Product inspection method and product inspection device

Abstract

To appropriately evaluate target blow holes to cause a computer to automatically inspect a defect in a cast product.SOLUTION: A product inspection method includes detecting blow holes on the basis of CT data obtained through X-ray photography of a cast product 100, and dividing an internal area of the cast product 100 into a first area 110 and a second area 120. The product inspection method includes performing classification determination for each blow hole present in the first area 110, and performing regression prediction of evaluating the total amount of blow holes present in the second area 120.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention relates to a product inspection method and a product inspection device. [Background technology]

[0002] Patent Document 1 discloses an X-ray inspection device, an X-ray inspection method, and an X-ray inspection program. In Patent Document 1, to detect defects in a cast product, X-rays are irradiated onto the cast product and transmitted X-rays that pass through the cast product are detected. Based on the detected transmitted X-rays, transmitted X-rays that would be transmitted if the cast product were defect-free are calculated. Defects in the cast product are detected by comparing the detected transmitted X-rays with the calculated transmitted X-rays that would be transmitted if the cast product were defect-free. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2006-105794 Summary of the Invention [Problem to be solved by the invention]

[0004] Some porosity will occur inside the manufactured casting products, and while some porosity does not affect the product's functionality, others do. In order to automatically inspect casting products for defects using a computer, the porosity in question must be properly evaluated.

[0005] The present invention has been made in view of the above-mentioned problems, and its object is to provide a product inspection method and product inspection device that can automatically inspect defects in cast products using a computer by appropriately evaluating target blowholes. [Means for solving the problem]

[0006] A product inspection method according to one embodiment of the present invention includes detecting blowholes based on image data obtained by X-raying a cast product, dividing the interior region of the cast product into a first region and a second region, classifying blowholes present in the first region, and performing regression prediction to evaluate the total amount of blowholes present in the second region. [Effects of the Invention]

[0007] According to the present invention, the target blowhole can be appropriately evaluated, and therefore defects in cast products can be automatically inspected by computer. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a block diagram showing the configuration of a product inspection device according to this embodiment. [Figure 2] FIG. 2 is a diagram illustrating the interior region of the cast product. [Figure 3] FIG. 3 is a flowchart showing the steps of the product inspection method. [Figure 4] FIG. 4 is a flowchart showing the details of the classification determination process. [Figure 5] FIG. 5 is a flowchart showing the details of the regression prediction process. [Figure 6] FIG. 6 is a flowchart showing the details of the image conversion process. [Figure 7] FIG. 7 is an explanatory diagram of a shape projection image. [Figure 8] FIG. 8 is an explanatory diagram of a boundary distance projection image. [Figure 9] FIG. 9 is a diagram showing the relationship between the total number of NG blowholes and the NG product. [Figure 10] FIG. 10 is a diagram showing the distribution of product unacceptable levels and thresholds. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the description of the drawings, the same parts are designated by the same reference numerals and the description thereof will be omitted.

[0010] A product inspection device 10 according to this embodiment will be described with reference to FIG. 1. The product inspection device 10 is a device that inspects defects in cast products. For example, the cast product may be a cylinder block for an engine mounted on a vehicle, which is formed by pouring molten material (e.g., aluminum alloy) into a mold. Porosity, which is a void, occurs in the interior region of the cast product. The product inspection device 10 detects the porosity in the interior region of the cast product and inspects the cast product for defects based on the detected porosity. Here, the interior region of the cast product refers to an area that is filled with material according to design, and does not include any space formed as the interior shape of the cast product.

[0011] The product inspection device 10 receives CT data generated by a CT data generation device 5. The CT data generation device 5 includes an X-ray irradiation device, a sensor, and a CT data generation unit.

[0012] The X-ray irradiation device is placed in a plane parallel to the horizontal plane (xy plane) so as to face the sensor across the casting. X-rays emitted from the X-ray irradiation device penetrate the casting in a cross-sectional manner and enter the sensor.

