X-ray inspection method and X-ray inspection apparatus

Abstract

The present invention provides an X-ray inspection method and X-ray inspection apparatus capable of acquiring three-dimensional data of defects in an object in a short amount of time. [Solution] The method includes: an X-ray image acquisition step of acquiring a two-dimensional X-ray transmission image 3a by performing X-ray imaging of the object 1; a model generation step of generating a three-dimensional virtual model 4 corresponding to the object 1 including defects 2 using known structural parameters for the object 1; a virtual transmission image generation step of generating a two-dimensional virtual transmission image 6a from the three-dimensional virtual model; and a parameter modification step of modifying the structural parameters of the three-dimensional virtual model 4 so that the two-dimensional virtual transmission image 6a approaches the two-dimensional X-ray transmission image 3a, using the comparison results of the two-dimensional X-ray transmission image 3a and the two-dimensional virtual transmission image 6a.

Description

Technical Field

[0001] The present invention relates to an X-ray inspection method and an X-ray inspection apparatus, and particularly to an X-ray inspection method and an X-ray inspection apparatus for performing X-ray imaging of an object.

Background Art

[0002] Conventionally, a defect inspection method (X-ray inspection method) and a defect inspection apparatus (X-ray inspection apparatus) for performing X-ray imaging of a subject (object) are known (for example, see Patent Document 1).

[0003] Patent Document 1 discloses a method for detecting internal defects of a subject by using spatial discrete data such as X-ray CT data obtained by performing tomography on the subject. In Patent Document 1, it is disclosed that the result of detecting a defect in the X-ray CT data of the subject is displayed as a three-dimensional rendering image. Further, in Patent Document 1, in the three-dimensional rendering image, detection of the defect is performed according to feature amounts such as the size, shape or existence position of the defect, and color-coded display or enlarged display is performed. That is, in Patent Document 1, all of the subject and the defect are generated as a three-dimensional rendering image.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] As described above, Patent Document 1 generates all of the subject and defects as three-dimensional rendering images. However, although not described in Patent Document 1, in order to display the subject as a three-dimensional rendering image (three-dimensional model), it is necessary to collect a large amount of X-ray transmission images of the object, and the volume calculation to construct the three-dimensional model requires a very long time. For this reason, it takes a very long time to acquire three-dimensional data of defects in the object, and in cases where defect detection in a short time is required, such as in line inspection, the defect detection method that constructs a three-dimensional model as disclosed in Patent Document 1 cannot be used. Therefore, there is a need for an X-ray inspection method and X-ray inspection apparatus that can acquire three-dimensional data of defects in an object in a short time.

[0006] This invention was made to solve the above-mentioned problems, and one of its objectives is to provide an X-ray inspection method and an X-ray inspection apparatus that can acquire three-dimensional data of defects in an object in a short amount of time. [Means for solving the problem]

[0007] To achieve the above objective, the X-ray inspection method in the first aspect of this invention comprises: an X-ray transmission image acquisition step of acquiring a two-dimensional X-ray transmission image by X-ray imaging of an object; a model generation step of generating a three-dimensional virtual model corresponding to an object containing defects using known structural parameters for the object; a virtual transmission image generation step of generating a two-dimensional virtual transmission image from the three-dimensional virtual model; and a parameter modification step of modifying the structural parameters of the three-dimensional virtual model so that the two-dimensional virtual transmission image approaches the two-dimensional X-ray transmission image, using the comparison result between the two-dimensional X-ray transmission image and the two-dimensional virtual transmission image.

[0008] To achieve the above objective, the X-ray inspection apparatus in the second aspect of this invention comprises an imaging unit that performs X-ray imaging of an object and a control unit that controls the imaging unit, wherein the control unit is configured to perform the following: control to acquire a two-dimensional X-ray transmission image of the object; control to generate a three-dimensional virtual model corresponding to the object containing defects using known structural parameters for the object; control to generate a two-dimensional virtual transmission image from the three-dimensional virtual model; and control to modify the structural parameters of the three-dimensional virtual model so that the two-dimensional virtual transmission image approaches the two-dimensional X-ray transmission image using the comparison result between the two-dimensional X-ray transmission image and the two-dimensional virtual transmission image. [Effects of the Invention]

[0009] By using the known structural parameters for the object described above to generate a 3D virtual model corresponding to the object containing defects, it becomes possible to eliminate the need to generate a 3D model from X-ray transmission images. This eliminates the need to collect a large number of X-ray transmission images of the object to generate the 3D virtual model, and also reduces the time required for volume calculations to generate the 3D model. As a result, 3D data of defects in the object can be obtained in a short time. [Brief explanation of the drawing]

[0010] [Figure 1] This is a block diagram showing the overall configuration of an X-ray inspection apparatus according to the first embodiment of the present invention. [Figure 2] This figure shows an object to be inspected for defects according to the first embodiment of the present invention. [Figure 3] This figure shows an X-ray transmission image of an object according to the first embodiment of the present invention. [Figure 4] This figure shows a virtual model and a modified virtual model according to the first embodiment of the present invention. [Figure 5] This figure shows a defect-free virtual model according to the first embodiment of the present invention. [Figure 6]This figure shows a virtual transmission image generated from a virtual model and a modified virtual transmission image generated from a modified virtual model according to the first embodiment of the present invention. [Figure 7] This is a flowchart illustrating the defect inspection performed by the control unit according to the first embodiment of the present invention. [Figure 8] This is a subflowchart for obtaining known structural parameters according to the first embodiment of the present invention. [Figure 9] This figure shows a defect-free virtual transmission image generated from a defect-free virtual model according to the first embodiment of the present invention. [Figure 10] This is a block diagram showing the overall configuration of an X-ray inspection apparatus according to a second embodiment of the present invention. [Figure 11] This figure illustrates the linear attenuation coefficient distribution and the first training projection data according to a second embodiment of the present invention. [Figure 12] This is a diagram illustrating the content that a learning model learns according to a second embodiment of the present invention. [Figure 13] This is a flowchart illustrating the defect inspection performed by the control unit according to a second embodiment of the present invention. [Figure 14] This is a subflowchart for detecting a defect according to a second embodiment of the present invention. [Figure 15] This figure illustrates the defect projection data output using a learning model according to a second embodiment of the present invention. [Figure 16] This figure illustrates the generation of defective tomography data according to a second embodiment of the present invention. [Figure 17] This is a block diagram showing the overall configuration of an X-ray inspection apparatus according to a third embodiment of the present invention. [Figure 18] This figure illustrates a method for generating first training defect projection data, training defect tomography data, and second training defect projection data according to a third embodiment of the present invention. [Figure 19] This figure illustrates a method for generating third learning defect projection data according to a third embodiment of the present invention. [Figure 20]A diagram for explaining the content learned by the learning model according to the second embodiment of the present invention. [Figure 21] A flowchart for performing a defect inspection by a control unit according to the third embodiment of the present invention. [Figure 22] A sub - flowchart for detecting a defective part according to the third embodiment of the present invention. [Figure 23] A diagram for explaining the output of defective projection data using a learning model. [Figure 24] A diagram for explaining a method of generating first defect detection projection data, first defect tomography data, and second defect detection projection data according to the third embodiment of the present invention. [Figure 25] A diagram for explaining a method of generating third defect detection projection data according to the third embodiment of the present invention. [Figure 26] A diagram for explaining the generation of defect tomography data according to the third embodiment of the present invention.

Embodiments for Carrying out the Invention

[0011] [First Embodiment] Hereinafter, a first embodiment embodying the present invention will be described based on the drawings.

[0012] (Overall Configuration of X - ray Inspection Apparatus) Referring to FIGS. 1 to 6, a defect inspection apparatus 100 according to the first embodiment of the present invention will be described.

