Coating inspection system and coating inspection auxiliary system

Abstract

A coating inspection system (100) is used for coating an object (1) by using a coating process comprising a plurality of partial processes. A coating inspection system (100) includes an inspection unit (20) and a determination unit (30). The inspection unit (20) determines whether or not the coating state is good after the object (1) is coated by the coating step. On the basis of data (31) in which "sensing data in which the states of at least two or more of the plurality of partial steps have been acquired" and "the coating state after coating the object (1)" have been associated in advance, and the coating state in which the inspection unit (20) has determined that a defect has occurred, the determination unit (30) identifies a defect step that causes the defect among the plurality of partial steps.

Description

Technical Field

[0001] The present disclosure relates to a coating inspection system and a coating inspection auxiliary system. Background Art

[0002] Until now, in appearance inspections to check the painted state of painted objects, workers have performed visual inspections on all objects. If a visually detectable coating defect is found, the worker will determine it as a defective product and remove it from the production line.

[0003] However, if the presence of coating defects is confirmed by visual inspection by workers, the inspection criteria may vary depending on the worker, resulting in unstable quality of the object.

[0004] Patent Document 1 discloses a surface defect detection device that acquires an image of an inspection site, inputs feature quantities of the image into a neural network, and uses inspectors' inspection results as training data for learning, thereby automatically determining defects in the inspection object. Prior art literature Patent Literature

[0005] Patent Document 1: Japanese Patent Application Laid-Open No. 6-76069 Summary of the Invention -Technical problem to be solved by the invention-

[0006] However, the painted state of an object after painting varies, and in the appearance inspection of the painted state, a technology for automatically identifying the type of painting defects by performing image processing using machine learning on an image of the painted state has not yet been established.

[0007] Furthermore, when a large number of defects occur, workers rely on past experience to determine where the cause lies in the painting process and make corrections to the process. This leads to significant deviations from the workers, and time is wasted trying to identify the cause of the defect. Alternatively, the cause cannot be determined, leading to the recurrence of the defect. These are all issues.

[0008] The purpose of the present disclosure is to improve the accuracy of coating inspection. -Technical solutions to solve technical problems-

[0009] The first aspect is a coating inspection system 100 for coating an object 1 using a coating process composed of a plurality of partial processes. The coating inspection system 100 includes an inspection unit 20 and a judgment unit 30. The inspection unit 20 judges whether the coating state of the object 1 after coating using the coating process is good or bad. The judgment unit 30 determines the bad process in the plurality of partial processes that is the cause of the bad coating state based on data 31 that has previously associated "sensor data of the states of at least two or more partial processes among the plurality of partial processes" with "the coating state after coating the object 1", and the coating state that has been judged as bad by the inspection unit 20.

[0010] In the first aspect, data 31 is prepared that associates sensor data capturing the status of each sub-process in the painting process with the coating status of the object 1 after painting. The coating status determined to be defective is compared with this data 31 to identify the defective process. This allows painting inspections to be performed without relying on operator experience, improving inspection accuracy.

[0011] The second aspect is based on the first aspect, wherein the coating inspection system includes an imaging unit 10 for photographing the object 1 , and the inspection unit 20 determines whether the coating condition of the object 1 is good or bad based on image data of the object 1 photographed by the imaging unit 10 .

[0012] In the second aspect, since the quality / defect of the painted state of the object 1 is determined based on the image data of the object 1 , defect determination can be easily performed.

[0013] According to a third aspect, based on the second aspect, the image data includes at least a normal image and a shape image.

[0014] In the third aspect, defects such as scratches, peeling, low gloss, coagulation, and thin coating thickness can be detected as image density (chroma, brightness, RGB, etc.) using normal images, and defects such as uneven coating can be detected as tiny bumps and depressions using shape images.

[0015] According to a fourth aspect, based on the third aspect, the image data is obtained using stripe pattern illumination.

[0016] In the fourth aspect, the normal image and the shape image can be simultaneously acquired by stripe pattern illumination. In addition, the unevenness of the painted surface can be acquired from the wavy pattern of stripes generated on the painted surface of the object 1 by stripe pattern illumination.

[0017] According to a fifth aspect, in addition to any one of the first to fourth aspects, the inspection unit 20 uses machine learning models 21 and 22 to determine whether the painted state of the object 1 is good or bad.

[0018] In the fifth aspect, for example, the machine learning models 21 and 22 obtained by learning using images of poorly painted surfaces as training data can be used to judge the quality of the painted surface with high accuracy based on image data of the painted surface of the object 1.

[0019] The sixth aspect is that based on any one of the first to fifth aspects, the coating inspection system also includes a notification unit 40, which notifies that a defect has occurred in the defective process based on the determination of the defective process by the judgment unit 30, or notifies abnormal information related to the defective process.

[0020] In the sixth aspect, the operator or the like can be informed of the occurrence of a failure in the coating process and detailed information of the failure (abnormality of the coating apparatus 2 , ie, equipment, parameters, etc.) through the notification unit 40 .

[0021] The seventh aspect is that, based on any one of the first to sixth aspects, the multiple partial processes include a degreasing process and an electrophoretic coating process, and the sensing data includes at least two data of the temperature and pressure of the processing liquid in the degreasing process, the paint temperature in the electrophoretic coating process, and the processing time of the object 1 in each of the multiple partial processes.

[0022] In the seventh aspect, it is possible to determine whether or not a failure has occurred in the degreasing process, the electrophoretic coating process, or the like.

[0023] The eighth aspect is any one of the first to seventh aspects, wherein the judgment of the good / bad coating condition targets at least one of aggregation, scratches, thin coating thickness, peeling, low gloss, and unevenness.

