JP7571504B2 - Information processing device, visual inspection device, printed circuit board manufacturing system and program - Google Patents

Description

The present invention relates to an information processing device, an appearance inspection device, a printed circuit board manufacturing system, and a program.

Conventionally, in the manufacture of printed circuit boards on which a large number of mounted components are mounted, visual inspection systems have been proposed to inspect the mounting quality of mounted components. With such visual inspection systems, it is necessary to accurately capture the shape of the solder joints in order to determine whether the electrodes of the mounted components and the lands on the printed circuit board have been soldered properly. As a method for acquiring the shape of a solder joint that has specular reflection, a method is known in which three color light sources are irradiated at different angles of incidence and the reflected light is captured to acquire the shape as color information. Then, conventional visual inspection systems use parameters such as the area and length of a specific color area to judge the quality of the shape of the solder joint acquired as color information in this way.

Patent Document 1 discloses a technology that uses images of good and defective products to set logic parameters using feature thresholds, inspection areas, and color ranges to separate good and defective products from the image features of each product, with the aim of automatically generating parameters for a judgment logic.

However, with conventional visual inspection systems, the appearance of the captured images varies due to changes in the solder surface condition caused by changes in production equipment over time or changes in mounting conditions, and there is a problem in that the accuracy of judgment decreases over time when using image inspections using the same logic.

The present invention has been made in consideration of the above, and aims to suppress the deterioration of the judgment accuracy of image inspection caused by changes in production equipment over time or changes in mounting conditions.

In order to solve the above-mentioned problems and achieve the object, the present invention comprises a feature extraction unit that uses machine learning to extract image features from a captured image of the appearance of a printed circuit board, a judgment unit that judges whether the product is good or bad based on the image features extracted by the feature extraction unit, a change detection unit that detects changes in the image features of good product images over time, and a re-learning unit that corrects the judgment conditions of the judgment unit based on the detection results of the change detection unit, using defective product images and good product images at the time when a change in the features has occurred , so that good products can be separated from defective products , and is characterized in that the re-learning unit corrects the extraction conditions of the feature extraction unit .

The present invention has the advantage of being able to suppress deterioration in judgment accuracy due to changes in production equipment over time or changes in mounting conditions.

FIG. 1 is a schematic diagram showing the main devices of a printed circuit board manufacturing system according to a first embodiment. FIG. 2 is a schematic diagram of the appearance inspection device. FIG. 3 is a diagram for explaining the image capturing process that is the premise of the pass/fail determination process. FIG. 4 is a diagram showing an example of an inspection target image. FIG. 5 is a flowchart showing an example of the flow of a quality determination method. FIG. 6 is a diagram showing a conventional visual inspection. FIG. 7 is a diagram showing the appearance inspection of this embodiment. FIG. 8 is a block diagram showing a hardware configuration of the information processing device. FIG. 9 is a functional block diagram showing functions related to the appearance inspection process. FIG. 10 is a diagram showing an example of removing positional variations. FIG. 11 is a diagram showing an example of removing the glare of reflected light from an electrode portion. FIG. 12 is an explanatory diagram of the change detection logic (scatter diagram) used in the change detection unit. FIG. 13 is an explanatory diagram of the change detection logic (change over time) used in the change detection unit. FIG. 14 is a flowchart showing the flow of the learning process of the information processing device. FIG. 15 is a flowchart showing a modified example of the flow of the learning process of the information processing device.

The following describes in detail embodiments of an information processing device, an appearance inspection device, a printed circuit board manufacturing system, and a program with reference to the accompanying drawings. In the following, an embodiment is described in which the present invention is applied to a printed circuit board manufacturing system including an appearance inspection device that optically determines whether the soldering of SMDs (Surface Mount Devices), which are elements such as capacitor chips and ICs mounted on a printed circuit board, is good or bad.

(First embodiment)
1 is a schematic diagram showing the main devices of a printed circuit board manufacturing system 100 according to a first embodiment. As shown in FIG. 1, the printed circuit board manufacturing system 100 includes a solder printer 20, an element mounter 30, a reflow oven 40, and an appearance inspection device 50.