[0013] As shown in Figure 1, the defect inspection apparatus 100 according to the first embodiment comprises an X-ray imaging unit 10 and an inspection unit 20. The defect inspection apparatus 100 is a non-destructive testing device for detecting whether or not a defect 2 (see Figure 2) exists inside an object 1 whose interior cannot be visually inspected by performing CT (Computed Tomography) imaging on the object 1. In the defect inspection apparatus 100, each of the X-ray imaging unit 10 and the inspection unit 20 has a communication module and transmits and receives information from each other via a network or the like. Note that the X-ray imaging unit 10 is an example of an "imaging unit" in the claims, and the defect inspection apparatus 100 is an example of an "X-ray inspection apparatus" in the claims.

[0014] As shown in Figure 2, object 1 contains a defect 2 inside. Object 1 is made of a material from which an internal transmission image can be obtained by X-ray imaging, such as resin, metal, or ceramic. Defect 2 is an unintended part of object 1, such as a void. Defect 2 may adversely affect the performance of object 1, such as its durability.

[0015] As shown in Figure 1, the X-ray imaging unit 10 performs X-ray imaging of the object 1. The X-ray imaging unit 10 includes an X-ray irradiation unit 11 and an X-ray detection unit 12. The X-ray irradiation unit 11 is configured to irradiate X-rays. In the first embodiment, the X-ray irradiation unit 11 irradiates the object 1, which includes a defect 2 (see Figure 2), with X-rays. The object 1 is placed on a rotating table (not shown) or the like, and is irradiated with X-rays from various angles. The X-ray irradiation unit 11 also includes an X-ray tube that irradiates X-rays when power is supplied from a power supply device (not shown).

[0016] The X-ray detection unit 12 is configured to detect X-rays emitted from the X-ray irradiation unit 11. The X-ray detection unit 12 outputs an electrical signal corresponding to the detected X-rays. The X-ray detection unit 12 includes, for example, an FPD (Flat Panel Detector), which is an X-ray detector. The X-ray irradiation unit 11 and the X-ray detection unit 12 are located inside a housing (not shown) of the X-ray imaging unit 10.

[0017] As shown in Figure 1, the inspection unit 20 has a control unit 21, a storage unit 22, and a display unit 23. The inspection unit 20 is, for example, a personal computer that is communicatively connected to the X-ray imaging unit 10. The control unit 21 includes a processor or circuit, such as a CPU (Central Processing Unit), and ROM (Read Only Memory) and RAM (Random Access Memory). The control unit 21 may also include a processor such as a GPU (Graphics Processing Unit) or an FPGA (Field-Programmable Gate Array) configured for image processing.

[0018] Furthermore, the control unit 21 controls the operation of each part of the X-ray imaging unit 10. The control unit 21 controls the irradiation of X-rays by the X-ray irradiation unit 11, for example, by controlling a power supply unit (not shown). As a result, the control unit 21 causes the X-ray imaging unit 10 to perform CT imaging, such as acquiring multiple projection data by irradiating the object 1 (see Figure 2) with X-rays from multiple directions.

[0019] The storage unit 22 is configured to store various programs executed by the control unit 21, and various structural parameters for generating a 3D model of the object 1. Specifically, the storage unit 22 stores, for example, known structural parameters 22a of the object and known structural parameters 22b of defects. The storage unit 22 includes, for example, non-volatile memory such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive).

[0020] Here, the known structural parameters 22a of the object and the known structural parameters 22b of the defect are, for example, the three-dimensional data that constitutes the 3D-CAD data used to design the object 1, or data acquired empirically by workers, etc. In the manufacturing process of the object 1, for example, the defect 2 may appear in similar locations depending on the manufacturing lot of the object 1, and workers, etc., may be able to acquire this data empirically and prepare it in advance. Note that the 3D-CAD data used to design the object 1 generally has a smaller (lighter) data capacity compared to the three-dimensional data generated based on the X-ray projection data 3 (see Figure 3) obtained by CT imaging.

[0021] Furthermore, a display unit 23 is connected to the inspection unit 20. The display unit 23 includes, for example, a liquid crystal monitor. The display unit 23 displays images and text information under the control of the control unit 21. The inspection unit 20 also includes an operation unit (not shown) that accepts input operations from the operator. The operation unit includes, for example, a keyboard and a pointing device such as a mouse.

[0022] The control unit 21 includes an X-ray transmission image generation unit 21a, a model generation unit 21b, a virtual transmission image generation unit 21c, and a transmission image calculation unit 21d. These are configured in software as functional blocks that are realized when the control unit 21 executes a program (not shown) stored in the memory unit 22. In other words, the program is configured to cause the control unit 21 to execute each of the above functional blocks. These functional blocks may be configured with separate hardware, each with its own dedicated processor (processing circuit).

[0023] The X-ray transmission image generation unit 21a generates X-ray projection data 3 as shown in Figure 3 based on the X-rays detected by the X-ray detection unit 12. Specifically, the X-ray transmission image generation unit 21a generates X-ray projection data 3 in which the object 1, including the defect 2, is captured as a two-dimensional X-ray transmission image 3a. The X-ray transmission image generation unit 21a is also configured to generate a three-dimensional image (rendered image) of the object 1, including the defect 2, by synthesizing multiple X-ray projection data 3 captured from multiple angles while rotating the object 1.

[0024] The model generation unit 21b generates a three-dimensional virtual model of the object 1 based on the object known structure parameters 22a and defect known structure parameters 22b, which are stored in the storage unit 22 as known structure parameters for generating a 3D model. Specifically, the model generation unit 21b generates a three-dimensional virtual model 4 corresponding to the object 1, as shown in Figure 4. The model generation unit 21b can also generate a defect-free virtual model 5, as shown in Figure 5, which includes an ideal defect-free object 5a that does not contain defect 2, based on the object known structure parameters 22a. The virtual model 4 is generated as voxel volume data that includes the virtual object 4a and the virtual defect 4b.

[0025] The virtual transmission image generation unit 21c generates virtual X-ray projection data 6 as shown in Figure 6, based on the virtual model 4. At this time, the virtual transmission image generation unit 21c generates the virtual X-ray projection data 6 by performing a simulation that reflects the imaging parameters of the X-ray imaging unit 10 used by the X-ray transmission image generation unit 21a to generate the X-ray projection data 3. The imaging parameters include the positions of the X-ray irradiation unit 11 and the X-ray detection unit 12, the tube voltage, tube current and exposure time of the X-ray irradiation unit 11, and the number of views. This results in a two-dimensional virtual transmission image 6a corresponding to the three-dimensional virtual model 4.

[0026] The transmission image calculation unit 21d is configured to compare the X-ray projection data 3 shown in Figure 3 with the virtual X-ray projection data 6 shown in Figure 6 and calculate the difference between them. As comparison result data, the transmission image calculation unit 21d calculates difference data, for example, the difference in brightness per pixel between the X-ray projection data 3 and the virtual X-ray projection data 6. Furthermore, based on the comparison results such as the difference data, the transmission image calculation unit 21d calculates data to modify the structural parameters for generating the virtual model 4 in order to bring the virtual transmission image 6a projected on the virtual X-ray projection data 6 closer to the X-ray transmission image 3a projected on the X-ray projection data 3.

[0027] (Defect detection method) Next, the method by which the control unit 21 (see Figure 1) detects a defect 2 contained in the object 1 will be explained according to the flowcharts in Figures 7 and 8.

[0028] (If known structural parameters have already been obtained) First, in the X-ray transmission image acquisition process of step S1, the control unit 21 (see Figure 1) acquires multiple X-ray projection data 3 (see Figure 3) captured by the X-ray imaging unit 10. Then, the process proceeds to step S2.

[0029] Next, in the determination step S2, a determination is made as to whether or not the object has known structural parameters. In this determination step S2, if it is confirmed that at least one of the object's known structural parameters 22a or the defect's known structural parameters 22b is not stored in the storage unit 22 (see Figure 1), it is determined that the object does not have known structural parameters, and the process proceeds to step S3. However, in this case, in the determination step S2, since the object's known structural parameters 22a and the defect's known structural parameters 22b are stored in the storage unit 22, it is determined that the object has known structural parameters, and the process proceeds to step S4. The process of step S3 will be described later.

