WO2020202332A1 - 検査装置及び検査方法 - Google Patents
検査装置及び検査方法 Download PDFInfo
- Publication number
- WO2020202332A1 WO2020202332A1 PCT/JP2019/014229 JP2019014229W WO2020202332A1 WO 2020202332 A1 WO2020202332 A1 WO 2020202332A1 JP 2019014229 W JP2019014229 W JP 2019014229W WO 2020202332 A1 WO2020202332 A1 WO 2020202332A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- inspection
- threshold value
- type
- numerical data
- inspection object
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
- G06V10/7747—Organisation of the process, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/776—Validation; Performance evaluation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8883—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges involving the calculation of gauges, generating models
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/95—Investigating the presence of flaws or contamination characterised by the material or shape of the object to be examined
- G01N21/9515—Objects of complex shape, e.g. examined with use of a surface follower device
- G01N2021/9518—Objects of complex shape, e.g. examined with use of a surface follower device using a surface follower, e.g. robot
Definitions
- the present invention relates to an inspection device and an inspection method using a learning model.
- AI Artificial Intelligence
- a learning model is generated by inputting multiple teacher data, input data is given to the generated learning model to perform calculations, and the result of machine learning is reflected.
- the AI processing data is output (Japanese Patent Laid-Open No. 2019-0398774).
- the present invention provides an inspection device and an inspection method capable of efficiently performing an inspection without deteriorating the inspection accuracy by performing the inspection by utilizing the AI process.
- At least a part of the classification result of classifying a plurality of inspected objects of the same type as the inspection object into a plurality of types is used as teacher data to perform the inspection.
- a learning unit that generates a learning model by performing learning to determine the type of an object, or acquires the learning model.
- a calculation unit that outputs numerical data that quantifies the high classification accuracy of the type of the inspection object based on the result calculated by inputting the inspection object into the learning model.
- a determination unit for automatically determining the type of the inspection object or manually determining the type of the inspection object based on the result of comparing the numerical data with one or more types of threshold values is provided.
- Inspection equipment is provided.
- a threshold value calculation unit that calculates one or more types of threshold values based on the plurality of numerical data calculated by inputting a plurality of inspection objects into the learning model may be provided.
- the threshold value calculation unit may calculate one or more types of threshold values by statistical processing of the plurality of numerical data.
- the one or more types of threshold values include a first threshold value and a second threshold value larger than the first threshold value.
- the determination unit may manually determine the type of the inspection object.
- the determination unit When the numerical data is smaller than the first threshold value or the numerical data is larger than the second threshold value, the determination unit does not manually determine the type of the inspection target object and performs the inspection. You may decide to automatically determine the type of object.
- re-learning is performed based on the unique information of the inspection object to generate a re-learning model, or the learning model is used.
- Re-learning department to acquire and It is provided with a recalculation unit that re-outputs the numerical data based on the result of inputting the inspection object into the re-learning model and calculating.
- the determination unit automatically determines the type of the inspection object based on the result of comparing the numerical data with the first threshold value and the second threshold value while taking into consideration the unique information of the inspection object. Or you may decide whether to manually determine the type of the inspection object.
- the determination unit determines the type of the inspection target for each type of unique information of the inspection target. You may decide whether to perform the automatic determination of the above or to manually determine the type of the inspection object.
- the plurality of types include a good product type and a defective product type.
- the unique information may include defect sizes of good and defective products.
- a practical level determination unit for determining that the learning model has reached a practical level may be provided.
- the determination unit manually determines the type of the inspection object. You may decide.
- a photographing unit for photographing the inspection object from a plurality of directions is provided.
- the learning unit may use a plurality of captured images of the inspection object captured by the photographing unit as the teacher data.
- It may be provided with a visualization unit that visualizes the numerical data calculated by inputting a plurality of inspection objects into the learning model.
- Another aspect of the present invention is an inspection method for inspecting an inspection object with a computer.
- a learning model in which at least a part of the classification results obtained by classifying a plurality of inspected objects of the same type as the inspection object into a plurality of types is used as teacher data to perform learning for discriminating the type of the inspection object. Or get the learning model Based on the result calculated by inputting the inspection object into the learning model, numerical data in which the high classification accuracy of the inspection object type is quantified is output. Based on the result of comparing the numerical data with one or more types of threshold values, it is determined whether to automatically determine the type of the inspection object or to manually determine the type of the inspection object.
- the computer is connected to a network
- the teacher data and the data of the inspection object are transmitted to the computer via the network.
- Information on whether to automatically determine the type of the inspection object or to manually determine the type of the inspection object, which is determined by the computer, may be received via the network.
- the computer may be made to calculate one or more types of threshold values based on the plurality of numerical data calculated by inputting a plurality of inspection objects into the learning model.
- the computer may be made to calculate the one or more types of threshold values by statistical processing of the plurality of numerical data.
- the one or more types of threshold values include a first threshold value and a second threshold value larger than the first threshold value.
- the computer may manually determine the type of the inspection object.
- the inspection target is not manually determined as the type of the inspection target. You may decide to automatically determine the type of object.
- the numerical data exists between the first threshold value and the second threshold value
- re-learning is performed based on the unique information of the inspection object to generate a re-learning model, or the re-learning model is generated.
- the numerical data is output based on the result calculated by inputting the inspection object into the re-learning model.
- the type of the inspection target object is automatically determined or manually described. You may decide whether to determine the type of object to be inspected.
- the type of the inspection target It may be decided whether to perform automatic discrimination or to manually discriminate the type of the inspection object.
- the plurality of types include a good product type and a defective product type.
- the unique information may include defect sizes of good and defective products.
- the computer is made to determine whether or not the ratio of the numerical data included between the first threshold value and the second threshold value is less than the third threshold value, and it is determined that the ratio is less than the third threshold value. Then, it may be judged that the learning model has reached a practical level.
- a plurality of captured images of the inspection object taken from a plurality of directions may be used as the teacher data.
- the computer may be made to visualize the numerical data calculated by inputting a plurality of inspection objects into the learning model.
- the inspection by performing the inspection by utilizing the AI process, the inspection can be efficiently performed without deteriorating the inspection accuracy.
- the block diagram which shows the internal structure of the AI processing part. A plot diagram showing the inspection results of a plurality of inspection objects. The plot figure which set the 1st and 2nd thresholds.
- the flowchart which shows the processing operation of the inspection apparatus by 1st Embodiment. The graph which shows how the inspection ratio by a person decreases by repeating learning based on the flowchart of FIG.