[0024] In the eighth aspect, a defective process can be identified based on each defect pattern.

[0025] A ninth aspect is directed to a coating inspection assistance system 150, which is used when a defect occurs in the coating state of an object 1 after coating using a coating process consisting of multiple partial processes. The coating inspection assistance system 150 includes a generation unit 160, which generates a machine learning model 161 using sensor data obtained for the states of at least two or more of the multiple partial processes and the coating state of the object 1 after coating as learning data. The machine learning model 161 is configured to identify a defective process among the multiple partial processes that is the cause of the defect based on the coating state of the object 1 after coating.

[0026] In the ninth aspect, a defective process can be identified with high accuracy based on the coating state of the object 1 after coating, using the machine learning model 161 obtained by learning sensor data corresponding to various coating defects as training data.

[0027] The tenth aspect is directed to a coating inspection assistance system 150, which is used when a defect occurs in the coating state of an object 1 after being coated using a coating process composed of multiple partial processes. The coating inspection assistance system 150 includes an acquisition unit 170, which uses multivariate analysis based on a physical model 171 of a coating device 2 used in the multiple partial processes to acquire parameters of the physical model 171. The parameters of the physical model 171 represent a correlation between sensor data that has acquired the states of at least two or more of the multiple partial processes and the coating state after coating the object 1. The physical model 171 is configured to be able to determine a defective process that is the cause of the defect among the multiple partial processes based on the coating state after coating the object 1.

[0028] In the tenth aspect, multivariate analysis can be used to obtain parameters of the physical model 171 that represent the correlation between sensor data of each partial process and various coating defects, and using these parameters, the defective process can be accurately determined based on the coating state of the object 1 after coating.

[0029] The eleventh aspect is based on the ninth or tenth aspect, wherein the coating inspection auxiliary system further includes a sensor unit 180, and the sensor unit 180 senses the status of at least two or more partial processes among the multiple partial processes to obtain sensor data.

[0030] In the eleventh aspect, a machine learning model 161 or a physical model 171 can be obtained, which is configured to be able to identify a defective process based on the coating state of the object 1 using the acquired sensor data.

[0031] The twelfth aspect is based on any one of the ninth to eleventh aspects, and the multiple partial processes include a degreasing process and an electrophoretic coating process, and the sensing data includes at least two data of the temperature and pressure of the processing liquid in the degreasing process, the paint temperature in the electrophoretic coating process, and the processing time of the object 1 of each partial process in the multiple partial processes.

[0032] In the twelfth aspect, the machine learning model 161 or the physical model 171 can be used to determine whether or not a defect has occurred in the degreasing process, the electrophoretic coating process, or the like.

[0033] A thirteenth aspect is any one of the ninth to twelfth aspects, wherein the defect in the coating state is at least one of aggregation, scratches, thin coating thickness, peeling, low gloss, and unevenness.

[0034] In the thirteenth aspect, the machine learning model 161 or the physical model 171 can be used to identify defective processes based on each defect pattern. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 An example of a schematic diagram of each of the paint inspection system and the paint inspection support system according to the embodiment is shown; Figure 2 An example of a process flow of a paint inspection performed by the paint inspection system according to the embodiment is shown; Figure 3 shows multiple images of the same specimen obtained using fringe pattern illumination; Figure 4 An example of a normal image is shown; Figure 5 An example of a shape image is shown; Figure 6 Images showing various coating defects occurring on the painted surface of a compressor; Figure 7 An example of sensor data obtained for the states of a plurality of sub-processes constituting a coating process and an example of an inspection result of the coating state after coating an object are shown; Figure 8 Show the Figure 7 An example of parameters of a physical model obtained by performing multivariate analysis on the data shown; Figure 9 Show the basis Figure 8 An example of a defective process identified by the parameters shown. DETAILED DESCRIPTION

[0036] (Implementation Method) Below, embodiments of the present disclosure are described in detail with reference to the accompanying drawings. It should be noted that the present disclosure is not limited to the embodiments shown below, and various modifications can be made without departing from the technical concept of the present disclosure. Since the drawings are used to conceptually illustrate the present disclosure, dimensions, proportions, or quantities may be exaggerated or simplified as needed for ease of understanding. The specific numerical values ​​of the various parameters described in the following embodiments are merely examples and are not limited thereto.

[0037] <Structure of the coating inspection system> A coating inspection system 100 according to an embodiment is used to coat an object 1 using a coating process consisting of multiple sub-processes. Object 1 is, for example, a compressor. The sub-processes, in order of performance, include degreasing, water washing, ultrasonic water washing, surface conditioning, zinc phosphate conversion coating, water washing, electrophoretic coating, water washing, and baking and drying.

[0038] like Figure 1 As shown, the paint inspection system 100 mainly includes an inspection unit 20 and a determination unit 30. The paint inspection system 100 may further include an imaging unit 10 and a notification unit 40.

[0039] Inspection unit 20 determines whether the painted state of object 1 after painting in a painting process is good or bad. If inspection unit 20 detects at least one of the following on the painted surface of object 1, for example, aggregation, scratches, thin coating thickness, flaking, low gloss, or unevenness, inspection unit 20 determines the painted state as bad. Inspection unit 20 uses machine learning models 21 and 22 to determine whether the painted state of object 1 is good or bad.

[0040] The determination unit 30 identifies a defective process among the multiple sub-processes that is the cause of the defect based on data 31 (hereinafter referred to as associated data 31) that pre-associates "sensor data obtained regarding the states of at least two or more sub-processes among the multiple sub-processes" with "the coating state of the object 1 after coating" and the coating state determined to be defective by the inspection unit 20. For example, the determination unit 30 compares the sensor data of each sub-process of the object 1 that is a qualified product without coating defects with the sensor data of each sub-process of the object 1 that has been determined to have coating defects, and identifies the sub-process that matches the difference between the two data as a defective process.