[0030] Next, in the registration process of step S4, rigid body registration of known structural parameters is performed. A process is carried out to align the virtual model 4 (see Figure 4) generated using the known structural parameters with the imaged object 1 (see Figure 2). Specifically, in the registration process, the known structural parameters for generating the virtual model 4 are read from the storage unit 22, and information such as the position (coordinates), orientation of the object 1, and the positions and angles of the X-ray irradiation unit 11 and X-ray detection unit 12 relative to the object 1 are reflected as structural parameters. After that, the process proceeds to step S5.

[0031] Next, in the model generation process of step S5, a three-dimensional virtual model 4 (see Figure 4) is generated based on the known structural parameters for which rigid body registration has been performed. In this first embodiment, the virtual object 4a is generated based on 3D-CAD data as the known structural parameters 22a of the object, and the virtual defect 4b is generated based on 3D-CAD data in which the coordinates and size are set based on the past knowledge of the worker, etc., as the known structural parameters 22b of the defect. Since the virtual model 4 is generated based on 3D-CAD data, there are no artifacts associated with CT imaging. Therefore, at this stage, the virtual object 4a and object 1 may not be exactly the same, and the virtual defect 4b and defect 2 may not be exactly the same at this stage either. After that, the process proceeds to step S6.

[0032] Next, in the virtual transmission image generation process of step S6, a two-dimensional virtual transmission image 6a (see Figure 6) is generated based on a three-dimensional virtual model 4 (see Figure 4). The virtual transmission image 6a is generated by a simulation applying imaging parameters corresponding to the X-ray transmission image 3a, and is the image projected onto the virtual X-ray projection data 6, which is generated in a number corresponding to the X-ray projection data 3 on which the X-ray transmission image 3a is projected. The simulation is performed, for example, by an X-ray tomography simulator such as ASTRA Toolbox. After that, the process proceeds to step S7.

[0033] Next, in the difference calculation step S7, difference data between the X-ray transmission image 3a and the virtual transmission image 6a is calculated. For example, difference data is calculated by determining the difference in brightness per pixel between the X-ray projection data 3 (see Figure 3) and the virtual X-ray projection data 6 (see Figure 6). After that, the process proceeds to step S8.

[0034] Next, in the determination step S8, it is determined whether or not modification of the virtual model 4 (see Figure 4) is necessary. Note that modification of the virtual model 4 includes calculating the voxel values ​​of the 3D voxel volumes that constitute the virtual model 4.

[0035] In the determination step S8, for example, the L2 norm (sum of the squares of the differences between the X-ray transmission image 3a and the virtual transmission image 6a) is calculated, which indicates the similarity between the X-ray transmission image 3a and the virtual transmission image 6a. If the L2 norm is smaller than a predetermined threshold, it is determined that modification of the virtual model 4 is unnecessary, and the process proceeds to step S13. In the first embodiment, the virtual object 4a is generated based on the 3D-CAD data used to design the object 1 and is quite similar, but the virtual defect 4b is set based on empirical information such as that of an operator, so the similarity is not very high. Therefore, an example is described in which the process proceeds to step S9 when the L2 norm indicating the similarity between the X-ray transmission image 3a and the virtual transmission image 6a is greater than or equal to a predetermined threshold and modification of the virtual model 4 is necessary.

[0036] Next, in the parameter modification step S9, the structural parameters for generating the virtual model 4 are modified so that the virtual model 4 is closer to the object 1. Specifically, the control unit 21 refers to the difference data and performs an optimization process to modify the structural parameters so that the shape, position (coordinates), material, and size of the virtual object 4a and the shape, position (coordinates), material, size, and number of virtual defects 4b in the virtual model 4 are closer to the object 1. At this time, it is not necessary to modify the parameters of the entire 3D virtual model, but rather the parameters of the parts that differ between the X-ray transmission image 3a and the virtual transmission image 6a are modified based on the difference data. After that, the process proceeds to step S10.

[0037] Next, in the modified model generation process of step S10, a three-dimensional modified virtual model 7, as shown in Figure 4, is generated based on the structural parameters that have been modified based on the difference data. In the generated modified virtual model 7, the modified virtual object 7a and modified virtual defect 7b each have their parameters corrected in the parts that differ between the X-ray transmission image 3a and the virtual transmission image 6a, thus improving their similarity to object 1 and defect 2. After that, the process proceeds to step S11.

[0038] Next, in the modified virtual transmission image generation step S11, modified virtual projection data 8, including a two-dimensional modified virtual transmission image 8a, is generated based on the three-dimensional modified virtual model 7 (see Figure 4). The modified virtual transmission image 8a is also generated by simulation applying imaging parameters corresponding to the X-ray transmission image 3a (see Figure 3), similar to the generation of the virtual transmission image 6a, and is generated in a number corresponding to the X-ray transmission image 3a. After that, the process proceeds to step S12.

[0039] Next, in the difference calculation step S12, similar to the difference calculation step S7, difference data between the X-ray transmission image 3a (see Figure 3) and the modified virtual transmission image 8a (see Figure 6) is calculated. After that, the process returns to step S8. By repeating the process from step S8 to step S12, the virtual model 4 (see Figure 4) is brought closer to the object 1. Similarly, the position (coordinates), size, and shape of the virtual defect 4b in the virtual model 4 are brought closer to the position (coordinates), size, and shape of defect 2 in the object 1.

[0040] Next, we will explain an example in which, by repeating the processes from steps S8 to S12, it is determined in the difference calculation process of step S8 that there is no need to modify the modified virtual model 7, and the process proceeds to the defect identification process of step S13.

[0041] In the defect identification step S13, at least one of the location (coordinates), shape, and size of defect 2 relative to object 1 is identified. In the first embodiment, the location (coordinates), shape, and size of defect 2 relative to object 1 are all identified. Here, since the modified virtual model 7 has a high degree of similarity to object 1, by identifying the location (coordinates), shape, and size of modified virtual defect 7b, the location (coordinates), shape, and size of defect 2 in object 1 are effectively identified. In the first embodiment, as described above, since the modified virtual defect 7b is directly identified, there is no need to perform a separate process to extract defect 2 from the entire 3D model of object 1 including defect 2, as in the prior art.

[0042] In step S13, the defect identification process, the display unit 23 (see Figure 1) displays at least one or all of the location (coordinates), shape, and size of defect 2 relative to object 1 as list data. Additionally, a corrected virtual defect 7b corresponding to defect 2 is displayed on the three-dimensional corrected virtual model 7 in a way that allows it to be distinguished from the corrected virtual object 7a. This enables not only the control unit 21 but also operators to identify (detect) defect 2. The defect inspection (X-ray inspection) process is completed with the above steps.

[0043] (If known structural parameters have not been obtained) Next, we will explain an example in which, in the determination step S2 described above, it is determined that there are no known structural parameters, and the process proceeds to step S3. Note that we will omit explanations of parts that are common to steps S1 to S13 described above.

[0044] In the determination step S2 shown in Figure 7, a determination is made as to whether or not the object has known structural parameters. In this determination step S2, if it is confirmed that at least one of the object's known structural parameters 22a or the defect's known structural parameters 22b is not stored in the storage unit 22 (see Figure 1), it is determined that the object does not have known structural parameters, and the process proceeds to step S3. Here, the storage unit 22 determines that the object's known structural parameters 22a are stored as known structural parameters, but the defect's known structural parameters 22b are not stored, and proceeds to step S3.

