JP2021001774A - Non-destructive inspection system, manufacturing method of learned exhaust gas treatment filter inspection model, and generation method of learning data - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the accuracy of determining whether or not an exhaust gas treatment filter can be used. A non-destructive inspection system for an exhaust gas treatment filter includes a non-destructive test device 200 for non-destructive testing of the exhaust gas treatment filter and a computer. The computer has a trained exhaust gas treatment filter inspection model for determining the availability of a non-destructive tested exhaust gas treatment filter. The computer acquires non-destructive test result data 320 representing the result of the non-destructive test of the exhaust gas treatment filter by the non-destructive test apparatus 200, and uses the trained exhaust gas treatment filter inspection model to obtain the exhaust gas treatment filter from the non-destructive test result data 320. The usability estimation result 120, which estimates the usability of the above, is output. [Selection diagram] FIG. 21

Description

The present disclosure relates to a non-destructive inspection system, a method of manufacturing a trained exhaust gas treatment filter inspection model, and a method of generating learning data.

Conventionally, US Pat. No. 7,849,747 (Patent Document 1) describes a technique for ultrasonically detecting a ceramic monolith used in a diesel particulate filter.

U.S. Pat. No. 7,849,747

Work machines generally use a diesel engine as a power source. Particulate matter (PM) emitted from a diesel engine is removed from the exhaust gas by a diesel particulate filter (DPF) or the like, as in a normal automobile. Since the removed particulate matter is mainly soot produced by combustion, it is burned in the DPF.

Since PM contains incombustibles in addition to the main component soot, the incombustibles accumulate in the DPF. Therefore, the DPF is replaced after a predetermined time has elapsed from the start of use. In ordinary automobiles, it is replaced with a new DPF, but in work machines, recycled DPFs that have been washed after use are widespread. The DPF after use is inspected for defects such as cracks by non-destructive inspection such as ultrasonic flaw detection before the washing treatment.

The output data of the ultrasonic flaw detection test is shown by a chart of the elapsed time since the ultrasonic wave was transmitted to the subject and the intensity of the ultrasonic wave reflected and returned. The examiner looks at this chart and judges whether or not there is a defect in the filter, but the frequency of erroneous judgment is high depending on the person. Therefore, the criteria for determining whether or not the product can be recycled is excessively strict to prevent the reproduction of defective products, and it is highly possible that the recyclable products are actually discarded.

The present disclosure provides a non-destructive inspection system, a method of manufacturing a trained exhaust gas treatment filter inspection model, and a method of generating learning data in order to improve the accuracy of determining the availability of an exhaust gas treatment filter.

According to certain aspects of this disclosure, a non-destructive inspection system for exhaust gas treatment filters is provided. The non-destructive inspection system includes a non-destructive test device for non-destructive testing of the exhaust gas treatment filter and a computer. The computer has a trained exhaust gas treatment filter inspection model for determining the availability of a non-destructive tested exhaust gas treatment filter. The computer acquires test result data representing the results of the non-destructive test of the exhaust gas treatment filter by the non-destructive test device, and estimates the usability of the exhaust gas treatment filter from the test result data using the trained exhaust gas treatment filter inspection model. Output the possibility estimation result.

According to certain aspects of the present disclosure, a method of manufacturing a trained exhaust gas treatment filter inspection model is provided. The manufacturing method includes the following processing. The first process is learning including a combination of test result data showing the result of the non-destructive test of the exhaust gas treatment filter and the usability judgment result data for determining the usability of the exhaust gas treatment filter subjected to the non-destructive test. Is to get the data for. The second process is to generate an exhaust gas treatment filter inspection model that inputs test result data and outputs a value related to the availability of the exhaust gas treatment filter based on the acquired plurality of learning data.

According to a certain aspect of the present disclosure, there is provided a method of generating learning data for training an exhaust gas treatment filter inspection model for determining whether or not an exhaust gas treatment filter to be inspected can be used. The generation method includes the following processes. The first treatment is to acquire test result data representing the result of the non-destructive test of the exhaust gas treatment filter. The second process is to acquire the usability determination result data for determining the usability of the exhaust gas treatment filter subjected to the non-destructive test.

According to certain aspects of the present disclosure, a method of manufacturing a trained exhaust gas treatment filter inspection model is provided. The manufacturing method includes the following processes. The first treatment is to acquire test result data representing the result of the non-destructive test of the exhaust gas treatment filter. The second process is to input the test result data into the trained first exhaust gas treatment filter inspection model and obtain the output of the usability estimation result that estimates the usability of the exhaust gas treatment filter. The third process is to train the second exhaust gas treatment filter inspection model from the learning data including the test result data and the usability estimation result.

According to the present disclosure, the accuracy of determining the availability of the exhaust gas treatment filter can be improved.

It is a schematic diagram which shows the schematic structure of the exhaust gas aftertreatment device. It is a block diagram which shows an example of a computer structure. It is a block diagram which shows an example of the structure of a learning computer. It is a functional block diagram for demonstrating the functional configuration of a learning computer. It is a flowchart which shows the processing procedure of the learning process in a learning computer. It is a schematic diagram which shows the process for learning the exhaust gas treatment filter inspection model. It is a flowchart which shows the processing procedure of the preprocessing for the data of the test result acquired by the ultrasonic flaw detection tester. It is a graph which shows an example of a point cloud data. It is a graph which shows an example of the extracted data. It is a graph which shows an example of the extraction data of a rejected product. It is a graph which shows an example of the extraction data of the conditional acceptance product. It is a flowchart which shows the processing procedure of the preprocessing for the image data acquired by the image pickup apparatus. It is a figure which shows an example of an entrance image. It is a figure which shows an example of an exit image. It is a flowchart which shows the processing procedure of the preprocessing for the X-ray image data acquired by an X-ray image pickup apparatus. It is a figure which shows an example of the transmission image data of a passing product. It is a figure which shows an example of the transmission image data of a conditional acceptance product. It is a figure which shows an example of the cross-sectional image data of a rejected product. It is a functional block diagram for demonstrating the functional configuration of the determination computer. It is a flowchart which shows the processing procedure of the estimation processing in a determination computer. It is the schematic which shows the process for estimating the useability of the exhaust gas treatment filter by the exhaust gas treatment filter inspection model. It is a flowchart which shows the process for generating a distillation model.

Hereinafter, embodiments will be described with reference to the drawings. In the following description, the same parts are designated by the same reference numerals. Their names and functions are the same. Therefore, the detailed description of them will not be repeated.

