Inspection device, inspection method, and inspection program
Determination of presence or absence of a defect having irregular position, size, shape, and/or the like in an image are made automatically. An inspection device includes: an inspection image obtaining section that obtains an inspection image used to determine presence or absence of an internal defect in an inspection target; and a defect presence/absence determining section that determines presence or absence of a defect with use of a restored image generated by inputting the inspection image into a generative model constructed by machine learning that uses, as training data, an image of an inspection target in which a defect is absent, the generative model being constructed so as to generate a new image having a similar feature to that of an image input into the generative model.
1 . An inspection device comprising a processor,
the processor carrying out:
an inspection image obtaining step of obtaining an inspection image, which is an image used to determine presence or absence of a defect inside an inspection target;
a defect presence/absence determination step of determining presence or absence of a defect in the inspection target with use of a restored image generated by inputting the inspection image into a generative model, the generative model being constructed by machine learning so that the generative model generates a new image having a similar feature to that of an image input into the generative model, the generative model being constructed by using, as training data, an image of an inspection target in which a defect is absent; and
an inspection image generating step of extracting, as an inspection target area, an area sandwiched between two peripheral echo areas from an ultrasonic testing image so as to generate the inspection image, the ultrasonic testing image being an image of an echo of an ultrasonic wave propagated in the inspection target, each of the two peripheral echo areas being an area where an echo coming from a periphery of an inspection target portion of the inspection target appears repeatedly,
wherein, in the inspection image obtaining step, the processor obtains the inspection image generated in the inspection image generating step.
2 . The inspection device as set forth in claim 1 , wherein, in the defect presence/absence determination step, in a case where a variance of pixel values of pixels constituting a difference image between the inspection image and the restored image exceeds a given threshold, the processor determines that a defect is present in the inspection target.
3 . The inspection device as set forth in claim 2 , wherein the processor further carries out a defect area detecting step of detecting, as a defect area, an area of the difference image which area is constituted by pixels having a pixel value not less than a threshold.
4 . The inspection device as set forth in claim 3 , wherein the defect is a defect in a welded portion of the inspection target; and
wherein the processor further carries out a defect type determination step of determining a type of the defect related to the defect area, in accordance with a position in an image area of the difference image at which position the defect area is detected.
5 . The inspection device as set forth in claim 2 , wherein the processor further carries out:
a heat map generating step of generating a heat map representing, by colors or gradations, the pixel values of the pixels constituting the difference image; and
a defect type determination step of determining a type of the defect in accordance with an output value obtained by inputting, into a type decision model, the heat map generated by the heat map generating step, the type decision model being constructed by machine learning that uses, as training data, a heat map of a difference image generated from an inspection image of an inspection target having a defect of a known type.
6 . The inspection device as set forth in claim 1 , wherein, in the inspection image generating step, the processor extracts the inspection target area in accordance with an output value obtained by inputting the ultrasonic testing image into an extraction model for the inspection target area, the extraction model being constructed by machine learning that uses, as correct data, an area including (i) an inspection target portion and (ii) at least a part of an area where an echo coming from the periphery of the inspection target portion appears.
7 . The inspection device as set forth in claim 1 , wherein, in the defect presence/absence determination step, the processor determines presence or absence of a defect in the inspection target, with respect to a remaining image area obtained by removing, from an image area of the restored image, the area where the echo coming from the periphery of the inspection target portion appears.
8 . The inspection device as set forth in claim 1 , wherein, in the defect presence/absence determination step, the processor carries out an integrative detection step of, in a case where the processor determines that a defect is present in a plurality of ultrasonic testing images corresponding to parts of the inspection target which parts are adjacent to each other, detecting, as a single defect, the defects captured in the plurality of ultrasonic testing images.
9 . An inspection method involving use of an inspection device, comprising the steps of:
obtaining an inspection image, which is an image used to determine presence or absence of a defect inside an inspection target;
determining presence or absence of a defect in the inspection target with use of a restored image generated by inputting the inspection image into a generative model, the generative model being constructed by machine learning so that the generative model generates a new image having a similar feature to that of an image input into the generative model, the generative model being constructed by using, as training data, a first ultrasonic testing image of an inspection target in which a defect is absent; and
extracting, as an inspection target area, an area sandwiched between two peripheral echo areas from a second ultrasonic testing image so as to generate the inspection image, the second ultrasonic testing image being an image of an echo of an ultrasonic wave propagated in the inspection target, each of the two peripheral echo areas being an area where an echo coming from a periphery of an inspection target portion of the inspection target appears repeatedly,
wherein, in the step of obtaining the inspection image, the inspection image is generated by the step of extracting.
10 . A non-transitory computer readable medium storing an inspection program configured to cause a computer to function as an inspection device, the inspection program causing the computer to carry out:
an inspection image obtaining step of obtaining an inspection image, which is an image used to determine presence or absence of a defect inside an inspection target;
a defect presence/absence determination step of determining presence or absence of a defect in the inspection target with use of a restored image generated by inputting the inspection image into a generative model, the generative model being constructed by machine learning so that the generative model generates a new image having a similar feature to that of an image input into the generative model, the generative model being constructed by using, as training data, an ultrasonic image of an inspection target in which a defect is absent; and
an inspection image generating step of extracting, as an inspection target area, an area sandwiched between two peripheral echo areas from an ultrasonic testing image so as to generate the inspection image, the ultrasonic testing image being an image of an echo of an ultrasonic wave propagated in the inspection target, each of the two peripheral echo areas being an area where an echo coming from a periphery of an inspection target portion of the inspection target appears repeatedly, in the inspection image obtaining step, the inspection program causes the computer to obtain the inspection image generated in the inspection image generating step.