The printed circuit board manufacturing system 100 transports a printed circuit board without components mounted thereon into the solder printer 20. The solder printer 20 prints solder at a predetermined position on the printed circuit board without components mounted thereon. The printed circuit board with the solder printed thereon is transported to the element mounter 30. The element mounter 30 is a so-called chip mounter and QFP (Quad Fiat Package) mounter. The element mounter 30 places SMD mounted components such as capacitors and ICs on the printed circuit board. The printed circuit board with the mounted components such as capacitors and ICs mounted thereon is then transported into the reflow furnace 40. The reflow furnace 40 heats the printed circuit board with the mounted components such as capacitors and ICs mounted thereon, and melts the solder on the printed circuit board printed by the solder printer 20. The melted solder on the printed circuit board is cooled to fix the mounted components on the printed circuit board.

The printed circuit board with the mounted components fixed in the above manner is then carried into the visual inspection device 50 by a transport means such as a belt conveyor, or manually by an operator. The visual inspection device 50 performs automated optical inspection (AOI), which optically judges whether the soldering is good or bad using a method described below. Then, the printed circuit board that is judged to be good by the visual inspection device 50 is transferred to a subsequent processing step (not shown).

Examples of devices that perform subsequent processing steps include a component mounter that mounts further components on the printed circuit board, and a packaging device that packages the printed circuit board. The device that performs subsequent processing steps may also be, for example, a sorting device that separates non-defective products from defective products based on the result of the pass/fail judgment by the visual inspection device 50. Note that, although the printed circuit board manufacturing system 100 is provided with a device that performs processing steps after the pass/fail judgment by the visual inspection device 50, this device is not necessarily required.

In addition, in this embodiment, a correction terminal 60, which is an information processing device, is connected to the appearance inspection device 50. The correction terminal 60 has a data input/output means therein for writing the judgment result data output from the appearance inspection device 50 and reading the written data.

The correction terminal 60 allows an operator to visually re-inspect whether the pass/fail judgment results from the visual inspection device 50 are correct, and to correct the inspection result data if an error is found in the inspection results. In this embodiment, individual barcodes are formed on the printed circuit board, and the barcode information is read by an automatic barcode reader. The barcode information read by the automatic barcode reader and the judgment result data from the visual inspection device 50 are sent to the correction terminal 60 via a serial communication cable and stored.

As will be described in detail later, the repair terminal 60 applies an image inspection algorithm to images judged to be defective by the visual inspection device 50 to judge whether they are good or bad, thereby reducing the man-hours required for visual inspection. Therefore, the repair terminal 60 further judges whether the stored judgment result data is good or bad, and then displays the judgment result that it is judged to be bad on the display 60a, which is a display means. Then, when the judgment result that it is judged to be bad is displayed on the display 60a, the operator visually checks the actual printed circuit board to confirm whether it is really a defective product. If it is judged not to be a defective product as a result, the operator inputs corrections using a data correction input means such as a keyboard (not shown) of the repair terminal 60.

The repair terminal 60 of this embodiment is configured as a stand-alone computer connected to the visual inspection device 50, but it may also be connected to the visual inspection device 50 via a network such as a LAN. Also, the repair terminal 60 of this embodiment may be provided integrally with the visual inspection device 50.

Next, we will explain the appearance inspection device 50.

Figure 2 is a schematic diagram of the visual inspection device 50. As shown in Figure 2, the visual inspection device 50 is equipped with three circular light sources 1R, 1G, and 1B that irradiate light onto a PCB 4, which is a printed circuit board on which SMDs are mounted. Therefore, the visual inspection device 50 is equipped with a total of six light sources. Each of the light sources 1R, 1G, and 1B can be an inverter-controlled ring-shaped fluorescent lamp with a color filter attached, or a light-emitting diode (LED) that emits light of each color.