[0045] In step S3, the process of acquiring known structural parameters, the control unit 21 is configured to acquire known structural parameters of the object 1 and the defect 2. The details of the process of acquiring known structural parameters will be explained using the flowchart in Figure 8. Note that this process of acquiring known structural parameters in step S3 is performed only once, when the object 1 is inspected for the first time, and once the parameters are stored in the storage unit 22, it is not necessary to perform it each time the object 1 is inspected thereafter.

[0046] As shown in Figure 8, in the determination step of step S31, it is first determined whether or not the object 1 has known structural parameters. Specifically, in this determination step of step S31, if it is confirmed that the object's known structural parameters 22a are not stored in the storage unit 22 (see Figure 1), it is determined that the object does not have known structural parameters 22a, and the process proceeds to step S32. In the object's known structural parameters acquisition step of step S32, 3D data is acquired using a method similar to that of the conventional technology, such as rendering a 3D model from CT imaging, and the object's known structural parameters 22a are acquired based on the obtained 3D data. In the first embodiment, the storage unit 22 assumes that the object's known structural parameters 22a are stored as known structural parameters, and the process proceeds to step S33.

[0047] Next, in the determination step S33, it is determined whether or not the defect 2 has known structural parameters. Specifically, in this determination step S33, if it is confirmed that the known structural parameters 22b of the defect are stored in the storage unit 22 (see Figure 1), it is determined that the defect has known structural parameters 22b, and the known structural parameter acquisition step ends. In the first embodiment, an example will be described in which it is confirmed that the known structural parameters 22b of the defect are not stored in the storage unit 22 (see Figure 1), and it is determined that the defect does not have known structural parameters 22b, and the process proceeds to step S34.

[0048] Next, in the defect-free model generation step S34, a three-dimensional defect-free virtual model 5 (see Figure 5) of object 1 is generated, assuming that defect 2 is not present. In the defect-free model generation step S34, for example, if CAD data used to design object 1 is available, a three-dimensional defect-free virtual model 5 is generated using that CAD data. If CAD data used to design object 1 is not available, for example, a worker or other person modifies the part corresponding to defect 2 from the three-dimensional data of object 1 obtained in step S32, and generates a three-dimensional defect-free virtual model 5 that eliminates the part corresponding to defect 2. After that, the process proceeds to step S35.

[0049] In the virtual transmission image generation process of step S35, a two-dimensional defect-free virtual transmission image 9a (see Figure 9) is generated based on a three-dimensional defect-free virtual model 5 (see Figure 5). The defect-free virtual transmission image 9a is generated by a simulation applying imaging parameters corresponding to the X-ray transmission image 3a, and is the image projected onto the defect-free virtual projection data 9, with the number of generated images corresponding to the X-ray projection data 3 on which the X-ray transmission image 3a is projected. The process then proceeds to step S36.

[0050] In step S36, the differential transmission image acquisition process, the difference between the X-ray transmission image 3a (see Figure 3) and the defect-free virtual transmission image 9a (see Figure 9) is acquired. This difference includes a differential transmission image (not shown), which includes, for example, an image composed of the difference in brightness values ​​per pixel between the X-ray transmission image 3a and the defect-free virtual transmission image 9a. Here, since the defect-free virtual transmission image 9a does not have a portion corresponding to defect 2, the differential transmission image with the X-ray transmission image 3a will effectively display defect 2. However, at this point, sufficient accuracy for defect inspection cannot be obtained due to the influence of noise, for example, when acquiring the X-ray transmission image 3a. The process then proceeds to step S37.

[0051] Next, in the differential volume calculation step S37, a volume calculation is performed to generate a three-dimensional differential virtual model (not shown) based on the differential transmission image obtained in step S36. That is, data such as the position (coordinates), size, and shape of the part corresponding to defect 2, which are necessary to generate a three-dimensional differential virtual model (not shown), are calculated. After that, the process proceeds to step S38.

[0052] Next, in the parameter extraction step S38, data such as the location (coordinates), size, and shape of the part corresponding to defect 2, which are necessary to generate a three-dimensional difference virtual model (not shown), are extracted as known defect structural parameters 22b (see Figure 1). This completes the known structural parameter acquisition step.

[0053] Furthermore, the above process acquires the known structural parameters 22a of the object and the known structural parameters 22b of the defect, making them available for use as known structural parameters for defect inspection. Therefore, the steps S4 to S13 shown in Figure 7 can be performed, and defect inspection (X-ray inspection) can be carried out based on the known structural parameters.

[0054] [Second Embodiment] Next, a second embodiment of the present invention will be described with reference to Figures 10 to 16. First, the configuration of the defect inspection device 200 will be described using Figure 10. In this second embodiment, as shown in Figure 10, an example will be described in which the storage unit 22 of the defect inspection device 200 stores the known material parameters 22c of the target object and the first learning model 22d. Note that in the second embodiment, the explanation of points common to the first embodiment will be omitted.

[0055] The object's known material parameters 22c are information obtained in advance by the worker or others regarding the materials constituting object 1 (see Figure 2). These parameters include, for example, information such as the material properties, specific gravity, and linear attenuation coefficient of object 1. If object 1 is formed from a combination of multiple different materials, the object's known material parameters 22c will include information such as the material properties, specific gravity, and linear attenuation coefficient for each of these materials.

[0056] Here, because the known material parameters 22c of the object are stored, the model generation unit 21b can generate a virtual model 4 (see Figure 4) that reflects the linear attenuation coefficients, which represent the different degrees of X-ray attenuation for each material constituting the object 1 (see Figure 2). In the second embodiment, the object 1 is formed by combining, for example, a resin material and a metal material. Therefore, the virtual object 4a, which models the object 1, is also a model that combines a resin material and a metal material. The virtual model 4 also includes virtual defects 4b.

[0057] At this time, the virtual transmission image generation unit 21c can generate a virtual tomographic image 30a, which is a tomographic diagram of the virtual model 4 that reflects the linear attenuation coefficient, as shown in Figure 11. The virtual tomographic image 30a includes singular areas 30b, where the linear attenuation coefficient differs significantly from that of other areas. Singular areas 30b are, for example, parts where metal materials (not shown) included in the virtual model 4 are located. The virtual transmission image generation unit 21c generates first training projection data 40 as a sinogram by projecting the virtual tomographic image 30a forward on the simulation. The first training projection data 40 includes the first training transmission image 40a, which contains information about defects as well as metal artifacts generated by performing a (virtual) projection process on the metal materials (not shown).

[0058] Next, the first learning model 22d will be described. The first learning model 22d performs inference processing using a fully convolutional neural network. The first learning model 22d is a learning model that takes two-dimensional X-ray projection data 3 (see Figure 3) and two-dimensional virtual X-ray projection data 6 (see Figure 6) as input data and outputs two-dimensional detection defect projection data 50 (see Figure 15). Note that the first learning model 22d is an example of a "learning model" in the claims. The first learning model 22d has been pre-trained to output detection defect projection data 50, etc.

[0059] Specifically, as shown in Figure 12, the first learning model 22d is pre-trained on a dataset consisting of the first training projection data 40, the second training projection data 41, and the first training defect projection data 42. Note that in Figure 12, for simplification, the first learning model 22d is shown with one instance each of the first training projection data 40, the second training projection data 41, and the first training defect projection data 42. In reality, the first learning model 22d is trained on multiple instances of each of the first training projection data 40, the second training projection data 41, and the first training defect projection data 42.

[0060] The second learning projection data 41 is generated by the virtual transmission image generation unit 21c, for example, based on a defect-free virtual model 5. In the second embodiment, the defect-free virtual model 5 is generated by the virtual transmission image generation unit 21c in such a way that it does not include the portion of the virtual model 4 relating to the virtual defect 4b. Therefore, the second learning transmission image 41a of the generated second learning projection data 41 includes metal artifacts derived from a metal material (not shown), but does not include information about defects.

