foreign matter inspection, orientation discrimination inspection, front and back inspection, shape inspection, foreign matter contamination inspection, character recognition / character collation inspection, dimension inspection, color identification inspection, and positioning inspection. Including. In the present embodiment, it is assumed that a plurality of types of objects are transported on a production line.

  First, a schematic configuration of an inspection system 100 according to the present embodiment will be described. FIG. 1 is a diagram showing a schematic configuration of the inspection system 100. The inspection system 100 in FIG. 1 includes an inspection device 1, a learning device 2, a management device 3, and an imaging device 4. The inspection device 1, the learning device 2, the management device 3, and the imaging device 4 are connected by a network such as a LAN (Local Area Network) or the Internet. Hereinafter, as shown in FIG. 1, three types of objects P1 to P3 are transported on the manufacturing line, an inspection T1 is performed on the object P1, an inspection T2 is performed on the object P2, and A case where the inspections T3 and T4 are performed on the object P3 will be described as an example.

  The inspection device 1 is a computer that executes an inspection T of the object P based on image data input from the imaging device 4. The inspection device 1 is a PC (Personal Computer), a server, or a dedicated terminal, but is not limited thereto. Further, the inspection device 1 may be configured by a plurality of computers connected by a network. The inspection device 1 performs an inspection T set in advance for each type of the object P. The inspection apparatus 1 may execute a plurality of inspections T on one type of target P as in the inspections T3 and T4 described above. When the inspection device 1 executes the inspection T, the inspection device 1 outputs inspection result data. The inspection result data is data including image data of the inspection target object P and data indicating the inspection result (label).

  As described above, when the inspection T1 is set as the inspection of the object P1, the inspection T2 is set as the inspection of the object P2, and the inspections T3 and T4 are set as the inspection of the object P3, the inspection apparatus 1 performs four types of inspections. Execute In this case, when receiving the image data of the object P1 from the photographing device 4, the inspection device 1 executes the inspection T1 and outputs the inspection result data of the inspection T1. The inspection result data output from the inspection device 1 is input to the learning device 2.

  The learning device 2 learns the relationship between the image data of the object P and the test result of the test T based on the test result data of the test T input from one or more test devices 1 (supervised learning). , A computer that generates an inspection model for performing the inspection T on the object P. The learning device 2 is a PC, a server, or a dedicated terminal, but is not limited thereto. Further, the learning device 2 may be configured by a plurality of computers connected by a network.

  The inspection model is a mathematical model that outputs an inspection result according to the image data. Generation of the inspection model corresponds to optimization of the parameters of the inspection model. The learning device 2 can use any machine learning method (algorithm) that can optimize the parameters of the inspection model. Machine learning methods include support vector machines, decision trees, random forests, neural networks, deep learning, and k-nearest neighbors. The learning device 2 generates an inspection model for each type of the object P and each type of the inspection T.

  Upon receiving the test result data, the learning device 2 selects learning data from the test result data. The learning data is test result data for which the test result is considered to be correct. The learning device 2 inputs the image data included in the test result data to the test model of the test T prepared in advance for selecting the learning data, and acquires the test result of the test T. It is preferable that the test model for selecting learning data is a test model different from the test model used by the test apparatus 1. When the test result obtained by the test model for selecting the learning data and the test result included in the test result data match, the learning device 2 determines that the test result included in the test result data is correct, and Inspection result data is selected as learning data. This is because when the test results obtained by a plurality of test models are the same, the test results are considered to be correct. The method of selecting learning data is not limited to the above example. For example, the learning data may be manually selected by the user of the learning device 2 from the test result data. The selection of the learning data may be performed by the management device 3. When the learning data is updated, the learning device 2 generates a test model based on the updated learning data, and outputs the test model. The test model output by the learning device 2 is input to the management device 3.

  The management device 3 is a computer that manages the test model input from the learning device 2. The management device 3 is a PC, a server, or a dedicated terminal, but is not limited thereto. Further, the management device 3 may be configured by a plurality of computers connected by a network.

  When a new test model is input from the learning device 2, the management device 3 calculates the test accuracy of the new test model based on test data prepared in advance. The test data is data including image data of the object P and data indicating an inspection result considered to be correct. The management device 3 compares the inspection result obtained by inputting the image data of the object P included in the test data into the inspection model, and the inspection result corresponding to the image data of the object P included in the test data. By performing the comparison, the detection accuracy can be calculated. Further, the management device 3 assigns a version to the new inspection model. Then, the management device 3 stores the new inspection model, the inspection accuracy of the new inspection model, and the version of the new inspection model in association with each other.