- a plot diagram showing the inspection results of a plurality of inspection objects The flowchart which shows the processing operation of the inspection apparatus according to 3rd Embodiment.
- the plot figure which shows the result which the worker discriminated a good product / defective product a plurality of times for a plurality of inspected objects.
- FIG. 1 is a block diagram showing a schematic configuration of the inspection device 1 according to the first embodiment.
- the inspection device 1 of FIG. 1 performs a visual inspection of the inspection object 5.
- the type of the inspection object 5 is not particularly limited.
- a typical example is a plurality of products manufactured to predetermined specifications.
- a more specific example is a forged product obtained by pressing a metal material or the like with a die, or a casting formed by pouring a metal material or the like into a mold.
- the shape, size, material, etc. of the inspection object 5 are also arbitrary, and may be formed of not only metal but also resin or the like.
- the inspection device 1 of FIG. 1 includes a control unit 2, an AI processing unit 3, and an information processing unit 4.
- the control unit 2, the AI processing unit 3, and the information processing unit 4 have a communication function for transmitting and receiving information to and from each other.
- This communication function may be a wireless communication function such as wireless LAN or proximity wireless communication, or may be a wired communication function such as Ethernet (registered trademark) or USB (Universal Serial Bus).
- at least two of the control unit 2, the AI processing unit 3, and the information processing unit 4 may be integrated into one housing or a SoC (Silicon on Chip).
- SoC Silicon on Chip
- at least a part of the processing operations performed by the control unit 2, the AI processing unit 3, and the information processing unit 4 may be executed by either hardware or software.
- the control unit 2 uses the captured image captured by the photographing unit 6 to generate teacher data to be given to the AI processing unit 3 and controls to generate the inspection target data of the inspection target 5. Since the present embodiment is intended to perform an appearance inspection of the inspection object 5, the control unit 2 performs AI processing using the photographed image of the appearance of the inspection object 5 taken by the photographing unit 6 as the inspection target data. It is transmitted to the part 3. Further, the control unit 2 transmits the photographed image of the appearance of the inspection object 5, which has been determined to be a non-defective product or a defective product, taken by the photographing unit 6 to the AI processing unit 3.
- the teacher data is at least a part of the classification result of classifying a plurality of inspected objects of the same type as the inspection object into a plurality of types.
- the plurality of types refer to each classification when the inspected object and the features such as the shape, characteristics, and size of the inspected object are classified into a plurality of types. More specifically, the teacher data may be supervised data including captured images for which good or defective products have been determined, or unsupervised data including captured images of only one of good and defective products.
- the control unit 2 of FIG. 1 has a function of controlling a robot 9 that sequentially grips the inspection object 5 from the storage body 7 that stores the inspection object 5 and conveys the inspection object 5 to the rotation stage 8.
- the robot 9 does not necessarily have to place the inspection object 5 on the rotating stage 8, and the operator may manually place the inspection object 5 on the rotating stage 8.
- the photographing unit 6 is installed diagonally above the rotating stage 8, for example.
- the position and number of the photographing units 6 are arbitrary.
- a plurality of captured images are generated in order to perform an appearance inspection of one inspection object 5.
- teacher data to which information indicating the determination result of non-defective product or defective product is added is generated for each photographed image.
- each captured image captured by the photographing unit 6 will be the inspection target data.
- the entire appearance of the inspection object 5 may be photographed with only one photographed image.
- one teacher data and one inspection target data are generated for each inspection target 5.
- the AI processing unit 3 inspects the inspection object 5 by AI processing.
- the AI processing refers to outputting the AI processing data obtained by giving input data to the learning model generated by machine learning and performing an operation.
- various learning methods have been proposed for machine learning, any learning method can be applied to the AI processing of the present embodiment.
- the information processing unit 4 automatically generates a program executed by the control unit 2 and a program executed by the AI processing unit 3.
- the information processing unit 4 has a display unit 4a for displaying a UI screen having a plurality of input fields input by an operator.
- a program executed by the control unit 2 and a program executed by the AI processing unit 3 are automatically generated.
- the automatically generated program is transmitted to the control unit 2 and the AI processing unit 3, respectively, via the communication function of the information processing unit 4.
- the control unit 2 controls the robot 9 described above, controls the photographing of the inspection target object 5, controls the inspection target data to be transmitted to the AI processing unit 3, and the like. I do.
- the AI processing unit 3 executes the program transmitted from the information processing unit 4 to control the reception of the inspection target data transmitted from the control unit 2 and perform AI processing on the inspection target data.
- FIG. 2 is a block diagram showing the internal configuration of the AI processing unit 3.
- the AI processing unit 3 has a learning unit 11, a calculation unit 12, and a determination unit 13.
- the learning unit 11 uses at least one of a plurality of non-defective products and defective products of the same type as the inspection object 5 as teacher data, performs learning for discriminating between non-defective products and defective products, and generates a learning model.
- the learning model can be generated by controlling the weighting coefficient of the model formula prepared in advance, but the specific model formula used to generate the learning model is not limited, and any model formula can be applied. ..
- the calculation unit 12 inputs the inspection object 5 into the learning model and outputs numerical data quantifying the high possibility of a non-defective product or a defective product based on the calculated result.
- Numerical data is used for relative evaluation, not data having a physical unit.
- the determination unit 13 automatically determines (determines) a non-defective product or a defective product based on the numerical data based on the result of comparing the numerical data with one or more types of threshold values, or manually inspects (determines) a non-defective product or a defective product. Decide if you want to. That is, the determination unit 13 determines to perform automatic determination only when the AI processing by the AI processing unit 3 can determine highly reliable of a non-defective product or a defective product, and to perform a manual inspection in other cases. As a result, the inspection accuracy by the inspection device 1 is not inferior to the inspection accuracy by hand.
- the determination result by the determination unit 13 is displayed on, for example, the display unit 4a of the AI processing unit 3 or the information processing unit 4. Based on the display of the display unit 4a, the worker determines whether to make an automatic judgment or to perform the inspection by the worker himself / herself.
- the AI processing unit 3 or the information processing unit 4 may have a visualization unit 14.
- the visualization unit 14 inputs a plurality of inspection objects 5 into the learning model and visualizes the calculated numerical data. As will be described later, for example, each numerical data is displayed as a plot on a two-dimensional coordinate plane in which the horizontal axis is the numerical data and the vertical axis is the work number of the inspection object 5, and the distribution of each plot is visually grasped. You may be able to do it. Further, since the visualization unit 14 can distinguish between the plot determined to be a non-defective product and the plot determined to be a defective product by a human being, it becomes easy to grasp the correlation between the non-defective product / defective product and the numerical data.