[Structure of exhaust gas aftertreatment device 1]
FIG. 1 is a schematic view showing a schematic configuration of an exhaust gas aftertreatment device 1. The exhaust gas aftertreatment device 1 is a device that purifies the exhaust gas by treating the residual substances contained in the exhaust gas of the diesel engine (hereinafter referred to as "engine") 10. The residual substance is, for example, particulate matter (PM), nitrogen oxide (NOx) and the like. The exhaust gas aftertreatment device 1 purifies the exhaust gas by performing treatments such as collecting PM and reducing NOx.

The exhaust gas aftertreatment device 1 includes a diesel particulate filter (DPF) device 2 and a selective catalytic reduction (SCR) device 3. The DPF device 2 and the SCR device 3 are connected to an exhaust pipe 11 through which the exhaust gas of the engine 10 flows. The DPF device 2 and the SCR device 3 are arranged in this order in the direction in which the exhaust gas flows, which is indicated by the arrow in FIG. The SCR device 3 is arranged on the downstream side of the DPF device 2 in the flow direction of the exhaust gas.

The DPF device 2 includes a Diesel Oxidation Catalyst (DOC) 21 and a Catalyzed Soot Filter (CSF) 22.

The DOC 21 is a catalyst that oxidizes and generates heat of the dosing fuel supplied to the exhaust gas as needed to raise the exhaust gas temperature to a predetermined high temperature range. By utilizing the exhaust gas whose temperature has risen, PM accumulated in CSF22, which will be described later, is self-burned to be incinerated and removed, and CSF22 is regenerated.

When the internal combustion engine is a diesel engine, the dosing fuel is, for example, the same light oil as the engine fuel. The dosing fuel is supplied into the exhaust gas by, for example, a fuel injection device for dosing (not shown) provided in the exhaust pipe 11, and flows into the DPF device 2 together with the exhaust gas.

CSF22 is a filter that collects PM in exhaust gas. The CSF 22 has an inlet end 22A in which the exhaust gas flows into the CSF 22 and an outlet end 22B in which the exhaust gas flows out from the CSF 22. CSF 22 corresponds to the exhaust gas treatment filter in the embodiment.

Although detailed illustration is omitted, the CSF 22 has a honeycomb structure having a large number of small holes. The small hole of the CSF 22 extends from the inlet end 22A toward the outlet end 22B. The cross section of the small hole of the CSF 22 is formed in a polygonal shape (for example, a quadrangular shape, a hexagonal shape, etc.).

The small holes of the CSF 22 are alternately arranged, one that is opened at the inlet end 22A and closed at the outlet end 22B, and the other that is closed at the inlet end 22A and opened at the outlet end 22B. Exhaust gas flows into the CSF 22 through a small hole opened at the inlet end 22A. The exhaust gas passes through the boundary wall separating the adjacent small holes, flows to the small holes opened at the outlet end 22B, and flows out from the CSF 22 at the outlet end 22B. Then, PM is collected at the boundary wall.

The material of CSF22 is, for example, ceramics such as cordierite and silicon carbide, and is appropriately determined according to the application. The inlet side of the CSF 22 may be coated with an oxidation catalyst made of a material different from that of the DOC 21 with a wash coat or the like.

The SCR device 3 includes a denitration catalyst 31 and an ammonia oxidation catalyst 32.
The denitration catalyst 31 is a catalyst that reduces and removes nitrogen oxides (NOx) in the exhaust gas by using ammonia obtained by decomposition of urea water injected from the reducing agent supply device into the exhaust pipe 11 as a reducing agent. ..

The ammonia oxidation catalyst 32 arranged on the downstream side of the denitration catalyst 31 is a catalyst that oxidizes ammonia that was not used in the reduction reaction of the denitration catalyst 31 to make it harmless, and further reduces harmful components in the exhaust gas. ..

[Configuration of computer 100]
Next, the configuration of the computer 100 included in the non-destructive inspection system of the embodiment for non-destructive inspection of the exhaust gas treatment filter (CSF22) will be described. FIG. 2 is a block diagram showing an example of the configuration of the computer 100.

As shown in FIG. 2, the computer 100 includes a display 102, a processor 104, a memory 106, a network controller 108, a storage 110, an optical drive 122, and an input device 126 as main hardware elements. Including. The input device 126 includes a keyboard 127 and a mouse 128. The input device 126 may include a touch panel.

The display 102 displays information necessary for processing on the computer 100. The display 102 is composed of, for example, an LCD (Liquid Crystal Display) or an organic EL (Electroluminescence) display.

The processor 104 is an arithmetic unit that executes processing necessary for realizing the computer 100 by executing various programs. The processor 104 is composed of, for example, one or a plurality of CPUs (Central Processing Units) or GPUs (Graphics Processing Units). A CPU or GPU having a plurality of cores may be used.

The memory 106 provides a storage area for temporarily storing a program code, a work memory, and the like when the processor 104 executes a program. As the memory 106, for example, a volatile memory device such as DRAM (Dynamic Random Access Memory) or SRAM (Static Random Access Memory) may be used.

The network controller 108 sends and receives data to and from any external device, including the non-destructive testing device 200 and the learning computer 300, which will be described later. The network controller 108 may be compatible with any communication method such as Ethernet (registered trademark), wireless LAN (Local Area Network), and Bluetooth (registered trademark).

The storage 110 stores an OS (Operating System) 112 executed by the processor 104, an application program 114 for realizing a predetermined functional configuration, a learned model 116, and the like. As the storage 110, for example, a non-volatile memory device such as a hard disk or SSD (Solid State Drive) may be used.

A part of the library or functional module required when the application program 114 is executed by the processor 104 may be a library or functional module provided as standard by the OS 112. In this case, the application program 114 alone does not include all the program modules necessary to realize the corresponding functions, but by being installed under the execution environment of OS 112, a predetermined functional configuration can be obtained. It will be possible. Therefore, even a program that does not include some such libraries or functional modules may be included in the technical scope of the present invention.

The optical drive 122 reads information such as a program stored in an optical disc 124 such as a CD-ROM (Compact Disc Read Only Memory) or a DVD (Digital Versatile Disc). The optical disk 124 is an example of a non-transitory recording medium, and is distributed in a non-volatile state in which an arbitrary program is stored. A computer 100 according to this embodiment can be configured by the optical drive 122 reading a program from the optical disk 124 and installing it in the storage 110. Therefore, the subject of the present invention may be the program itself installed in the storage 110 or the like, or a recording medium such as an optical disk 124 containing a program for realizing a function or process according to the present embodiment.

FIG. 2 shows an optical recording medium such as an optical disk 124 as an example of a non-transient recording medium, but the present invention is not limited to this, and a semiconductor recording medium such as a flash memory or a magnetic recording medium such as a hard disk or a storage tape is shown. , MO (Magneto-Optical disk) or the like may be used.

Alternatively, the program for realizing the computer 100 may be distributed not only by being stored in an arbitrary recording medium as described above and distributed, but also by downloading from a server device or the like via the Internet or an intranet.