  The management device 3 includes a test model management table that manages a test model for each version of the test T. The test model management table is provided on the HDD of the management device 3. FIG. 2 is a diagram illustrating an example of the inspection model management table. In the example of FIG. 2, the inspection T, the inspection model version (ver), the inspection model (parameter value), and the inspection accuracy (Ac) are associated with each other. In FIG. 2, a1 and b1 are parameters of the inspection model of the inspection T1, a2 and b2 are parameters of the inspection model of the inspection T2, and a3 and b3 are parameters of the inspection model of the inspection T3. For example, version 1 of the test model of test T1 is a test model of a1 = 1 and b1 = 2, and has a test accuracy of 70%. The management device 3 manages, for each version, test models of a plurality of types of tests T as in the example of FIG.

  In addition, the management device 3 includes an inspection device management table that manages the version of the inspection model used by the inspection device 1 for each inspection T. The inspection device management table is provided on the HDD of the management device 3. FIG. 3 is a diagram illustrating an example of the inspection device 1 management table. In the example of FIG. 3, the identification information (ID) of the inspection apparatus 1, the identification information (id) of the inspection processing unit 123, the type of the inspection T, and the version (ver) of the inspection model are associated with each other. In FIG. 3, the inspection apparatus D1 (the inspection apparatus 1 whose identification information is D1) is configured such that the inspection processing units 123a to 123d (the inspection processing units 123 whose identification information is a to d) perform inspections T1 to T1 on the object P, respectively. The inspection apparatus 1 executes T4, and the inspection processing unit 123a uses version 1 as an inspection model of the inspection T1. The management device 3 manages the version of the inspection model used by each inspection processing unit 123 for each inspection T for a plurality of inspection devices 1 as in the example of FIG.

  Further, the management device 3 updates the inspection model used by the inspection device 1 at a predetermined timing. The predetermined timing is a timing at which a new inspection model generated by the learning device 2 is input to the management device 3, a timing at which the inspection device 1 is activated, a timing at which an inspection board 12 of the inspection device 1 described later is activated, and This includes the timing requested by the user of the device 3. Further, the management device 3 may periodically update the inspection model used by the inspection device 1.

  Specifically, the management device 3 updates the test model of the test T used by the test device 1 to a version having the highest test accuracy among the test models of the test T. 2 and 3, the test model of the test T1 used by the test apparatus D1 is updated to version 2, the test model of the test T2 is updated to version 2, and the test model of the test T3 is maintained at version 1. You. That is, the test model management table in FIG. 3 is updated as shown in FIG. When the management device 3 updates the test model, the test device 1 can execute the test T using the test model with the highest test accuracy generated by the learning device 2.

  The image capturing device 4 is a camera that captures an image of the object P, and is installed on a production line so that the image of the object P can be captured. The imaging device 4 is connected to the inspection device 1 by wire or wirelessly. The photographing device 4 may perform photographing periodically (for example, every one second) or at a predetermined timing, and output image data obtained by photographing, or continuously photograph a moving image, and perform photographing. The moving image data (image data) obtained by the above may be output. The predetermined timing is, for example, a timing at which the target object P arrives at a shooting position of the shooting device 4. A sensor (an image sensor, an infrared sensor, or the like) for detecting that the object P has arrived at the image capturing position is provided on the manufacturing line, and the sensor notifies the image capturing device 4 of the arrival of the object P. Photographing can be performed at the timing. Image data obtained by the photographing of the photographing device 4 is input to the inspection device 1. The photographing device 4 is a camera provided with a CCD image sensor or a CMOS image sensor as an image sensor, but is not limited to this. Note that a plurality of photographing devices 4 may be installed on the production line, and image data may be input to the inspection device 1 from each photographing device.

  In the above, the case where the inspection device 1, the learning device 2, and the management device 3 are independent computers has been described as an example, but at least two of the inspection device 1, the learning device 2, and the management device 3 are: It may be constituted by one computer.

  Next, a hardware configuration of the inspection system 100 will be described. FIG. 5 is a diagram illustrating an example of a hardware configuration of the inspection device 1. The inspection apparatus 1 of FIG. 5 includes a main board (main board) 11 and an inspection board (inspection board) 12. The main board 11 is a board that executes general-purpose processing, and the inspection board 12 is a board that executes an inspection T. By configuring the main board 11 and the inspection board 12 separately, the inspection apparatus 1 can be easily and inexpensively configured simply by connecting the inspection board 12 to a general-purpose computer (such as a PC) having the main board 11. .