[0013] The CT data generation unit generates image data (image data) by X-raying the cast product based on the output from the sensor. The X-ray irradiation device X-rays the cast product from all directions, thereby obtaining cross-sectional image data of the cast product at a predetermined position in the vertical direction (z direction). In addition, the X-ray irradiation device and sensor can move relative to the cast product in the Z direction. The CT data generation unit can obtain CT data (three-dimensional image data) of the cast product from the cross-sectional image data obtained at multiple positions in the Z direction.

[0014] The product inspection device 10 is composed of a computer 20, a data acquisition unit 30, a display 40, an input device 50, and a storage device 60. The CT data generation device 5 described above may constitute a part of the product inspection device 10.

[0015] The data acquisition unit 30 acquires CT data from the CT data generation device 5. For example, the data acquisition unit 30 is a communication interface circuit that transmits and receives communication signals to and from the CT data generation device 5. Alternatively, the data acquisition unit 30 is an interface circuit that inputs and outputs data to and from a recording medium on which the CT data generated by the CT data generation device 5 is recorded.

[0016] The display 40 is a device that displays information from the computer 20. The input device 50 is a device for performing input operations on the computer 20, such as a keyboard and a mouse. The computer 20 can generate a graphical user interface (GUI) required for inspecting cast products and display it on the display 40. Furthermore, an operator can input necessary information according to the GUI displayed on the display 40.

[0017] The storage device 60 is an external storage device connectable to the computer 20, and stores the CT data acquired from the CT data generation device 5 and various data generated by the computer 20. The storage device 60 is a hard disk, a solid state drive (SSD), or other non-volatile storage device. The storage device 60 may be configured as a recording area on the cloud.

[0018] The computer 20 is composed of a hardware processor such as a CPU (Central Processing Unit), memory, and various interfaces. The processor reads various computer programs stored in the memory or the like and executes various instructions contained in the programs. By executing the programs, the processor functions as multiple information processing circuits provided in the computer 20. The computer 20 realizes its functions using software, but its functions can also be realized using dedicated hardware.

[0019] The computer 20 includes a blowhole detection unit 21, an area setting unit 22, a classification determination unit 23, a regression prediction unit 24, and a data processing unit 25 as one or more information processing circuits.

[0020] The blowhole detection unit 21 detects blowholes present in the interior region of the cast product based on the CT data.

[0021] The region setting unit 22 divides the interior region of the cast product to be inspected into multiple regions. FIG. 2 shows a cross section of a cast product 100, schematically illustrating the interior region within a certain range from the outer edge 101 of the cast product. The region setting unit 22 divides the interior region of the cast product 100 into a first region 110 having a thickness T and a second region 120 located inward from the first region based on the distance from the outer edge 101 of the cast product. The region setting unit 22 sets the first region 110 a specified distance X inward from the outer edge 101 of the cast product 100, via a third region 130 adjacent to the outer edge 101 of the cast product 100.

[0022] The classification determination unit 23 performs a classification determination for each blowhole present in the first region 110 to determine whether or not the blowhole will become a defect in the cast product 100. Specifically, the classification determination unit 23 is equipped with a classification determination model made up of a deep neural network, and performs the classification determination using this classification determination model. When CT data of a blowhole is input, the classification determination model outputs a classification as to whether the blowhole is a blowhole that will become a defect in the cast product 100 or a blowhole that will not become a defect in the cast product 100.

[0023] The classification judgment model is trained by machine learning using training data consisting of CT data of blowholes that will become defects in the cast product 100 and CT data of blowholes that will not become defects in the cast product 100, which have been selected in advance by an operator, and the parameters of the deep neural network are optimized. A blowhole that will become a defect in the cast product 100 refers to a blowhole that will affect the product's functions, while a blowhole that will not become a defect in the cast product 100 refers to a blowhole that will not affect the product's functions. The CT data input to the classification judgment model is converted into projection image data, which will be described later, to improve calculation efficiency.

