A device for managing defects in inspected objects.

Abstract

A control device (10) calculates the luminance of the transmitted light in each of a plurality of regions included in an object to be inspected (S) on the basis of information acquired from a camera (201). The control device (10) estimates the shape of a first surface and the shape of a second surface on the basis of information acquired from sensors (301, 401). The control device (10) estimates density information indicating a density distribution of each of the plurality of regions on the basis of the shape of the first surface and the shape of the second surface, and the luminance of the transmitted light in each of the plurality of regions.

Description

[Technical Field]

[0001] The present disclosure relates to an apparatus for managing defects of an inspection object, a machine learning method, and a method for manufacturing a trained model. [Background Art]

[0002] In quality inspection of manufactured products, it is required to determine defects on the surface and inner layer of an inspection object by non-destructive inspection. For determining the presence or absence of defects, there is a determination method in which an inspection object is irradiated with light, the state thereof is imaged by a camera, and a location where the luminance of the acquired image becomes high is determined to have a defect.

[0003] For example, Japanese Patent Laying-Open No. 2014-163694 (Patent Document 1) discloses a defect inspection method in which a sheet-shaped inspection object is irradiated with first light from one side, the inspection object is irradiated with second light from the other side, imaging data corresponding to the intensities of the reflected light of the first light and the transmitted light of the second light is acquired, and an inner layer defect of the inspection object is detected based on the acquired intensities of the reflected light and the transmitted light. In this defect inspection method, whether a defect occurring in the inspection object is a surface defect or an inner layer defect is determined by using a luminance ratio obtained by dividing the luminance value of the acquired reflected light or transmitted light by the luminance value when there is no defect. [Prior Art Literature] [Patent Literature]

[0004] [Patent Document 1] Japanese Patent Laying-Open No. 2014-163694 [Summary of the Invention] [Problem to be Solved by the Invention]

[0005] The defect inspection method described in Patent Document 1 can determine whether or not there are defects in a sheet-like object under inspection, but it cannot estimate the volume of the area where a defect occurs (for example, the volume of vacuoles created by the defect) or the density of the area containing the defect (for example, the density of some areas decreases due to the inclusion of vacuoles) in an object under inspection that has thickness.

[0006] For example, in foamed polyurethane resin molded products, which are formed into a three-dimensional shape by injecting resin into a mold and going through a foaming process, defects such as chips or voids that result in scratches may occur on the surface or in the inner layer. In foamed polyurethane resin molded products where the molding results are irregular in this way, it is difficult to control the density of the molded product, and it has not been possible to create an optimal molding process that uniformizes the density.

[0007] This disclosure is made to solve the problems described above, and its purpose is to provide a technology that can appropriately control the density of an object under inspection in relation to defects in the object under inspection.

[0008] The apparatus for managing defects in an object under inspection according to this disclosure comprises an illumination device, an imaging device, a sensor, and a processing device. The illumination device irradiates transmitted light onto a first surface of the object under inspection. The imaging device is positioned opposite the illumination device and images a second surface of the object under inspection that is opposite the first surface. The sensor measures information about the external shape of the object under inspection. The processing device processes the information obtained from the sensor and the imaging device. Based on the information obtained from the imaging device, the processing device calculates the luminance of transmitted light in each of a plurality of regions contained in the object under inspection. Each of the plurality of regions is a three-dimensional region between a two-dimensional minute first region contained in the first surface and a two-dimensional minute second region contained in the second surface. The second region is the region reached by the transmitted light irradiated onto the first region after passing through the object under inspection. Based on the information obtained from the sensor, the processing device estimates the shape of the first surface and the shape of the second surface. Based on the shape of the first surface and the shape of the second surface, and the luminance of transmitted light in each of the plurality of regions, the processing device estimates density information indicating the density distribution of each of the plurality of regions.

[0009] The apparatus for managing defects in an object under inspection according to this disclosure comprises a data acquisition unit and a model generation unit. The data acquisition unit acquires training data including density information showing the density distribution of each of a plurality of regions contained in the object under inspection, and injection path information of the object under inspection corresponding to the density information. The model generation unit uses the training data to generate a trained model for inferring injection path information from density information. The injection path information includes an injection path in which a robot that generates an object under inspection by injecting the material of the object under inspection into a mold for forming the object under inspection moves while injecting the material of the object under inspection.

[0010] The apparatus for managing defects in an object under inspection according to this disclosure comprises a data acquisition unit and an inference unit. The data acquisition unit acquires density information showing the density distribution of each of a plurality of regions contained in the object under inspection. The inference unit outputs injection path information from the density information acquired by the data acquisition unit, using a trained model for inferring injection path information of the object under inspection from the density information. The injection path information includes information including the injection path of a robot that generates the object under inspection by injecting the material of the object under inspection into a mold for forming the object under inspection, as the robot moves while injecting the material of the object under inspection.

