Defect determination device

Abstract

To solve the problem that it is difficult to prepare teacher data on a measurement result including a defect when automatically determining the presence or absence of a defect pertaining to a bead of welding on the basis of a measurement result of a laser ultrasonic method.SOLUTION: A defect determination device 1 includes: a reception section 11 that receives a measurement result of a laser ultrasonic method pertaining to a bead of welding; an image conversion section 12 that converts a plurality of measurement results into an image for each measurement position in a longitudinal direction of the bead; a feature acquisition section 13 that acquires feature information, including a feature quantity of the image, for each image; a clustering section 15 that clusters a plurality of pieces of feature information; and a determination section 16 that determines that a defect is present in the bead when the clustering section 15 clusters the pieces of feature information into two or more clusters. Thus, the defect determination device can determine without preparing teacher data on a measurement result, including a defect, by determining by clustering.SELECTED DRAWING: Figure 2A

Description

Technical Field

[0001] The present invention relates to a defect determination device that determines the presence or absence of defects in a weld bead using the measurement results of the laser ultrasonic method.

Background Art

[0002] Conventionally, an apparatus for inspecting a weld using the Laser Ultrasonic Technique has been known (see, for example, Patent Document 1). By using the laser ultrasonic method, the state of the welded portion can be measured non - contact, so that, for example, measurement of the bead immediately after welding is also possible.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] There is a desire to automatically find welding defects by applying the measurement results by the laser ultrasonic method to a learning device such as a neural network. However, in order to train such a learning device, it is necessary to prepare the measurement results at defect - free locations and the measurement results at defective locations as teacher data. However, there is a problem that it is difficult to prepare the measurement results at defective locations of the weld in an amount required for learning.

[0005] The present invention has been made to solve the above problems, and an object of the present invention is to provide a defect determination device that can automatically determine the presence or absence of defects related to a weld bead based on the measurement results of the laser ultrasonic method without using teacher data.

Means for Solving the Problems

[0006] To achieve the above object, a defect determination device according to an aspect of the present invention includes a reception unit that receives measurement results of the laser ultrasonic method regarding a weld bead, an image conversion unit that converts a plurality of measurement results for each measurement position in the longitudinal direction of the bead into an image, a feature acquisition unit that acquires feature information including feature amounts of the image for each image, a clustering unit that clusters a plurality of pieces of feature information, and a determination unit that determines that a defect exists in the bead when it is clustered into two or more clusters by the clustering unit.

Advantages of the Invention

[0007] According to the defect determination device according to an aspect of the present invention, by clustering a plurality of pieces of feature information acquired from measurement results of the laser ultrasonic method regarding a weld bead, it is possible to automatically determine the presence or absence of a defect without using teacher data.

Brief Description of the Drawings

[0008]

Figure 1

Figure 2A

Figure 2B

Figure 3

Figure 4

Figure 5

Modes for Carrying Out the Invention

[0009] Hereinafter, a defect determination apparatus according to the present invention will be described using embodiments. In the following embodiments, components and steps denoted by the same reference numerals are the same or corresponding, and repeated descriptions may be omitted. The defect determination apparatus according to the present embodiment clusters feature information obtained using the measurement results of the laser ultrasonic method, and determines the presence or absence of welding defects using the clustering result.

[0010] FIG. 1 is a schematic diagram showing the configuration of a robot control system 100 according to the present embodiment, and FIG. 2A is a block diagram showing the configuration of a defect determination apparatus 1 according to the present embodiment. As shown in FIG. 1, the robot control system 100 includes a defect determination apparatus 1, a manipulator 2, a robot control apparatus 3, and a welding power source 4. Note that, in FIG. 1, the defect determination apparatus 1 is shown as an independent apparatus, but the defect determination apparatus 1 may be incorporated in another apparatus such as the robot control apparatus 3, for example.

[0011] The defect determination apparatus 1 determines the presence or absence of defects in a welded portion using the measurement results of the laser ultrasonic method. As shown in FIG. 2A, the defect determination apparatus 1 includes a reception unit 11, an image conversion unit 12, a feature acquisition unit 13, a storage unit 14, a clustering unit 15, a determination unit 16, and an output unit 17. The defect determination apparatus 1 may be connected to, for example, the laser ultrasonic measurement apparatus 2a of the manipulator 2 or the robot control apparatus 3. Note that each configuration of the defect determination apparatus 1 will be described later.