FIG. 2 shows a configuration example in which the general-purpose computer (processor 104) realizes the computer 100 by executing the application program 114, but all or a part of the functions necessary for realizing the computer 100 are integrated circuits. It may be realized by using a hard-wired circuit such as. For example, it may be realized by using an ASIC (Application Specific Integrated Circuit) or an FPGA (Field-Programmable Gate Array).

[Configuration of learning computer 300]
Next, the configuration of the learning computer 300 for learning the exhaust gas treatment filter inspection model for determining whether or not the exhaust gas treatment filter (CSF22) to be inspected can be recycled will be described. FIG. 3 is a block diagram showing an example of the configuration of the learning computer 300.

As shown in FIG. 3, the learning computer 300 includes a display 302, a processor 304, a memory 306, a network controller 308, a storage 310, and an input device 330 as main hardware elements. ..

The display 302 displays information necessary for processing by the learning computer 300. The display 302 is composed of, for example, an LCD or an organic EL display.

The processor 304 is an arithmetic unit that executes processing necessary for realizing the learning computer 300 by executing various programs. The processor 304 is composed of, for example, one or a plurality of CPUs or GPUs. A CPU or GPU having a plurality of cores may be used. In the learning computer 300, it is preferable to adopt a GPU or the like suitable for the learning process for generating the trained model.

The memory 306 provides a storage area for temporarily storing a program code, a work memory, and the like when the processor 304 executes a program. As the memory 306, for example, a volatile memory device such as DRAM or SRAM may be used.

The network controller 308 sends and receives data to and from any external device, including the computer 100 and the non-destructive testing device 200. The network controller 308 may be compatible with any communication method such as Ethernet, wireless LAN, and Bluetooth.

The storage 310 is for generating a learning data set 324 from the OS 312 executed by the processor 304, the application program 314 for realizing a predetermined functional configuration, the non-destructive test result data 320, and the usability determination result data 322. The preprocessing program 316 and the training program 318 for generating the trained model 326 using the training data set 324 are stored.

For convenience of explanation, the trained model stored in the computer 100 and the trained model generated by the learning computer 300 are designated by different reference numerals (116,326). However, since the trained model 116 stored in the computer 100 is a trained model transmitted (distributed) from the learning computer 300, the two trained models 116 and 326 are substantially the same. Specifically, the trained model 116 and the trained model 326 have substantially the same network structure and trained parameters.

The learning data set 324 is a training data set in which the non-destructive test result data 320 is assigned the usability determination result data 322 as a label (or tag). Further, the trained model 326 is an estimation model obtained by executing a training process using the training data set 324.

As the storage 310, for example, a non-volatile memory device such as a hard disk or SSD may be used.

Some of the libraries and functional modules required when executing the application program 314, the preprocessing program 316, and the learning program 318 on the processor 304 may use the libraries or functional modules provided as standard by the OS 312. .. In this case, each of the application program 314, the preprocessing program 316, and the learning program 318 does not include all the program modules necessary to realize the corresponding functions, but under the execution environment of OS 312. By being installed in, it is possible to realize a predetermined functional configuration. Therefore, even a program that does not include some such libraries or functional modules may be included in the technical scope of the present invention.

The application program 314, the preprocessing program 316, and the learning program 318 are optical recording media such as optical disks, semiconductor recording media such as flash memory, magnetic recording media such as hard disks or storage tapes, and magneto-optical such as MO. It may be stored in a non-transient recording medium such as a recording medium, distributed, and installed in the storage 310. Therefore, the subject of the present invention may be the program itself installed in the storage 310 or the like, or a recording medium in which a program for realizing a function or process according to the present embodiment is stored.

Alternatively, the program for realizing the learning computer 300 is not only stored and distributed in an arbitrary recording medium as described above, but also distributed by downloading from a server device or the like via the Internet or an intranet. Good.

The input device 330 accepts various input operations. As the input device 330, for example, a keyboard, a mouse, a touch panel, or the like may be used.

FIG. 3 shows a configuration example in which the general-purpose computer (processor 304) realizes the learning computer 300 by executing the application program 314, the preprocessing program 316, and the learning program 318, and realizes the learning computer 300. All or part of the functions required for this may be realized by using a hard-wired circuit such as an integrated circuit. For example, it may be realized by using ASIC, FPGA or the like.

[Learning process]
The learning process executed by the learning computer 300 will be described. Specifically, a method of manufacturing the trained model 326 will be described. FIG. 4 is a functional block diagram for explaining the functional configuration of the learning computer 300.

As shown in FIG. 4, the learning computer 300 includes an input receiving unit 350, a control unit 360, and a communication IF (Interface) unit 370. The control unit 360 includes a learning unit 362. The learning unit 362 includes a learning model 366 and a learning program 368. The model 366 for learning is composed of a network structure 366N and a parameter 366P. The network structure 366N is constructed in advance and stored in the learning computer 300.

The input receiving unit 350 receives the input of the learning data set 324. The learning data set 324 includes non-destructive test result data for learning and usability determination result data as teacher data. Specifically, the learning data set 324 includes a plurality of data sets (learning data), and each set (each learning data) can be used as a non-destructive test result data for learning and a teacher data. Includes data. In addition, a part of a set of a plurality of data sets of the training data set may be used to evaluate the accuracy of the trained model.

The control unit 360 controls the overall operation of the learning computer 300.
The learning unit 362 in the control unit 360 generates the trained model 326. The generation of the trained model 326 will be described below.

The learning unit 362 updates the value of the parameter 366P of the learning model 366 by machine learning using the learning data set 324. Specifically, the learning unit 362 updates the value of the parameter 366P by using the learning program 368. The update of parameter 366P is repeated for the number of sets of data used for training (excluding sets of data for evaluation).

When the training is completed, the trained model 326 is obtained. The trained model 326 has a network structure of 366N and trained parameters. The parameter 366P for which the update is completed corresponds to the learned parameter.

The generated trained model 326 is transmitted to the computer 100 via the communication IF unit 370. As described above, the trained model 326 transmitted to the computer 100 is referred to as "trained model 116" for convenience of explanation.

FIG. 5 is a flowchart showing a processing procedure of the learning process in the learning computer 300. Each step shown in FIG. 5 is typically realized by the processor 304 of the learning computer 300 executing the OS 312, the application program 314, the preprocessing program 316, and the learning program 318 (all see FIG. 3). May be done.

As shown in FIG. 5, in step S1, the learning computer 300 acquires the non-destructive test result data 320.