[0061] The first training defect projection data 42 is generated by obtaining the difference between the first training projection data 40, which includes information on artifacts and defects originating from the metallic material, and the second training projection data 41, which includes artifacts originating from the metallic material but does not include information on defects. In other words, the training first defect transmission image 42a included in the first training defect projection data 42 will only contain the portion relating to the virtual defect 4b of the virtual model 4, and will not include metal artifacts originating from the metallic material that are not shown. The first training defect projection data 42 may also be generated by virtually back-projecting (reconstructing) the virtual defect 4b, using only the virtual defect 4b of the virtual model 4 used to generate the first training projection data 40 as the model. Furthermore, the first training model 22d may be trained even if the position and angle of the arrangement of the first training projection data 40 and the second training projection data 41 are not precisely matched.

[0062] (Defect detection method) Next, the method by which the control unit 21 (see Figure 10) detects defects 2 contained in the object 1 will be explained according to the flowcharts in Figures 13 and 14. Note that in the flowchart of Figure 13, steps with the same names as steps in the flowchart of the first embodiment shown in Figure 7 are processed in the same way as in the first embodiment, so some of the explanation will be omitted.

[0063] In step S101, the control unit 21 (see Figure 10) acquires the X-ray projection data 3 (see Figure 15). In step S102, the control unit 21 determines whether or not it has known parameters. In the second embodiment, the case in which the known parameters include the object's known structural parameter 22a, the defect's known structural parameter 22b, and the object's known material parameter 22c will be described. In this case, the control unit 21 determines that it has known parameters (step S102: Yes) and proceeds to step S104.

[0064] Next, in the model generation process of step S104, a three-dimensional virtual model 4 (see Figure 4) is generated based on the known structural parameters 22a of the object, the known structural parameters 22b of the defect, and the known material parameters 22c of the object. Here, the virtual model 4 is a model that combines a metal material (not shown) and a resin material. Note that at this stage, the virtual object 4a and object 1 may not be exactly the same, and the virtual defect 4b and defect 2 may not be exactly the same either. After that, the process proceeds to step S105.

[0065] In step S105, the virtual transmission image generation unit 21c (see Figure 10) generates virtual X-ray projection data 6 (see Figure 15) based on the virtual model 4. At this time, since the virtual X-ray projection data 6 is generated based on the linear attenuation coefficient distribution 30 as a tomographic diagram of the 3D-CAD data (virtual model 4), artifacts (metal artifacts) associated with the virtual X-ray projection exist.

[0066] In the defect information detection step S106, the transmission image calculation unit 21d (see Figure 10) detects information to identify defect 2 in the object 1. Here, in the defect information detection step S106, the control unit 21 (transmission image calculation unit 21d) executes the process according to the subflowchart shown in Figure 14.

[0067] In the defect transmission image generation step S106a shown in Figure 14, the transmission image calculation unit 21d calls the first learning model 22d from the storage unit 22, as shown in Figure 15, and calculates the detection defect projection data 50 as output data based on the X-ray projection data 3 and virtual X-ray projection data 6 as input data. At this time, the first learning model 22d is learning with the position and angle of the first learning projection data 40 (see Figure 12) and the first learning defect projection data 42 (see Figure 12) not precisely matched. Therefore, even if the position of the X-ray projection data 3 and virtual X-ray projection data 6 is slightly misaligned, the first learning model 22d can appropriately calculate the detection defect projection data 50. As a result, a detection defect transmission image 50a, which contains information about defect 2 and excludes the influence of metal artifacts, is obtained. The process then proceeds to step S106b.

[0068] In the defect detection step S106b, the transmission image calculation unit 21d generates the defect tomography data 51 for detection by virtually back-projecting (reconstructing) the defect projection data 50 for detection through simulation, as shown in Figure 16. The defect tomography data 51 for detection is, for example, a tomographic image in which the defect 51a is displayed, and indicates the location of defect 2 in the tomographic image. The process then proceeds to step S107.

[0069] In the determination step S107, it is determined whether or not the virtual model 4 (see Figure 4) needs to be modified. The control unit 21 determines whether or not the virtual model 4 needs to be modified based on, for example, the number and size of the defective parts 51a. If the virtual model 4 needs to be modified, the process proceeds to step S108; if it does not need to be modified, the process proceeds to step S112. The processes from step S108 to step S112 are the same as the processes from step S8 to step S13 in the flowchart shown in Figure 7. Through the above process, it becomes possible to perform defect inspection with the influence of artifacts suppressed.

[0070] [Third Embodiment] Next, a third embodiment of the present invention will be described with reference to Figures 17 to 26. First, the configuration of the defect inspection device 300 will be described using Figure 17. In this third embodiment, an example will be described in which the storage unit 22 of the defect inspection device 300 stores the second learning model 22e, as shown in Figure 17. Note that in this third embodiment, the points that are common to the first and second embodiments will not be explained.

[0071] The second learning model 22e is described below. The second learning model 22e performs inference processing using a fully convolutional neural network. As shown in Figure 2, the second learning model 22e is a learning model that takes two-dimensional X-ray projection data 3 (see Figure 3) and two-dimensional virtual X-ray projection data 6 (see Figure 6) as input data and outputs two-dimensional first detection defect projection data 60 (see Figure 23). Note that the second learning model 22e is an example of a "learning model" in the claims. The second learning model 22e has been pre-trained to output the first detection defect transmission image 60a, etc.

[0072] Here, the second learning model 22e learns not only the learning data (first learning projection data 40, second learning projection data 41, and first learning defect projection data 42) learned by the first learning model 22d as described in the second embodiment, but also another set of data generated based on the first learning defect projection data 42. Specifically, as shown in Figure 18, the second learning model 22e also learns learning defect tomography data 43, which is obtained by virtually back-projecting the first learning defect projection data 42 through simulation, and second learning defect projection data 44, which is obtained by virtually forward-projecting the learning defect tomography data 43 through simulation. Furthermore, as shown in Figure 19, the second learning model 22e also learns third learning defect projection data 45, which is generated by integrating the first learning defect projection data 42 and the second learning defect projection data 44. The third training defect projection data 45 is generated, for example, by averaging the first training defect projection data 42 and the second training defect projection data 44.

[0073] In other words, as shown in Figure 20, the second learning model 22e learns from a dataset consisting of the first learning projection data 40, the second learning projection data 41, the first learning defect projection data 42, the learning defect tomography data 43, the second learning defect projection data 44, and the third learning defect projection data 45. Multiple copies of each of these learning data sets are prepared. Furthermore, the second learning model 22e learns by weighting each of the first learning defect projection data 42, the learning defect tomography data 43, the second learning defect projection data 44, and the third learning defect projection data 45 according to a ratio pre-set by the user or other means. Note that the weighting (ratio) for each learning data set may be automatically set through training or other means.

[0074] (Defect detection method) Next, the method by which the control unit 21 (see Figure 17) detects defects 2 contained in the object 1 will be explained according to the flowcharts in Figures 21 and 22. Note that in the flowchart of Figure 21, steps with the same names as the steps in the flowcharts of the first and second embodiments shown in Figures 7 and 13 are processed in the same way as in the first and second embodiments, so some of the explanation will be omitted.

[0075] In steps S201 to S205 of the flowchart in Figure 21, the same processing as in steps S101 to S105 of the flowchart of the second embodiment shown in Figure 13 is performed. In the subsequent defect information detection step S206, the transmission image calculation unit 21d (see Figure 17) detects information for identifying defect 2 in the object 1. Here, in the defect information detection step S206, the control unit 21 (transmission image calculation unit 21d) executes the processing according to the flow of the subflowchart shown in Figure 22.