The appearance inspection device 50 also includes a high-speed camera 2 as an imaging device that receives light irradiated from these light sources 1R, 1G, and 1B and captures images. The high-speed camera 2 is capable of capturing images with a high-speed shutter of about 1/100,000th of a second. The high-speed camera 2 may be configured with an imaging element such as a CCD. In this embodiment, a CCD camera with a resolution that can be changed between 13 μm, 15 μm, 20 μm, 25 μm, 30 μm, 35 μm, 50 μm, and 70 μm is used as the high-speed camera 2. The high-speed camera 2 is positioned directly above the inspection area as shown in FIG. 2.

The visual inspection device 50 also includes an XY stage 3 as a driving means for two-dimensionally moving the PCB 4 placed on the mounting table in the horizontal direction so as to move relative to the high-speed camera 2, and a control unit 10 as a control means. The XY stage 3 supports the PCB 4 horizontally.

The control unit 10 includes a CPU (Central Processing Unit), a memory unit, an interface unit, etc. The control unit 10 controls the shutter of the high-speed camera 2, drives and controls the XY stage 3 via the stage drive unit 3a, and performs pass/fail judgment processing.

Next, we will explain the image capture process that is the premise for the pass/fail judgment process.

Figure 3 is a diagram explaining the image capture process that is the premise of the pass/fail judgment process. The diagram shown in Figure 3 is a diagram showing the state in which light is irradiated onto a solder fillet 6, which is a solder joint that connects an electrode 5a of a capacitor 5, which is an example of an SMD, to a pad 4a on a PCB 4. As shown in Figure 3, light source 1R is a red light source that irradiates light (red light R) at the smallest irradiation angle with respect to the vertical direction. Light source 1B is a blue light source that irradiates light (blue light B) at the largest irradiation angle. Light source 1G is a green light source that irradiates light (green light G) at an irradiation angle intermediate between light source 1R and light source 1B.

As shown in FIG. 3, the high-speed camera 2 captures an image of the PCB 4 illuminated with red light R, green light G, and blue light B from light sources 1R, 1G, and 1B. In this case, the high-speed camera 2 captures an image of the inspection object that is color-coded into three colors: R, G, and B. The color distribution of these three colors makes it possible to observe the soldering state, i.e., the outer shape of the solder fillet 6.

The imaging area (inspection area) of the high-speed camera 2 is narrower than the PCB 4. Therefore, when the PCB 4 is imaged by the high-speed camera 2, the PCB 4 on the XY stage 3 moves smoothly and continuously in the horizontal direction. Specifically, the high-speed camera 2 sequentially images the PCB 4 for one or more SMDs that fit within the imaging area. In this way, it is possible to grasp the outer shape of the solder fillet 6 for the electrode 5a of the capacitor 5, which is an example of each SMD.

Here, FIG. 4 shows an example of an image to be inspected. As shown in FIG. 4, the image to be inspected is an image captured so that the outer shape of the solder fillet 6, which is the solder joint, can be observed as a color distribution, as described above.

Next, we will explain the pass/fail judgment process in the control unit 10 of the visual inspection device 50.

Figure 5 is a flow chart showing an example of the flow of a quality determination method. As shown in Figure 5, the control unit 10 of the visual inspection device 50 first controls the drive of the XY table 3 to move the capacitor 5 on the PCB 4 into the imaging area of the high-speed camera 2 (step S1).

Next, the control unit 10 uses the high-speed camera 2 to capture the reflected light of each color of light irradiated from the light sources 1R, 1G, and 1B onto the PCB 4 (step S2). This makes it possible to obtain an image like that shown in FIG. 4 based on the captured image data. At this time, if the solder fillet 6 is formed properly on the pad 4a, the area (position, size) of the solder fillet 6 will be shown in blue, as shown in FIG. 4. If the solder fillet 6 is formed properly, the area of the blue hue will be large in the solder fillet 6, as shown in FIG. 4. In contrast, the area of hues other than blue will be large in the solder fillet 6 of a defective product.