[0011] The machine learning method for managing defects in an object under inspection according to this disclosure comprises the steps of acquiring training data including density information showing the density distribution of each of several regions contained in the object under inspection, and injection path information of the object under inspection corresponding to the density information, and generating a trained model for inferring injection path information from density information using the training data. The injection path information includes information including an injection path in which a robot that generates an object under inspection by injecting the material of the object under inspection into a mold for forming the object under inspection moves while injecting the material of the object under inspection.

[0012] The method for manufacturing a trained model according to this disclosure comprises an acquisition step of acquiring training data including density information indicating the density distribution of each of a plurality of regions contained in the object to be inspected, and injection path information of the object to be inspected corresponding to the density information, and a generation step of generating a trained model for inferring injection path information from density information using the training data. The injection path information includes information including an injection path in which a robot that generates the object to be inspected by injecting the material of the object to be inspected into a mold for forming the object to be inspected moves while injecting the material of the object to be inspected. [Effects of the Invention]

[0013] According to this disclosure, the density of the inspected object can be appropriately controlled in relation to defects in the inspected object. [Brief explanation of the drawing]

[0014] [Figure 1] This diagram shows the overall schematic configuration of the defect management device according to Embodiment 1. [Figure 2] This figure shows the cross-sectional shape of the object being inspected. [Figure 3] This figure shows the cross-sectional shape of the object being inspected. [Figure 4] This figure shows the cross-sectional shape of the object being inspected. [Figure 5] This figure shows the cross-sectional shape of the object being inspected. [Figure 6] This graph shows a virtual cross-section of the object being inspected. [Figure 7] This graph shows a virtual cross-section of the object being inspected. [Figure 8] This graph shows the brightness and cross-sectional density of the object being inspected. [Figure 9] This graph shows the corrected cross-sectional density of the inspected object. [Figure 10] This is a flowchart showing the processes performed by the control unit. [Figure 11] This is a flowchart showing the defect detection process. [Figure 12] This diagram shows the hardware configuration of the control unit. [Figure 13]It is a diagram showing a schematic overall configuration of a defect management apparatus according to Embodiment 2. [Figure 14] It is a diagram showing a process in which an object to be inspected is generated by a robot. [Figure 15] It is a diagram showing how an object to be inspected is measured. [Figure 16] It is a diagram showing how an object to be inspected is measured. [Figure 17] It is a diagram showing how an object to be inspected is measured. [Figure 18] It is a diagram showing how an object to be inspected is measured. [Figure 19] It is a configuration diagram of a learning apparatus. [Figure 20] It is a flowchart showing a learning process of the learning apparatus. [Figure 21] It is a configuration diagram of an inference apparatus. [Figure 22] It is a flowchart showing an inference procedure performed by the inference apparatus. MODE FOR CARRYING OUT THE INVENTION

[0015] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings. In the drawings, the same or corresponding portions are denoted by the same reference symbols, and the description thereof will not be repeated in principle. However, unless otherwise specified, the dimensions, materials, shapes, relative arrangements, and the like of the constituent components described in the embodiments are not intended to limit the scope of the present invention only thereto.

[0016] Embodiment 1. FIG. 1 is a diagram showing a schematic overall configuration of a defect management apparatus 1 according to Embodiment 1. The defect management apparatus 1 is an apparatus that manages defects in an object to be inspected S. The defect management apparatus 1 inspects the object to be inspected S conveyed by a conveyor 501.

[0017] The object under inspection S is, for example, a foamed polyurethane resin molded product used as thermal insulation. Foamed polyurethane resin molded products are formed by injecting resin into a mold having the desired shape and then forming it into a three-dimensional shape through a foaming process. Depending on how the foaming progresses, defects such as chips or voids that result in scratches may occur on the surface or inner layer of the finished resin molded product.

[0018] The defect management device 1 according to this embodiment 1 obtains the accurate density of the object to be inspected S by acquiring the external shape of the object S and detecting the amount of defects present in the internal layer. The defect management device 1 can determine the presence or absence of defects based on the density of the object S, rather than determining the presence or absence of scratches or cavities that may occur.

[0019] The defect management device 1 comprises a light source 101 as an illumination device, a camera 201 as an imaging device, sensors 301 and 401, a transport conveyor 501, a control device 10 as a processing device, and a storage unit 660 for storing data output by the control device 10.

[0020] Sensors 301 and 401 measure information regarding the external shape of the object S under inspection. The sensors (sensors 301 and 401) include a first sensor (also called the "lower sensor") 401 that measures the first surface (bottom surface) of the lower part of the object S under inspection, and a second sensor (also called the "upper sensor") 301 that measures the second surface (top surface) of the upper part of the object S under inspection. By using sensors 301 and 401, the external shape and dimensions of the object S under inspection can be obtained.