FIG. 6 is a schematic view showing a process for learning an exhaust gas treatment filter inspection model. As shown in FIG. 6, the non-destructive inspection system includes a non-destructive testing device 200. The non-destructive testing device 200 includes an ultrasonic flaw detection tester 210, an imaging device 220, and an X-ray CT (computerized tomography) scanner 230. The non-destructive test apparatus 200 may include an X-ray transmission apparatus 240 in place of or in addition to the X-ray CT scanner 230.

The ultrasonic flaw detector 210 has a flaw detector. The flaw detector emits ultrasonic pulses and receives reflected echoes. When ultrasonic waves are transmitted from the flaw detector located at the inlet end 22A of the CSF 22, the ultrasonic waves that do not hit the defect are reflected and returned at the exit end 22B, and the reflected echo echoes at a distance corresponding to the length of the CSF 22. A waveform appears. When a defect exists inside the CSF 22, the apex of the waveform of the reflected echo reflected by the defect appears at a position shorter than the length of the CSF 22. In this way, the presence of defects inside the CSF 22 and the positions of the defects are detected based on the positions of the vertices of the waveform of the reflected echo.

The image pickup apparatus 220 acquires an optical image of the image pickup target and detects the outer shape of the image pickup target. The image pickup device includes an image pickup device such as a CCD (Charge-Coupled Device) image sensor or a CMOS (Complementary Metal Oxide Semiconductor) image sensor. The imaging device 220 of the embodiment images the CSF 22, more specifically, the inlet end 22A and the outlet end 22B of the CSF 22.

The X-ray CT scanner 230 irradiates the CSF 22 to be imaged with X-rays and processes the detection signal of the X-rays transmitted through the CSF 22 to generate cross-sectional image data inside the CSF 22. From the generated cross-sectional image data, defects such as cracks or melting have occurred inside the CSF 22, and the small holes of the CSF 22 are clogged with soot or incombustibles, and the small holes are blocked. It is detected that the CSF 22 cannot be reproduced and used as it is.

The X-ray transmission device 240 irradiates the CSF 22 to be imaged with X-rays and processes the detection signal of the X-rays transmitted through the CSF 22 to generate transmission image data of the entire inside of the CSF 22. From the generated transmission image data, defects such as cracks or melting have occurred inside the CSF 22, or soot or incombustibles are clogged in the small holes of the CSF 22 and the small holes are blocked. It is detected that the CSF 22 cannot be reproduced and used as it is.

The non-destructive test apparatus 200 transmits the non-destructive test result data 320 representing the result of the non-destructive test to the learning computer 300, specifically, the network controller 308. Specifically, the ultrasonic flaw detection tester 210 transmits the ultrasonic flaw detection test result data 320A to the learning computer 300. The image pickup apparatus 220 transmits the entrance / exit image data 320B to the learning computer 300. The X-ray CT scanner 230 transmits the cross-sectional image data 320C to the learning computer 300. The X-ray transmission device 240 transmits the transmission image data 320D to the learning computer 300.

Ultrasonic flaw detection test result data 320A contains 10 data for one CSF22. These 10 data are 5 data in which flaw detectors are arranged at 5 locations at the entrance end 22A of one CSF 22, typically 1 location at the center of the entrance end 22A and 4 locations around the entrance end 22A at equal intervals. including. The 10 data also include 5 flaw detectors placed at 5 locations on the exit end 22B of the same CSF 22, typically at 1 location in the center of the exit end 22B and 4 locations on the periphery evenly spaced. Contains data.

The preprocessing program 316 shown in FIG. 3 executes preprocessing for each of a plurality of data input from the ultrasonic flaw detection tester 210 to the learning computer 300. FIG. 7 is a flowchart showing a processing procedure of preprocessing for the data of the test result acquired by the ultrasonic flaw detection tester 210.

As shown in FIG. 7, in step S101, the learning computer 300 acquires one ultrasonic flaw detection test result data 320A from the ultrasonic flaw detection tester 210. Next, in step S102, the learning computer 300 generates point cloud data obtained by drawing the ultrasonic flaw detection test result data 320A acquired in step S101 on the Cartesian coordinate plane.

FIG. 8 is a graph showing an example of point cloud data. The horizontal axis of the graph shown in FIG. 8 indicates the length of the CSF 22. Of both ends of the CSF 22, the end on the side where the flaw detector is arranged corresponds to the coordinate zero on the horizontal axis. The vertical axis of the graph shown in FIG. 8 indicates the echo level received by the flaw detector. The learning computer 300 creates point cloud data, which is a graph obtained by plotting ultrasonic flaw detection test result data 320A on the Cartesian coordinate plane shown in FIG. The data at each point drawn in FIG. 8 shows the magnitude of the reflected echo reflected at one point in the length direction of the CSF 22.

Returning to FIG. 7, in step S103, the learning computer 300 generates polygonal line data in which a straight line connecting the points of the point cloud data shown in FIG. 8 is drawn. In step S104, the learning computer 300 colors a part of the generated polygonal line data. In step S105, the learning computer 300 generates extraction data obtained by extracting a part of the data in the length direction of the CSF 22 from the polygonal line data generated in step S103 and partially colored in step S104.

FIG. 9 is a graph showing an example of the extracted data. The extracted data shown in FIG. 9 is generated by extracting the data in the region A1 shown in FIG. 8 and drawing a straight line connecting the point clouds included in the region A1. Since the vicinity of the coordinate zero on the horizontal axis is a dead zone and the accuracy of the data is low, by removing the data in the dead zone, it is possible to increase the resolution of a high-value part in the judgment of whether or not the CSF 22 can be used. The accuracy of is improved.

Compared with the scatter plot shown in FIG. 8, by generating the line data connecting each plot with a straight line, it becomes possible to easily recognize the apex of the waveform of the reflected echo. As a result, the accuracy of determining whether or not the product can be used is improved.

FIG. 9 shows the echo level threshold T. The threshold value T is defined as, for example, 50% of the maximum echo level of the echo levels of each data included in the ultrasonic flaw detection test result data 320A. The extracted data shown in FIG. 9 is colored in a range where the echo level is equal to or higher than the threshold value T. Coloring in this way makes it easier to accurately recognize the vertices of the reflected echo waveform, so that it is possible to more accurately determine whether or not the CSF 22 has an internal defect, and therefore the availability of the CSF 22 can be determined. The accuracy of the judgment is improved.

In the extracted data shown in FIG. 9, the apex of the reflected echo waveform is located at a position corresponding to the length of the CSF 22. The ultrasonic waves transmitted from the flaw detector at one end of the CSF 22 (for example, the inlet end 22A) propagate to the other end of the CSF 22 (for example, the exit end 22B), are reflected at the other end, and are received by the flaw detector. It is shown that it was done. In this case, there are no defects such as cracks or melting inside the CSF 22, and it is determined that the CSF 22 is a accepted product.