Next, the control unit 10 calculates the sum _S1 of the number of pixels in the blue region present on both sides of the electrode 5a of a single capacitor 5 from the image data captured by the high-speed camera 2 (step S3 ₎ . The control unit 10 then judges whether the calculated sum S1 of the number of pixels is greater than a predetermined number of pixels (e.g., 8 pixels) (step S4). If the sum _S1 is greater than 8 pixels, the control unit 10 judges that the capacitor 5 is a good product (step S5), and if the sum S1 is 8 pixels or less, the control unit 10 judges that the capacitor 5 is a defective product (step S6). This makes it possible to judge the quality of the soldering for each capacitor 5.

In the inspection of the inspection target image in the appearance inspection device 50, as described above, a rule-based inspection is performed on the color features of the inspection target image. However, in the case of production data such as the inspection target image, slight changes in production conditions over time cause the tendency of the acquired data to change, so it is not possible to continuously inspect using a once-determined judgment criterion. As a result, there are many false reports (overdetections) in which the appearance inspection device 50 judges a good product to be defective. Figure 6 is a diagram showing a conventional appearance inspection. As shown in Figure 6, conventionally, inspection is performed in two stages, the appearance inspection device 50 and visual inspection. Therefore, if there are many false reports (overdetections) in which the appearance inspection device 50 judges a good product to be defective, there is a problem that the visual inspection requires a lot of man-hours.

Figure 7 is a diagram showing the visual inspection of this embodiment. As shown in Figure 7, in the repair terminal 60 of this embodiment, the details will be described later, an image inspection algorithm is introduced to images judged as defective by the visual inspection device 50 to judge whether they are good or bad, thereby reducing the man-hours for visual inspection. The image inspection algorithm is an incremental learning type image inspection algorithm that accumulates data during operation and performs incremental learning while inspecting the image to be inspected, and uses machine learning. According to the image inspection algorithm, the judgment results can be used to improve the accuracy of a pre-constructed judger.

Here, FIG. 8 is a block diagram showing the hardware configuration of the editing terminal 60. As shown in FIG. 8, the editing terminal 60 is constructed by a computer. As shown in FIG. 8, the editing terminal 60 is equipped with a CPU 601, a ROM 602, a RAM 603, a HD 604, a HDD (Hard Disk Drive) controller 605, a display 606, an external device connection I/F (Interface) 608, a network I/F 609, a data bus 610, a keyboard 611, a pointing device 612, a DVD-RW (Digital Versatile Disc Rewritable) drive 614, and a media I/F 616.

Of these, the CPU 601 controls the operation of the entire editing terminal 60. The ROM 602 stores programs used to drive the CPU 601, such as IPL. The RAM 603 is used as a work area for the CPU 601. The HD 604 stores various data such as programs. The HDD controller 605 controls the reading or writing of various data from the HD 604 according to the control of the CPU 601. The display 606 displays various information such as a cursor, a menu, a window, characters, or an image. The external device connection I/F 608 is an interface for connecting various external devices. In this case, the external device is, for example, an appearance inspection device 50, a USB (Universal Serial Bus) memory, a printer, etc. The external device connection I/F 608 functions as a receiving unit 61 and an output unit 62 (see FIG. 9). The network I/F 609 is an interface for data communication using a communication network. The bus line 610 is an address bus, a data bus, etc., for electrically connecting each component such as the CPU 601 shown in FIG. 8.

The keyboard 611 is a type of input means equipped with multiple keys for inputting characters, numbers, various instructions, etc. The pointing device 612 is a type of input means for selecting and executing various instructions, selecting a processing target, moving the cursor, etc. The DVD-RW drive 614 controls the reading and writing of various data from a DVD-RW 613, which is an example of a removable recording medium. Note that this is not limited to a DVD-RW, and may be a DVD-R, etc. The media I/F 616 controls the reading and writing (storing) of data from a recording medium 615, such as a flash memory.