- the AI processing unit 3 may have a threshold value calculation unit 15.
- the threshold value calculation unit 15 calculates one or more types of threshold values based on a plurality of numerical data calculated by inputting a plurality of inspection objects 5 into the learning model. For example, when the numerical data of the inspection object 5 judged to be a non-defective product by the worker and the numerical data of the inspection object 5 judged to be a defective product by the worker are close to each other, the threshold value calculation unit 15 uses these numerical values. Thresholds may be set between the data.
- the threshold value calculation unit 15 may calculate one or more types of threshold values by statistical processing of a plurality of numerical data.
- the statistical processing may be averaging or distributed processing of a plurality of numerical data, or may be an MT (Mahalanobis Taguchi) method or the like.
- the threshold value calculated by the threshold value calculation unit 15 may include, for example, a first threshold value and a second threshold value larger than the first threshold value.
- the determination unit 13 may decide to manually inspect the non-defective product or the defective product. That is, when the numerical data is smaller than the first threshold value or the numerical data is larger than the second threshold value, the determination unit 13 is performed by the AI processing unit 3 without manually inspecting the non-defective product or the defective product. It may be decided to automatically determine a non-defective product or a defective product, and if numerical data exists between the first threshold value and the second threshold value, manually inspect the non-defective product or the defective product.
- the inspection process of the inspection device 1 of FIG. 1 will be described.
- the control unit 2 takes an image with the photographing unit 6 while rotating the inspection object 5 placed on the rotating stage 8, and for example, 36 images are taken for one inspection object 5.
- the inspection device 1 of FIG. 1 inspects a non-defective product or a defective product for each inspection target data. Therefore, 288 types of inspection target data are inspected for one inspection target 5.
- the number of inspection target data per one inspection target 5 is arbitrary.
- FIG. 3 is a plot diagram showing the inspection results of a plurality of inspection objects 5.
- the AI processing unit 3 calculates numerical data for 288 inspection target data for each inspection target 5, and the plot ⁇ in which the operator determines that each inspection target data is a good product, and a defective product. It is displayed separately from the judged plot ⁇ .
- a separate work number is assigned to each data to be inspected, and the vertical axis of FIG. 3 indicates the work number.
- the horizontal axis of FIG. 3 is the numerical data calculated by the AI processing unit 3, and the right side indicates that the value of the numerical data is larger.
- the numerical data of the inspection target data judged to be non-defective by the worker is solidified in the right direction of the horizontal axis in FIG. 3, whereas the worker considers it to be defective.
- the numerical data of the determined inspection target data is dispersed in a wide range on the left side of the horizontal axis in FIG.
- the AI processing unit 3 compares the numerical data with the threshold value to determine whether the product is a non-defective product or a defective product, the inspection accuracy of the AI processing unit 3 may decrease in a region where non-defective products and defective products coexist. ..
- the AI processing unit 3 of the present embodiment sets the first threshold value and the second threshold value calculated by the threshold value calculation unit 15 in the region where the non-defective product and the defective product coexist. Then, the numerical data may be compared with the first and second threshold values to determine whether or not to make an automatic determination. More specifically, the AI processing unit 3 automatically determines that the product is defective if the numerical data is less than the first threshold value, and automatically determines that the product is non-defective if the numerical data is larger than the second threshold value. Further, when the numerical data exists between the first threshold value and the second threshold value, the AI processing unit 3 does not perform automatic determination by the AI processing unit 3, but a good product / defective product by a person (worker). Decide to do the inspection.
- determining a non-defective product may be referred to as "OK”, and determining a defective product may be referred to as "NG".
- FIG. 5 is a flowchart showing the processing operation of the inspection device 1 according to the first embodiment.
- a learning model is generated by learning a plurality of inspection objects 5 for which a person (worker) has determined OK or NG (step S1).
- the process of step S1 is performed by the learning unit 11.
- Step S1 is premised on supervised learning, but when unsupervised learning is performed, for example, clustering processing of inspection target data corresponding to a plurality of inspection objects 5 and principal component analysis are performed instead of step S1.
- a learning model is generated by performing learning by.
- step S1 When the process of step S1 is completed, next, the inspection target data photographed by the photographing unit 6 for the inspection object 5 for which OK or NG has not been determined is input to the learning model generated in step S1 to determine OK or NG. Generate numerical data for (step S2). Next, a distribution of numerical data corresponding to the plurality of inspection objects 5 is generated (step S3). This process is performed, for example, by the determination unit 13. The distribution is a distribution of plots on a two-dimensional coordinate plane as shown in FIGS. 3 and 4.
- step S4 the first threshold value and the second threshold value for evaluating the numerical data are generated based on the generated distribution.
- the process of step S4 is performed by the threshold value calculation unit 15.
- step S5 if the numerical data generated in step S2 exists between the first threshold value and the second threshold value, it is decided to perform a human inspection, and the numerical data is less than the first threshold value or the second threshold value. If it is larger than that, it is determined that the AI processing unit 3 automatically determines whether the product is non-defective or defective (step S5).
- the process of step S5 is performed by the determination unit 13. More specifically, the determination unit 13 determines that the product is defective if the numerical data is less than the first threshold value, and determines that the product is non-defective if the numerical data is larger than the second threshold value.
- step S5 the learning model is updated by inputting the result of AI processing or human judgment of non-defective product / defective product into the learning unit 11 together with the numerical data (step S6).
- the learning model is repeatedly updated, and the number of plots existing between the first threshold value and the second threshold value shown in FIG. 4 can be reduced.
- the inspection rate can be reduced.
- FIG. 6 is a graph showing how the inspection rate by humans decreases by repeating learning based on the flowchart of FIG.
- the horizontal axis of the graph of FIG. 6 is the number of processes of the flowchart of FIG. 5, and the vertical axis is the inspection rate [%] by a person.
- the judgment result of non-defective product / defective product by AI processing and the inspection result of non-defective product / defective product by humans become closer to each other.
- the distance between the 1st threshold and the 2nd threshold can be narrowed, and the inspection rate by humans can be reduced.
- the first embodiment whether to automatically determine a non-defective product or a defective product based on the numerical data based on the result of inputting the inspection object 5 into the learning model and comparing the calculated numerical data with the threshold value. , Determine whether to manually inspect non-defective or defective products. That is, in the present embodiment, since the manual inspection is performed only when the AI processing cannot accurately and automatically determine whether the product is a good product or a defective product, the manual inspection ratio can be gradually reduced as the learning model is updated. .. As described above, in the present embodiment, when the inspection process is performed, the inspection accuracy is lowered because the inspection rate is changed manually according to the degree of update of the learning model, instead of performing all the inspection processing by the AI process. However, the inspection accuracy of the AI process can be gradually improved, and the manual inspection ratio can be gradually reduced accordingly.