  The main board 11 corresponds to a control unit that controls the recognition processing unit 11 and the inspection processing unit 123 described below. The main board 11 includes a CPU (Central Processing Unit) 111, a RAM (Random Access Memory) 112, a ROM (Read Only Memory) 113, a HDD (Hard Disk Drive) 114, a connection interface 115, an input device 116, , A display device 117, a communication interface 118, and a bus 119.

  The CPU 111 controls each component of the inspection apparatus 1 by executing the program, and realizes the function of the inspection apparatus 1. The program executed by the CPU 111 can be recorded on an arbitrary computer-readable recording medium such as a CD (Compact Disk), a DVD, and a flash memory.

  The RAM 112 provides the CPU 111 with a work area.

  The ROM 113 stores programs executed by the CPU 111 and various data.

  The HDD 114 stores programs executed by the CPU 111 and various data (image data and the like). The inspection device 1 may include an SSD together with or instead of the HDD 114, but is not limited thereto.

  The connection interface 115 connects the main board 11 to the inspection board 12 and controls communication between the boards.

  The input device 116 inputs information according to a user operation to the inspection device 1. The input device 116 is a touchpad, a keyboard, a mouse, or a hardware button, but is not limited thereto.

  The display device 117 displays a screen according to a user operation. The display device 117 is a liquid crystal display or an organic EL display, but is not limited thereto. The inspection device 1 may include a touch panel in which the touch pad and the display device 117 are integrated.

  The communication interface 118 connects the inspection device 1 to a network and controls communication with an external device on the network. The inspection device 1 communicates with an external device such as the imaging device 4 and the learning device 2 via the communication interface 118.

  The bus 119 connects the CPU 111, the RAM 112, the ROM 113, the HDD 114, the connection interface 115, the input device 116, the display device 117, and the communication interface 118 to each other.

  The inspection board 12 includes a connection interface 121, recognition processing units 122a to 122c, inspection processing units 123a to 123d, and a bus 124.

  The connection interface 121 connects the inspection board 12 to the main board 11 and controls communication between the boards.

  The recognition processing unit 122 is a dedicated processor that executes recognition processing of the corresponding target object P based on image data input from the imaging device 4. Specifically, the recognition processing unit 122 detects and extracts image data of the target object P from the image data input from the imaging device 4. The inspection device 1 can execute the recognition processing by any known method such as pattern matching and deep learning. The recognition processing unit 122 is a CPU, a GPU, a field programmable gate array (FPGA), or an application specific integrated circuit (ASIC), but is not limited thereto. The recognition processing unit 122 may have a built-in ROM and RAM, or a ROM and a RAM corresponding to the recognition processing unit 122 may be mounted on the inspection board 12.

  The recognition processing unit 122 is provided, for example, for each type of the target object P transported on the manufacturing line, that is, for each type of the target object P to be recognized. In the present embodiment, since three types of objects P1 to P3 are transported on the production line, three recognition processing units 122 are provided. The recognition processing unit 122a executes a recognition process of the object P1, the recognition processing unit 122b executes a recognition process of the object P2, and the recognition processing unit 122c executes a recognition process of the object P3. Upon extracting the image data of the object P, the recognition processing unit 122 outputs the image data. The image data output by the recognition processing unit 122 is input to the inspection processing unit 123. The number of the recognition processing units 122 can be arbitrarily designed according to the type of the target object P. In addition, the recognition processing of a plurality of types of objects P can be executed by one recognition processing unit 122.

  The inspection processing unit 123 is a dedicated processor (mainly a hardware accelerator, which executes an inspection T corresponding to the object P based on the image data of the object P input from the recognition processing unit 122. It is not limited). Specifically, the inspection processing unit 123 inputs the image data of the object P to the inspection model of the inspection T, and outputs an inspection result. The inspection processing unit 123 is a CPU, a GPU, an FPGA, or an ASIC, but is not limited thereto. The inspection processing unit 123 may include a ROM and a RAM, or a ROM and a RAM corresponding to the inspection processing unit 123 may be mounted on the inspection board 12.

  The inspection processing unit 123 is provided, for example, for each inspection T performed by the inspection device 1. In the present embodiment, four inspection processing units 123 are provided because the inspection device 1 performs the inspections T1 to T4. The inspection processing unit 123a executes the inspection T1, the inspection processing unit 123b executes the inspection T2, the inspection processing unit 123c executes the inspection T3, and the inspection processing unit 123d executes the inspection T4. When performing the test T, the test processing unit 123 outputs test result data. The inspection result data output by the inspection processing unit 123 is input to the learning device 2. The number of the inspection processing units 123 can be arbitrarily designed according to the type of the inspection T. Further, a plurality of types of inspections T can be executed by one inspection processing unit 123.