[0012] The manipulator 2 has a plurality of arms connected by joints driven by motors. A welding torch is attached to the tip of the manipulator 2. When a welding wire is used for welding, the manipulator 2 may be equipped with a wire feeding device for feeding the welding wire. Further, a laser ultrasonic measuring device 2a is attached to the manipulator 2. The attachment position of the laser ultrasonic measuring device 2a on the manipulator 2 is not particularly limited, but it is preferably attached at a position where the welding portion can be sensed from a desired direction by the laser ultrasonic measuring device 2a. Therefore, the laser ultrasonic measuring device 2a may be attached, for example, to the tip side of the manipulator 2. The tip side of the manipulator 2 may be, for example, a position on the base end side of the welding torch as shown in FIG. 1. The manipulator 2 is not particularly limited, but may be, for example, a vertically articulated robot.

[0013] The laser ultrasonic measuring device 2a measures the weld bead by the laser ultrasonic method. The laser ultrasonic measuring device 2a may include, for example, an ultrasonic generating unit 2b, an ultrasonic detecting unit 2c, and the like. In the laser ultrasonic measuring device 2a, ultrasonic waves may be generated in the measurement target by condensing the pulsed laser light emitted from the ultrasonic generating unit 2b onto the measurement target. Then, for example, the ultrasonic detecting unit 2c, which is a laser interferometer, irradiates the measurement target with laser light for ultrasonic detection, and captures the Doppler effect caused by the surface vibration generated when the generated ultrasonic waves reach the irradiation position of the laser light for detection by the laser interferometer to detect the ultrasonic waves. In addition, ultrasonic waves may be detected by a method other than using a laser interferometer, for example, the knife-edge method.

[0014] In this embodiment, as shown in FIG. 4, the case where laser light for generating ultrasonic waves is irradiated at a plurality of locations and the ultrasonic waves generated accordingly are detected by irradiating the laser light at one location will be mainly described. FIG. 4 is a diagram showing the result of fillet welding of the overlap joint, that is, the result of welding the joint portion between the upper plate work piece 5a and the lower plate work piece 5b. In FIG. 4, the x-axis is provided in the longitudinal direction of the bead 6. And at each position where the value of x is B1, B2, B3, etc., measurement using the laser ultrasonic method is being performed. Hereinafter, each position where the value of x is B1, B2, etc. may also be referred to as measurement positions B1, B2, etc. In this embodiment, for example, for one measurement position B1, laser light for generating ultrasonic waves is sequentially irradiated at a plurality of irradiation positions T1, and the ultrasonic waves generated by the laser light are detected by irradiating laser light for ultrasonic wave detection at one irradiation position R1. The measurement by the laser ultrasonic measurement device 2a may be performed, for example, in response to a measurement instruction from the defect determination device 1, or not. In the latter case, the laser ultrasonic measurement device 2a may repeat, for example, irradiating laser light for generating ultrasonic waves at a plurality of irradiation positions at predetermined intervals (for example, for each of the measurement positions B1, B2) along the longitudinal direction of the bead 6 to obtain a plurality of measurement results. As shown in FIG. 4, the laser light may be irradiated, for example, on the bead 6, or may be irradiated on the work pieces 5a, 5b. Also, in this embodiment, the case where the bead 6 to be measured by the laser ultrasonic method corresponds to the bead of one welding pass will be mainly described, but the bead to be measured may be, for example, a set of beads corresponding to a plurality of welding passes in multi-layer multi-pass welding, etc.

[0015] Note that the irradiation position of the laser beam in the laser ultrasonic method is not limited to the above description. For example, the laser beam for generating ultrasonic waves may be irradiated at one location, and the ultrasonic waves generated accordingly may be detected by irradiating the laser beam at a plurality of locations. In this way, the irradiation position of the laser beam for generating ultrasonic waves may be, for example, one location or a plurality of locations, and the irradiation position of the laser beam for detecting ultrasonic waves may be, for example, one location or a plurality of locations. As a result, as long as a plurality of measurement results in a direction perpendicular to the longitudinal direction of the bead 6 can be obtained for each measurement position, the measurement method by the laser ultrasonic method is not limited. Note that the laser ultrasonic measurement device 2a is already known, and a detailed description thereof is omitted.