- the second embodiment determines whether or not the learning model has reached a practical level.
- the learning model is repeatedly updated and the plot existing between the first threshold value and the second threshold value in FIG. It is necessary to reduce the number of to the extent that there is no problem in practical use.
- the inspection device 1 according to the second embodiment has the same block configuration as that of FIG. 1, but the internal configuration of the AI processing unit 3 is partially different from that of FIG.
- FIG. 7 is a block diagram showing an internal configuration of the AI processing unit 3 according to the second embodiment.
- the AI processing unit 3 of FIG. 7 has a practical level determination unit 16 in addition to the configuration of FIG.
- the practical level determination unit 16 determines whether or not the ratio of the numerical data included between the first threshold value and the second threshold value in the distribution of the plot as shown in FIG. 4 is less than the third threshold value, and determines whether or not the ratio is less than the third threshold value. If it is determined that the value is less than, it is determined that the learning model has reached the practical level, and if it is determined that the value is less than the third threshold value, it is determined that the learning model has not yet reached the practical level.
- the ratio is the ratio of the number of numerical data existing between the first threshold value and the second threshold value to the total number of numerical data.
- FIG. 8 is a flowchart showing the processing operation of the inspection device 1 according to the second embodiment.
- Steps S11 to S16 are the same as steps S1 to S6 in FIG.
- the distribution of the numerical data of the inspection object 5 is regenerated using the updated learning model, and the first threshold value and the second threshold value are reset based on the regenerated distribution.
- Step S17 The process of step S17 is performed by, for example, the determination unit 13 and the threshold value calculation unit 15.
- the distance between the first threshold and the second threshold is reset so as to be narrowed. This reduces the number of plots that exist between the first and second thresholds.
- step S18 it is determined whether or not the ratio of the numerical data included between the first threshold value and the second threshold value is less than the third threshold value (step S18). If the ratio is still equal to or higher than the third threshold value, the process returns to step S16 and the learning model is continuously updated. On the other hand, when it is determined in step S18 that the ratio is less than the third threshold value, it is determined that the learning model has reached the practical level (step S19). The processing of steps S18 and S19 is performed by the practical level determination unit 16.
- the ratio of the numerical data existing between the first threshold value and the second threshold value in the distribution of the plot becomes less than the third threshold value, it is determined that the learning model has reached the practical level. Therefore, it is possible to easily and accurately determine whether or not the learning model should be used for the inspection of the actual product.
- step S19 of FIG. 7 If it is determined in step S19 of FIG. 7 that the learning model has reached a practical level, the actual product is set as the inspection target 5, and the processes of steps S4 to S6 of FIG. 1 are performed. That is, even if it is determined that the learning model has reached a practical level, the learning model can be updated every time a new inspection object 5 is inspected, so that the inspection accuracy of the learning model can be further improved and the inspection accuracy is manually improved. The inspection rate can be further reduced.
- the third embodiment regarding the numerical data existing between the first threshold value and the second threshold value, it is determined whether the product is a non-defective product or a defective product by adding defect information.
- the surface of the inspection object 5 has a defect such as a scratch, it is usually judged as a defective product when the defect size exceeds a predetermined size. However, if the defect does not interfere with the operation or function of the inspection object 5, it may be treated as a non-defective product.
- the numerical data existing between the first threshold and the second threshold in the plot diagram as shown in FIG. 4 is re-learned in consideration of the defect information such as the defect size.
- a learning model is generated, the inspection target data is input to the generated re-learning model, and the numerical data is output again based on the calculated result. That is, the determination unit 13 of the present embodiment is based on the first threshold value and the second threshold value set for each type of unique information of the inspection target, and for each type of unique information of the inspection target, the inspection target Decide whether to automatically determine the type or to manually determine the type of the inspection object.
- the unique information is arbitrary information that characterizes the inspection target object, and is a concept that includes defect information such as the defect size described above.
- the inspection device 1 according to the third embodiment has the same block configuration as that of FIG. 1, but the internal configuration of the AI processing unit 3 is partially different from that of FIG.
- FIG. 9 is a block diagram showing an internal configuration of the AI processing unit 3 according to the third embodiment.
- the AI processing unit 3 of FIG. 9 has a re-learning unit 17 and a recalculation unit 18 in addition to the configuration of FIG.
- the re-learning unit 17 relearns based on the defect information of the non-defective product and the defective product to generate a re-learning model.
- the defect information is, for example, the defect size of the inspection object 5.
- the defect size of the inspection object 5 can be obtained from a photographed image taken by the photographing unit 6. More specifically, the difference image between the photographed image as a reference without defects and the photographed image of the inspection object 5 can be regarded as a defect, and the size thereof can be set as the defect size. Alternatively, the operator may measure the defect size in the inspection object 5 in advance and input the measured defect size to the relearning unit 17 separately from the captured image to generate a relearning model. ..
- the recalculation unit 18 inputs the inspection target data into the re-learning model and re-outputs the numerical data based on the calculated result.
- the recalculation unit 18 identifies a defect included in the captured image of the inspection object 5 by the above-mentioned method, inputs the defect size into the re-learning model, and calculates numerical data.
- FIG. 10 is a plot diagram showing the inspection results of a plurality of inspection objects 5.
- the horizontal axis of FIG. 10 is the numerical data calculated by the calculation unit 12, and the vertical axis is the work number of the inspection object 5.
- a plot ⁇ judged to be a non-defective product by the operator a plot ⁇ judged to be a defective product due to a large-sized defect, a plot ⁇ judged to be a defective product due to a medium-sized defect, and a small-sized defect are not acceptable.
- Four types of plots are shown, including plots that are judged to be non-defective.
- the numerical data judged to be a good product or a defective product differs depending on the defect size, and the range of the numerical data judged to be a good product or a defective product also differs.
- FIG. 10 shows an example in which the first threshold value and the second threshold value are individually set for each of the three defect sizes, large, medium, and small. For each defect size, numerical data below the first threshold is automatically determined to be defective, and numerical data larger than the second threshold is automatically determined to be non-defective, and numerical data from the first threshold to the second threshold. Indicates that good / defective products are manually inspected without automatic judgment by AI processing.
- the inspection target 5 including a large-sized or medium-sized defect does not have a large ratio to the total number of numerical data that are judged to be non-defective or defective.