  The bus 124 connects the connection interface 121, the recognition processing unit 122, and the inspection processing unit 123 to each other.

  The hardware configuration of the inspection device 1 is not limited to the example in FIG. As described above, the inspection device 1 may be configured by a plurality of computers. Further, the inspection device 1 may be configured by a single substrate.

  FIG. 6 is a diagram illustrating an example of a hardware configuration of the learning device 2. The learning device 2 in FIG. 6 includes a CPU 21, a RAM 22, a ROM 23, an HDD 24, an input device 25, a display device 26, a communication interface 27, and a bus 28.

  The CPU 21 controls each configuration of the learning device 2 by executing the program, and realizes the function of the learning device 2. The program executed by the CPU 21 can be recorded on an arbitrary computer-readable recording medium such as a CD, a DVD, and a flash memory. The learning device 2 may include a plurality of CPUs 21 so as to execute machine learning at high speed, or may include one or a plurality of GPUs (Graphics Processing Units) instead of the CPU 21.

  The RAM 22 provides the CPU 21 with a work area.

  The ROM 23 stores programs executed by the CPU 21 and various data.

  The HDD 24 stores programs executed by the CPU 21 and various data (such as inspection result data). The learning device 2 may include an SSD together with or instead of the HDD 24.

  The input device 25 inputs information according to a user operation to the learning device 2. The input device 25 is a touchpad, a keyboard, a mouse, or a hardware button, but is not limited thereto.

  The display device 26 displays a screen according to a user operation. The display device 26 is a liquid crystal display or an organic EL display, but is not limited thereto. The learning device 2 may include a touch panel in which the touch pad and the display device 26 are integrated.

  The communication interface 27 connects the learning device 2 to a network and controls communication with an external device on the network. The learning device 2 communicates with an external device such as the inspection device 1 and the management device 3 via the communication interface 27.

  The bus 28 connects the CPU 21, the RAM 22, the ROM 23, the HDD 24, the input device 25, the display device 26, and the communication interface 27 to each other.

  The hardware configuration of the learning device 2 is not limited to the example in FIG. As described above, the learning device 2 may be configured by a plurality of computers. Further, the hardware configuration of the management device 3 is the same as that of the learning device 2, and a description thereof will be omitted.

  Next, operations of the inspection device 1 and the imaging device 4 will be described. FIG. 7 is a flowchart illustrating an example of the operation of the inspection device 1 and the imaging device 4. Hereinafter, a case will be described as an example in which the image capturing apparatus 4 periodically performs image capturing or captures a moving image. The inspection device 1 and the imaging device 4 execute the operation of FIG. 7 at predetermined time intervals.

  First, the photographing device 4 photographs the production line (step S101), and outputs the obtained image data. When the object P is included in the imaging range of the imaging device 4, image data including image data of the object P is output. On the other hand, when the object P is not included in the imaging range of the imaging device 4, image data that does not include the image data of the object P is output. The output image data is input to the main board 11.

  When the image data is input to the inspection device 1, the main board 11 performs preprocessing on the image data (Step S102). The pre-processing is an image processing performed in advance to improve the accuracy of the recognition processing and the inspection T, and includes a filter processing and a mask processing. After executing the preprocessing, the main board 11 inputs the preprocessed image data to the recognition processing unit 122a, and instructs the recognition processing unit 122a to execute the recognition processing.

  Upon receiving the image data from the main board 11, the recognition processing unit 122a executes a recognition process for the target P1 (Step S103). After executing the recognition process, the recognition processing unit 122a inputs the detection result of the target P1 (whether or not the target P1 is detected) to the main board 11. When detecting the target object P1, the recognition processing unit 122a extracts the image data of the target object P1 from the input image data and inputs the image data to the main board 11. When detecting a plurality of objects P1, the recognition processing unit 122a may input a detection result and image data to the main board 11 for each object P1. When the detection result is input from the recognition processing unit 122a, the main board 11 inputs the preprocessed image data to the recognition processing unit 122b, and instructs the recognition processing unit 122b to execute the recognition processing.

  Upon receiving the image data from the main board 11, the recognition processing unit 122b executes a recognition process for the target P2 (Step S104). After executing the recognition process, the recognition processing unit 122b inputs the detection result of the target object P2 (whether or not the target object P2 has been detected) to the main board 11. Further, when detecting the target P2, the recognition processing unit 122b extracts the image data of the target P2 from the input image data and inputs the image data to the main board 11. When detecting a plurality of targets P2, the recognition processing unit 122b may input a detection result and image data to the main board 11 for each target P2. When the detection result is input from the recognition processing unit 122b, the main board 11 inputs the preprocessed image data to the recognition processing unit 122c, and instructs the recognition processing unit 122c to execute the recognition processing.