[0041] Sensor data is acquired by sensing the status of the equipment used in each of the multiple sub-processes, namely, the coating device 2, during the sub-process. Examples of sensing items include the concentration, temperature, and pressure of the degreasing liquid and the degreasing nozzle flow rate in the degreasing process as a pre-process; the pressure, flow rate, and conductivity of the purified water in the purified water rinse as a pre-process; the flow rate of the air blow as a pre-process; the temperature and conductivity of the paint and the coating voltage in the electrophoretic coating process; the pressure and flow rate of the UF (ultrafiltration) wash in the UF rinse as a post-process; the pressure and conductivity of the purified water in the purified water rinse as a post-process; and the temperature of the baking and drying process as a post-process.

[0042] The related data 31 will be described in detail in the later-described <Painting Inspection Support System>.

[0043] The inspection unit 20 and the determination unit 30 are each composed of, for example, a processor and a memory, the memory storing programs and information for operating the processor. The inspection unit 20 and the determination unit 30 are configured to be able to exchange data such as images with each other. The inspection unit 20 may include, for example, a storage unit such as a hard disk to store captured images and other data. The determination unit 30 may also include, for example, a storage unit such as a hard disk to store images and other data transmitted from the inspection unit 20. The inspection unit 20 and the determination unit 30 may also share a common storage unit.

[0044] The coating inspection system 100 may include a notification unit 40, which notifies that a defect has occurred in a defective process, or notifies abnormal information related to the defective process, based on the determination of the defective process by the judgment unit 30. The notification unit 40 can be, for example, a terminal used by an operator, such as a personal computer, a tablet computer, a smart phone, or a dedicated terminal for inspection. The judgment unit 30 and the notification unit 40 are connected to each other via wired or wireless communication so that information can be sent from the judgment unit 30 to the notification unit 40. The inspection unit 20, the judgment unit 30, and the notification unit 40 can be connected to each other via a network or integrated.

[0045] In this example, the inspection unit 20 uses an image of the object 1 in order to determine whether the coating state of the object 1 is good or bad after the object 1 is coated using a coating process. Therefore, the coating inspection system 100 may include an imaging unit 10 that captures the object 1. The imaging unit 10 is, for example, a camera that captures a still image of the object 1. In order to capture the object 1 using one or more imaging units 10, the object 1 may be held on a rotating device that rotates the object 1. The imaging unit 10 and the inspection unit 20 are connected to each other via wired or wireless connections so that image data can be sent from the imaging unit 10 to the inspection unit 20. The imaging unit 10, the inspection unit 20, and the judgment unit 30 may be connected to each other via a network or may be integrated. The imaging unit 10 may be configured to capture the coating state of the object 1 after coating based on instructions from the inspection unit 20 or the judgment unit 30.

[0046] In this example, the camera unit 10 acquires image data that includes at least a normal image and a shape image. Therefore, the paint inspection system 100 may include a lighting device 15 that performs stripe pattern lighting on the object 1. The lighting device 15 may be a single device specifically used for stripe pattern lighting, or a plurality of devices that are constructed and arranged to perform stripe pattern lighting. The camera unit 10 may also be a dedicated camera that is constructed so as to acquire a normal image and a shape image of the painted surface of the object 1 obtained by stripe pattern lighting by the lighting device 15. The normal image and the shape image will be described in detail in the <Processing flow of paint inspection> described later.

[0047] The inspection unit 20 determines whether the painted state of the object 1 is good or bad based on the image data of the object 1 captured by the imaging unit 10. The inspection unit 20 may also use machine learning models 21 and 22 learned using images of the painted object 1 as training data to determine whether the painted state of the object 1 is good or bad based on the image data of the painted state of the object 1. The machine learning models 21 and 22 may include a first model 21 corresponding to a normal image and a second model 22 corresponding to a shape image.

[0048] The machine learning models 21 and 22 can be, for example, neural networks such as multilayer perceptrons capable of deep learning, support vector machines, discriminant functions, or Bayesian networks, but are not particularly limited. When the machine learning models 21 and 22 are configured as neural networks, the machine learning models 21 and 22 have, for example, three layers: an input layer, an intermediate layer, and an output layer. Each layer includes one or more neurons, and each neuron in the input layer is connected to each neuron in the intermediate layer, and each neuron in the intermediate layer is connected to each neuron in the output layer. Image data of the object 1 captured by the imaging unit 10 is input to each neuron in the input layer.

[0049] Before using the machine learning models 21 and 22 to determine whether the coating condition of the object 1 is good or bad, the inspection unit 20 or other information processing device performs machine learning on the machine learning models 21 and 22, i.e., calculates model data, such as model data representing the settings of the neural network. The model data includes, for example, the number of layers in the neural network, the number of neurons (nodes) contained in each layer, and the coupling coefficient (coupling load) between neurons. The inspection unit 20 uses the model data to set the machine learning models 21 and 22, and uses the set machine learning models 21 and 22 (i.e., the learned models) to determine whether the coating condition of the object 1 is good or bad.

[0050] When machine learning models 21 and 22 are constructed as neural networks, the neural network's machine learning process begins by first inputting image data of an object 1 with poorly painted surfaces into the input layer as training data. This data is then transferred from the input layer to the output layer, whereupon output information is obtained. Next, using the obtained output information and the training data for the output information, the coupling coefficient between the input and output layers and the bias assigned to the neurons in the intermediate layers are calculated. For example, the coupling coefficient and bias are adjusted to minimize the difference between the output information and the training data.