- the inspection object 5 including a small-sized defect has a very large ratio to the total number of numerical data that is determined to be a non-defective product or a defective product. Therefore, the processing may be divided according to whether or not the defect contained in the inspection object 5 has a small size. That is, if the defect size is not small, it is determined whether to perform automatic discrimination based on AI processing or manual discrimination based on the comparison result of the first threshold value and the second threshold value determined in advance, and the defect size is determined. If the size is small, the first threshold value and the second threshold value may be set again.
- FIG. 11 is a flowchart showing the processing operation of the inspection device 1 according to the third embodiment.
- Steps S21 to S25 are the same as steps S1 to S5 in FIG.
- the defect information of the inspection object 5 corresponding to the numerical data included between the first threshold value and the second threshold value is acquired (step S26).
- the defect size can be acquired from the difference image between the photographed image without defects and the photographed image of the inspection object 5. Alternatively, the operator may enter the defect size.
- the operator determines whether the inspection object 5 corresponding to the numerical data included between the first threshold value and the second threshold value is a non-defective product or a defective product in consideration of the defect information (step S27).
- the re-learning unit 17 performs re-learning to generate a re-learning model (step S28).
- step S29 the distribution of the numerical data of the inspection object 5 is generated using the updated learning model in consideration of the defect information.
- the process of step S29 is performed by the determination unit 13, and for example, a plot diagram as shown in FIG. 10 is generated.
- step S30 the first threshold value and the second threshold value are reset based on the distribution generated in consideration of the defect information.
- the process of step S30 is performed by the determination unit 13 and the threshold value calculation unit 15, and for example, the first threshold value and the second threshold value shown by the broken line as shown in FIG. 10 are reset.
- step S31 it is determined whether or not the ratio of the numerical data included between the first threshold value and the second threshold value is less than the third threshold value. If it is not less than the third threshold value, the processing after step S28 is repeated, and if it is less than the third threshold value, it is determined that the learning model has reached the practical level (step S32).
- the inspection object 5 having a defect size larger than a certain level 5
- a first threshold value and a second threshold value for discriminating between a non-defective product and a defective product can be set based on the above. Therefore, the ratio of the numerical data included between the first threshold value and the second threshold value can be reduced, and the manual inspection ratio can be reduced without lowering the inspection accuracy.
- the inspection device 1 according to the fourth embodiment has the same block configuration as that of FIG. 1, and the AI processing unit 3 has the same block configuration as that of FIG. 2 or FIG.
- the processing operation of the determination unit 13 is different from the processing operation of the determination unit 13 according to the first to third embodiments.
- each inspection target 5 is classified a plurality of times based on a plurality of inspection target data obtained by performing a plurality of times of photographing.
- the determination unit 13 according to the fourth embodiment classifies the same inspection object 5 a plurality of times, the frequency of classification into a specific type is equal to or more than the fourth threshold value and less than the fifth threshold value. It is determined that the type of the inspection object is manually determined.
- each of the plurality of inspected objects is photographed a plurality of times (for example, 15 times), and a non-defective product / a defective product is discriminated based on the plurality of captured image data of each inspection object 5.
- the horizontal axis of FIG. 12 is the number of times the product is determined to be defective, and the vertical axis is the identification number (work number) of each inspected object.
- Each plot of FIG. 12 represents a different inspected object, and plots the number of times a defective product is determined as a result of performing a plurality of non-defective / defective product discriminations.
- the determination unit 13 of the present embodiment determines that the inspected object that is determined to be defective is defective more than a predetermined number of times with respect to the total number of times that the good product / defective product is determined for each inspected object. Since there is no problem in determining that, it is decided to perform automatic determination by AI processing for the inspected object judged to be defective more than a predetermined number of times. On the other hand, it is decided to manually inspect the inspected object that is judged to be defective less than the predetermined number of times. Determining whether or not to perform manual inspection at a predetermined number of times with respect to the total number of times means determining whether or not to perform manual inspection at a frequency of determining defective products or non-defective products.
- two or more threshold values are set in order to determine whether or not to perform the manual inspection according to the frequency of variation. It is possible to decide whether or not to perform a manual inspection without it.
- At least a part of the inspection device 1 and the inspection method described in the above-described embodiment may be configured by hardware or software.
- a program that realizes at least a part of the functions of the inspection device 1 and the inspection method may be stored in a recording medium such as a flexible disk or a CD-ROM, read by a computer, and executed.
- the recording medium is not limited to a removable one such as a magnetic disk or an optical disk, and may be a fixed recording medium such as a hard disk device or a memory.
- a program that realizes at least a part of the functions of the inspection device 1 and the inspection method may be distributed via a communication line (including wireless communication) such as the Internet. Further, the program may be encrypted, modulated, compressed, and distributed via a wired line or wireless line such as the Internet, or stored in a recording medium.
- the AI processing unit 3 is connected to a predetermined network such as a public line such as the Internet or a dedicated line, and teacher data and inspection target data are transmitted to the AI processing unit 3 via the network.
- the AI processing result executed by the AI processing unit 3 may be received via the network.
- at least a part of the components in the inspection device 1 may be provided in the cloud environment.
- the defect size of the inspection object 5 is illustrated as the defect information, but the present invention is not limited to this, and the defect position (the position of the defect in the inspection object 5) or the like is defective. It may be used as information.
- the present invention is not limited to this, and for example, the captured image captured by the photographing unit 6 is AI. It may be transmitted to the processing unit 3 and the AI processing unit 3 may generate the teacher data and the inspection target data. In this case, since the captured image captured by the photographing unit 6 is transmitted to the AI processing unit 3 without going through the control unit 2, the teacher data and the inspection target data can be easily generated as compared with each of the above-described embodiments. It can be done at high speed.
- the present invention is not limited to this, and for example, AI.
- the learning model and the re-learning model generated by other than the processing unit 3 may be acquired by the learning unit 11 or the re-learning unit 17.
- the processing load of the AI processing unit 3 can be reduced as compared with each of the above-described embodiments.