  Upon receiving the image data from the main board 11, the recognition processing unit 122c executes a recognition process for the target object P3 (Step S105). After executing the recognition process, the recognition processing unit 122c inputs the detection result of the target P3 (whether or not the target P3 is detected) to the main board 11. When detecting the target P3, the recognition processing unit 122c extracts the image data of the target P3 from the input image data and inputs the image data to the main board 11. When detecting a plurality of targets P3, the recognition processing unit 122c may input a detection result and image data to the main board 11 for each target P3.

  As described above, the inspection device 1 executes the recognition processing of one or more corresponding objects P by the respective recognition processing units 122. Therefore, the inspection apparatus 1 can cope with a change in the type of the target object P only by replacing the recognition processing unit 122. In the example of FIG. 7, the recognition processes of the objects P1 to P3 are sequentially executed by the recognition processing units 122a to 122c. However, the recognition processes of the objects P1 to P3 are performed in parallel by the recognition processing units 122a to 122c. May be executed. Thereby, the time required for the recognition processing of the objects P1 to P3 can be reduced.

  When the detection result is input from the recognition processing unit 122c, the main board 11 performs the following processing according to the detection results of the objects P1 to P3.

  When the object P1 is detected by the recognition processing unit 122a (step S106: YES), the main board 11 inputs the image data of the object P1 input from the recognition processing unit 122a to the inspection processing unit 123a, and performs the inspection processing. The unit 123a is instructed to execute the test T1.

  When the image data of the object P1 is input from the main board 11, the inspection processing unit 123a executes the inspection T1 on the object P1 (step S107). That is, the inspection processing unit 123a inputs the image data of the object P1 to the inspection model of the inspection T1, and outputs an inspection result. When performing the test T1, the test processing unit 123a inputs data indicating the test result to the main board 11. When image data of a plurality of objects P1 is input, the inspection processing unit 123a performs an inspection T1 on each of the objects P1, and inputs data indicating an inspection result for each of the objects P1 to the main board 11. I just need. When the data indicating the detection result is input from the inspection processing unit 123a, the main board 11 generates the inspection result data of the inspection T1 including the data indicating the inspection result and the image data of the object P1. If the target P1 is not detected by the recognition processing unit 122a (step S106: NO), the main board 11 does not receive image data of the target P1 from the recognition processing unit 122a. Therefore, the inspection T1 (Step S107) is omitted.

  On the other hand, when the target P2 is detected by the recognition processing unit 122b (step S108: YES), the main board 11 inputs the image data of the target P2 input from the recognition processing unit 122b to the inspection processing unit 123b, Instruct the inspection processing unit 123b to execute the inspection T2.

  Upon receiving the image data of the object P2 from the main board 11, the inspection processing unit 123b performs an inspection T2 on the object P2 (step S109). That is, the inspection processing unit 123b inputs the image data of the object P2 to the inspection model of the inspection T2, and outputs an inspection result. When performing the test T2, the test processing unit 123b inputs data indicating the test result to the main board 11. When image data of a plurality of objects P2 is input, the inspection processing unit 123b executes an inspection T2 on each of the objects P2, and inputs data indicating an inspection result for each of the objects P2 to the main board 11. I just need. When the data indicating the detection result is input from the inspection processing unit 123b, the main board 11 generates the inspection result data of the inspection T2 including the data indicating the inspection result and the image data of the object P2. When the target P2 is not detected by the recognition processing unit 122b (step S108: NO), the main board 11 does not receive image data of the target P2 from the recognition processing unit 122b. Therefore, the inspection T2 (Step S109) is omitted.

  On the other hand, when the object P3 is detected by the recognition processing unit 122c (step S110: YES), the main board 11 inputs the image data of the object P3 input from the recognition processing unit 122c to the inspection processing unit 123c. Instruct the inspection processing unit 123c to execute the inspection T3.

  Upon receiving the image data of the object P3 from the main board 11, the inspection processing unit 123c performs an inspection T3 on the object P3 (step S111). That is, the inspection processing unit 123c inputs the image data of the object P3 to the inspection model of the inspection T3, and outputs an inspection result. When performing the test T3, the test processing unit 123c inputs data indicating a test result to the main board 11. When image data of a plurality of objects P3 is input, the inspection processing unit 123c executes an inspection T3 on each of the objects P3, and inputs data indicating an inspection result for each of the objects P3 to the main board 11. I just need. When the data indicating the detection result is input from the inspection processing unit 123c, the main board 11 generates the inspection result data of the inspection T3 including the data indicating the inspection result and the image data of the object P3. Further, the main board 11 inputs the image data of the target object P3 to the inspection processing unit 123d, and instructs the inspection processing unit 123d to execute the inspection T4.