[0051] As a result of the machine learning described above, model data including the adjusted coupling coefficient and deviation is generated. The model data includes, for example, the number of layers in the neural network, the number of neurons belonging to each layer, the coupling coefficient, and the deviation. The generated model data is stored in, for example, the storage unit of the inspection unit 20. Before judging whether the coating condition of the object 1 is good or bad, the inspection unit 20 sets the neural network based on the stored model data. That is, the inspection unit 20 sets the number of layers, the number of neurons, the coupling coefficient, and the deviation in the neural network constituting the machine learning models 21 and 22 to the values ​​assigned to the model data. After doing so, the inspection unit 20 uses the machine learning models 21 and 22 using the model data to judge whether the coating condition of the object 1 is good or bad.

[0052] <Structure of the coating inspection support system> The coating inspection support system 150 according to the embodiment is used when a defect occurs in the coating state of an object 1 after coating using a coating process composed of multiple sub-processes. The coating inspection support system 150 generates the associated data 31 used by the determination unit 30 of the coating inspection system 100 .

[0053] like Figure 1 As shown, the paint inspection support system 150 mainly includes a generator 160. The paint inspection support system 150 may further include a sensor 180. As described in detail in the "Modification of the Paint Inspection Support System" below, the paint inspection support system 150 may further include an acquirer 170 instead of the generator 160.

[0054] The generation unit 160 generates a machine learning model (third model) 161 using "the sensor data of the states of at least two or more partial processes among the multiple partial processes constituting the painting process" and "the painting state after the object 1 is painted" as learning data. The learned machine learning model 161 is configured to be able to determine the defective process that is the cause of the defect among the multiple partial processes based on the painting state after the object 1 is painted. In other words, the machine learning model 161 obtained by learning using the sensor data corresponding to various painting defects as training data is used to determine the defective process based on the painting state of the object 1 that has been judged as defective by the judgment unit 30 of the painting inspection system 100. The learned machine learning model 161 can, for example, compare the sensor data of each partial process of the object 1 that is a qualified product without painting defects with the sensor data of each partial process of the object 1 that has been judged as having painting defects, and determine the partial process that matches the difference between the two data as a defective process.

[0055] The machine learning model 161, with the exception of its input and output, can be constructed similarly to the machine learning models 21 and 22 used in the inspection unit 20 of the paint inspection system 100. The data associated with the learned machine learning model 161 (such as the model body and model data) is an example of the associated data 31 used by the determination unit 30 of the paint inspection system 100.

[0056] The generation unit 160 is composed of, for example, a processor and a memory, and the memory stores programs and information for making the processor work. The generation unit 160 may have, for example, a storage unit such as a hard disk to store data such as sensor data. The generation unit 160 and the judgment unit 30 of the coating inspection system 100 may be connected to each other via wired or wireless communication so that data related to the machine learning model 161 can be sent from the generation unit 160 to the judgment unit 30. The generation unit 160 and the judgment unit 30 may be connected to each other via a network or may be integrated. Alternatively, it may be that the coating inspection system 100 and the coating inspection auxiliary system 150 are configured to be non-connected, and the data related to the machine learning model 161 obtained in the generation unit 160 is saved in a storage medium, and the data related to the machine learning model 161 is read out from the storage medium by the judgment unit 30.

[0057] The generation unit 160 can obtain the "painting state of the object 1 after painting" (for example, image data of the object 1 that has been judged to be poorly painted) from the inspection unit 20 of the coating inspection system 100 as learning data. The generation unit 160 and the inspection unit 20 can be connected to each other via wired or wireless so that the learning data can be sent from the inspection unit 20 to the generation unit 160. The generation unit 160 and the inspection unit 20 can be connected to each other via a network or integrated. Alternatively, it can be such that the coating inspection system 100 and the coating inspection auxiliary system 150 are configured to be non-connected, and the learning data obtained in the inspection unit 20 is saved in a storage medium, and the generation unit 160 reads the learning data from the storage medium.

[0058] The sensing unit 180 senses the progress of a portion of the process being processed in real time, and acquires sensing data used in the generating unit 160. The sensing unit 180 and the generating unit 160 may be connected to each other via a wired or wireless connection so that sensing data can be sent from the sensing unit 180 to the generating unit 160. The sensing unit 180 and the generating unit 160 may be connected to each other via a network, or may be integrated. Alternatively, the sensing unit 180 and the generating unit 160 may be configured to be non-connected, and the sensing data acquired by the sensing unit 180 may be stored in a storage medium, and the sensing data may be read from the storage medium by the generating unit 160. The sensing unit 180 may be configured to sense the progress of a portion of the process being processed based on an instruction from the generating unit 160.

[0059] <Painting inspection process> Below, we take the case of using normal image and shape image to judge the defect as an example. Figures 2 to 6 The process flow of the paint inspection performed by the paint inspection system 100 will be described.

[0060] After the object 1 is painted in the painting process, in step S1, the imaging unit 10 captures the object 1 to obtain image data of the painted surface of the object 1. The painting process is composed of a plurality of sub-processes including, for example, a degreasing process and an electrophoretic coating process.