[0021] The light source 101 irradiates light (transmitted light) onto the underside of the object S under inspection. The camera 201 is positioned opposite the light source 101 and images the upper surface of the object S under inspection, which is on the opposite side of the underside. The camera 201 images the transmitted light that has been irradiated from the light source 101 and passed through the object S under inspection.

[0022] The control device 10 processes the information acquired from sensors 301, 401 and camera 201 to estimate the density information of the object to be inspected S (described later) and to determine whether or not the object to be inspected S has defects.

[0023] The conveyor belt 501 moves the object to be inspected S in the X-axis direction. Sensors 301 and 401 continuously measure the shape of the object to be inspected S as it moves along the conveyor belt 501. Sensors 301 and 401 measure the distance in the Z-direction from their respective installation positions to the object to be inspected S, with the Y-direction component being the amount of measurement required. Sensors 301 and 401 are selected to have an appropriate resolution that affects measurement accuracy within the range covering the object to be inspected S.

[0024] The upper shape acquisition unit 611 acquires information from the sensor (upper sensor) 301. The upper shape acquisition unit 611 acquires information on the height z1(y,t) for the Y-direction component in the transport time unit t, as a measurement result from the upper sensor 301.

[0025] Similarly, the lower shape acquisition unit 621 acquires information from the sensor (lower sensor) 401. The lower shape acquisition unit 621 acquires height z2(y,t) information for the Y-direction component in the transport time unit t, as a measurement result from the lower sensor 401.

[0026] After measurements are completed by sensors 301 and 401, inspection is performed using the light source 101 and camera 201. If there is a cavity in the object S under inspection, the amount of light transmitted from the light source 101 increases compared to when there is no cavity, resulting in higher brightness in the image captured by camera 201. The transmitted light processing unit 631 acquires the captured data from camera 201. Based on the brightness of the captured data, it becomes possible to detect internal defects in the object S under inspection.

[0027] Figures 2 to 5 show the cross-sectional shapes of the object S under inspection. The ideal cross-section 11 shown in Figure 2 is the cross-section of the object S under inspection if it is ideally formed. Figures 3 to 5 show the cross-sections of the object S under inspection that were actually produced.

[0028] As shown in Figure 3, the upper shape acquisition unit 611 acquires the shape of the cross section 12 within the sensor detection range 302 of the upper sensor 301. Specifically, the upper shape acquisition unit 611 can acquire the shape of the upper surface of the object S under inspection. A surface defect 21 exists on the upper part of the cross section 12.

[0029] As shown in Figure 4, the lower shape acquisition unit 621 acquires the shape of the cross section 13 within the sensor detection range 402 of the lower sensor 401. Specifically, the lower shape acquisition unit 621 can acquire the shape of the lower surface of the object S under inspection. Surface defects 21 are present on the upper part of the cross section 13, and surface defects 22 are present on the lower part of the cross section 13.

[0030] For example, if the cross-sectional shape of the object S under inspection is circular, the upper sensor 301 will detect the shape of the upper part of the circle, and the lower sensor 401 will detect the shape of the lower part of the circle.

[0031] As shown in Figure 5, the light source 101 irradiates light onto the lower surface of the object S under inspection. The camera 201 images the upper surface of the object S under inspection, allowing observation of the state of the transmitted light that has passed through the object S. The transmitted light processing unit 631 acquires the image captured by the camera 201. In addition to surface defects 21 and 22, an internal layer defect 23 is present in the cross section 14.

[0032] Figures 6 and 7 are graphs showing a virtual cross-section of the object S under inspection. As shown in Figure 6, in the cross-section 12 shown in Figure 3, the outer shape of the upper surface of the object S under inspection is measured as the measurement height z1(y,t). When the ideal cross-section 11 is measured, the measurement height zT1 is obtained.

[0033] The upper shape comparison unit 612 calculates the difference (increase) δz1(y,t) [=z1(y,t)-zT1] between the acquired measured height z1(y,t) and the measured height zT1 of the ideal cross-section 11.

[0034] However, if a surface defect is detected at δz1(y,t), the surface defect is removed to obtain a virtual cross-section A1(y,t). For example, at δz1(y,t), "0" is used as the threshold, and any defective portion 31 (detected as a surface defect 21) that is less than or equal to 0 is removed from δz1(y,t) (if it is less than 0, it is corrected to 0). In this way, the surface defect 21 is corrected as the measured height zT1.

[0035] As shown in Figure 7, in the cross-section 13 shown in Figure 4, the outer shape of the lower surface of the object S under inspection is measured as the measurement height z2(y,t). When the ideal cross-section 11 is measured, the measurement height zT2 is obtained.

[0036] The lower shape comparison unit 622 calculates the difference (decrease) δz2(y,t) [=z2(y,t)-zT2] between the acquired measured height z2(y,t) and the measured height zT2 of the ideal cross-section 11.

[0037] However, if a surface defect is detected at δz1(y,t), the surface defect is removed to obtain a virtual cross-section A2(y,t). For example, at δz2(y,t), the defective portion 32 (detected as surface defect 22) that is below a threshold (a negative value in this example) is removed from δz2(y,t).