The program executed by the editing terminal 60 of this embodiment is provided as a file in an installable or executable format recorded on a computer-readable recording medium such as a CD-ROM, a flexible disk (FD), a CD-R, or a DVD (Digital Versatile Disc).

The program executed by the editing terminal 60 of this embodiment may be stored on a computer connected to a network such as the Internet and provided by downloading it via the network. The program executed by the editing terminal 60 of this embodiment may be provided or distributed via a network such as the Internet. The program executed by the editing terminal 60 of this embodiment may be provided by being pre-installed in a ROM or the like.

Next, we will explain the functions related to visual inspection processing.

Figure 9 is a functional block diagram showing the functions related to the appearance inspection process. As shown in Figure 9, the editing terminal 60 includes a receiving unit 61, a preprocessing unit 63, a feature extraction unit 64, a judgment unit 65, a change detection unit 66, a re-learning unit 67, and an output unit 62. As described above, the receiving unit 61 and the output unit 62 are the external device connection I/F 608. In addition, the program executed by the editing terminal 60 of this embodiment has a modular configuration including each unit (preprocessing unit 63, feature extraction unit 64, judgment unit 65, change detection unit 66, re-learning unit 67), and as actual hardware, the CPU (processor) reads and executes the program from the storage medium, loading each unit onto the main storage device, and the preprocessing unit 63, feature extraction unit 64, judgment unit 65, change detection unit 66, and re-learning unit 67 are generated on the main storage device.

The receiving unit 61 receives from the visual inspection device 50 images that have been determined to be defective by the visual inspection device 50, i.e., visual information of the printed circuit board captured by the high-speed camera 2.

The preprocessing unit 63 performs preprocessing appropriate to the inspection content on the image received by the receiving unit 61. Specifically, the preprocessing unit 63 performs extraction processing to extract a specified color (e.g., only B) and extract only the area required for judgment from the image.

The pre-processing unit 63 also functions as a noise removal unit that performs processing to remove information that becomes noise in the quality inspection of solder joints based on rule-based image processing. Examples of information that becomes noise in the quality inspection include positional variation and glare from the electrode part.

Figure 10 shows an example of removing positional variation. As shown in Figure 10, removing positional variation involves detecting solder joints while taking into account detection deviations.

Figure 11 is a diagram showing an example of removing the glare of reflected light from the electrode part. As shown in Figure 11, the glare of reflected light from the electrode part is removed by masking the original image and trimming the vicinity of the defective part to detect the solder joint.

The feature extraction unit 64 uses machine learning to extract image features.

The determination unit 65 converts the image feature amount extracted by the feature extraction unit 64 into an index value indicating whether the product is good or bad. The determination unit 65 then determines whether the product is good or bad based on the index value indicating whether the product is good or bad using a threshold value.

The output unit 62 outputs the judgment result from the judgment unit 65.

The change detection unit 66 detects changes in the features of the non-defective images using the output of the feature extraction unit 64 or the determination unit 65. The change detection logic used by the change detection unit 66 is explained below.

Here, FIG. 12 is an explanatory diagram of the change detection logic (scatter plot) used by the change detection unit 66. FIG. 12 shows a method for detecting fluctuations in image features based on the distribution of image features. The change detection unit 66 compares the distribution 10a of image features of good-quality images after a change in manufacturing conditions such as a lot change, or after a specified production volume or elapsed period, output from the feature extraction unit 64 and the judgment unit 65, with the distribution 9a of image features of the acquired good-quality images. In FIG. 12, the reference numerals 9b and 10b indicate the image features of one image. The change detection unit 66 performs the comparison using statistics such as the average and variance of each group of image features.

Note that while FIG. 12 shows a two-dimensional image feature space, the change detection unit 66 may perform evaluation in a higher- or lower-dimensional image feature space. In addition, the change detection unit 66 may perform change detection using index values indicating pass/fail used by the judgment unit 65, rather than image features.