[0024] The regression prediction unit 24 performs regression prediction to determine defects in the cast product 100 by quantitatively evaluating the total amount of voids present in the second region 120. Specifically, the regression prediction unit 24 is equipped with a regression prediction model consisting of a deep neural network, and performs regression prediction using this regression prediction model. When CT data of a void is input, the regression prediction model outputs a void rejection degree for that void. The regression prediction unit 24 calculates the sum of the void rejection degrees for all voids present in the second region 120, and calculates the product void rejection degree from this total value. This product void rejection degree corresponds to a value for quantitatively evaluating the total amount of voids present in the second region 120.

[0025] The regression prediction model uses machine learning to optimize the parameters of the deep neural network using the CT data of the pores and the pore NG level corresponding to the pores as training data. The training data is data obtained from multiple casting products. The CT data input into the regression prediction model is converted into projection image data to improve calculation efficiency.

[0026] The NG degree of a blown cavity, which serves as training data, is given by, for example, the following formula (1): The NG degree of a blown cavity is a value used to evaluate one blown cavity, and the larger this value, the higher the NG degree. Blowhole NG degree=α·V / Vmax+β·L / Lmax···(1)

[0027] V is the total number of porosity present in the second region 120 of the cast product 100. L is the length of the porosity present in the second region 120 of the cast product 100 (the length of the long side of the bounding box described below). Vmax is obtained by tripling the standard deviation of the total number of porosity present in each cast product, plus the average value of the total number of porosity present in each cast product. Lmax is obtained by tripling the standard deviation of the length of porosity present in each cast product 100, plus the average value of the length of porosity present in each cast product 100. α and β are parameters for adjustment.

[0028] Porosity in CT data is recognized by a bounding box. A bounding box is created by determining whether or not a porosity exists for each voxel, and then connecting voxels that are determined to contain porosity. In other words, an individual consisting of multiple connected voxels becomes a bounding box that indicates one porosity.

[0029] The data processing unit 25 performs processing necessary for inspecting the cast product 100. For example, the data processing unit 25 performs image conversion to convert CT data (three-dimensional image data) of the blowhole into projection image data (two-dimensional image data) projected onto a two-dimensional plane.

[0030] The steps of the product inspection method executed by the computer 20 will be described below with reference to Fig. 3. First, the computer 20 acquires CT data of the cast product 100 to be inspected via the data acquisition unit 30 (S10).

[0031] The computer 20 detects any blowholes present in the interior region of the cast product 100 based on the CT data (S11). For example, the computer 20 detects blowholes by comparing the CT data acquired in step S10 with reference data showing no blowholes in the interior region of the cast product 100. In detecting blowholes, the computer 20 identifies the size (volume and length) and location of each blowhole. The computer 20 then extracts, from the detected blowholes, those that satisfy a criterion as the blowholes to be evaluated. The criterion is that the blowhole is present in a specific location in the cast part and is equal to or larger than a specific size (volume and length). The location of the cast product 100 and the size of the blowhole that serve as the criterion are registered in advance by an operator using the input device 50 to input information into the GUI displayed on the display 40.

[0032] The computer 20 divides the interior region of the cast product 100 (S12). Specifically, as shown in FIG. 2 , the computer 20 divides the interior region into a third region 130, a first region 110, and a second region 120 based on the distance from the outer edge 101 of the cast product 100. The outer edge 101 of the cast product 100 is defined by the design values ​​of the cast product 100. The parameters (specified distance X and thickness T) defining each of the regions 110, 120, and 130 are registered in advance by an operator using the input device 50 to input information into the GUI displayed on the display 40.

[0033] The computer 20 determines whether or not classification determination has been performed (S13). If classification determination has not been performed, the process proceeds to step S14. On the other hand, if classification determination has already been performed, the process proceeds to step S20.

[0034] The computer 20 performs classification determination processing (S14). As shown in Fig. 4, the computer 20 extracts blowholes that exist in the first region 110 from among the blowholes to be determined (S30).

[0035] The computer 20 performs image conversion to convert the CT data of the blowholes into projection image data (S31). The projection image data is two-dimensional image data generated by projecting the CT data, which is three-dimensional image data, onto a two-dimensional plane. The targets of the image conversion are all blowholes present in the first region 110, and the image conversion is performed for each blowhole. Details of the image conversion will be described later.