[0038] Figure 8 is a graph showing the brightness and cross-sectional density of the object S under inspection. In the imaging data from camera 201, the amount of transmitted light increases in areas containing the inner layer defects 23 compared to areas without the inner layer defects 23, resulting in higher brightness values. Similarly, areas containing surface defects 21 and 22 also show higher brightness values.

[0039] The brightness in an ideal cross-section 11 with a uniform inner layer state is brightness LT. The brightness of the cross-section 14 of the inspected object S shown in Figure 5 is L(y,t). The brightness value is high due to surface defects 21, inner layer defects 23, and surface defects 22.

[0040] The defect detection unit 632 calculates the cross-sectional density A3(y,t) in the Y-axis direction. The cross-sectional density A3(y,t) is obtained by correcting and transforming the value of L(y,t) based on the correlation with the brightness LT. The cross-sectional density A3(y,t) is determined based on the density of the inspected object S when it is an ideal cross-section 11 (the value based on brightness LT), and the amount of defects due to the internal layer defects (the value expressed in L(y,t)).

[0041] Figure 9 is a graph showing the corrected cross-sectional density of the object S under inspection. The cross-sectional density calculation unit 641 combines and adds the cross-sectional densities in the Y-axis direction to obtain the cross-sectional density A(y,t). Specifically, as shown in Figure 9, the cross-sectional density A(y,t) is calculated using the formula A(y,t) = A1(y,t) + A2(y,t) + A3(y,t).

[0042] The value A(t), obtained by integrating the cross-sectional density A(y,t) in the Y-axis direction, can be derived as the cross-sectional density in a small time unit. Furthermore, integrating the value A(t) in the X-axis direction can be derived as the total density A of the object S being inspected. The output unit 651 outputs information such as the cross-sectional density A(t) in a small time unit and the value of the total density A as results.

[0043] The cross-sectional density calculation unit 641 can calculate the cross-sectional area and total volume of the object S under inspection, excluding any cavities. If there are no cavities, the density of the object S is homogeneous. If there are cavities, the density of the object S decreases locally. For example, based on the distance between the top and bottom surfaces and density information, it is possible to calculate the cross-sectional area and total volume of the object S excluding any cavities. Furthermore, the total weight of the object S can be calculated from the total density A and total volume.

[0044] The following describes the processes performed by the control device 10 using a flowchart. Figure 10 is a flowchart of the processes performed by the control device 10. Hereafter, steps will also be simply referred to as "S". This process can be activated, for example, each time the camera 201, sensors 301 and 401 detect an object S to be inspected moving on the conveyor belt 501 (at each sampling time).

[0045] When this process starts, in S11, the control device 10 acquires information (distance to the bottom surface) detected by the lower sensor 401. In S12, the control device 10 estimates the shape of the bottom surface based on the information acquired from the lower sensor 401. As explained using Figures 4 and 7, for example, the outer shape of the bottom surface of the object S under inspection is calculated as the measured height z2(y,t), and a virtual cross section A2(y,t) is obtained.

[0046] In S13, the control device 10 acquires information (distance to the top surface) detected by the upper sensor 301. In S14, the control device 10 estimates the shape of the top surface based on the information acquired from the upper sensor 301. As explained using Figures 3 and 6, for example, the outer shape of the top surface of the object S under inspection is calculated as the measured height z1(y,t), and a virtual cross section A1(y,t) is obtained.

[0047] In S15, the control device 10 acquires the information detected by the camera 201. In S16, the control device 10 calculates the luminance of transmitted light in each region based on the information acquired from the camera 201. As explained using Figures 5 and 8, for example, L(y,t) is calculated as luminance, and furthermore, the cross-sectional density A3(y,t) is obtained.

[0048] In S17, the control device 10 estimates density information based on the shape of the lower surface, the shape of the upper surface, and the luminance of the transmitted light, and terminates this process. As described above, the lower sensor 401 provides the measurement height z2(y,t), the upper sensor 301 provides the measurement height z1(y,t), and the camera 201 provides the luminance L(y,t). Finally, the cross-sectional density A(y,t) is obtained from the calculated virtual cross-section A2(y,t), virtual cross-section A1(y,t), and cross-sectional density A3(y,t). The value A(t) is obtained by integrating the cross-sectional density A3(y,t) in the Y-axis direction.

[0049] This process is performed at predetermined intervals (sampling periods) while the object S under inspection is moving in the X-axis direction. While the object S under inspection is being transported, the length of the object S in the transport direction (X-axis direction) can be calculated based on the inspection start and end timings of the sensor 301, etc. The total density A is obtained by integrating the value A(t) in the X-axis direction. In addition, as described above, the cross-sectional area, total volume, and total weight of the object S under inspection, excluding the cavity, can be calculated.