[0076] In the first defect transmission image generation step S206a shown in Figure 18, the transmission image calculation unit 21d calls the second learning model 22e from the storage unit 22, as shown in Figure 23, and calculates the first detection defect projection data 60 as output data based on the X-ray projection data 3 and virtual X-ray projection data 6 as input data. At this time, the second learning model 22e has learned the learning defect tomography data 43, the second learning defect projection data 44, and the third learning defect projection data 45 in addition to the first learning defect projection data 42, so it is possible to generate the first detection defect projection data 60 that represents the transmission image (first detection defect transmission image 60a) with higher accuracy than the detection defect projection data 50 generated in step S106a of the second embodiment. Furthermore, if the object 1 has a complex structure, the contrast of the X-ray projection data 3 input to the second learning model 22e may be low, and the first detection defect transmission image 60a output by the second learning model 22e may not be able to reflect the delicate defects (discontinuities may occur). After that, the process proceeds to step S206b. Note that the first detection defect transmission image 60a is an example of the "first defect transmission image" in the claims.

[0077] In the first defect detection step S206b, the transmission image calculation unit 21d generates first defect tomography data 61 by virtually back-projecting (reconstructing) the first detection defect projection data 60 through simulation, as shown in Figure 24. The first defect tomography data 61 is, for example, a two-dimensional tomography image in which the defect 61a is displayed. The process then proceeds to step S206c.

[0078] In the second defect transmission image generation step S206c, the transmission image calculation unit 21d generates second detection defect projection data 62 by virtually forward-projecting the first defect tomography data 61 through simulation, as shown in Figure 24. The second detection defect projection data 62 displays the second detection defect transmission image 62a. Although the second detection defect transmission image 62a has less discontinuity than the first detection defect transmission image 60a due to back-projection and forward projection onto the first defect tomography data 61, artifacts such as blurring may remain. The process then proceeds to step S206d. The second detection defect transmission image 62a is an example of the "second defect transmission image" in the claims.

[0079] In the third defect transmission image generation step S206d, the transmission image calculation unit 21d generates the third detection defect projection data 63 by integrating (for example, averaging) the first detection defect projection data 60 and the second detection defect projection data 62, as shown in Figure 25. The third detection defect transmission image 63a displayed in this third detection defect projection data 63 maintains continuity while suppressing artifacts such as blurring. The process then proceeds to step S206e. Note that the third detection defect transmission image 63a is an example of the "third defect transmission image" in the claims.

[0080] In the second defect detection step of step S206e, the transmission image calculation unit 21d generates the second defect tomography data 64 by virtually back-projecting (reconstructing) the third detection defect projection data 63 through simulation, as shown in Figure 26. The second defect tomography data 64 is, for example, a tomographic image in which the defect 64a is displayed. The process then proceeds to step S207. The process from step S207 to step S212 is the same as the process from step S107 to step S112 in the flowchart shown in Figure 13. Through the above process, it becomes possible to perform defect inspection with the influence of artifacts suppressed.

[0081] [Differentiation] It should be noted that the embodiments disclosed herein are illustrative and not restrictive in all respects. The scope of the present invention is defined by the claims rather than by the description of the embodiments above, and further includes all modifications (exceptions) within the meaning and scope equivalent to the claims. For example, the known structural parameters may be configured to include at least one of the object's known structural parameters 22a, or the defect's known structural parameters 22b of a known defect 2 present in object 1. Furthermore, for example, in the parameter modification steps of steps S9, S108, and S209, not only the structural parameters of the different parts between the two-dimensional X-ray transmission image 3a and the two-dimensional virtual transmission image 6a may be modified, but the overall structural parameters of the three-dimensional virtual model 4 may also be modified. Furthermore, for example, after the parameter modification steps of steps S9, S108, and S209, there may be a defect identification step that displays the location, shape, or size of defect 2 relative to object 1, or the system may be configured so that the defect 2 is identified by an operator or the like based on their own judgment, without having a defect identification step itself. Furthermore, for example, the defect identification steps S13, S112, and S212 may be configured to display the location, shape, and size of defect 2 relative to object 1 as list data, or in a display format that allows identification of the part corresponding to the object on the modified virtual model 7. Furthermore, for example, after the defect identification steps of steps S13, S112, and S212, there may be a step to determine whether object 1 is a good product or a defective product based on the location, shape, or size of the defect 2 relative to object 1. Furthermore, for example, the virtual transmission image generation steps in steps S6, S105, and S205 may be configured to generate a two-dimensional virtual transmission image 6a from a three-dimensional virtual model 4 under the same conditions each time, without performing a simulation that reflects the imaging parameters used in the X-ray transmission image acquisition steps in steps S1, S101, and S201. In that case, it is preferable to perform a step to correct the discrepancy between the imaging parameters and the simulation conditions (parameters). Furthermore, for example, the process for acquiring known structural parameters may be configured to input known structural parameters 22b obtained from past experimental data and inspection results from operators, etc., rather than using the difference obtained by comparing the X-ray transmission image 3a with a two-dimensional defect-free virtual transmission image 9a that does not contain defect 2. Furthermore, for example, the three-dimensional virtual model 4, the modified virtual model 7, and the differential virtual model (not shown) generated by the model generation unit 21b may be generated as mesh models with a lighter data size, rather than voxel volume models. Furthermore, the index used in the determination step of step S8 to indicate the similarity between the X-ray transmission image 3a and the virtual transmission image 6a may be a similarity measure other than the L2 norm. For example, the cosine similarity between the X-ray transmission image 3a and the virtual transmission image 6a may be used. Furthermore, in the determination step S8, the determination of whether or not the virtual model 4 (see Figure 4) needs to be modified may be made not by whether or not the similarity is less than a predetermined threshold, but by whether or not the number of modification iterations set in advance by an operator or the like has been met. Furthermore, the object 1 may be formed from only one type of material, such as resin material. Also, the model generation unit 21b may generate a virtual model 4 without using known material parameters. Furthermore, the model generation unit 21b may generate a two-dimensional virtual transmission image based on a tomographic map generated using an index other than the linear attenuation coefficient distribution, which represents the degree of X-ray attenuation that differs for each material. Furthermore, even when a virtual model 4 is generated using known material parameters and a defect identification process is performed, as in the second and third embodiments, the process may be carried out without using a learning model. Also, even when a defect identification process is performed as in the first embodiment, the process may be carried out using a learning model. Furthermore, in the defect information acquisition steps of steps S106 and S206, defect information for object 1 may be detected based on the second defect transmission image. Furthermore, the learning model may learn only one or two of the learning defect tomography data 43, the second learning defect projection data 44, and the third learning defect projection data 45, rather than learning all of them. Alternatively, the learning model may be trained using the first training defect projection data 42, the training defect tomography data 43, the second training defect projection data 44, and the third training defect projection data 45 without weighting each of them. Furthermore, the third detection defect projection data 63 may be generated by linearly adding the first detection defect projection data 60 and the second detection defect projection data 62 without averaging them. Alternatively, the third detection defect projection data 63 may be generated by loading the first detection defect projection data 60 and the second detection defect projection data 62 into a learning model on a pixel-by-pixel basis, or on a pixel-by-pixel filtered basis (pixel feature basis), and making inferences using a neural network. Furthermore, when calculating virtual X-ray projection data 6 using known material parameters, if the positions of different materials in the X-ray transmission image 3a and the virtual transmission image 6a differ significantly, the known material parameters may be readjusted to correspond to the difference in position.

[0082] [Pattern] Those skilled in the art will understand that the exemplary embodiments described above are specific examples of the following embodiments.