FIG. 10 is a graph showing an example of extracted data of rejected products. In the extracted data shown in FIG. 10, the apex of the reflected echo waveform is located in the middle of the CSF 22 in the length direction. The ultrasonic waves transmitted from the flaw detector at one end of the CSF 22 (for example, the inlet end 22A) do not propagate to the other end of the CSF 22 (for example, the exit end 22B), but are reflected back at an intermediate position and returned to the flaw detector. It is shown that it was received at. In this case, the propagation of ultrasonic waves is blocked at a position in the middle, and there is a defect such as a crack at the position in the middle, so that the CSF 22 is not connected. Therefore, it is determined that the CSF 22 is a rejected product having a defect inside.

FIG. 11 is a graph showing an example of extraction data of conditional accepted products. In the extracted data shown in FIG. 11, there are no clear vertices in the reflected echo waveform. It means that the apex of the waveform of the reflected echo was not taken because the non-combustible component clogged in the small hole of the CSF 22 solidified and became solid, and the amount of scattering attenuation of the ultrasonic wave became large. In this case, if the clogging in the small holes can be cleared by washing, the CSF 22 can be regenerated and used, so that it is determined that the CSF 22 is a conditionally accepted product.

Washing may be performed using an appropriate solution to dissolve the non-combustible component, or may be performed by blowing off the non-combustible component with an air blow. In addition to or instead of washing, a treatment may be performed in which the non-combustible component is burned by applying heat to the CSF 22 to clear the clogging in the small holes.

The ultrasonic flaw detection test result data 320A subjected to the pretreatment in this way is stored in the storage 310 and constitutes the non-destructive test result data 320.

FIG. 12 is a flowchart showing a processing procedure of preprocessing for the image data acquired by the image pickup apparatus 220. As shown in FIG. 12, in step S111, the learning computer 300 acquires the captured image from the imaging device 220. Specifically, the learning computer 300 acquires an entrance image of the entrance end 22A of the CSF 22 and an exit image of the exit end 22B of the CSF 22.

FIG. 13 is a diagram showing an example of an entrance image. FIG. 13 shows a still image of the inlet end 22A of the CSF 22. The CSF 22 is held by a stainless steel holding tube 28 and is housed in the holding tube 28. Therefore, in the entrance image shown in FIG. 13, the holding tube 28 is reflected around the CSF 22. FIG. 14 is a diagram showing an example of an exit image. FIG. 14 shows a still image of the exit end 22B of the CSF 22.

Returning to FIG. 12, in step S112, trimming of the entrance image and the exit image is executed. Objects other than CSF22, such as the holding tube 28 and the photographer's feet, are also reflected in the images of CSF 22 shown in FIGS. 13 and 14. Therefore, the entrance image and the exit image are trimmed to cut out the outer portion of the region A2 shown in FIGS. 13 and 14, leaving the portion in the region A2 in which only the CSF 22 is shown. By removing an object other than the CSF 22 from the entrance image and the exit image, it is possible to increase the resolution of a high-value portion in the determination of the availability of the CSF 22, thereby improving the accuracy of the determination of the availability.

The entrance / exit image data 320B for which the preprocessing has been executed in this way is stored in the storage 310 and constitutes the non-destructive test result data 320.

FIG. 15 is a flowchart showing a processing procedure of preprocessing for the X-ray image data acquired by the X-ray imaging apparatus. If the X-ray imaging device is an X-ray CT scanner 230, the X-ray image data is the cross-sectional image data 320C. If the X-ray image pickup device is an X-ray transmission device 240, the X-ray image data is the transmission image data 320D. As shown in FIG. 15, in step S121, the learning computer 300 acquires X-ray image data from the X-ray imaging device.

FIG. 16 is a diagram showing an example of transmission image data of a passing product as an example of X-ray image data. FIG. 16 shows one transmission image data of CSF22. The transmission image data shown in FIG. 16 includes an inlet end 22A and an outlet end 22B.

Returning to FIG. 15, in step S122, trimming of the X-ray image data is executed. The transmission image data shown in FIG. 16 includes other than CSF22. Therefore, the transmission image data is trimmed to cut out the outer portion of the region A3 shown in FIG. 16 to leave a portion in the region A3 in which only the CSF 22 is shown. By removing an object other than the CSF 22 from the transmission image data, it is possible to increase the resolution of a high-value portion in the determination of the availability of the CSF 22, and thereby improve the accuracy of the determination of the availability.

In the transmission image data shown in FIG. 16, there are no defects such as cracks or melting inside the CSF 22 from the inlet end 22A to the outlet end 22B, and it is determined that the CSF 22 is a accepted product.

FIG. 17 is a diagram showing an example of transmission image data of a conditionally accepted product. In the transmission image data shown in FIG. 17, a state in which incombustibles are clogged in the region A4 on the outlet end 22B side is shown. This CSF 22 is determined to be a conditionally acceptable product if the CSF 22 can be recycled by washing the incombustible material.

FIG. 18 is a diagram showing an example of cross-sectional image data of a rejected product. In the cross-sectional image data shown in FIG. 18, a melted portion exists in the middle of the CSF 22, and a crack exists in the vicinity of the outlet end 22B. Therefore, it is determined that the CSF 22 is a rejected product having a defect inside.

The X-ray image data (cross-sectional image data 320C and transmission image data 320D) for which the preprocessing has been executed in this way is stored in the storage 310 and constitutes the non-destructive test result data 320.

Returning to FIGS. 5 and 6, then in step S2, the learning computer 300 acquires the usability determination result data 322. The usability determination result data 322 should be discarded because each CSF 22 that has been non-destructively tested can be recycled as it is, can be recycled if washed, or cannot be recycled. Or, it contains the determined result.

The usability determination result data 322 as the teacher data can be obtained, for example, by inspecting the cross-sectional image data 320C acquired by the X-ray CT scanner 230. Also, insert the pin into the small hole through the opening of the inlet end 22A (or outlet end 22B) of the CSF 22 and check whether the tip of the pin reaches the closed outlet end 22B (or inlet end 22A). Based on the result of the clogging confirmation test, the usability determination result data 322 can be obtained.

Further, by cutting the non-destructive test CSF 22 for generating the non-destructive test result data 320 for learning and visually observing the inside, the usability determination result data 322 can be obtained.

Next, in step S3, the learning computer 300 generates the learning data set 324 by associating the non-destructive test result data 320 with the usability determination result data 322. In step S4, the learning computer 300 selects one set of data (learning data) from the generated learning data set 324.

Next, in step S5, the learning computer 300 inputs the learning data selected in step S4 into the learning model 366, and obtains an output of the usability estimation result.

As described with reference to FIG. 4, the learning model 366 has a network structure of 366N. The network structure 366N includes the neural network shown in FIG.