[0050] According to the configuration described above, the control device 10 divides the object under inspection S into multiple regions and estimates the density of each of these regions. Each of these regions is a minute region in the Y-axis direction, as shown by y(t) in Figure 6 or Figure 7, and a minute region in the X-axis direction due to the movement of the object under inspection S at each sampling period.

[0051] Each of the multiple regions is a region demarcated by a small width in the Y-axis direction × a small width in the Z-axis direction, and a density can be obtained for each of these minute regions. Here, the region in contact with the bottom surface is called the first region, and the region in contact with the top surface is called the second region.

[0052] In other words, each of the multiple regions is a three-dimensional region between a two-dimensional minute first region contained in the lower surface and a two-dimensional minute second region contained in the upper surface. The second region is the region reached by transmitted light that has passed through the object S being inspected after being irradiated into the first region.

[0053] Sensor 401 measures the distance to the first region on the bottom surface. Sensor 301 measures the distance to the second region on the top surface. Sensors 301 and 401 make it possible to measure the shape of the object S under inspection by repeating these measurements.

[0054] The control device 10 estimates the shape of the bottom surface based on the information acquired from the sensor 401. The control device 10 estimates the shape of the top surface based on the information acquired from the sensor 301. Furthermore, the control device 10 calculates the luminance of transmitted light in each of the multiple regions contained in the object S under inspection based on the information acquired from the camera 201. Based on the shape of the bottom surface, the shape of the top surface, and the luminance of transmitted light in each of the multiple regions, the control device 10 estimates density information that shows the density distribution of each of the multiple regions.

[0055] In this way, it is possible to determine whether the density of multiple regions is homogeneous or not. If there are regions where the density decreases, it can be determined that there are defects in the inspected object, such as vacuoles. In this embodiment, not only is the presence or absence of an inner layer of the inspected object determined, but the quantitative density of the inspected object can also be obtained, making it possible to numerically grasp the amount of defects and judge whether the quality is acceptable or not. This makes it possible to appropriately manage the density of the inspected object in relation to defects in the inspected object.

[0056] Figure 11 is a flowchart of the defect detection process. The defect detection process should be initiated, for example, when the inspection shown in Figure 10 is completed for the entire object S under inspection.

[0057] When the defect detection process starts, the control device 10 detects vacuoles in S21 based on the density information obtained in the process shown in Figure 10. For example, if there are regions where the density decreases across multiple areas, it can be determined that vacuoles exist. For example, the inner layer 23 shown in Figures 5 and 9 falls into this category.

[0058] In S22, the control device 10 determines whether or not it has detected a vacuole of a predetermined volume or larger. For example, the volume of a vacuole can be calculated according to the amount of density reduction and the area over which the density is reduced.

[0059] If the control device 10 determines that it has detected a vacuole (an unacceptably large inner layer defect) exceeding a predetermined volume (YES in S22), it determines that the inspected object has a defect (S23) and terminates the defect determination process.

[0060] On the other hand, if the control device 10 does not detect any vacuoles exceeding a predetermined volume (NO in S22), it determines that there are no defects in the inspected object (S24) and terminates the defect detection process. There is no problem if the internal defects are within acceptable quality limits (for example, minute vacuoles).

[0061] In this way, the control device 10 detects vacuoles in the object S under inspection based on density information. If the control device 10 detects vacuoles exceeding a predetermined volume, it determines that the object S under inspection has a defect. In this way, it is possible to determine whether or not the object under inspection has a defect due to vacuoles. This makes it possible to appropriately manage the density of the object under inspection in relation to defects in the object under inspection.

[0062] Figure 12 shows the hardware configuration of the control device 10. The control device 10 can perform corresponding operations using either digital circuit hardware or software. When the functions of the control device 10 are implemented using software, the control device 10 can, for example, include a processor 51 and a memory 52 connected by a bus 53, as shown in Figure 12, and the processor 51 can execute a program stored in the memory 52.

[0063] Embodiment 2. Figure 13 is a diagram showing the overall schematic configuration of the defect management device 2 according to Embodiment 2. The defect management device 2 is a device that manages defects in the object S under inspection.

[0064] The defect management device 1 according to Embodiment 1 comprises a light source 101, a camera 201, sensors 301 and 401, a transport conveyor 501, and a control device 10. The defect management device 2 according to Embodiment 2 further comprises a trained model storage unit 661, a learning device 701, an inference device 761, and a robot 801.

[0065] As a result, the defect management device 2 inspects the object to be inspected S being transported by the transport conveyor 501, generates (manufactures) a trained model by the learning device 701, performs inference using the trained model by the inference device 761, and generates the object to be inspected S by the robot 801.

[0066] Figure 14 shows the process by which the robot 801 generates the object to be inspected S. The robot 801 generates the object to be inspected S by injecting the material (base material) of the object to be inspected S into a mold 821 for forming the object to be inspected S.