[0036] The computer 20 performs classification of the blowholes using the classification model (S32). The classification is performed on all blowholes present in the first region 110, and the classification is performed for each blowhole. In the classification, the projection image data of the blowhole is input into the classification model, and a classification is output as to whether the blowhole will be a defect in the cast product 100 or whether it will not be a defect in the cast product 100.

[0037] The computer 20 obtains the classification result from the classification determination model (S33). All porosity present in the first region 110 is classified as either a porosity that will become a defect in the cast product 100 or a porosity that will not become a defect in the cast product 100.

[0038] 3, the computer 20 determines whether or not there are any NGs, i.e., whether or not there are any blowholes that will become defects in the cast product 100 (S15). If there is at least one blowhole classified as a blowhole that will become a defect in the cast product 100 among the blowholes present in the first region 110, the process proceeds to step S17. On the other hand, if there is not even one blowhole classified as a blowhole that will become a defect in the cast product 100 among the blowholes present in the first region 110, the process proceeds to step S16.

[0039] The computer 20 associates the cast product 100 to be inspected with the judgment result and writes the association in the storage device 60 (S16), and returns to the process of step S13.

[0040] The computer 20 outputs projection image data of the entire product and displays it on the display 40 (S17). By visually checking the projection image data displayed on the display 40, the worker can confirm any blowholes present in the first region 110 and any blowholes that have been determined to be NG. The worker then determines whether or not to determine that the inspected cast product 100 is an NG product. The worker's determination result can be registered by the worker inputting information into the GUI displayed on the display 40 using the input device 50.

[0041] If the worker determines that the product is unacceptable (S18: YES), the computer 20 instructs the casting product 100 to be scrapped (S19).

[0042] If the worker has not determined that the product is NG (S18: NO), the process proceeds to step S16. Then, the computer 20 associates the cast product 100 to be inspected with the determination result and writes the association to the storage device 60 (S16), and the process returns to step S13.

[0043] After such classification and determination has been performed, the computer 20 then performs a regression prediction process (S20). As shown in Fig. 5, the computer 20 extracts, from among the determination target blowholes, blowholes that exist in the second region 120 (S40).

[0044] The computer 20 performs image conversion to convert the CT data of the blowholes into projection image data (S41). The target of this image conversion is all blowholes present in the second region 120, and the image conversion is performed for each blowhole.

[0045] The computer 20 predicts the NG degree of blown holes using the regression prediction model (S42). The NG degree of blown holes prediction targets all blown holes present in the second region 120, and the NG degree of blown holes prediction is performed for each blown hole. In the regression prediction, projection image data of the blown holes is input into the regression prediction model, and the NG degree of blown holes is output.

[0046] The computer 20 calculates the product NG degree (S43). Specifically, the computer 20 calculates the total value of the predicted NG degree of all the blowholes present in the second region 120. The computer 20 holds a map showing the relationship between the total value of the NG degree of blowholes and the product NG degree as shown in Fig. 9, and calculates the product NG degree from the total value of the NG degree of blowholes. The product NG degree increases as the total value of the NG degree of blowholes increases.

[0047] As shown in Fig. 3, the computer 20 determines whether the product rejection rate is equal to or lower than a threshold value (S21). This threshold value can be set in advance by an operator using the input device 50 to input information into the GUI displayed on the display 40. If the product rejection rate is greater than the threshold value, the process proceeds to step S23. On the other hand, if the product rejection rate is equal to or lower than the threshold value, the process proceeds to step S22. The computer 20 then associates the cast product 100 to be inspected with the determination result and writes them to the storage device 60 (S22), and ends this process.

[0048] The computer 20 outputs the projection image data of the entire product and displays it on the display 40 (S23).

[0049] The computer 20 outputs the distribution graph shown in Fig. 10 and displays it on the display 40 (S24). This distribution graph shows the results of the regression prediction performed on the same type of cast product 100, i.e., the product NG degree and frequency of the cast product 100.