[0016] The robot control device 3 controls the manipulator 2 and may control the manipulator 2 and the welding power source 4 so that welding is performed along the welding lines of the workpieces 5a and 5b, for example. Note that the configuration of the robot control device 3 is already known, and a detailed description thereof is omitted.

[0017] The welding power source 4 supplies the high voltage used in welding to the welding torch and the workpieces 5a and 5b. Further, when a welding wire is used for welding, the welding power source 4 may perform control related to the feeding of the welding wire. Note that the configuration of the welding power source 4 is already known, and a detailed description thereof is omitted.

[0018] The reception unit 11 of the defect determination device 1 receives the measurement results of the laser ultrasonic method regarding the weld bead. The reception unit 11 may receive, as one measurement result, the ultrasonic waves measured by the laser light for ultrasonic wave detection irradiated at another location when, for example, the laser light for generating ultrasonic waves is irradiated at a certain location. In the present embodiment, this case will be mainly described. The reception unit 11 may receive, in real time, the measurement results regarding the bead formed by welding from the laser ultrasonic measurement device 2a when welding is being performed or after welding is completed, or may receive a plurality of measurement results previously measured by the laser ultrasonic measurement device 2a in a batch. The reception of the batch of measurement results may be, for example, the reading of a plurality of measurement results from a recording medium or the reception of a plurality of measurement results.

[0019] The reception unit 11 may receive measurement results from, for example, the ultrasonic detection unit 2c, may receive measurement results transmitted via a wired or wireless communication line, or may receive measurement results read from a predetermined recording medium (such as an optical disk, magnetic disk, semiconductor memory, etc.). The reception unit 11 may receive information other than measurement results. For example, the reception unit 11 may receive the measurement position of the laser ultrasonic measurement device 2a from the robot control device 3. Also, the measurement results may be received in response to the defect determination device 1 outputting a measurement instruction to the laser ultrasonic measurement device 2a. The measurement instruction may be output, for example, when the measurement position received from the robot control device 3 becomes a predetermined value (such as B1 or B2). In response to the measurement instruction, the laser ultrasonic measurement device 2a may irradiate a plurality of irradiation points at the measurement position with laser light, thereby obtaining a plurality of measurement results corresponding to the plurality of irradiation points and transmitting them to the defect determination device 1. And the plurality of measurement results may be received by the reception unit 11. Note that the reception unit 11 may or may not include a device (such as a modem or network card) for reception. Also, the reception unit 11 may be realized by hardware or may be realized by software such as a driver for driving a predetermined device.

[0020] The image conversion unit 12 converts a plurality of measurement results into an image for each measurement position in the longitudinal direction of the bead. The measurement positions along the longitudinal direction of the bead may be provided, for example, at equal intervals. Also, the image conversion unit 12 may convert, for example, eight measurement results measured at the measurement position B1 shown in FIG. 4 into an image. The converted image may be, for example, an image of a B-scan, an image of a C-scan, or other two-dimensional data. FIG. 5 is a diagram showing an example of an image of a B-scan. The converted B-scan image shown in FIG. 5 has, for example, the horizontal axis indicating the transmission position (i.e., the vertical position shown in FIG. 4) and the vertical axis indicating the arrival time (i.e., the propagation distance of the ultrasonic wave from the irradiation position for ultrasonic wave generation shown in FIG. 4 to the irradiation position for ultrasonic wave detection). This B-scan image corresponds to, for example, an image of a cross-section in a direction orthogonal to the longitudinal direction of the bead at the measurement position. The image conversion unit 12 may convert, for example, a plurality of measurement results at one measurement position into one image. In this case, one image is obtained for one measurement position. Note that the method of converting the measurement results into an image is already known, and a detailed description thereof is omitted.

[0021] The feature acquisition unit 13 acquires feature information including the feature amount of the image for each image. Therefore, one piece of feature information is obtained for one image. The feature information may have, for example, a plurality of feature amounts obtained from one image. The feature amount may be, for example, HOG (Histogram of Oriented Gradients), SIFT (Scale-Invariant Feature Transform), or a feature amount obtained using a neural network. Note that HOG and SIFT are already known, and a detailed description thereof is omitted. The feature acquisition unit 13 may store the acquired feature information in the storage unit 14. When storing the feature information in the storage unit 14, for example, the feature information may be stored in association with the measurement position. The measurement position is the measurement position of the measurement result corresponding to the image used for acquiring the feature information, and may be, for example, the measurement position received by the reception unit 11.