Here, FIG. 13 is an explanatory diagram of the change detection logic (time change) used by the change detection unit 66. FIG. 13 shows a method for detecting variations in image features based on time-series changes in image features. FIG. 13 expresses the distribution of image features of non-defective images as time-series data. The change detection unit 66 uses a time-series data anomaly detection method to detect changes in the distribution 10c of image features of non-defective images after a change in manufacturing conditions such as a change in lot, or after a specified number of productions or after a certain period of time, relative to the distribution 9c of image features of acquired non-defective images. That is, the change detection unit 66 checks for the presence or absence of changes in the image by comparing data after a change in manufacturing conditions such as a change in lot, or after a specified number of productions or after a specified period of time. Note that, although FIG. 13 shows changes in image features in a time series, the present invention is not limited to this, and may be changes in image features according to changes in the number of productions.

When the change detection unit 66 detects a change, it modifies the logic of the feature extraction unit 64 and the judgment unit 65. Using the defective product image and the good product image at the time of the change occurrence, the change detection unit 66 modifies the extraction conditions of the feature extraction unit 64 and the judgment conditions of the judgment unit 65 so that good products and bad products can be best separated.

The re-learning unit 67 re-learns the extraction conditions of the feature extraction unit 64 and the judgment conditions of the judgment unit 65 based on the detection results of the change detection unit 66. The re-learning unit 67 sends the re-learning results to the feature extraction unit 64 and the judgment unit 65 and corrects the logic. Note that when the change detection unit 66 and the re-learning unit 67 detect a change, they may correct the logic of only the judgment unit 65.

Next, we will explain the learning process of the editing terminal 60.

Here, FIG. 14 is a flowchart showing the flow of the learning process of the editing terminal 60. As shown in FIG. 14, when the feature extraction unit 64 determines that a predetermined production volume or a predetermined period of time has elapsed since the previous learning, it determines that change detection needs to be performed (Yes in step S11) and proceeds to step S12.

The feature extraction unit 64 extracts image features of the trained image (step S12). Furthermore, the feature extraction unit 64 extracts image features of the image after a specified production run or a specified period of time (step S13).

Next, the change detection unit 66 extracts the amount of change from the image features extracted in steps S12 and S13, respectively, using the change detection logic shown in FIG. 12 or FIG. 13 (step S14), and performs a threshold determination (step S15).

If the change detection unit 66 detects that the amount of change is equal to or greater than the threshold and a change is detected (Yes in step S15), the re-learning unit 67 uses the image after the change to update the extraction conditions of the feature extraction unit 64 (step S16) and update the judgment conditions of the judgment unit 65 (step S17), thereby re-learning.

On the other hand, if the amount of change is not greater than the threshold value and no change is detected by the change detection unit 66 (No in step S15), the re-learning unit 67 ends the process.

As described above, according to this embodiment, the image features of good product images are analyzed over time, and when a change in the image features is detected, the image features of the good product images after the change are re-learned, and the logic is modified to enable separation of good products from defective products. This makes it possible to suppress deterioration in the accuracy of logic judgment due to changes in production equipment over time or changes in mounting conditions, and enables stable image inspection to be performed.

The judgment unit 65 may classify the image received by the receiving unit 61, the judgment result of the judgment unit 65, and the judgment result of an inspector's visual inspection as good or bad using machine learning. In this way, an inspection close to the judgment result of an inspector can be realized. In other words, the inspector's visual judgment is performed only in the initial stage, and after accuracy has been improved, the process of visual judgment by the inspector can be eliminated.

In this embodiment, the re-learning unit 67 re-learns when the amount of change in the change detection unit 66 is equal to or greater than the threshold and a change is recognized (Yes in step S15), but this is not limited to the above.

Here, FIG. 15 is a flowchart showing a modified example of the flow of the learning process of the editing terminal 60. As shown in FIG. 15, when the number of erroneous judgments or the erroneous judgment rate of the visual inspection is equal to or greater than a threshold (Yes in step S21), the re-learning unit 67 may re-learn by correcting the extraction conditions of the feature extraction unit 64 (step S22) and correcting the judgment conditions of the judgment unit 65 (step S23) using the changed image. The threshold for the erroneous judgment rate is, for example, 10 to 20%.