The network structure 366N includes a deep neural network such as a convolutional neural network (CNN). The neural network includes an input layer 366Na, an intermediate layer (hidden layer) 366Nb, and an output layer 366Nc. The intermediate layer 366Nb is multi-layered. The input layer 366Na, the intermediate layer 366Nb and the output layer 366Nc have one or more units (neurons). The number of units of the input layer 366Na, the intermediate layer 366Nb, and the output layer 366Nc can be appropriately set.

Units in adjacent layers are bonded together, and weights are set for each bond. A bias is set for each unit. A threshold is set for each unit. The output value of each unit is determined by whether or not the value obtained by adding the bias to the sum of the products of the input values and the weights for each unit exceeds the threshold value.

The model 366 for learning is learned from the non-destructive test result data 320 of the CSF 22 so as to determine whether or not the CSF 22 subjected to the non-destructive test can be regenerated. The model 366 for training includes the parameter 366P obtained by training. The parameter 366P includes, for example, the number of layers of the neural network, the number of units in each layer, the connection relationship between units, the weight of the connection between each unit, the bias associated with each unit, and the threshold value of each unit. There is.

The learning computer 300 uses the non-destructive test result data 320 as an input of the input layer 366Na to perform arithmetic processing of forward propagation of the neural network of the network structure 366N. As a result, the learning computer 300 obtains a usability estimation result that estimates the usability of the CSF 22 as an output value output from the output layer 366Nc of the neural network.

Next, in step S6, the learning computer 300 calculates an error between the usability determination result data of the selected learning data and the usability estimation result obtained in step S5.

Next, in step S7, the learning computer 300 updates the learning model 366 (CSF inspection model). The learning computer 300 determines the weight of the coupling between each unit, the bias of each unit, the threshold value of each unit, and the like, based on the error of the usability estimation result with respect to the usability determination result data 322 calculated in step S6. The parameter 366P of the model 366 for training is updated.

In this way, the learning computer 300 inputs the non-destructive test result data 320 into the learning model 366, and the usability estimation result output is labeled with the non-destructive test result data 320. The process of optimizing the parameter 366P of the learning model 366 is executed so as to approach the determination result data 322. Then, when the same non-destructive test result data 320 is input to the input layer 366Na, an output value closer to the usability determination result data 322 can be output.

Next, in step S8, the learning computer 300 determines whether or not all of the learning data sets 324 generated in step S3 have been processed. If all of the training data set 324 has not been processed (NO in step S8), the processing of step S4 and subsequent steps is repeated. If all of the training data set 324 has been processed (YES in step S8), the process proceeds to step S9, and the learning computer 300 stores the current parameter 366P. With the above, the learning process is completed (end).

The initial value of the parameter 366P of the model 366 for learning may be given by the template. Alternatively, the initial value of parameter 366P may be manually given by human input. When retraining the learning model 366, the learning computer 300 prepares the initial value of the parameter 366P based on the value stored as the parameter 366P of the learning model 366 to be retrained. You may.

[Use of learning model]
The trained model 326 defined by the current parameter 366P, which has been trained as described above, is transmitted to the computer 100 (determination computer). Hereinafter, the use of the trained model 116 in the computer 100 will be described. Specifically, the process of estimating the availability of the CSF 22 executed by the computer 100 will be described.

FIG. 19 is a functional block diagram for explaining the functional configuration of the computer 100 (determination computer). As shown in FIG. 19, the computer 100 includes an input receiving unit 150, a control unit 160, and a display unit 170. The control unit 160 includes a usability estimation unit 161 and a display control unit 162. The usability estimation unit 161 includes a trained model 116.

The input receiving unit 150 receives the input of the non-destructive test result data 320.
The control unit 160 controls the overall operation of the computer 100.

The usability estimation unit 161 in the control unit 160 has a trained model 116. The trained model 116 is composed of a network structure 116N and a trained parameter 116P. The network structure 116N is substantially identical to the network structure 366N (see FIG. 4) and includes an input layer 116Na, an intermediate layer (hidden layer) 116Nb, and an output layer 116Nc (see FIG. 21 below). ..

The usability estimation unit 161 estimates from the non-destructive test result data 320 whether or not the non-destructive tested exhaust gas treatment filter (CSF22) is reproducible by using the trained model 116. The usability estimation unit 161 sends the usability estimation result to the display control unit 162.

The display control unit 162 causes the display unit 170 to display the usability estimation result. The display unit 170 corresponds to the display 102 (see FIG. 2).

FIG. 20 is a flowchart showing a processing procedure of the estimation process in the computer 100 (determination computer). Each step shown in FIG. 20 may typically be implemented by processor 104 of computer 100 executing OS 112 and application program 114 (both see FIG. 2). FIG. 21 is a schematic view showing a process for estimating the availability of the exhaust gas treatment filter by the exhaust gas treatment filter inspection model.

As shown in FIG. 20, in step S201, the computer 100 acquires the non-destructive test result data 320. The non-destructive test apparatus 200 transmits the non-destructive test result data 320 representing the result of the non-destructive test to the computer 100, specifically, the input receiving unit 150. Specifically, the ultrasonic flaw detection tester 210 transmits the ultrasonic flaw detection test result data 320A to the computer 100. The image pickup apparatus 220 transmits the entrance / exit image data 320B to the computer 100. The X-ray CT scanner 230 transmits the cross-sectional image data 320C to the computer 100. The X-ray transmission device 240 transmits the transmission image data 320D to the computer 100.

Next, in step S202, the computer 100 inputs the non-destructive test result data 320 into the trained model 116. The non-destructive test result data 320 acquired in step S201 is used as input data to the input layer 116Na of the network structure 116N of the trained model 116. The usability estimation unit 161 inputs the non-destructive test result data 320 to each unit included in the input layer 116Na of the network structure 116N.

In step S203, the computer 100 outputs the usability estimation result 120. The usability estimation unit 161 outputs a usability estimation result 120 that estimates the reproducibility of the CSF 22 that has been subjected to the non-destructive test from the output layer 116Nc of the network structure 116N of the trained model 116. More specifically, the usability estimation unit 161 determines whether the CSF 22 subjected to the non-destructive test using the trained model 116 can be regenerated as it is, or can be regenerated if it is washed. Generates an estimation result that estimates whether it is non-renewable and should be discarded.

Finally, in step S204, the computer 100 displays the usability estimation result 120 on the display 102. Then, the process ends (end).

As described above, in the non-destructive inspection system according to the embodiment, the computer 100 has a trained model 116 for determining whether or not the exhaust gas treatment filter (CSF22) subjected to the non-destructive test can be recycled. ing. As shown in FIGS. 20 and 21, the computer 100 acquires the non-destructive test result data 320 representing the result of the non-destructive test of the CSF 22 by the non-destructive test device 200, and uses the trained model 116 to perform the non-destructive test. It is programmed to output the usability estimation result 120, which estimates the reproducibility of the CSF 22 from the result data 320.