[0083] (Item 1) The process involves X-ray imaging of the object and obtaining a two-dimensional X-ray transmission image, and A model generation step of generating a model corresponding to the object, including defects, as a three-dimensional virtual model using known structural parameters for the object; A virtual transmission image generation process that generates a two-dimensional virtual transmission image from the aforementioned three-dimensional virtual model, An X-ray inspection method comprising: a parameter modification step of modifying the structural parameters of a three-dimensional virtual model so that the two-dimensional virtual transmission image approaches the two-dimensional X-ray transmission image, using the results of a comparison between the two-dimensional X-ray transmission image and the two-dimensional virtual transmission image. By generating a 3D virtual model of an object containing defects using known structural parameters for the object, the need to generate a 3D model from X-ray transmission images is eliminated. This eliminates the need to collect a large number of X-ray transmission images of the object to generate the 3D virtual model, and also reduces the time required for volume calculations to generate the 3D model. As a result, 3D data of defects in the object can be obtained in a short time. (Item 2) The X-ray inspection method according to item 1, wherein the known structural parameters include known structural parameters of the object and known structural parameters of defects present in the object. In this case, the similarity between the virtual model and the object increases, resulting in a smaller difference between the 2D virtual transmission image generated using the 3D virtual model and the 2D X-ray transmission image obtained from the object. This reduces the amount of modification required for the structural parameters of the 3D virtual model during the modification parameter process. As a result, 3D data of defects in the object can be obtained in a shorter time. (Item 3) The X-ray inspection method according to item 1, wherein the parameter modification step modifies the structural parameters of the different parts between the two-dimensional X-ray radiograph and the two-dimensional virtual radiograph without modifying the structural parameters of the entire three-dimensional virtual model. In this case, since the structural parameters of the entire 3D virtual model are not modified during the parameter modification process, the amount of modification required for the structural parameters of the 3D virtual model can be reduced. Furthermore, because the structural parameters of the parts that differ between the 2D X-ray radiograph and the 2D virtual radiograph are modified, the 3D virtual model can be modified appropriately. (Item 4) The X-ray inspection method according to item 1, further comprising a defect identification step after the parameter correction step, which identifies at least one of the location, shape, and size of the defect in the object. In this case, the location, shape, or size of the defect relative to the object can be identified using a virtual model modified to resemble the object by adjusting its structural parameters, thus enabling proper identification of the defect. (Item 5) The X-ray inspection method according to item 4, wherein the defect identification step displays at least one of the location, shape, and size of the defect on the object as list data. In this case, the defect can be easily identified because at least one of its location, shape, or size can be easily recognized visually as list data. (Item 6) The system further comprises a modified model generation step, which generates the modified 3D virtual model as a 3D modified virtual model using the structural parameters modified by the parameter modification step, The X-ray inspection method according to item 4, wherein the defect identification step displays the defect in the object on the three-dimensional modified virtual model in a manner that is identifiable to the object. In this case, defects can be more easily identified visually because they are displayed in a identifiable manner on a virtual model whose structural parameters have been modified to resemble the actual object. (Item 7) The X-ray inspection method according to item 1, wherein the virtual transmission image generation step generates the two-dimensional virtual transmission image from the three-dimensional virtual model by performing a simulation that reflects the imaging parameters used in the X-ray transmission image acquisition step. In this case, a two-dimensional virtual transmission image is generated by a simulation that reflects the imaging parameters used in the X-ray transmission image acquisition process, thus obtaining a virtual transmission image similar to the X-ray transmission image generated from the object. This allows for a more accurate comparison between the X-ray transmission image and the two-dimensional virtual transmission image. As a result, structural parameters can be corrected more accurately. (Item 8) The X-ray inspection method according to item 1, further comprising a step of obtaining known structural parameters, which involves obtaining the structural parameters of the defect known as known structural parameters using the difference obtained by comparing the aforementioned X-ray transmission image with a two-dimensional defect-free virtual transmission image that does not contain the defect. In this case, even if the known structural parameters of the defect are not prepared, the structural parameters related to the defect can be obtained by comparing the X-ray transmission image with a two-dimensional, defect-free virtual transmission image that does not contain the defect. As a result, the known structural parameters of the defect are always obtained before the virtual model is generated, making it possible to generate a virtual model that reflects the known structural parameters of the defect. (Item 9) The X-ray inspection method according to item 1, wherein the model generation step generates a three-dimensional virtual model using known material parameters, which include information about the material of the object, in addition to the known structural parameters. In this case, known material parameters can be reflected in the generation of the 3D virtual model, making it possible to generate a more appropriate virtual transmission image even for objects that combine dissimilar materials with different X-ray transmission rates, such as resin and metal. (Item 10) The X-ray inspection method according to item 9, wherein the virtual transmission image generation step generates a two-dimensional virtual transmission image based on a linear attenuation coefficient distribution that represents the degree of X-ray attenuation which differs for each material, generated from the three-dimensional virtual model generated using the known structural parameters and the known material parameters. Here, for example, when generating an X-ray transmission image from a material with low X-ray transmission, such as metals, noise called artifacts may occur in the X-ray transmission image. In this case, comparing the X-ray transmission image with artifacts with a virtual transmission image without artifacts may make it difficult to distinguish between the artifacts and the defect information that is actually needed, and it may not be possible to properly modify the structural parameters of the virtual model. Therefore, by configuring the system as described above, a two-dimensional virtual transmission image is generated based on a linear attenuation coefficient distribution that represents the degree of X-ray attenuation which differs for each material, making it possible to generate a virtual transmission image that reproduces the artifacts. Consequently, it becomes possible to compare the X-ray transmission image with artifacts with a virtual transmission image that reproduces the artifacts. As a result, the difference between the X-ray transmission image and the virtual transmission image becomes clear, and information about defects in the object contained in the X-ray transmission image can be obtained more appropriately. (Item 11) The X-ray inspection method according to item 1, further comprising a defect information detection step, which, prior to the parameter correction step, detects information about the defect in the object based on a detection defect transmission image as the two-dimensional defect transmission image, generated using a learning model that takes the two-dimensional X-ray transmission image and the two-dimensional virtual transmission image as input data and outputs a two-dimensional defect transmission image. In this case, a learning model is used that outputs a defective radiographic image using a two-dimensional X-ray radiographic image and a two-dimensional virtual radiographic image as input data. For example, if the learning model is trained to handle X-ray radiographic images containing artifacts, it can appropriately output a defective radiographic image even for X-ray radiographic images containing artifacts. In other words, by using a learning model, it is sometimes possible to obtain information about defects in the object contained in the X-ray radiographic image more appropriately than, for example, simply obtaining the difference between a two-dimensional X-ray radiographic image and a two-dimensional virtual radiographic image. (Item 12) The detection defect transmission image includes a first defect transmission image directly generated using the learning model, and a second defect transmission image generated by forward projection of a defect tomography image generated by back-projecting the first defect transmission image, and a third defect transmission image generated using these three methods. The X-ray inspection method according to item 11, wherein the defect information detection step detects information about the defect in the object based on the third defect transmission image. Here, if the object has a complex structure, or if beam hardening occurs with a high average X-ray energy, the contrast from the X-ray transmission image input to the learning model becomes low, making it difficult to reflect subtle defects in the first defect transmission image output by the learning model. Therefore, by projecting forward onto the defect tomographic image obtained by back-projecting the first defect transmission image, a second defect transmission image can be generated that has been corrected to reflect subtle defects in the first defect transmission image. On the other hand, the second defect transmission image may suffer from image blurring. Therefore, it is effective to generate a third defect transmission image that is less blurred and reflects subtle defects by combining the first and second defect transmission images in a way that compensates for the shortcomings of both. Thus, by configuring it as described above, defect information can be detected based on the third defect transmission image, which is less blurred and reflects subtle defects, allowing for more appropriate acquisition of defect information of the object contained in the X-ray transmission image. (Item 13) The learning model, in addition to the first defective transmission image acquired in advance as a training image for outputting the first defective transmission image, The X-ray inspection method according to item 12, which is generated by learning at least one of the following: a learning defect tomography image generated by back-projecting the learning first defect tomography image; a learning second defect tomography image generated by forward-projecting the learning defect tomography image; and a learning third defect tomography image generated using the learning first defect tomography image and the learning second defect tomography image. In this case, since the learning model learns at least one of the following in addition to the first training defective transmission image: a training defective tomographic image, a training second defective transmission image, and a training third defective transmission image, it can output the first defective transmission image with higher accuracy compared to when the learning model learns only the first training defective transmission image. (Item 14) The X-ray inspection method described in item 13, wherein the learning model is generated by learning each of the first learning defective radiograph, the learning defective tomographic image, the learning second defective radiograph, and the learning third defective radiograph by assigning weights to each of them. Here, by adjusting the weights of the training data—the first training defect transmission image, the training defect tomography, the second training defect transmission image, and the third training defect transmission image—which the learning model learns from, the degree to which each training data influences the output is adjusted. Therefore, by configuring it as described above, for example, the learning model can be trained appropriately according to the degree to which it is important for the learning model to generate the optimal first defect transmission image. (Item 15) An imaging unit that performs X-ray imaging of the object, The system includes a control unit that controls the imaging unit, The control unit, Control for acquiring a two-dimensional X-ray transmission image of the object, A control system that generates a 3D virtual model of the object, including defects, using known structural parameters for the object; Control for generating a two-dimensional virtual transmission image from the aforementioned three-dimensional virtual model, An X-ray inspection apparatus configured to perform control such as modifying the structural parameters of a three-dimensional virtual model so that the two-dimensional virtual transmission image approaches the two-dimensional X-ray transmission image, using the comparison result between the two-dimensional X-ray transmission image and the two-dimensional virtual transmission image. The same technical effects as in item 1 can be obtained in the present invention. [Explanation of Symbols]