[0022] Here, the neural network used to acquire the feature amount will be described. The input to this neural network is one image converted by the image conversion unit 12. That is, each pixel value of the image may be input to the neural network. This neural network may be, for example, a CNN (Convolutional Neural Network), a neural network composed of fully connected layers, or other neural networks. The CNN used to acquire the feature amount may be, for example, a CNN before learning, a CNN learned using ImageNet, or a CNN learned using the image converted by the image conversion unit 12 cut at a predetermined intermediate layer (for example, the Nth layer). And the output of each node of the Nth layer may be used as the feature amount respectively. In this case, for example, the set of outputs of all the nodes of the Nth layer may be the feature information. The output used as the teacher data in the learning using the image converted by the image conversion unit 12 may be, for example, all without defects, or the determination result (for example, with defects or without defects) determined using radiation or the like. Usually, since there are few defective parts in the welded part, it may be learned as in the former case.

[0023] In the storage unit 14, as described above, the measurement position corresponding to the image and the feature information acquired from the image may be associated and stored. Also, information other than the feature information may be stored in the storage unit 14. Note that the storage unit 14 is preferably realized by a non-volatile recording medium, but may be realized by a volatile recording medium. The recording medium may be, for example, a semiconductor memory, a magnetic disk, an optical disk, or the like.

[0024] The clustering unit 15 clusters a plurality of pieces of feature information acquired by the feature acquisition unit 13. Since one piece of feature information is acquired from one image, there is a one-to-one relationship between the images and the feature information. Therefore, it can also be considered that this clustering clusters a plurality of images. The clustering unit 15 may perform clustering of a plurality of pieces of feature information using, for example, the Ward method, the group average method, the k-means method, etc., or may perform clustering of a plurality of pieces of feature information by other methods. Note that the Ward method, the group average method, and the k-means method are already known, and detailed descriptions thereof are omitted.

[0025] Note that it is necessary to set a threshold value in the Ward method and the group average method. This threshold value is set such that when there are no defects at each measurement position corresponding to a plurality of pieces of feature information that are the objects of clustering, the clustering result becomes one cluster, and when there is a defect at any measurement position, the clustering result becomes two clusters.

[0026] Also, in the k-means method, it is necessary to set the number of clusters for classification destination. However, the number of clusters may be set to two. Then, the clustering unit 15 obtains the distances between two clusters that are the clustering results of a plurality of pieces of feature information using the k-means method. When the distance is greater than a preset threshold value, the clustering result of the k-means method is directly used as the clustering result. When the distance is smaller than the preset threshold value, all the feature information may be clustered into one cluster. In addition, when the distance of one cluster is equal to the threshold value, the clustering unit 15 may cluster the plurality of pieces of feature information into two clusters or may cluster them into one cluster. Thus, when the k-means method is used, the final clustering may be performed using the clustering result of the k-means method. Also in this case, as the threshold value, it is preferably set such that when there is no defect at each measurement position, the final clustering result is one cluster, and when there is a defect at any measurement position, the final clustering result is two clusters.

[0027] Thus, when setting the threshold value used in clustering, it is necessary to prepare a set of multiple pieces of feature information regarding beads without defects and a set of multiple pieces of feature information regarding beads with defects. However, the number of sets of multiple pieces of feature information required to determine the threshold value may be less than the number of pieces of teacher data that need to be prepared in supervised machine learning. Therefore, even if it is not possible to prepare a sufficient amount of teacher data regarding welding defects, it is possible to set the threshold value used in clustering. Also, by setting the threshold value to a larger value, the possibility of being classified into one cluster can be increased, and as a result, the possibility of misjudging a bead without defects as having a defect can be reduced. On the other hand, by setting the threshold value to a smaller value, the possibility of being classified into two or more clusters can be increased, and as a result, the possibility of misjudging a bead with a small defect as not having a defect can be reduced. Therefore, if it is desired to avoid misjudging a bead without defects as having a defect, the threshold value should be set to a larger value, and if it is desired to avoid misjudging and overlooking a defect, the threshold value should be set to a smaller value.