- step S3, S13, S17, S23, S29 the setting of the first threshold value and the second threshold value
- steps S4, S14 the setting of the first threshold value and the second threshold value
- step S17, S24, S30 the learning model update
- step S6, S16, S28 the learning model update
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
Abstract
Description
前記検査対象物を前記学習モデルに入力して演算された結果に基づいて、前記検査対象物のタイプの分類精度の高さを数値化した数値データを出力する算出部と、
前記数値データを一種類以上の閾値と比較した結果に基づいて、前記検査対象物のタイプを自動判別するか、人手により前記検査対象物のタイプを判別するかを決定する決定部と、を備える、検査装置が提供される。
前記決定部は、前記数値データが前記第1閾値と前記第2閾値との間に存在する場合には、人手により前記検査対象物のタイプを判別することを決定してもよい。
前記検査対象物を前記再学習モデルに入力して演算された結果に基づいて、前記数値データを出力し直す再算出部と、を備え、
前記決定部は、前記検査対象物の固有情報を考慮に入れつつ、前記数値データを前記第1閾値及び前記第2閾値と比較した結果に基づいて、前記検査対象物のタイプの自動判別を行うか、人手による前記検査対象物のタイプの判別を行うかを決定してもよい。
前記固有情報は、良品及び不良品の欠陥サイズを含んでもよい。
前記学習部は、前記撮影部で撮影された前記検査対象物の複数の撮影画像を前記教師データとして用いてもよい。
前記コンピュータに、
前記検査対象物と同種の複数の検査済対象物を複数のタイプに分類した分類結果の少なくとも一部を教師データとして用いて、前記検査対象物のタイプを判別するための学習を行って学習モデルを生成させるか、或いは前記学習モデルを取得させ、
前記検査対象物を前記学習モデルに入力して演算された結果に基づいて、前記検査対象物のタイプの分類精度の高さを数値化した数値データを出力させ、
前記数値データを一種類以上の閾値と比較した結果に基づいて、前記検査対象物のタイプを自動判別するか、人手により前記検査対象物のタイプを判別するかを決定させる。
前記ネットワークを介して、前記コンピュータに対して、前記教師データと、前記検査対象物のデータとを送信し、
前記コンピュータにより決定させた、前記検査対象物のタイプを自動判別するか、人手により前記検査対象物のタイプを判別するかの情報を、前記ネットワークを介して受信してもよい。
前記コンピュータに、前記数値データが前記第1閾値と前記第2閾値との間に存在する場合には、人手により前記検査対象物のタイプを判別することを決定させてもよい。
前記数値データが前記第1閾値と前記第2閾値との間に存在する場合に、前記検査対象物の固有情報に基づいて再学習を行って再学習モデルを生成させるか、或いは前記再学習モデルを取得させ、
前記検査対象物を前記再学習モデルに入力して演算された結果に基づいて、前記数値データを出力させ、
前記検査対象物の固有情報を考慮に入れつつ、前記数値データを前記第1閾値及び前記第2閾値と比較した結果に基づいて、前記検査対象物のタイプの自動判別を行うか、人手による前記検査対象物のタイプの判別を行うかを決定させてもよい。
前記固有情報は、良品及び不良品の欠陥サイズを含んでもよい。
図1は第1の実施形態による検査装置1の概略構成を示すブロック図である。図1の検査装置1は、検査対象物5の外観検査を行うものである。検査対象物5の種類は特に限定されない。典型的な一例としては、予め決められた仕様で製造された複数の製造物である。より具体的な一例としては、金型で金属材料等をプレスした鍛造物や金型に金属材料等を流し込んで成形した鋳造物などである。検査対象物5の形状やサイズ、材料なども任意であり、金属だけでなく、樹脂等で形成されたものでもよい。
第2の実施形態は、学習モデルが実用レベルに達したか否かを判断するものである。第1の実施形態による学習部11で生成された学習モデルを実製品の検査に用いるには、学習モデルを繰り返し更新して、図4の第1閾値と第2閾値との間に存在するプロットの数が実用上問題ない程度まで少なくなる必要がある。
第3の実施形態は、第1閾値と第2閾値の間に存在する数値データについては、欠陥情報を加味して、良品か不良品かを判断するものである。検査対象物5の表面に傷等の欠陥がある場合、通常は欠陥サイズが所定の大きさを超える場合に不良品と判断することが多い。しかしながら、検査対象物5の動作や機能に全く支障のない欠陥であれば、良品として扱ってよい場合がある。
上述した第1~第3の実施形態では、数値データが第1閾値と第2閾値の間に存在する場合には、人手による良品又は不良品の検査を行う例を説明したが、数値データが良品又は不良品に分類される頻度に応じて、人手による良品又は不良品の検査を行うか否かを決定してもよい。
さらに、上述した各実施形態では、検査装置1の処理動作において、数値データの分布の生成(ステップS3、S13、S17、S23、S29)、第1閾値と第2閾値の設定(ステップS4、S14、S17、S24、S30)及び学習モデルの更新(ステップS6、S16、S28)をそれぞれ実行する例を説明したが、これに限られるものではなく、例えば、これらのステップを省略し、予め設定された閾値に基づいて、自動判断するのか、人手により検査するのかを決定してもよい。
Claims (25)
- 検査対象物と同種の複数の検査済対象物を複数のタイプに分類した分類結果の少なくとも一部を教師データとして用いて、前記検査対象物のタイプを判別するための学習を行って学習モデルを生成するか、或いは前記学習モデルを取得する学習部と、
前記検査対象物を前記学習モデルに入力して演算された結果に基づいて、前記検査対象物のタイプの分類精度の高さを数値化した数値データを出力する算出部と、
前記数値データを一種類以上の閾値と比較した結果に基づいて、前記検査対象物のタイプを自動判別するか、人手により前記検査対象物のタイプを判別するかを決定する決定部と、を備える、検査装置。 - 複数の検査対象物を前記学習モデルに入力して演算された複数の前記数値データに基づいて、前記一種類以上の閾値を算出する閾値算出部を備える、請求項1に記載の検査装置。
- 前記閾値算出部は、前記複数の数値データの統計処理により、前記一種類以上の閾値を算出する、請求項2に記載の検査装置。
- 前記一種類以上の閾値は、第1閾値と、前記第1閾値よりも大きい第2閾値とを含んでおり、
前記決定部は、前記数値データが前記第1閾値と前記第2閾値との間に存在する場合には、人手により前記検査対象物のタイプを判別することを決定する、請求項1乃至3のいずれか一項に記載の検査装置。 - 前記決定部は、前記数値データが前記第1閾値よりも小さいか、又は前記数値データが前記第2閾値よりも大きい場合には、人手により前記検査対象物のタイプを判別せずに、前記検査対象物のタイプを自動判別することを決定する、請求項4に記載の検査装置。
- 前記数値データが前記第1閾値と前記第2閾値との間に存在する場合に、前記検査対象物の固有情報に基づいて再学習を行って再学習モデルを生成するか、或いは前記再学習モデルを取得する再学習部と、
前記検査対象物を前記再学習モデルに入力して演算された結果に基づいて、前記数値データを出力し直す再算出部と、を備え、
前記決定部は、前記検査対象物の固有情報を考慮に入れつつ、前記数値データを前記第1閾値及び前記第2閾値と比較した結果に基づいて、前記検査対象物のタイプの自動判別を行うか、人手による前記検査対象物のタイプの判別を行うかを決定する、請求項4又は5に記載の検査装置。 - 前記決定部は、前記検査対象物の固有情報の種類ごとに設定される前記第1閾値及び前記第2閾値に基づいて、前記検査対象物の固有情報の種類ごとに、前記検査対象物のタイプの自動判別を行うか、人手による前記検査対象物のタイプの判別を行うかを決定する、請求項6に記載の検査装置。
- 前記複数のタイプは、良品のタイプと、不良品のタイプを含んでおり、