[0084] 1. Object 2. Defects 3. X-ray projection data 3a X-ray transmission image (2D X-ray transmission image) 4. Virtual Model (3D Virtual Model) 4a Virtual Object Model 4b Virtual Defect Model 5. Defect-free virtual model 6. Virtual X-ray projection data 6a Virtual transmission image 7. Modified Virtual Model 7a Modified virtual object 7b Modified virtual defect model 8. Modified virtual projection data 8a Modified virtual transmission image 9. Defect-free virtual projection data 9a Defect-free virtual transmission image 10. X-ray imaging unit (imaging unit) 11 X-ray irradiation section 12 X-ray detection unit 20. Inspection Department 21 Control Unit 21a X-ray image generation section 21b Model generation unit 21c Virtual transmission image generation unit 21d Transmission image calculation section 22a Known structural parameters of the object 22b Defective known structural parameters 22c Object Known Material Parameters (Known Material Parameters) 22d First Learning Model (Learning Model) 22e Second Learning Model (Learning Model) 30. Distribution of linear attenuation coefficients 40. First learning projection data 40a First transmission image for learning 41. Second learning projection data 41a Second transmission image for learning 42. Defect projection data for the first training. 42a First learning defective transmission image 43. Defective fault data for training 43a Defects for learning 44. Second training defect projection data 44a Second learning defective transmission image 45. Defective projection data for the third learning stage 45a Third learning defective transmission image 50 Defect projection data for detection 50a Defect detection transmission image 51. Defect fault data for detection 51a Defective part 60. Projection data for the first detection of defects 60a First detection defect transmission image (detection defect transmission image, first defect transmission image) 61. Data on the first defective fault 61a Defect area (detected tomographic image) 62 Second Defect Projection Data for Detection 62a Second detection defect transmission image (detection defect transmission image, second defect transmission image) 63. Third detection defect projection data 63a Third Defect Detection Image (Detection Defect Detection Image, Third Defect Detection Image) 64. Data on the second defective fault 64a Defective part 100, 200, 300 Defect Inspection Equipment (X-ray Inspection Equipment)

Claims

1. The process involves X-ray imaging of the object and obtaining a two-dimensional X-ray transmission image, and A model generation step of generating a model corresponding to the object, including defects, as a three-dimensional virtual model using known structural parameters for the object; A virtual transmission image generation process that generates a two-dimensional virtual transmission image from the aforementioned three-dimensional virtual model, An X-ray inspection method comprising: a parameter modification step of modifying the structural parameters of a three-dimensional virtual model so that the two-dimensional virtual transmission image approaches the two-dimensional X-ray transmission image, using the results of comparing the two-dimensional X-ray transmission image with the two-dimensional virtual transmission image.

2. The X-ray inspection method according to claim 1, wherein the known structural parameters include known structural parameters of the object and known structural parameters of defects present in the object.

3. The X-ray inspection method according to claim 1, wherein the parameter modification step modifies the structural parameters of different parts between the two-dimensional X-ray transmission image and the two-dimensional virtual transmission image without modifying the structural parameters of the entire three-dimensional virtual model.

4. The X-ray inspection method according to claim 1, further comprising a defect identification step after the parameter modification step, which identifies at least one of the location, shape, and size of the defect in the object.

5. The X-ray inspection method according to claim 4, wherein the defect identification step displays at least one of the location, shape, and size of the defect on the object as list data.

6. The system further comprises a modified model generation step, which generates the modified three-dimensional virtual model as a three-dimensional modified virtual model using the structural parameters modified by the parameter modification step, The X-ray inspection method according to claim 4, wherein the defect identification step displays the defect in the object on the three-dimensional modified virtual model in a manner that is identifiable with respect to the object.

7. The X-ray inspection method according to claim 1, wherein the virtual transmission image generation step generates the two-dimensional virtual transmission image from the three-dimensional virtual model by performing a simulation that reflects the imaging parameters used in the X-ray transmission image acquisition step.

8. The X-ray inspection method according to claim 1, further comprising a step of obtaining known structural parameters, which involves obtaining the structural parameters of the defect known as known structural parameters using the difference obtained by comparing the X-ray transmission image with a two-dimensional defect-free virtual transmission image that does not contain the defect.

9. The X-ray inspection method according to claim 1, wherein the model generation step generates a three-dimensional virtual model using known material parameters, which include information about the material of the object, in addition to the known structural parameters.

10. The X-ray inspection method according to claim 9, wherein the virtual transmission image generation step generates a two-dimensional virtual transmission image based on a linear attenuation coefficient distribution that represents different X-ray attenuation rates for each material, generated from the three-dimensional virtual model generated using the known structural parameters and the known material parameters.

11. The X-ray inspection method according to claim 1, further comprising a defect information detection step, which, prior to the parameter correction step, detects information about the defect in the object based on a detection defect transmission image as the two-dimensional defect transmission image, generated using a learning model that takes the two-dimensional X-ray transmission image and the two-dimensional virtual transmission image as input data and outputs a two-dimensional defect transmission image.

12. The detection defect transmission image includes a first defect transmission image directly generated using the learning model, and a second defect transmission image generated by forward projection of a defect tomography image generated by back-projecting the first defect transmission image, and a third defect transmission image generated using these three methods. The X-ray inspection method according to claim 11, wherein the defect information detection step detects information about the defect in the object based on the third defect transmission image.

13. The learning model, in addition to the first defective transmission image acquired in advance as a training image for outputting the first defective transmission image, The X-ray inspection method according to claim 12, which is generated by learning at least one of the following: a learning defect tomography image generated by back-projecting the learning first defect tomography image; a learning second defect tomography image generated by forward-projecting the learning defect tomography image; and a learning third defect tomography image generated using the learning first defect tomography image and the learning second defect tomography image.

14. The X-ray inspection method according to claim 13, wherein the learning model is generated by learning each of the first learning defective transmission image, the learning defective tomographic image, the learning second defective transmission image, and the learning third defective transmission image with weights assigned to each of them.

15. An imaging unit that performs X-ray imaging of the object, The system includes a control unit that controls the imaging unit, The control unit, Control for acquiring a two-dimensional X-ray transmission image of the object, A control system that generates a three-dimensional virtual model of the object, including defects, using known structural parameters for the object; Control for generating a two-dimensional virtual transmission image from the aforementioned three-dimensional virtual model, An X-ray inspection apparatus configured to perform control such as modifying the structural parameters of a three-dimensional virtual model so that the two-dimensional virtual transmission image approaches the two-dimensional X-ray transmission image, using the comparison result between the two-dimensional X-ray transmission image and the two-dimensional virtual transmission image.

Images