Therefore, it is possible to estimate the reproducibility of the CSF 22 by using the learned model 116 of the artificial intelligence suitable for estimating the usability of the CSF 22. As a result, the availability of the CSF 22 can be easily estimated by the computer 100 using artificial intelligence, and the accuracy of determining the availability can be improved.

As shown in FIGS. 5 and 6, the computer 100 includes the usability estimation result of estimating the usability of the CSF 22 from the non-destructive test result data 320 using the learning model 366, and the learning data set 324. The learning model 366 is programmed to be updated based on the error from the availability determination result data 322 of the CSF 22. By doing so, it is possible to sufficiently train the learning model 366 in advance and create a highly accurate model. By transmitting the learning model 366 to the computer 100 and using it as the trained model 116, it is possible to accurately determine whether or not the CSF 22 can be used.

As shown in FIGS. 6 and 7-11, the non-destructive test of CSF 22 includes an ultrasonic flaw detection test. The non-destructive test result data 320 includes ultrasonic flaw detection test result data 320A representing the result of the ultrasonic flaw detection test. Thereby, the result of the ultrasonic flaw detection test can be used for learning the model 366 for learning, and the availability of the CSF 22 can be determined by using the result of the ultrasonic flaw detection test.

As shown in FIG. 8, a graph is created in which the results of the ultrasonic flaw detection test are plotted with the length of the CSF 22 as the horizontal axis and the echo level as the vertical axis. Using this graph, it is possible to determine the availability of CSF22 that has undergone an ultrasonic flaw detection test.

As shown in FIGS. 9 and 10, a part of the graph is colored. By recognizing a colored part of the graph, it is possible to improve the accuracy of determining whether or not the CSF 22 can be used.

As shown in FIGS. 9 to 11, the extracted data obtained by extracting a part of the data in the length direction of the CSF 22 is generated. By removing low-value parts such as dead zones as data, it is possible to increase the resolution of high-value parts in the determination of whether or not the CSF 22 can be used, so that the accuracy of the determination of availability can be improved.

As shown in FIGS. 6 and 12-14, the non-destructive test of CSF 22 includes image analysis of an inlet image of the inlet end 22A of the CSF 22 and an exit image of the exit end 22B. The non-destructive test result data 320 includes entrance / exit image data 320B representing the result of image analysis. As a result, the result of the image analysis of the entrance image and the exit image can be used for learning the model 366 for learning, and the availability of the CSF 22 can be determined more accurately by using the result of the image analysis. it can.

As shown in FIGS. 13 and 14, a part of the entrance image is trimmed and a part of the exit image is trimmed. By removing objects other than the CSF 22 from the entrance image and the exit image, it is possible to increase the resolution of a high-value portion in the determination of the availability of the CSF 22, and thus the accuracy of the determination of the availability can be improved.

As shown in FIGS. 6 and 18, the non-destructive test of CSF 22 includes an X-ray computed tomography test. The non-destructive test result data 320 includes cross-sectional image data 320C representing the result of the X-ray computed tomography test. As a result, the results of the X-ray computed tomography test can be used for learning the model 366 for learning, and the availability of the CSF 22 can be determined more accurately by using the results of the X-ray computed tomography test. Can be done.

As shown in FIGS. 6 and 16 and 17, the non-destructive test of CSF 22 may include, or in addition to, an X-ray computed tomography test. In that case, the non-destructive test result data 320 includes transmission image data 320D showing the result of the X-ray transmission test. Thereby, the result of the X-ray transmission test can be used for learning the model 366 for learning, and the availability of the CSF 22 can be determined more accurately by using the result of the X-ray transmission test.

As shown in FIG. 16, a part of the transmission image data 320D is trimmed. By removing an object other than the CSF 22 from the transmission image data 320D, the resolution of a high-value portion in the determination of the availability of the CSF 22 can be increased, so that the accuracy of the determination of the availability can be improved.

[Manufacturing method of distillation model]
The trained model 116 described above is not limited to the model trained by the learning computer 300 by machine learning using the non-destructive test result data 320, and may be a model generated by using the trained model. For example, the trained model 116 may be another model (distillation model) trained based on the result obtained by repeating the input / output of data to the trained model. FIG. 22 is a flowchart showing a process for generating a distillation model.

As shown in FIG. 22, first, in step S301, non-destructive test result data is acquired. Similar to step S1 shown in FIG. 5, the learning computer acquires non-destructive test result data 320 representing the result of the non-destructive test of the exhaust gas treatment filter (CSF22) from the non-destructive test apparatus 200.

Next, in step S302, the learning computer inputs the non-destructive test result data 320 into the trained first exhaust gas treatment filter (CSF22) inspection model, and the CSF 22 subjected to the non-destructive test is reused. Estimate if it is possible. In step S303, the learning computer outputs the usability estimation result.

The learning computer inputs the non-destructive test result data 320 acquired in step S301 into the input layer of the trained first CSF inspection model. From the output layer of the first CSF inspection model that has been trained, the usability estimation result, specifically, the CSF22 that has been subjected to the non-destructive test can be reused as it is or if it is washed. The estimation result of estimating whether the product is non-renewable and should be discarded is output.

Next, in step S304, the learning computer saves the non-destructive test result data 320 acquired in step S301 and the usability estimation result output in step S303 as learning data.

Next, in step S305, the learning computer learns the second exhaust gas treatment filter (CSF22) inspection model. The learning computer inputs the non-destructive test result data 320 stored in step S304 into the input layer of the second CSF inspection model. The learning computer outputs an output value indicating the usability estimation result of the CSF 22 subjected to the non-destructive test from the output layer of the second CSF inspection model. The error between the usability estimation result output from the second CSF inspection model and the usability estimation result output from the first CSF inspection model output in step S303 is obtained. Based on this error, the learning computer updates the parameters of the second CSF checking model. In this way, learning of the second CSF test model is performed.

Finally, in step S306, the updated parameters of the second CSF inspection model are saved as trained parameters. Then, the process ends (end).

As described above, the second CSF inspection model (distillation model) uses the non-destructive test result data and the usability estimation result of estimating the usability of CSF22 using the first CSF inspection model as learning data. By learning, the computer for learning can estimate whether or not the CSF subjected to the non-destructive inspection can be regenerated by using the first CSF inspection model and the simple second CSF inspection model. This makes it possible to reduce the load on the determination computer for estimating the availability of the CSF.

In the above embodiment, the trained model 116 and the training model 366 include a neural network. Not limited to this, the trained model 116 and the training model 366 can recycle the exhaust gas treatment filter from the non-destructive inspection result of the exhaust gas treatment filter using machine learning such as a support vector machine and a decision tree. It may be a model that can accurately estimate whether or not.