  When the image data of the object P3 is input from the main board 11, the inspection processing unit 123d performs an inspection T4 on the object P3 (Step S112). That is, the inspection processing unit 123d inputs the image data of the object P3 to the inspection model of the inspection T4, and outputs an inspection result. When performing the test T4, the test processing unit 123d inputs data indicating a test result to the main board 11. When image data of a plurality of objects P3 is input, the inspection processing unit 123d executes an inspection T4 on each of the objects P3, and inputs data indicating an inspection result for each of the objects P3 to the main board 11. I just need. When the data indicating the detection result is input from the inspection processing unit 123d, the main board 11 generates the inspection result data of the inspection T4 including the data indicating the inspection result and the image data of the object P3. If the target P3 is not detected by the recognition processing unit 122c (step S110: NO), the main board 11 does not receive the image data of the target P3 from the recognition processing unit 122c. Therefore, the inspections T3 and T4 (steps S111 and S112) are omitted.

  When all the inspection result data of the inspection T of the object P detected by the recognition processing unit 122 is generated, the main board 11 outputs the generated inspection result data (Step S113). The main board 11 may temporarily store the inspection result data and output the inspection result data collectively at predetermined time intervals. The inspection result data output by the main board 11 is input to the learning device 2.

  As described above, the inspection apparatus 1 executes each inspection T by each inspection processing unit 123. Therefore, the inspection apparatus 1 can cope with a change in the type of the inspection T only by replacing the inspection processing unit 123.

  Further, even when a plurality of types of objects P are simultaneously imaged by the imaging device 4, the inspection device 1 performs a simple process of inputting image data of each object P to the corresponding inspection processing unit 123. The inspection T of each object P can be executed, and each inspection T can be executed by a dedicated processor. Therefore, the inspection device 1 can quickly execute the inspections T of the plurality of types of objects P. Further, by using the inspection apparatus 1, there is no need to align the objects P so that a plurality of types of objects P are not simultaneously photographed on the production line, so that the configuration of the production line can be simplified. .

  In the example of FIG. 7, the tests T1 to T4 are sequentially performed by the test processing units 123a to 123d, but the tests T1 to T4 may be performed in parallel by the test processing units 123a to 123d. Thereby, the time required for the inspections T1 to T4 can be reduced. Further, the main board 11 may cause the display device 117 to display the inspection results of the inspections T1 to T4 and the images of the objects P1 to P3 to be inspected.

  Next, the operation of the learning device 2 will be described. FIG. 8 is a flowchart illustrating an example of the operation of the learning device 2. The learning device 2 performs the operation of FIG. 8 at predetermined time intervals.

  First, the learning device 2 checks whether the learning timing has come (step S201). The learning timing may be at a predetermined time interval, a timing at which a fixed number of test result data is input, or a timing at which an instruction from a user is received. If the learning timing has not arrived (step S201: NO), the learning device 2 ends the operation.

  On the other hand, when the learning timing has arrived (step S201: YES), the learning device 2 executes the following operation for each test T.

  First, the learning device 2 checks whether new test result data of the test T has been input (step S202). The new test result data is test result data input to the learning device 2 after the test model of the test T was generated last time. When new test result data has not been input (step S202: NO), the learning device 2 ends the process for the test T.

  On the other hand, when new test result data is input (step S202: YES), the learning device 2 selects new learning data from the new test result data (step S203). The method of selecting the learning data is as described above.

  Upon selecting the new learning data, the learning device 2 learns the relationship between the image data of the object P and the inspection result of the inspection T based on the new learning data, and generates a new inspection model (Step S10). S204). At this time, the learning device 2 may use existing learning data together with new learning data. Upon generating a new test model of the test T, the learning device 2 outputs the test model (Step S205). The test model output by the learning device 2 is input to the management device 3.

  The learning device 2 executes the processing of steps S202 to S205 for each test T, thereby generating a new test model for each test T.

  As described above, the learning device 2 can generate a test model of the test T based on the test result data of the test T input from the test device 1.

  Next, the operation of the management device 3 will be described. FIG. 9 is a flowchart illustrating an example of the operation of the management device 3. The management device 3 executes the operation of FIG. 9 every predetermined time.