[0067] Robot 801 controls the trajectory (injection path) and the amount of base material injected over time to form the desired shape (mold 821) of the object to be inspected S, thereby generating a molding trajectory 811. Methods for injecting the base material include robot 801 and other means. Robot 801 is controllable in both horizontal and angular orientation, which affect the quality of the object to be inspected S.

[0068] Figures 15 to 18 show how the object under inspection S is measured. The equipment configuration in Figure 15 is the same as the equipment configuration shown in Figure 1. The molded product S produced after the base material injection is placed on the conveyor belt 501 and inspected by the defect control device 2.

[0069] As shown in Figures 16 to 18, first, the generated object S is transported by the transport conveyor 500, then the object S is inspected on the transport conveyor 501, and finally, the inspected object S is transported by the transport conveyor 502.

[0070] The shape of the object S to be inspected is estimated using sensors 301 and 401 installed between conveyor belts 500 and 501. Next, the brightness of the transmitted light passing through the object S is measured using a light source 101 and a camera 201 installed between conveyor belts 501 and 502. In addition, the defect management device 1 according to Embodiment 1 may also be configured using conveyor belts 500 to 502 in this manner.

[0071] The following steps are performed based on the cross-sectional density information acquired by the defect control device 2: a learning phase in which the process of injecting the base material into the inspected object S is learned, and an utilization phase in which the molding process is used to bring the cross-sectional density closer to the standard.

[0072] <Learning Phase> Figure 19 is a diagram showing the configuration of the learning device 701. The learning device 701 comprises a data acquisition unit 711 and a model generation unit 721.

[0073] The data acquisition unit 711 acquires data as training data, which includes density information (same as in Embodiment 1) showing the density distribution of each of the multiple regions contained in the object under inspection S, and injection path information of the object under inspection S corresponding to the density information. The density information also indicates the presence or absence of defects (surface / internal layer defects) at each position in the object under inspection S and the size of the defects. The injection path information includes the injection path and the amount of material (base material) injected at the injection position on the injection path. The injection path is the path along which the robot 801 moves while injecting material into the object under inspection S.

[0074] The model generation unit 721 uses training data to generate a trained model for inferring injection path information from the density information of the object under inspection S. In other words, a trained model is generated that infers injection path information (injection path information corresponding to the density information) in the input state from the density information of the object under inspection S. When density information (location and amount of surface / internal defects) is input to the trained model, appropriate injection path information (injection path and amount of material injected at injection positions on the injection path) with little deviation from the reference cross-sectional density is output.

[0075] The model generation unit 721 can use any known learning algorithm, such as supervised learning, unsupervised learning, or reinforcement learning. As an example, let's explain the case where reinforcement learning is applied. In reinforcement learning, an agent (an acting entity) in a given environment observes the current state (environmental parameters) and decides what action to take. The environment changes dynamically as a result of the agent's actions, and the agent is given a reward according to the changes in the environment. The agent repeats this process and learns the action strategy that yields the most rewards through a series of actions. Representative reinforcement learning methods include Q-learning and TD-learning. For example, in the case of Q-learning, the general update formula for the action-value function Q(s,a) is expressed in equation 1.

[0076]

number

[0077] In Math 1, st represents the state of the environment at time t, and at represents the action at time t. The action at changes the state to st+1. rt+1 represents the reward received for this change in state, γ represents the discount rate, and α represents the learning rate. Note that γ is in the range of 0 < γ ≤ 1 and α is in the range of 0 < α ≤ 1. The injection path information of the object under inspection S becomes the action at, and the density information of the molded product (location and amount of surface / internal defects) becomes the state st, and the best action at for state st at time t is learned.

[0078] The update formula, represented by equation 1, increases the action value Q of action a if the action value Q of action a with the highest Q value at time t+1 is greater than the action value Q of action a performed at time t, and decreases the action value Q if the opposite is true. In other words, the action value function Q(s,a) is updated so that the action value Q of action a at time t approaches the best action value at time t+1. As a result, the best action value in a given environment is sequentially propagated to the action values ​​in previous environments.

[0079] As described above, when a trained model is generated by reinforcement learning, the model generation unit 721 includes a reward calculation unit 722 and a function update unit 723.

[0080] The reward calculation unit 722 calculates a reward based on the injection path information and density information of the object S under inspection. The reward calculation unit 722 calculates a reward r based on the deviation value from the reference cross-sectional density. For example, if the deviation value from the reference cross-sectional density decreases, the reward r is increased (for example, a reward of "1" is given), and on the other hand, if the deviation value from the reference cross-sectional density increases, the reward r is decreased (for example, a reward of "-1" is given).

[0081] The function update unit 723 updates the function for determining injection path information in the input state according to the reward calculated by the reward calculation unit 722, and outputs it to the trained model storage unit 661. For example, in the case of Q-learning, the action-value function Q(st,at) represented by Equation 1 is used as the function for calculating injection path information in the input state.