[0061] In this example, to obtain image data containing at least a normal image and a shape image, the painted surface of the object 1 is illuminated with a stripe pattern using the lighting device 15 during imaging. The imaging unit 10 is a dedicated camera adapted for stripe pattern illumination, and can capture multiple different types of original images in a single shot. Figure 3 Eight original images of the same sample obtained using fringe pattern illumination are shown. Figure 3 In the image, X1 to X4 images are images captured by changing the phase of the illuminated stripe pattern along the X direction, and Y1 to Y4 images are images captured by changing the phase of the illuminated stripe pattern along the Y direction. Figure 3 The eight original images shown are synthesized to obtain five types of image data. Specifically, the five types of image data are a normal image, a mirror reflection image, a diffuse reflection image, a gloss ratio image, and a shape image. The normal image is an image obtained by averaging all the original images captured, and is used to confirm the overall appearance of the captured object. The mirror reflection image is an image obtained by extracting only the mirror reflection component in the stripe pattern, and is used to detect line scratches, abrasions, etc. on the glossy surface. The diffuse reflection image is an image obtained by extracting the diffuse reflection component based on the comparison between the normal image and the mirror reflection image, and is used to detect foreign matter, dirt, etc. The gloss ratio image is an image obtained by extracting the part with changing gloss based on the comparison between the mirror reflection image and the diffuse reflection image, and is used to detect the darkening of the surface, etc. The shape image is an image obtained by extracting the changing part such as concave and convex from the wavy shape of the stripe pattern, and is used to detect dents, shallower concave and convex, etc. Figure 4 、 Figure 5 Examples of various images, including a normal image and a shape image, are shown.

[0062] Next, in step S2, the inspection unit 20 uses a machine learning model (first model) 21, which is trained using normal images of coating defects as training data, to detect defects such as agglomeration, scratches, peeling, low gloss, and thin coating thickness as image shading (color, brightness, RGB, etc.) from the normal image of the coating state of the object 1. For example, the first model 21 may compare the normal image of the object to be judged with a normal image of a qualified product without coating defects, and judge the object as defective if the difference between the two is significant.

[0063] If no defects are detected in step S2, in step S3, the inspection unit 20 uses a machine learning model (second model) 22, which is trained using the shape image of the coating defect as training data, to detect defects such as uneven coating as minute bumps and unevenness from the shape image of the painted state of the object 1. For example, the second model 22 may compare the shape image of the object to be judged with the shape image of a qualified product without coating defects, and judge the object as defective if the difference between the two is significant.

[0064] If no defects are detected in step S3, the coating inspection system 100 notifies the operator in step S4 that the object 1 is a qualified product. The inspection result can be displayed on the notification unit 40. Then, in step S5, the object 1 without coating defects is conveyed to the next process.

[0065] If a defect is detected in step S2 or S3 , the paint inspection system 100 notifies the operator of the inspection result that a paint defect is detected in the object 1 in step S6 . The inspection result may be displayed on the notification unit 40 . Figure 6 Images showing various coating defects occurring on the coating surface of a compressor serving as the object 1 are shown.

[0066] In step S7 following step S6, the determination unit 30 identifies the defective process among the multiple sub-processes constituting the coating process that is the cause of the defect based on the associated data 31 and the coating state determined to be defective by the inspection unit 20 (in this example, the normal image or shape image determined to be defective). The associated data 31 is data that associates "sensor data obtained for the states of at least two or more of the multiple sub-processes" with "the coating state of the object 1 after coating." For example, the sensor data includes at least two of the following: the temperature and pressure of the treatment liquid in the degreasing process, the temperature of the coating material in the electrophoretic coating process, and the treatment time of the object 1 in each sub-process.

[0067] In this example, as the associated data 31, data (model body and model data, etc.) related to the machine learning model (third model) 161 obtained in the generation unit 160 of the paint inspection assistance system 150 is used. The third model 161 obtained by learning the sensor data corresponding to various paint defects as training data can be used to determine the defective process based on the paint state (normal image or shape image) of the object 1 that has been judged as defective by the judgment unit 30. It should be noted that, as described in the "Variation Example of the Paint Inspection Assistance System" described later, as the associated data 31, data (model formula and model parameters, etc.) related to the physical model 171 obtained in the acquisition unit 170 of the paint inspection assistance system 150 can also be used.

[0068] Next, in step S8, the notification unit 40 notifies that a defect has occurred in the defective process based on the determination of the defective process by the judgment unit 30 in step S7, or notifies abnormal information related to the defective process (such as abnormalities in the coating device 2, i.e., equipment, parameters, etc.).

[0069] <Features of the embodiment> The coating inspection system 100 of the embodiment is used to coat an object 1 using a coating process composed of a plurality of partial processes. The coating inspection system 100 includes an inspection unit 20 and a judgment unit 30. The inspection unit 20 judges whether the coating state of the object 1 after coating using the coating process is good or bad. The judgment unit 30 determines the bad process that is the cause of the bad coating state among the plurality of partial processes based on the associated data 31 that has previously associated "sensor data of the states of at least two or more partial processes among the plurality of partial processes" with "the coating state after the object 1 is coated" and the coating state that has been judged as bad by the inspection unit 20.

[0070] In the coating inspection system 100 of the embodiment, the correlation between the results of the coating inspection (thin coating thickness, scratches, coating peeling, etc.) and the sensor data of the coating process is pre-analyzed and prepared as the associated data 31, and the coating state that has been judged to be poor is compared with the associated data 31, thereby being able to determine where the problem occurred in the coating process or the coating device 2, i.e., the equipment. As a result, the coating inspection can be carried out without relying on the experience of the operator, so the inspection accuracy can be improved. In addition, by automating the coating appearance inspection, labor can be saved and the inspection cost can be reduced, and it is also possible to achieve a balance of inspection costs on a global scale without relying on the inspection location. In addition, by automating the coating appearance inspection, the difference in inspection standards that depend on the operator can be reduced, thereby stabilizing the quality, and it is possible to stabilize the quality on a global scale without relying on the inspection location. In addition, since the visual inspection of the coating appearance can be eliminated, repetitive work can be eliminated, thereby reducing the burden on the operator. Furthermore, the time required to identify the cause of coating defects can be reduced, and the cause can be identified without relying on the operator's experience, thus making the time required to identify the cause constant. Furthermore, by identifying the cause of coating defects, appropriate coating conditions can be proposed, which can reduce the defect rate in the future. By making the defect rate infinitely close to zero, the coating inspection itself can be eliminated, which is expected to bring about a significant cost reduction effect.