[0028] Also, when making a determination in real time, the clustering unit 15 may start clustering, for example, when a predetermined number of pieces of feature information are acquired. The predetermined number may be, for example, 3 or more. More specifically, the predetermined number may be 3, 5, or 10. By setting the predetermined number to a larger value, more accurate determination can be made. When making a determination in real time, the clustering unit 15 may repeat clustering, for example, after a predetermined number of pieces of feature information are acquired. This repetition of clustering may be, for example, a repetition each time feature information is acquired. In this case, for example, when N pieces of feature information are acquired, clustering of the N pieces of feature information is performed, and when the next new piece of feature information is acquired, clustering of N + 1 pieces of feature information is performed, and this may be repeated. N is an integer of 3 or more. Also, when the time taken for one clustering is longer than the acquisition period of the feature information, for example, when one clustering ends, immediately after the end, the next clustering using all the feature information acquired so far may be repeatedly started.

[0029] When a plurality of pieces of feature information are clustered into two or more clusters by the clustering unit 15, the determination unit 16 determines that there is a defect in the bead, and when the clustering unit 15 clusters the pieces of feature information into one cluster, the determination unit 16 determines that there is no defect in the bead. Usually, an image of a measurement position where there is no defect and the feature information obtained from the image are considered to be an image and feature information having similar features. On the other hand, an image and feature information of a measurement position where there is a defect are considered to be an outlier image and outlier feature information having features different from those of an image and feature information of a measurement position where there is no defect. Therefore, by performing clustering, the feature information having features different from those of other feature information, that is, the feature information of the measurement position where there is a defect, is classified into a cluster different from other feature information, and thus, as described above, it is possible to determine the presence or absence of a defect in the welded portion.

[0030] When there is one defect in the bead, usually, the clustering result is considered to be two clusters. On the other hand, when there are two or more defects in the bead, it is also conceivable that the clustering result will be three or more clusters. Therefore, as described above, the determination unit 16 may determine that there is a defect when clustering is performed into two or more clusters. Usually, since the characteristic information of the measurement position where the defect exists is small, the cluster corresponding to normal is the cluster with the largest number of characteristic information included in that cluster, and the cluster corresponding to the defect is a cluster other than the cluster corresponding to normal.

[0031] Also, when the determination unit 16 determines that there is a defect in the bead, it may specify the defect position, which is the position of the bead corresponding to the characteristic information classified into the cluster corresponding to the defect, using the characteristic information and the measurement position that are associated and stored in the storage unit 14. For example, when a certain characteristic information is classified into the cluster corresponding to the defect, the measurement position corresponding to the measurement result used to acquire the image corresponding to the characteristic information may be specified as the defect position. For example, as shown in FIG. 4, when there is a cavity V in the bead 6, the characteristic information obtained from the image corresponding to the measurement result of the measurement position B2 is classified into the cluster corresponding to the defect. Therefore, in this case, the measurement position B2 corresponding to the measurement result used to acquire the image corresponding to the characteristic information may be specified as the defect position. In this way, the defect position may be the position in the longitudinal direction of the bead 6.

[0032] In addition, when the reception unit 11 receives in real time the measurement results regarding the bead formed by the welding while the welding is being performed or after the welding is completed, the clustering unit 15 and the determination unit 16 may each repeatedly perform the clustering and determination processes, or they may not. In the former case, clustering and determination will be performed in real time. In the latter case, for example, after the real-time reception is completed, the clustering and determination processes may be performed only once.

[0033] The output unit 17 may output the determination result. When the output unit 17 outputs a determination result indicating the presence of a defect, it may output the defect position together with the determination result. When the real-time determination is repeatedly performed, the output unit 17 may, for example, repeatedly output the determination result. This output may be, for example, a display on a display device (such as a liquid crystal display or an organic EL display), a transmission via a communication line to a predetermined device, a print by a printer, an audio output by a speaker, a storage in a recording medium, or a transfer to other components. Note that the output unit 17 may or may not include a device for performing the output (such as a display device or a communication device). Also, the output unit 17 may be realized by hardware or by software such as a driver for driving those devices.

[0034] Next, the operation of the defect determination device 1 will be described with reference to the flowchart of FIG. 3. (Step S101) The reception unit 11 determines whether it has received the measurement results from the laser ultrasonic measurement device 2a. If it has received them, it proceeds to step S102; otherwise, it repeats the process of step S101 until it receives the measurement results. The received measurement results may be temporarily stored in a recording medium (not shown).