Machine learning is a technology that allows computers to acquire human-like learning capabilities, in which the computer autonomously generates algorithms necessary for judgments such as data identification from training data that is previously loaded, and applies these to new data to make predictions. The learning method for machine learning may be any of supervised learning, unsupervised learning, semi-supervised learning, reinforcement learning, and deep learning, or may be a combination of these learning methods; any learning method for machine learning is acceptable.

Reference Signs List 2: Imaging device 4: Printed circuit board 50: Appearance inspection device 60: Information processing device 63: Noise removal section 64: Feature extraction section 65: Determination section 66: Change detection section 67: Re-learning section 100: Printed circuit board manufacturing system

Patent No. 4492356

Claims (8)

a feature extraction unit that extracts image features from a captured image of the appearance of the printed circuit board using machine learning;
a determination unit for determining whether the product is good or bad based on the image feature amount extracted by the feature extraction unit;
A change detection unit that detects a change in an image feature amount of a non-defective image over time;
a re-learning unit that corrects the judgment conditions of the judgment unit so that good products can be separated from good products by using defective product images and good product images when a change in feature amount occurs based on the detection result of the change detection unit;
Equipped with
The re-learning unit corrects an extraction condition of the feature extraction unit.
23. An information processing apparatus comprising: The re-learning unit corrects the judgment conditions of the judgment unit based on the number of erroneous judgments or the erroneous judgment rate of the appearance inspection.
2. The information processing apparatus according to claim 1, the re-learning unit corrects the extraction conditions of the feature extraction unit based on the number or rate of erroneous determinations in the visual inspection.
2. The information processing apparatus according to claim 1 , a noise removal unit that removes information that becomes noise in appearance inspection from the captured image of the appearance of the printed circuit board;
4. The information processing apparatus according to claim 1, wherein the information processing apparatus further comprises: The determination unit classifies the captured image, the determination result of the determination unit as to whether the product is good or bad, and the determination result of the inspector as to whether the product is good or bad by visual inspection using machine learning.
5. The information processing apparatus according to claim 1, wherein the information processing apparatus further comprises: a first input section; A visual inspection apparatus for determining whether a printed circuit board is good or bad based on an image of the appearance of the printed circuit board obtained by an imaging device, comprising:
A feature extraction unit that extracts image features from a captured image of the appearance of the printed circuit board using machine learning;
a determination unit for determining whether the product is good or bad based on the image feature amount extracted by the feature extraction unit;
A change detection unit that detects a change in an image feature amount of a non-defective image over time;
a re-learning unit that corrects the judgment conditions of the judgment unit so that good products can be separated from good products by using defective product images and good product images when a change in feature amount occurs based on the detection result of the change detection unit;
Equipped with
The re-learning unit corrects an extraction condition of the feature extraction unit.
A visual inspection apparatus comprising: an imaging device for capturing an image of the appearance of the printed circuit board;
an appearance inspection device for determining whether a printed circuit board is good or bad based on an image of the appearance of the printed circuit board obtained by the imaging device;
5. The information processing device according to claim 1 , further comprising: a display device that displays a display image of the printed circuit board that has been determined to be defective by the appearance inspection device, and determines whether the printed circuit board is non-defective or defective;
A printed circuit board manufacturing system comprising: Computer,
a feature extraction unit that extracts image features from a captured image of the appearance of the printed circuit board using machine learning;
a determination unit for determining whether the product is good or bad based on the image feature amount extracted by the feature extraction unit;
A change detection unit that detects a change in an image feature amount of a non-defective image over time;
a re-learning unit that corrects the judgment conditions of the judgment unit so that good products can be separated from good products by using defective product images and good product images when a change in feature amount occurs based on the detection result of the change detection unit;
Function as a
The re-learning unit corrects an extraction condition of the feature extraction unit.
A program characterized by:

Images