[0050] The worker determines whether or not the cast product 100 being inspected is a non-compliant product by visually checking the distribution graph and the projection image data of the entire product displayed on the display 40. The worker's determination result can be registered by inputting information into the GUI displayed on the display 40 using the input device 50.

[0051] If the worker determines that the product is unacceptable (S25: YES), the computer 20 instructs the casting product 100 to be scrapped (S26).

[0052] If the worker does not determine that the product is NG (S25: NO), the process proceeds to step S22. Then, the computer 20 associates the cast product 100 to be inspected with the determination result and writes the association to the storage device 60 (S22), and ends this process.

[0053] Next, with reference to FIG. 6, the image conversion process for creating projection image data from the CT data of the cast hole to be determined will be described.

[0054] The computer 20 creates a shape projection image from the CT data (S50). The CT data of the blowhole is three-dimensional data expressed by voxels in an xyz coordinate system, as shown in FIG. 7(a). The computer 20 projects the CT data of the blowhole onto the xy plane, and assigns the number of voxels that overlap in the projection direction as a pixel value to each pixel of the projected image. This creates two-dimensional image data for the xy plane, as shown in FIG. 7(b). The computer 20 similarly creates two-dimensional image data for the yz plane and the xz plane. The shape projection image is composed of the two-dimensional image data projected onto the xy plane, yz plane, and xz plane, and is data that two-dimensionally represents the shape and size of each individual blowhole.

[0055] The computer 20 creates a boundary distance projection image from the CT data (S51). First, the computer 20 calculates the boundary distance for each voxel of the cavity. The boundary distance is the smallest value among the distances to the outer edge (boundary surface) of the cast product 100 around the voxel when viewed on the xy plane. As shown in FIG. 8(a), the computer assigns a boundary distance to each voxel of the cavity.

[0056] The computer 20 projects the CT data of the blowhole onto the xy plane, and assigns the smallest boundary distance among the voxels that overlap in the projection direction as a pixel value to each pixel of the projected image. This creates two-dimensional image data for the xy plane, as shown in Figure 8(b). The computer 20 similarly creates two-dimensional image data for the yz and xz planes. The boundary projection image is composed of the two-dimensional image data projected onto the xy, yz, and xz planes, and is data that two-dimensionally represents the location (distance) of the blowhole relative to the outer edge of the product.

[0057] As shown in FIG. 6, the computer 20 creates projection image data by combining the shape projection image and the boundary distance projection image (S52).

[0058] The computer 20 outputs the created projection image data (S53).

[0059] As described above, according to this embodiment, the computer 20 divides the interior region of the cast product 100 into a first region 110 and a second region 120 based on the distance from the outer edge 101 of the cast product 100. Then, the computer 20 performs a classification determination for each blowhole present in the first region 110 to determine whether or not the blowhole will become a defect in the cast product 100. The computer 20 also performs a regression prediction to determine defects in the cast product 100 by quantitatively evaluating the total amount of blowholes present in the second region 120.

[0060] Because the first region 110 is located on the outer edge (boundary) of the cast product 100, harmful porosity in this first region 110 may affect product functionality, such as by causing cracks. Therefore, individual porosity is evaluated in the first region 110 to determine whether it will cause defects in the cast product 100. Meanwhile, because the second region 120 is located more inward than the first region 110, the presence of such porosity is not a problem. However, the presence of many porosity porosity porosity porosity porosity porosity may be problematic and may affect the first region 110. Therefore, the total amount of porosity in the second region 120 is quantitatively evaluated using regression prediction. The regression prediction allows the total amount of porosity in the second region 120 to be determined by summing the porosity NG grade of each porosity. By determining the total amount of porosity in the second region 120, the quality of the cast product 100 can be determined. In this way, the target porosity can be appropriately evaluated, allowing the computer 20 to automatically inspect the cast product 100 for defects.