[0082] The learning process described above is repeated. The trained model memory unit 661 stores the action-value function Q(st,at) updated by the function update unit 723, i.e., the trained model.

[0083] Next, we will explain the learning process of the learning device 701 using Figure 20. Figure 20 is a flowchart of the learning process of the learning device 701.

[0084] In S101, the data acquisition unit acquires injection path information and density information of the object S under inspection as training data.

[0085] In S102, the model generation unit 721 calculates a reward based on the injection path information and density information. Specifically, the reward calculation unit 722 acquires the injection path information and density information and decides whether to increase the reward (S103) or decrease the reward (S104) based on the deviation value from a predetermined reference cross-sectional density.

[0086] For example, the model generation unit 721 increases the reward given to the learned model (action value function Q) when the difference between the predetermined standard density information and the density information of the object S under inspection is below a predetermined standard (for example, a state where there are no unacceptably large vacuoles).

[0087] If the reward calculation unit 722 determines that the reward should be increased, it increases the reward in S103. On the other hand, if the reward calculation unit determines that the reward should be decreased, it decreases the reward in S104.

[0088] In S105, the function update unit 723 updates the action-value function Q(st,at) represented by Equation 1, which is stored in the trained model storage unit 661, based on the reward calculated by the reward calculation unit 722.

[0089] The learning device repeatedly executes steps S101 to S105 above and stores the generated action-value function Q(st,at) as a trained model.

[0090] In this embodiment, the learning device stores the learned model in a learned model storage unit 661 located outside the learning device; however, the learned model storage unit 661 may be located inside the learning device 701.

[0091] Thus, the learning method includes the steps of acquiring training data including density information and injection path information, and generating a trained model for inferring injection path information from density information using the training data. Furthermore, the method for manufacturing the trained model includes the acquisition step of acquiring training data including density information and injection path information of the inspected object S corresponding to the density information, and the generation step of generating a trained model for inferring injection path information from density information using the training data.

[0092] <Utilization Phase> Figure 21 is a diagram showing the configuration of the inference device 761. The inference device 761 comprises a data acquisition unit 711 and an inference unit 771. The data acquisition unit 711 acquires density information (same as in Embodiment 1) that shows the density distribution of each of the multiple regions contained in the object S under inspection.

[0093] The inference unit 771 uses a trained model to infer injection path information of the object S under inspection from density information, and outputs injection path information from the density information acquired by the data acquisition unit 711. By inputting the density information acquired by the data acquisition unit 711 into this trained model, it is possible to infer injection path information (injection path information corresponding to the density information) in the input state.

[0094] In this embodiment, the injection path information for the input state is output using a trained model learned by the model generation unit 721 of the object under inspection S. However, it is also possible to obtain a trained model from another object under inspection S and output the injection path information for the input state based on this trained model.

[0095] Next, using Figure 22, we will explain the process for obtaining injection path information in the input state using the learning device 701. Figure 22 is a flowchart showing the inference procedure by the inference device 761.

[0096] In S201, the data acquisition unit 711 acquires density information. In S202, the inference unit 771 inputs density information into the trained model stored in the trained model storage unit 661 and obtains injection path information (injection path information corresponding to the density information) in the input state. The inference unit 771 outputs the obtained injection path information in the input state to the object under inspection S.

[0097] In S203, the inference unit 771 outputs injection path information for the input state obtained by the trained model to the robot 801.

[0098] In S204, the robot 801 uses the outputted injection path information in the input state to control the amount of material injected at the injection position along the injection path and mold the object to be inspected S. This makes it possible to create a shape with a uniform cross-sectional density that is close to the standard cross-sectional density.

[0099] In this embodiment, we have described a case where reinforcement learning is applied to the learning algorithm used by the inference unit, but this is not the only possible approach. In addition to reinforcement learning, supervised learning can also be applied to the learning algorithm.

[0100] Furthermore, the learning algorithm used in the model generation unit can be deep learning, which learns to extract the features themselves, or machine learning can be performed according to other known methods, such as neural networks, genetic programming, functional logic programming, or support vector machines.

[0101] Furthermore, the learning device 701 and the inference device 761 may be connected to the state of the object S under inspection via a network, for example, and may be separate devices from the molding environment of the object S under inspection. Also, the learning device 701 and the inference device 761 may be built into the object S under inspection. Moreover, the learning device 701 and the inference device 761 may reside on a cloud server.

[0102] Furthermore, the model generation unit 721 may learn injection path information in the input state using training data acquired from multiple objects S under inspection. The model generation unit 721 may acquire training data from multiple objects S under inspection used in the same area, or it may learn injection path information in the input state using training data collected from multiple objects S operating independently in different areas. In addition, a learning device that has learned injection path information in the input state for one object under inspection may be applied to another object under inspection, and the injection path information for that other object under inspection may be relearned and updated.