[0071] Alternatively, the paint inspection system 100 of the embodiment includes an imaging unit 10 for capturing an image of the object 1, and the inspection unit 20 determines whether the paint condition of the object 1 is good or bad based on image data of the object 1 captured by the imaging unit 10. In this manner, image processing performed by the inspection unit 20 facilitates the determination of a bad condition.

[0072] In the paint inspection system 100 of the embodiment, image data can include at least a normal image and a shape image. This allows defects such as scratches, peeling, low gloss, clumping, and thin paint thickness to be detected as image shading (color, brightness, RGB, etc.) using the normal image, while defects such as uneven paint can be detected as minute bumps and depressions using the shape image.

[0073] In the paint inspection system 100 of the embodiment, image data can be acquired using stripe pattern illumination. This eliminates the need for multiple illumination devices, allowing for simultaneous acquisition of both normal and shape images. Furthermore, the unevenness of the painted surface of the object 1 can be determined from the wavy pattern of stripes created by stripe pattern illumination.

[0074] In the paint inspection system 100 of the embodiment, the inspection unit 20 can use machine learning models 21 and 22 to determine whether the paint condition of the object 1 is good or bad. For example, machine learning models 21 and 22, trained using images of poorly painted surfaces as training data, can be used to accurately determine whether the paint condition of the object 1 is good or bad based on image data captured of the painted surface. In particular, by constructing the first model 21 and the second model 22 corresponding to the normal image and the shape image, respectively, the accuracy of paint inspection using image processing can be significantly improved.

[0075] The coating inspection system 100 of the embodiment may further include a notification unit 40 that, based on the determination of a defective process by the determination unit 30, a defect has occurred in the defective process, or abnormality information related to the defective process. In this way, the operator can be informed of the occurrence of a defect in the coating process and detailed information about the defect (abnormalities in the coating apparatus 2, i.e., equipment, parameters, etc.) through the notification unit 40.

[0076] In the coating inspection system 100 of the embodiment, the multiple sub-processes comprising the coating process may include a degreasing process and an electrophoretic coating process. The sensor data may include at least two of the following: the temperature and pressure of the treatment fluid in the degreasing process, the temperature of the coating material in the electrophoretic coating process, and the treatment time of the object 1 in each sub-process. This allows the determination of whether any defects have occurred in the degreasing process, the electrophoretic coating process, or the like.

[0077] In the coating inspection system 100 of the embodiment, the good / bad coating state can be determined based on at least one of coagulation, scratches, thin coating thickness, peeling, low gloss, and unevenness. This allows the defective process to be identified based on each defect pattern.

[0078] The coating inspection assistance system 150 of the embodiment is used when a defect occurs in the coating state of the object 1 after coating using a coating process composed of multiple partial processes. The coating inspection assistance system 150 includes a generation unit 160, which generates a machine learning model (third model) 161 using "sensor data of the states of at least two or more partial processes among the multiple partial processes constituting the coating process" and "the coating state after the object 1 is painted" as learning data. The machine learning model 161 is configured to be able to determine the defective process that is the cause of the defect among the multiple partial processes based on the coating state after the object 1 is painted. In this way, the machine learning model 161 obtained by learning the sensor data corresponding to various coating defects as training data can be used to determine the defective process with high accuracy based on the coating state after the object 1 is painted.

[0079] The coating inspection support system 150 of the embodiment may further include a sensor unit 180 that senses the status of at least two of the plurality of sub-processes and acquires sensor data. This allows for the generation of a machine learning model 161 that uses the acquired sensor data to identify defective processes based on the coating status of the object 1.

[0080] In the coating inspection support system 150 of the embodiment, the multiple sub-processes may include a degreasing process and an electrophoretic coating process, and the sensor data may include at least two of the following: the temperature and pressure of the treatment fluid in the degreasing process, the paint temperature in the electrophoretic coating process, and the treatment time of the object 1 in each sub-process. This allows the machine learning model 161 to determine whether a defect has occurred in the degreasing process, the electrophoretic coating process, or the like.

[0081] In the coating inspection support system 150 of the embodiment, the coating defect may be at least one of coagulation, scratches, thin coating thickness, peeling, low gloss, and unevenness. In this way, the machine learning model 161 can be used to identify defective processes based on each defect pattern.

[0082] <Modification of the coating inspection support system> The coating inspection support system 150 of the modified example is used when a defect occurs in the coating state of the object 1 after coating using a coating process composed of multiple sub-processes. The coating inspection support system 150 creates the associated data 31 used by the determination unit 30 in the coating inspection system 100 of the embodiment.

[0083] The difference between the coating inspection support system 150 of the modified example and the coating inspection support system 150 of the embodiment is that the coating inspection support system 150 of the modified example includes an acquisition unit 170 instead of the generation unit 160 (see Figure 1 The paint inspection support system 150 of the modified example may further include a sensor unit 180 , similar to the paint inspection support system 150 of the embodiment.

[0084] Acquisition unit 170 uses multivariate analysis based on a physical model 171 of coating apparatus 2 used in multiple sub-processes constituting the coating process to acquire parameters for physical model 171. These parameters represent the correlation between sensor data acquired regarding the states of at least two or more of the multiple sub-processes and the coating state of object 1 after coating. Physical model 171, with its parameters determined, is configured to identify a defective process among the multiple sub-processes that is the cause of a defect based on the coating state of object 1 after coating.