  First, the management device 3 checks whether a new test model has been input from the learning device 2 (step S301). If a new inspection model has not been input (step S301: NO), the process proceeds to step S305. On the other hand, when a new inspection model is input (step S301: YES), the management device 3 assigns a new version to the inspection model (step S302).

  Next, the management device 3 calculates the inspection accuracy of the new inspection model based on the test data (Step S303). When calculating the inspection accuracy, the management device 3 preferably performs the same pre-processing as the inspection device 1 on the image data of the object P included in the test data.

  Thereafter, the management device 3 stores the new inspection model, the inspection accuracy of the new inspection model, and the new inspection model version in association with each other (step S304). Thereby, as in the example of FIG. 2, the management device 3 can manage a plurality of versions of the test model of each test T.

  Subsequently, the management device 3 checks whether the update timing of the inspection model of the inspection device 1 has come (step S305). If the update timing of any of the inspection apparatuses 1 that manage the inspection model has not arrived (step S305: NO), the management apparatus 3 ends the operation.

  On the other hand, when the update timing of any of the inspection devices 1 whose inspection models are managed by the management device 3 has also come (step S305: YES), the management device 3 updates the inspection device 1 to be updated. Is obtained (step S306). For example, consider a case where the start time of the inspection apparatus 1 is set as the inspection model update timing. In this case, when activated, the inspection device 1 notifies the management device 3 of the fact together with the identification information. The management device 3 may acquire the identification information notified from the inspection device 1 as the identification information of the inspection device 1 to be updated. When the update time of each inspection device 1 is set in the management device 3, the management device 3 replaces the identification information of the inspection device 1 whose update time has arrived with the identification information of the inspection device 1 to be updated. Should be obtained as The method for acquiring the identification information of the inspection apparatus 1 is not limited to this.

  When acquiring the identification information of the inspection device 1 to be updated, the management device 3 executes the following operation for each inspection device 1 from which the identification information has been acquired.

  The management device 3 acquires the type of the test T executed by the test device 1 with reference to the test device management table (step S307). In the example of FIG. 3, the tests T1 to T4 are acquired as the types of the tests T executed by the test apparatus D1. When acquiring the type of the test T executed by the inspection device 1, the management device 3 executes the following operation for each acquired test T.

  The management device 3 checks whether the version of the inspection model of the inspection T used by the inspection device 1 has the highest inspection accuracy (step S308). Specifically, the management device 3 acquires the version of the test model of the test T used by the test device 1 with reference to the test device management table. Then, the management device 3 refers to the inspection model management table and checks whether the inspection accuracy of the acquired version is the highest among the versions of the inspection model of the inspection T. When the version of the inspection model of the inspection T used by the inspection apparatus 1 is the version having the highest inspection accuracy (step S308: YES), the management apparatus 3 ends the operation for the inspection T, and ends the next inspection T Start the operation for.

  On the other hand, when the version of the inspection model of the inspection T used by the inspection apparatus 1 is not the version with the highest inspection accuracy (step S308: NO), the management apparatus 3 refers to the inspection model management table and refers to the inspection model management table. Among the versions of the inspection model, the version having the highest inspection accuracy is selected (step S309).

  When selecting the version with the highest inspection accuracy, the management device 3 updates the inspection model of the inspection T used by the inspection device 1 to the selected version (Step S310). Specifically, the management device 3 transmits the selected version of the inspection model (parameter value) to the inspection device 1 and instructs the inspection device 1 to update the inspection model. The inspection device 1 updates the inspection model used by the inspection processing unit 123 that executes the inspection T according to the instruction. Then, the management device 3 updates the version of the test model of the test T of the test device 1 stored in the test device management table to the selected version. Thereafter, the management device 3 ends the operation for the test T, and starts the operation for the next test T.

  The management device 3 executes the processing of steps S307 to S310 on the test model of the test T of each test device 1, so that the test model of the test T of each test device 1 is updated.

  For example, in the examples of FIGS. 2 and 3, the inspection model of the inspection T1 of the inspection device D1 is updated from version 1 to version 2 (step S310). This is because the inspection accuracy (80%) of the inspection model version 2 of the inspection T1 is higher than the inspection accuracy (70%) of the inspection model version 1 used by the inspection apparatus D1 for the inspection T1 (step). S308: NO). By updating the inspection model of the inspection T1 from version 1 to version 2, the accuracy of the inspection T1 by the inspection device D1 can be improved.