[0103] The learning device 701 and the inference device 761, like the control device 10, can have their corresponding operations configured using digital circuit hardware or software. When the functions of the learning device 701 and the inference device 761 are implemented using software, the learning device 701 and the inference device 761 can, for example, include a processor 51 and a memory 52 connected by a bus 53, as shown in Figure 12, and the processor 51 can execute a program stored in the memory 52.

[0104] As described above, the learning device 701 of the defect management device 2, which manages defects in the object under inspection S, comprises a data acquisition unit 711 and a model generation unit 721. The data acquisition unit 711 acquires learning data including density information showing the density distribution of each of a plurality of regions contained in the object under inspection S, and injection path information of the object under inspection S corresponding to the density information. The model generation unit 721 uses the learning data to generate a trained model for inferring injection path information from density information. The injection path information includes an injection path in which a robot 801, which generates the object under inspection S by injecting the material of the object under inspection S into a mold 821 for molding the object under inspection S, moves while injecting the material of the object under inspection S. The model generation unit 721 increases the reward given to the learning model when the difference between the predetermined reference density information of the object under inspection S and the density information is less than or equal to a predetermined standard.

[0105] The defect management device 2 and the inference device 761 comprise a data acquisition unit 711 and an inference unit 771. The data acquisition unit 711 acquires density information showing the density distribution of each of a plurality of regions contained in the object under inspection S. The inference unit 771 uses a trained model to infer injection path information of the object under inspection S from the density information and outputs injection path information from the density information acquired by the data acquisition unit 711.

[0106] In conventional foamed urethane molded products, where the molding results are irregular, controlling the density of the molded product is difficult, the location of defects is unpredictable, and it has been impossible to create an optimal molding process. By configuring the product as described above, it is possible to produce a uniformly dense product S using an optimal molding process that does not produce internal layer defects. Furthermore, it becomes possible to control the amount of resin filled during molding in areas where there is a mass imbalance in the product S. This allows for appropriate control of the density of the product, which is related to defects in the product.

[0107] The embodiments disclosed herein should be considered in all respects to be illustrative and not restrictive. At least two of the embodiments disclosed herein can be combined, as long as they do not contradict each other. The technical scope provided herein is defined by the claims rather than by the description of the embodiments above, and all modifications within the meaning and scope of the claims are intended to be included. [Explanation of Symbols]

[0108] 1,2 Defect management device, 10 Control device, 11 Ideal cross section, 12~14 Cross section, 21,22 Surface defect, 23 Inner layer defect, 31,32 Measurement height, 51 Processor, 52 Memory, 53 Bus, 101 Light source, 201 Camera, 301 Sensor (upper sensor), 401 Sensor (lower sensor), 500~502 Conveyor, 611 Upper shape acquisition unit, 612 Upper shape comparison unit, 621 Lower shape acquisition unit, 622 Lower shape comparison unit, 631 Transmitted light processing unit, 632 Defect detection unit, 641 Cross section density calculation unit, 651 Output unit, 660 Storage unit, 661 Trained model storage unit, 701 Learning device, 711 Data acquisition unit, 721 Model generation unit, 722 Reward calculation unit, 723 Function update unit, 761 Inference device, 771 Inference unit, 801 robot, 810 control device, 811 molding trajectory, 821 mold, S workpiece, S1 workpiece after molding, S2 workpiece after defect inspection.

Claims

1. A device for managing defects in an object under inspection, An irradiation device that irradiates transmitted light onto the first surface of the object to be inspected, An imaging device is positioned opposite the irradiation device and images the second surface of the object to be inspected, which is opposite to the first surface. A sensor for measuring information regarding the external shape of the object to be inspected, The system comprises a sensor and a processing device for processing information acquired from the imaging device, The processing apparatus calculates the brightness of the transmitted light in each of the multiple regions included in the object to be inspected, based on the information acquired from the imaging device. Each of the aforementioned plurality of regions is a three-dimensional region between a two-dimensional minute first region included in the first surface and a two-dimensional minute second region included in the second surface. The second region is the region to which the transmitted light irradiated into the first region passes through the object to be inspected and reaches. The aforementioned processing apparatus is Based on the information obtained from the sensor, the shape of the first surface and the shape of the second surface are estimated. A device for estimating volume information corresponding to the volume excluding the vacuolar portion in each of the plurality of regions, based on the shape of the first surface and the shape of the second surface and the brightness of the transmitted light in each of the plurality of regions.

2. The sensor includes a first sensor for measuring the first surface and a second sensor for measuring the second surface. The aforementioned processing apparatus is Based on the information obtained from the first sensor, the shape of the first surface is estimated. The apparatus according to claim 1, which estimates the shape of the second surface based on information obtained from the second sensor.

3. The aforementioned processing apparatus is Based on the volume information, vacuoles are detected within the object to be inspected. The apparatus according to claim 1 or 2, which determines that the object to be inspected has a defect when a vacuole exceeding a predetermined volume is detected.

Images