It should be considered that the embodiments disclosed this time are exemplary in all respects and not restrictive. The scope of the present invention is shown by the scope of claims rather than the above description, and is intended to include all modifications within the meaning and scope equivalent to the scope of claims.

1 Exhaust gas aftertreatment device, 2 DPF device, 3 SCR device, 10 engine, 11 exhaust pipe, 21 DOC, 22 CSF, 22A inlet end, 22B outlet end, 28 holding pipe, 31 denitration catalyst, 32 ammonia oxidation catalyst, 100 computer , 102, 302 display, 104, 304 processor, 106, 306 memory, 108, 308 network controller, 110, 310 storage, 114, 314 application program, 116, 326 trained model, 116N, 366N network structure, 116Na, 366Na input Layer, 116Nb, 366Nb intermediate layer, 116Nc, 366Nc output layer, 116P trained parameter, 120 availability estimation result, 122 optical drive, 124 optical disk, 126,330 input device, 127 keyboard, 128 mouse, 150,350 input reception Unit, 160, 360 Control unit, 161 Usability estimation unit, 162 Display control unit, 170 Display unit, 200 Non-destructive test device, 210 Ultrasonic flaw detection tester, 220 Imaging device, 230 X-ray CT scanner, 300 Learning computer 316 preprocessing program, 318,368 learning program, 320 non-destructive test result data, 320A ultrasonic flaw detection test result data, 320B entrance / exit image data, 320C cross-sectional image data, 322 usability judgment result data, 324 learning data set , 362 learning unit, 366 learning model, 366P parameter, 370 communication IF unit.

Claims (18)

A non-destructive inspection system for exhaust gas treatment filters
A non-destructive test device for non-destructive testing of the exhaust gas treatment filter,
A computer with a trained exhaust gas treatment filter inspection model for determining the availability of the exhaust gas treatment filter subjected to the non-destructive test.
The computer acquires test result data representing the result of the non-destructive test of the exhaust gas treatment filter by the non-destructive test apparatus, and uses the trained exhaust gas treatment filter inspection model to obtain the exhaust gas treatment filter from the test result data. A non-destructive inspection system that outputs the usability estimation result that estimates the usability of. The trained exhaust gas treatment filter inspection model is subjected to learning processing using a training data set so as to output the usability estimation result from the test result data when the test result data is input. The non-destructive inspection system according to claim 1. The trained exhaust gas treatment filter inspection model is generated by learning processing using the training data set, and the training data set is the exhaust gas treatment filter obtained by performing the non-destructive test on the test result data. The non-destructive inspection system according to claim 1, which includes a plurality of learning data labeled with the usability determination result data for determining the availability of the data. The non-destructive inspection system according to any one of claims 1 to 3, wherein the non-destructive test includes an ultrasonic flaw detection test. The non-destructive inspection system according to claim 4, wherein the non-destructive test further includes at least one of an X-ray transmission test and an X-ray computed tomography test. The exhaust gas treatment filter has an inlet end at which the exhaust gas flows into the exhaust gas treatment filter and an outlet end at which the exhaust gas flows out from the exhaust gas treatment filter.
The non-destructive inspection system according to claim 4 or 5, wherein the non-destructive test further includes an image analysis for analyzing an entrance image of the entrance end and an exit image of the exit end. It is a method of manufacturing a trained exhaust gas treatment filter inspection model.
Acquisition of learning data including a combination of test result data showing the result of the non-destructive test of the exhaust gas treatment filter and the usability judgment result data for determining the usability of the exhaust gas treatment filter subjected to the non-destructive test. To do and
A manufacturing method comprising generating the exhaust gas treatment filter inspection model in which the test result data is input and the value related to the availability of the exhaust gas treatment filter is output based on the plurality of acquired learning data. .. Generating the exhaust gas treatment filter inspection model
The test result data is input to the exhaust gas treatment filter inspection model to obtain the output of the usability estimation result that estimates the usability of the exhaust gas treatment filter.
Obtaining a comparison result comparing the usability determination result data corresponding to the test result data input to the exhaust gas treatment filter inspection model and the usability estimation result.
The manufacturing method according to claim 7, wherein the exhaust gas treatment filter inspection model is updated based on the comparison result. It is a method of generating learning data for training an exhaust gas treatment filter inspection model for determining whether or not an exhaust gas treatment filter to be inspected can be used.
Obtaining test result data representing the results of the non-destructive test of the exhaust gas treatment filter, and
A method for generating learning data, which comprises acquiring usability determination result data for determining the usability of the exhaust gas treatment filter subjected to the non-destructive test. The non-destructive test includes an ultrasonic flaw detection test.
The method for generating learning data according to claim 9, wherein acquiring the test result data includes acquiring ultrasonic flaw detection test result data representing the result of the ultrasonic flaw detection test. 10. Acquiring the test result data includes creating a graph plotting the ultrasonic flaw detection test result data with the length of the exhaust gas treatment filter as the horizontal axis and the echo level as the vertical axis. The method of generating training data described in. The method for generating learning data according to claim 11, wherein acquiring the test result data includes coloring a part of the graph. Any of claims 10 to 12, wherein acquiring the test result data includes generating extraction data obtained by extracting a part of the data in the length direction of the exhaust gas treatment filter from the ultrasonic flaw detection test result data. The method for generating training data according to item 1. The exhaust gas treatment filter has an inlet end at which the exhaust gas flows into the exhaust gas treatment filter and an outlet end at which the exhaust gas flows out from the exhaust gas treatment filter.
The non-destructive test further includes image analysis for analyzing an inlet image of the inlet end and an exit image of the exit end.
The method for generating learning data according to claim 10, wherein acquiring the test result data includes acquiring the entrance image and acquiring the exit image. The claim that acquiring the test result data includes generating a partial entrance image obtained by trimming a part of the entrance image and generating a partial exit image obtained by trimming a part of the exit image. 14. The method for generating learning data according to 14. The non-destructive test includes an X-ray computed tomography test.
The method for generating learning data according to claim 10, wherein acquiring the test result data includes acquiring cross-sectional image data representing the result of the X-ray computed tomography test. The method for generating learning data according to claim 16, wherein acquiring the test result data includes acquiring partial cross-sectional image data obtained by trimming a part of the cross-sectional image data. Acquiring test result data showing the results of non-destructive testing of exhaust gas treatment filters,
The test result data is input to the trained first exhaust gas treatment filter inspection model to obtain the output of the usability estimation result that estimates the usability of the exhaust gas treatment filter.
A method for manufacturing a trained exhaust gas treatment filter inspection model, which comprises learning a second exhaust gas treatment filter inspection model from learning data including the test result data and the usability estimation result.