[0085] The physical model 171 is represented by, for example, the following regression equation (model equation): The model equation is created for each failure mode (aggregation, scratches, peeling, low gloss, thin coating thickness, unevenness, etc.).

[0086] Y m =a1·X1+a2·X2+a3·X3+a4·X4+…+a N ·X N +b Y m Indicates the target variable corresponding to each defect mode, X1~X N represents the explanatory variables corresponding to each sensing item (degreasing temperature, degreasing pressure, coating (paint) temperature, coating voltage, drying temperature, etc.), a1 to a N Represents the regression coefficient (correlation coefficient).

[0087] The acquisition unit 170 uses, for example, Figure 7 The "sensor data of the status of multiple partial processes constituting the painting process" and the "inspection results of the painting status after the object 1 is painted (appearance inspection results)" shown are data collected by the sensor unit 180 and the inspection unit 20 of the painting inspection system 100, and a multivariate analysis based on the physical model 171 is performed, thereby calculating the standard regression coefficient (model parameter) for each defect mode. Figure 8 An example of the model parameters calculated by the acquisition unit 170 is shown. Figure 7 In the table, the unit of temperature is °C, the unit of pressure is MPa, the unit of flow rate is L / min, and the unit of voltage is V.

[0088] The model parameters represent the contribution of each sensing item to each bad mode. Figure 9 As shown, the sensor items that should be monitored first when a defect occurs are known, thereby enabling the defective process to be identified. Figure 9 Shown based on Figure 8 The model parameters shown determine the monitoring priority of the sensing items when uneven coating occurs (order from high to low standardized regression coefficient).

[0089] The data (model formula, model parameters, etc.) related to the physical model 171 acquired by the acquisition unit 170 as described above is another example of the related data 31 used by the determination unit 30 of the paint inspection system 100 .

[0090] The acquisition unit 170 is composed of, for example, a processor and a memory, and the memory stores programs and information for making the processor work. The acquisition unit 170 may have, for example, a storage unit such as a hard disk to store data such as sensor data. The acquisition unit 170 and the judgment unit 30 of the coating inspection system 100 may be connected to each other via wired or wireless communication so that data related to the physical model 171 can be sent from the acquisition unit 170 to the judgment unit 30. The acquisition unit 170 and the judgment unit 30 may be connected to each other via a network or may be integrated. Alternatively, it may be that the coating inspection system 100 and the coating inspection auxiliary system 150 are configured to be non-connected, and the data related to the physical model 171 obtained in the acquisition unit 170 is saved in a storage medium, and the data related to the physical model 171 is read out from the storage medium by the judgment unit 30.

[0091] The acquisition unit 170 can acquire inspection data (appearance inspection results) about the "painting status of the object 1 after painting" from the inspection unit 20 of the coating inspection system 100. The acquisition unit 170 and the inspection unit 20 can be connected to each other via wired or wireless communication so that inspection data can be sent from the inspection unit 20 to the acquisition unit 170. The acquisition unit 170 and the inspection unit 20 can be connected to each other via a network or integrated. Alternatively, the coating inspection system 100 and the coating inspection auxiliary system 150 can be configured to be non-connected, and the inspection data obtained by the inspection unit 20 is saved in a storage medium, and the acquisition unit 170 reads the inspection data from the storage medium.

[0092] The sensing unit 180 senses the progress of a portion of the process being processed in real time, and acquires sensing data used in the acquisition unit 170. The sensing unit 180 and the acquisition unit 170 can be connected to each other via a wired or wireless connection so that sensing data can be sent from the sensing unit 180 to the acquisition unit 170. The sensing unit 180 and the acquisition unit 170 can be connected to each other via a network or integrated. Alternatively, the sensing unit 180 and the acquisition unit 170 can be configured to be non-connected, and the sensing data acquired by the sensing unit 180 can be stored in a storage medium, and the sensing data can be read from the storage medium by the acquisition unit 170. The sensing unit 180 can be configured to sense the progress of a portion of the process being processed based on an instruction from the acquisition unit 170.

[0093] In the modified example of the coating inspection assistance system 150 described above, multivariate analysis can be used to obtain parameters of the physical model 171 representing the correlation between the sensor data of each partial process and various coating defects, and using this parameter, the defective process can be determined with high precision based on the coating state of the object 1 after coating.

[0094] The modified coating inspection support system 150 may further include a sensor unit 180 that senses the status of at least two of the multiple sub-processes and acquires sensor data. This allows the creation of a physical model 171 that is configured to identify defective processes based on the coating status of the object 1 using the acquired sensor data.

[0095] In a modified example of the coating inspection support system 150, the multiple sub-processes may include a degreasing process and an electrophoretic coating process. The sensor data may include at least two of the following: the temperature and pressure of the treatment fluid in the degreasing process, the temperature of the coating material in the electrophoretic coating process, and the treatment time of the object 1 in each sub-process. This allows the physical model 171 to determine whether any defects have occurred in the degreasing process, the electrophoretic coating process, or the like.

[0096] In the modified coating inspection support system 150 , the coating defect may be at least one of coagulation, scratches, thin coating thickness, peeling, low gloss, and unevenness. Thus, the physical model 171 can be used to identify defective processes based on each defect pattern.