前記固有情報は、良品及び不良品の欠陥サイズを含む、請求項6又は7に記載の検査装置。 - 前記第1閾値と前記第2閾値との間に含まれる前記数値データの割合が第3閾値未満になったか否かを判定し、前記割合が前記第3閾値未満になったと判定されると、前記学習モデルが実用レベルに達したと判断する実用レベル判断部を備える、請求項4乃至8のいずれか一項に記載の検査装置。
- 前記決定部は、同一の検査対象物の分類分けを複数回行ったときに、特定のタイプに分類される頻度が第4閾値未満の場合には、人手により前記検査対象物のタイプを判別すると決定する、請求項1乃至9のいずれか一項に記載の検査装置。
- 前記検査対象物を複数の方向から撮影する撮影部を備え、
前記学習部は、前記撮影部で撮影された前記検査対象物の複数の撮影画像を前記教師データとして用いる、請求項1乃至10のいずれか一項に記載の検査装置。 - 複数の検査対象物を前記学習モデルに入力して演算された前記数値データを可視化する可視化部を備える、請求項1乃至11のいずれか一項に記載の検査装置。
- コンピュータにて、検査対象物の検査を行う検査方法であって、
前記コンピュータに、
前記検査対象物と同種の複数の検査済対象物を複数のタイプに分類した分類結果の少なくとも一部を教師データとして用いて、前記検査対象物のタイプを判別するための学習を行って学習モデルを生成させるか、或いは前記学習モデルを取得させ、
前記検査対象物を前記学習モデルに入力して演算された結果に基づいて、前記検査対象物のタイプの分類精度の高さを数値化した数値データを出力させ、
前記数値データを一種類以上の閾値と比較した結果に基づいて、前記検査対象物のタイプを自動判別するか、人手により前記検査対象物のタイプを判別するかを決定させる、検査方法。 - 前記コンピュータは、ネットワークに接続されており、
前記ネットワークを介して、前記コンピュータに対して、前記教師データと、前記検査対象物のデータとを送信し、
前記コンピュータにより決定させた、前記検査対象物のタイプを自動判別するか、人手により前記検査対象物のタイプを判別するかの情報を、前記ネットワークを介して受信する、請求項13に記載の検査方法。 - 前記コンピュータに、複数の検査対象物を前記学習モデルに入力して演算された複数の前記数値データに基づいて、前記一種類以上の閾値を算出させる、請求項13又は14に記載の検査方法。
- 前記コンピュータに、前記複数の数値データの統計処理により、前記一種類以上の閾値を算出させる、請求項15に記載の検査方法。
- 前記一種類以上の閾値は、第1閾値と、前記第1閾値よりも大きい第2閾値とを含んでおり、
前記コンピュータに、前記数値データが前記第1閾値と前記第2閾値との間に存在する場合には、人手により前記検査対象物のタイプを判別することを決定させる、請求項13乃至16のいずれか一項に記載の検査方法。 - 前記コンピュータに、前記数値データが前記第1閾値よりも小さいか、又は前記数値データが前記第2閾値よりも大きい場合には、人手により前記検査対象物のタイプを判別せずに、前記検査対象物のタイプを自動判別することを決定させる、請求項17に記載の検査方法。
- 前記コンピュータに、
前記数値データが前記第1閾値と前記第2閾値との間に存在する場合に、前記検査対象物の固有情報に基づいて再学習を行って再学習モデルを生成させるか、或いは前記再学習モデルを取得させ、
前記検査対象物を前記再学習モデルに入力して演算された結果に基づいて、前記数値データを出力させ、
前記検査対象物の固有情報を考慮に入れつつ、前記数値データを前記第1閾値及び前記第2閾値と比較した結果に基づいて、前記検査対象物のタイプの自動判別を行うか、人手による前記検査対象物のタイプの判別を行うかを決定させる、請求項17又は18に記載の検査方法。 - 前記コンピュータに、前記検査対象物の固有情報の種類ごとに設定される前記第1閾値及び前記第2閾値に基づいて、前記検査対象物の固有情報の種類ごとに、前記検査対象物のタイプの自動判別を行うか、人手による前記検査対象物のタイプの判別を行うかを決定させる、請求項19に記載の検査方法。
- 前記複数のタイプは、良品のタイプと、不良品のタイプを含んでおり、
前記固有情報は、良品及び不良品の欠陥サイズを含む、請求項19又は20に記載の検査方法。 - 前記コンピュータに、前記第1閾値と前記第2閾値との間に含まれる前記数値データの割合が第3閾値未満になったか否かを判定させ、前記割合が前記第3閾値未満になったと判定されると、前記学習モデルが実用レベルに達したと判断させる、請求項17乃至21のいずれか一項に記載の検査方法。
- 前記コンピュータに、同一の検査対象物の分類分けを複数回行ったときに、特定のタイプに分類される頻度が第4閾値未満の場合には、人手により前記検査対象物のタイプを判別すると決定させる、請求項13乃至22のいずれか一項に記載の検査方法。
- 複数の方向から撮影された前記検査対象物の複数の撮影画像を前記教師データとして用いる、
請求項13乃至23のいずれか一項に記載の検査方法。 - 前記コンピュータに、複数の検査対象物を前記学習モデルに入力して演算された前記数値データを可視化させる、請求項13乃至24のいずれか一項に記載の検査方法。
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CA3135320A CA3135320A1 (en) | 2019-03-29 | 2019-03-29 | Inspection device and inspection method |
JP2021511709A JP7229338B2 (ja) | 2019-03-29 | 2019-03-29 | 検査装置及び検査方法 |
US17/598,756 US20220198785A1 (en) | 2019-03-29 | 2019-03-29 | Inspection device and inspection method |
PCT/JP2019/014229 WO2020202332A1 (ja) | 2019-03-29 | 2019-03-29 | 検査装置及び検査方法 |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/JP2019/014229 WO2020202332A1 (ja) | 2019-03-29 | 2019-03-29 | 検査装置及び検査方法 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2020202332A1 true WO2020202332A1 (ja) | 2020-10-08 |
Family
ID=72666673
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2019/014229 WO2020202332A1 (ja) | 2019-03-29 | 2019-03-29 | 検査装置及び検査方法 |
Country Status (4)
Country | Link |
---|---|
US (1) | US20220198785A1 (ja) |
JP (1) | JP7229338B2 (ja) |
CA (1) | CA3135320A1 (ja) |
WO (1) | WO2020202332A1 (ja) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112485259A (zh) * | 2020-11-13 | 2021-03-12 | 湖南交通工程学院 | 一种自适应强的金属表面智能缺陷视觉检测装备及检测方法 |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPS63215954A (ja) * | 1987-03-04 | 1988-09-08 | Toshiba Corp | 表面検査装置 |
JPH09264856A (ja) * | 1996-03-29 | 1997-10-07 | Jieneshisu Technol Kk | 物品外観検査装置 |
JP2006266845A (ja) * | 2005-03-23 | 2006-10-05 | Fuji Xerox Co Ltd | 画像判定装置及び画像判定方法 |
JP2011145179A (ja) * | 2010-01-15 | 2011-07-28 | Panasonic Corp | 官能検査装置及び官能検査方法 |