  On the other hand, in the examples of FIGS. 2 and 3, the inspection model of the inspection T3 of the inspection device D1 is not updated from version 1 to version 2. That is, the inspection model of the inspection T3 of the inspection device D1 is maintained at version 1. This is because the inspection accuracy (95%) of version 1 of the inspection model used by the inspection apparatus D1 for the inspection T3 is higher than the inspection accuracy (90%) of version 2 of the inspection model of the inspection T3 (step S308: YES). By maintaining the inspection model of the inspection T3 as version 1, it is possible to suppress a decrease in accuracy of the inspection T3 by the inspection device D1.

  Thus, the management device 3 can manage the test model of the test T input from the learning device 2 for each version. In addition, the management device 3 can update the version of the inspection model of the inspection T used by the inspection device 1 to the version having the highest inspection accuracy for the plurality of inspection devices 1.

  As described above, according to the present embodiment, the inspection system 100 photographs the target P by the imaging device 4, executes the inspection T on the target P by the inspection device 1, and executes the inspection T on the target P by the learning device 2. A new inspection model is generated, and the version of the inspection model of the inspection T used by the inspection device 1 is updated by the management device 3. The inspection system 100 performs the inspection T on the object P by automatically and repeatedly executing the above processing by the inspection device 1, the learning device 2, the management device 3, and the imaging device 4, thereby performing the inspection T on the object P. Accuracy can be automatically improved.

  In the above, the case where the inspection system 100 performs the inspection T on a plurality of types of objects P has been described as an example, but the inspection system 100 performs the inspection T on a plurality of portions p of the object P. Is also good. In this case, the plurality of portions p to be inspected of the object P correspond to the plurality of objects P described above. Therefore, the recognition processing unit 122 of the inspection apparatus 1 may execute the recognition processing of each part p of the object P, and the inspection processing unit 123 of the inspection apparatus 1 may execute the inspection T on each part p of the object P. Further, the learning device 2 generates a test model of the test T for each part p of the object P, the management device 3 manages a version of the test model of the test T for each portion p of the target P, and What is necessary is just to update the version of the inspection model of the inspection T for each part p of the object P.

  Although the preferred embodiment of the present invention has been described in detail, the present invention is not limited to the above-described embodiment, and various modifications and substitutions can be made to the above-described embodiment without departing from the scope of the present invention. Can be added.

1: inspection apparatus 2: learning apparatus 3: management apparatus 4: imaging apparatus 11: main board 12: inspection board 100: inspection system 111: CPU
112: RAM
113: ROM
114: HDD
115: connection interface 116: input device 117: display device 118: communication interface 119: bus 121: connection interface 122: recognition processing unit 123: inspection processing unit 124: bus P: object T: inspection

Claims (12)

A plurality of recognition processing units for extracting image data of a corresponding object from the image data;
Based on the image data of the object, a plurality of inspection processing unit that performs an inspection corresponding to the object,
A control unit that controls the plurality of recognition processing units and the plurality of inspection processing units,
An inspection device comprising: The inspection device according to claim 1, wherein the control unit inputs the image data to each of the plurality of recognition processing units. The control unit, when the image data of the object is extracted by the recognition processing unit, inputs the image data of the object to the inspection processing unit that performs the inspection on the object. The inspection device according to claim 2. The inspection device according to any one of claims 1 to 3, wherein the recognition processing unit and the inspection processing unit are dedicated processors, respectively. The inspection device according to any one of claims 1 to 4, wherein the recognition processing unit and the inspection processing unit and the control unit are provided on different substrates. The inspection apparatus according to any one of claims 1 to 5, wherein the inspection processing unit inputs image data of the object to an inspection model and outputs an inspection result of the object. An inspection apparatus according to any one of claims 1 to 6,
The control unit outputs, the image data of the object, data indicating the inspection result of the object, based on the inspection result data including, based on the image data of the object, the relationship between the inspection result. A learning device that learns and generates an inspection model;
Inspection system comprising: The inspection system according to claim 7, further comprising a photographing device that inputs image data to the inspection device. An inspection apparatus according to any one of claims 1 to 6,
A management device that updates the version of the test model used by each of the plurality of test processing units,
Inspection system comprising: An inspection apparatus according to any one of claims 1 to 6,
An imaging device for inputting image data to the inspection device,
Inspection system comprising: An inspection apparatus according to any one of claims 1 to 6,
The control unit outputs, the image data of the object, data indicating the inspection result of the object, based on the inspection result data including, based on the image data of the object, the relationship between the inspection result. A learning device that learns and generates an inspection model;
A management device that manages the plurality of test models generated by the learning device and updates a version of the test model used by each of the plurality of test processing units;
Inspection system comprising: The inspection system according to claim 11, further comprising a photographing device that inputs image data to the inspection device.

Images