[0097] (Other embodiments) In the above-described embodiments (including variations, the same below), the object 1 is not limited to a compressor, but may also be various objects that can be painted. In addition, in the inspection unit 20 of the coating inspection system 100, an image is used to determine whether the coating condition of the object 1 is good or bad. However, the inspection unit 20 is not limited to images, and may also use a measurement value of the coating condition measured by a measuring device such as a film thickness meter for measuring the coating thickness to determine whether the coating condition of the object 1 is good or bad. In addition, the inspection unit 20 uses machine learning models 21 and 22 obtained by learning images of poor coating as training data to determine whether the coating condition is good or bad based on the image data of the coating condition of the object 1. However, the inspection unit 20 may not use a machine learning model, but may determine whether the coating condition is good or bad based on a predetermined threshold value (rule base) from the brightness value of the image of the coating condition of the object 1, the area of ​​the foreign matter in the image, and the like.

[0098] While the above describes embodiments and variations, it should be understood that various modifications to the embodiments and detailed structures may be made without departing from the spirit and scope of the claims. The above embodiments, variations, and other embodiments may be appropriately combined or substituted without affecting the functionality of the disclosed subject matter. The terms "first," "second," "third," and so on, described above, are used to distinguish between statements containing these terms and do not limit the number or order of such statements. Industrial Applicability

[0099] In summary, the present disclosure is useful for a paint inspection system and a paint inspection assistance system. - Explanation of symbols -

[0100] 1 Object 2 Coating equipment 10. Camera Department 15 lighting fixtures 20 Inspection Department 21. First Model (Machine Learning Model) 22 Second Model (Machine Learning Model) 30 Judgment Department 31 Linked Data (data) 40 Notification Department 100 Painting Inspection System 150 Painting Inspection Assistance System 160 Generation Department 161 The third model (machine learning model) 170 Acquisition Department 171 Physical Model 180 Sensor Department

Claims

1. A coating inspection system (100) for coating an object (1) using a coating process consisting of a plurality of partial processes, characterized in that: The coating inspection system (100) includes an inspection unit (20) and a judgment unit (30), wherein the inspection unit (20) judges whether the coating state of the object (1) after coating using the coating process is good or bad. The judgment unit (30) determines a defective process among the multiple partial processes that is the cause of the defect based on sensor data that has been previously acquired on the status of at least two or more partial processes among the multiple partial processes and data (31) that associates the coating status of the object (1) after coating, and the coating status that has been judged as defective by the inspection unit (20).

2. The coating inspection system according to claim 1, characterized in that: The coating inspection system includes an imaging unit (10) for imaging the object (1), The inspection unit (20) determines whether the coating state of the object (1) is good or bad based on the image data of the object (1) captured by the imaging unit (10).

3. The coating inspection system according to claim 2, characterized in that: The image data includes at least a normal image and a shape image.

4. The coating inspection system according to claim 3, characterized in that: The image data is acquired using fringe pattern illumination.

5. The coating inspection system according to any one of claims 1 to 4, characterized in that: The inspection unit (20) uses machine learning models (21, 22) to determine whether the coating state of the object (1) is good or bad.

6. The coating inspection system according to any one of claims 1 to 5, characterized in that: The coating inspection system includes a notification unit (40) which notifies that a defect has occurred in the defective process or abnormal information related to the defective process based on the determination of the defective process by the judgment unit (30).

7. The coating inspection system according to any one of claims 1 to 6, characterized in that: The multiple partial processes include a degreasing process and an electrophoretic coating process, The sensor data includes at least two data of the temperature and pressure of the treatment liquid in the degreasing process, the paint temperature in the electrophoretic coating process, and the treatment time of the object (1) in each of the plurality of partial processes.

8. The coating inspection system according to any one of claims 1 to 7, characterized in that: The judgment of whether the coating state is good or bad is based on at least one of aggregation, scratches, thin coating thickness, peeling, low gloss, and unevenness.

9. A coating inspection support system (150) for use when a coating state of an object (1) coated by a coating process consisting of a plurality of partial processes becomes poor, characterized in that: The coating inspection support system (150) includes a generating unit (160) that generates a machine learning model (161) using sensor data of states of at least two or more partial processes among the plurality of partial processes and a coating state of the object (1) after coating as learning data. The machine learning model (161) is configured to be able to identify a defective process that is a cause of the defect among the plurality of partial processes based on a coating state of the object (1) after coating.

10. A coating inspection support system (150) for use when a coating state of an object (1) coated by a coating process consisting of a plurality of partial processes becomes poor, characterized in that: The coating inspection auxiliary system (150) includes an acquisition unit (170), wherein the acquisition unit (170) acquires parameters of the physical model (171) using multivariate analysis based on a physical model (171) of a coating device (2) used in the plurality of partial processes, wherein the parameters of the physical model (171) represent a correlation between sensor data of states of at least two or more partial processes among the plurality of partial processes acquired and a coating state of the object (1) after coating. The physical model (171) is configured to be able to identify a defective process that is a cause of the defect among the plurality of partial processes based on a coating state of the object (1) after coating.

11. The coating inspection auxiliary system according to claim 9 or 10, characterized in that: The coating inspection auxiliary system further includes a sensing unit (180) configured to sense the states of at least two or more partial processes among the plurality of partial processes to obtain the sensing data.

12. The coating inspection auxiliary system according to any one of claims 9 to 11, characterized in that: The multiple partial processes include a degreasing process and an electrophoretic coating process, The sensor data includes at least two data of the temperature and pressure of the treatment liquid in the degreasing process, the paint temperature in the electrophoretic coating process, and the treatment time of the object (1) in each of the plurality of partial processes.

13. The coating inspection auxiliary system according to any one of claims 9 to 12, characterized in that: The defective coating state is at least one of aggregation, scratches, thin coating thickness, peeling, low gloss, and unevenness.

Images