US20170082555A1 (en) * | 2015-09-18 | 2017-03-23 | Kla-Tencor Corporation | Adaptive Automatic Defect Classification |
JP2019076819A (ja) * | 2017-10-23 | 2019-05-23 | ファナック株式会社 | 仕分けシステム |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP3140177B2 (ja) | 1992-06-02 | 2001-03-05 | 株式会社不二越 | 検査装置の検査方法 |
JP5767963B2 (ja) | 2011-12-28 | 2015-08-26 | 株式会社キーエンス | 外観検査装置、外観検査方法及びコンピュータプログラム |
JP2017211259A (ja) | 2016-05-25 | 2017-11-30 | 株式会社シーイーシー | 検査装置、検査方法、及びプログラム |
JP2018005639A (ja) | 2016-07-04 | 2018-01-11 | タカノ株式会社 | 画像分類装置、画像検査装置、及び、プログラム |
JP6542824B2 (ja) | 2017-03-13 | 2019-07-10 | ファナック株式会社 | 入力画像から検出した対象物の像の尤度を計算する画像処理装置および画像処理方法 |
-
2019
- 2019-03-29 WO PCT/JP2019/014229 patent/WO2020202332A1/ja active Application Filing
- 2019-03-29 CA CA3135320A patent/CA3135320A1/en active Pending
- 2019-03-29 JP JP2021511709A patent/JP7229338B2/ja active Active
- 2019-03-29 US US17/598,756 patent/US20220198785A1/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPS63215954A (ja) * | 1987-03-04 | 1988-09-08 | Toshiba Corp | 表面検査装置 |
JPH09264856A (ja) * | 1996-03-29 | 1997-10-07 | Jieneshisu Technol Kk | 物品外観検査装置 |
JP2006266845A (ja) * | 2005-03-23 | 2006-10-05 | Fuji Xerox Co Ltd | 画像判定装置及び画像判定方法 |
JP2011145179A (ja) * | 2010-01-15 | 2011-07-28 | Panasonic Corp | 官能検査装置及び官能検査方法 |
US20170082555A1 (en) * | 2015-09-18 | 2017-03-23 | Kla-Tencor Corporation | Adaptive Automatic Defect Classification |
JP2019076819A (ja) * | 2017-10-23 | 2019-05-23 | ファナック株式会社 | 仕分けシステム |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112485259A (zh) * | 2020-11-13 | 2021-03-12 | 湖南交通工程学院 | 一种自适应强的金属表面智能缺陷视觉检测装备及检测方法 |
Also Published As
Publication number | Publication date |
---|---|
US20220198785A1 (en) | 2022-06-23 |
CA3135320A1 (en) | 2020-10-08 |
JPWO2020202332A1 (ja) | 2020-10-08 |
JP7229338B2 (ja) | 2023-02-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP6560707B2 (ja) | 加工面品位評価装置 | |
JP6453805B2 (ja) | 製品の異常に関連する変数の判定値を設定する生産システム | |
WO2021000524A1 (zh) | 孔位保护门检测方法、装置、计算机设备和存储介质 | |
JP6530779B2 (ja) | 加工不良要因推定装置 | |
US20180264613A1 (en) | Abnormality detection apparatus and machine learning apparatus | |
US7933441B2 (en) | Method of inspection for inner defects of an object and apparatus for same | |
JP6707920B2 (ja) | 画像処理装置、画像処理方法、およびプログラム | |
JP2019095217A (ja) | 外観検査装置 | |
KR20210008352A (ko) | 촬상된 품목의 결함을 검출하기 위한 시스템 및 방법 | |
EP4036561A1 (en) | Data generation system, learning device, data generation device, data generation method, and data generation program | |
JP6795562B2 (ja) | 検査装置及び機械学習方法 | |
TWI642126B (zh) | 半導體晶圓檢測系統及其方法 | |
CA2964021A1 (en) | Determination of localised quality measurements from a volumetric image record | |
CN114037673B (zh) | 一种基于机器视觉的硬件衔接接口监测方法及*** | |
Schmitt et al. | Machine vision system for inspecting flank wear on cutting tools | |
US20210394247A1 (en) | Method and system for forming a stamped component using a stamping simulation model | |
JP6823025B2 (ja) | 検査装置及び機械学習方法 | |
WO2020202332A1 (ja) | 検査装置及び検査方法 | |
WO2019176988A1 (ja) | 検査システム、識別システム、及び識別器評価装置 | |
JP2020052596A (ja) | 品質予測装置、品質予測方法、及び品質予測プログラム | |
CN115362445A (zh) | 数据分类装置、数据分类方法以及数据分类程序 | |
JP6708695B2 (ja) | 検査装置 | |
CN113419502B (zh) | 基于人工智能的制壳生产过程的工序监测方法与*** | |
JP7327984B2 (ja) | 情報処理装置、情報処理方法、及びプログラム | |
Avendãno et al. | Image based inspection of concrete cracks using UAV photography |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 19922894 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2021511709 Country of ref document: JP Kind code of ref document: A |
|
ENP | Entry into the national phase |
Ref document number: 3135320 Country of ref document: CA |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 19922894 Country of ref document: EP Kind code of ref document: A1 |