[0061] In this embodiment, the first region 110 is separated by a third region 130 located within a specified distance X from the outer edge 101 of the cast product 100. This third region is an area where classification is not performed. Because the actual cast product 100 has tolerances, if classification were performed based on the designed outer edge 101, there is a possibility that an air layer outside the outer edge 101 would be erroneously detected as a blowhole. According to this embodiment, the area located near the outer edge 101 can be excluded from the classification, allowing for accurate detection of blowholes.

[0062] In this embodiment, the computer 20 extracts the target blowholes from among the blowholes present in the interior region of the cast product 100, based on the location of the cast product 100 and the size of the blowholes. According to this embodiment, by using the location of the cast product 100 and the size of the blowholes as parameters, blowholes that could be defects in the cast product 100 can be automatically extracted. This allows the computer 20 to automatically inspect the cast product 100 for defects.

[0063] In this embodiment, the classification is performed using a classification model in which neural network parameters are optimized by machine learning. According to this embodiment, the classification model that has undergone machine learning can accurately determine whether or not a defect is a blowhole that would be a defect in the cast product 100.

[0064] In this embodiment, the regression prediction is performed using a regression prediction model in which neural network parameters are optimized by machine learning. According to this embodiment, the regression prediction model that has undergone machine learning can accurately evaluate defects in the cast product 100.

[0065] In this embodiment, the image data is CT data obtained by three-dimensionally X-ray photographing the cast product 100, and the computer 20 converts the three-dimensional image data of the blowholes into two-dimensional image data projected onto a two-dimensional plane. According to this embodiment, the judgment can be made using the data converted from three-dimensional to two-dimensional, thereby reducing the calculation load.

[0066] Although the embodiments of the present invention have been described above, the descriptions and drawings that form part of this disclosure should not be understood to limit the present invention. Various alternative embodiments, examples, and operating techniques will become apparent to those skilled in the art from this disclosure. [Explanation of symbols]

[0067] 5 CT data generator 10 Product inspection equipment 20 Computer 21 Blowhole detection section 22 Area setting section 23 Classification judgment section 24 Regression prediction unit 25 Data Processing Unit 30 Data Acquisition Section 40 Display 50 Input Device 60 Storage device 100 Casting Products 101 outer edge 110 First Area 120 Second Domain 130 The Third Domain

Claims

1. 1. A computer-implemented product inspection method for inspecting a cast product for defects, comprising: detecting blowholes present in an internal region of the cast product based on image data obtained by X-ray photographing the cast product; dividing the internal region into a first region and a second region located more inward than the first region based on a distance from an outer edge of the cast product; performing a classification determination for each of the blowholes present in the first region to determine whether or not the blowholes are defects of the cast product; performing a regression prediction to determine defects in the cast product by quantitatively evaluating the total amount of the porosity present in the second region; Product inspection methods, including:

2. the first region is provided via a third region located within a specified distance from an outer edge of the cast product; The third region is a region where the classification determination is not performed. The product inspection method according to claim 1.

3. extracting, from the pores detected in the internal region, pores to be subjected to the classification determination and the regression prediction based on the location of the cast product and the size of the pores. The product inspection method according to claim 1.

4. The classification determination is performed by a classification determination model in which neural network parameters are set by machine learning. The product inspection method according to claim 1.

5. The regression prediction is performed by a regression prediction model in which neural network parameters are set by machine learning. The product inspection method according to claim 1.

6. the image data is three-dimensional image data obtained by three-dimensionally X-ray photographing the casting product, The three-dimensional image data of the blowhole is converted into two-dimensional image data projected onto a two-dimensional plane. The product inspection method according to claim 1.

7. A product inspection device for inspecting defects in a casting product, comprising: a data acquisition unit that acquires image data obtained by X-raying the casting product; a computer; The computer detecting blowholes present in an internal region of the cast product based on image data obtained by X-ray photographing the cast product; dividing the internal region into a first region and a second region located more inward than the first region based on a distance from an outer edge of the cast product; performing a classification determination for each of the blowholes present in the first region to determine whether or not the blowholes are defects of the cast product; performing a regression prediction to determine defects in the cast product by quantitatively evaluating the total amount of the porosity present in the second region; Product inspection equipment.

Images