[0035] (Step S102) The image conversion unit 12 determines whether the search for one column is completed. If the search for one column is completed, it proceeds to step S103; otherwise, it returns to step S101. Note that the image conversion unit 12 may determine that the search for one column is completed when the measurement results for all irradiation positions have been received for a certain measurement position. For example, when the measurement is performed as shown in FIG. 4, the image conversion unit 12 may determine that the search for one column is completed when eight measurement results have been received for a certain measurement position (for example, measurement position B1, etc.).

[0036] (Step S103) The image conversion unit 12 converts the measurement results for one column received so far into one image. The image may be, for example, the one shown in FIG. 5.

[0037] (Step S104) The feature acquisition unit 13 acquires feature information by obtaining feature amounts from the image converted by the image conversion unit 12. The feature acquisition unit 13 may store the acquired feature information in the storage unit 14.

[0038] (Step S105) The clustering unit 15 determines whether to perform clustering on a plurality of pieces of feature information. If clustering is to be performed, it proceeds to step S106; otherwise, it returns to step S101. Note that the clustering unit 15 may determine to perform clustering, for example, when the number of pieces of feature information stored in the storage unit 14 is equal to or more than a predetermined number.

[0039] (Step S106) The clustering unit 15 clusters a plurality of pieces of feature information.

[0040] (Step S107) The determination unit 16 determines whether there is a defect in the bead based on the result of clustering. More specifically, when a plurality of pieces of feature information are clustered into two or more clusters, it may be determined that there is a defect in the bead, and when this is not the case, it may be determined that there is no defect. Also, when there is a defect, for example, the position of the defect may be specified.

[0041] (Step S108) The output unit 17 outputs the determination result by the determination unit 16. Note that the position of the defect may be output together with the determination result.

[0042] (Step S109) The clustering unit 15 determines whether to end the clustering process. Then, when the clustering process is to be ended, the series of processes for determining the welding defect ends, and when this is not the case, the process returns to Step S101. Note that when the measurement using the laser ultrasonic method is completed up to the end of the bead, it may be determined to end the clustering process.

[0043] In the flowchart of FIG. 3, when it is determined that a defect exists, a series of processes such as acceptance of measurement results and image conversion may be terminated. Further, when the laser ultrasonic measurement device 2a performs measurement in response to a measurement instruction, the flowchart of FIG. 3 may include a process of outputting the measurement instruction to the laser ultrasonic measurement device 2a. This measurement instruction may be output, for example, by an instruction unit (not shown) of the defect determination device 1. In this case, the reception unit 11 may receive the measurement result in response to the output of the measurement instruction. Further, in the flowchart of FIG. 3, the output of the determination result may be performed only once when a series of processes are terminated. Further, the flowchart of FIG. 3 shows the case where determination is performed in real time during welding, but this is not necessary. When determination is not performed in real time, for example, the processes of steps S103 and S104 are performed for each measurement position, and after the processes are completed, the processes of S106 to S108 may be performed only once for a plurality of pieces of feature information. Further, the order of the processes in the flowchart of FIG. 3 is an example, and the order of each step may be changed as long as the same result can be obtained.

[0044] Next, the operation of the defect determination device 1 according to the present embodiment will be described using a specific example. In this specific example, the case where the defect determination device 1 determines the presence or absence of a defect using the measurement results obtained by the laser ultrasonic measurement device 2a while performing welding by the manipulator 2 will be described. In this specific example, it is assumed that clustering processing is performed when the number of pieces of feature information is three or more.

[0045] First, the reception unit 11 sequentially receives the measurement results for each of the irradiation positions T1 corresponding to the measurement position B1 (steps S101, S102). Then, when the number of received measurement results reaches 8, which corresponds to one row of exploration, the image conversion unit 12 converts the 8 measurement results into an image of a B-scan (step S103). Also, the feature acquisition unit 13 acquires feature information by obtaining feature amounts from the B-scan image and stores the feature information in the storage unit 14 (step S104). At this point, since there is only one piece of feature information, it is determined that clustering is not performed (step S105).

[0046] Next, the reception of each measurement result at the measurement position B2, the image conversion of those measurement results, and the acquisition of feature information are performed (steps S101 to S104). Also in this case, since there are only two pieces of feature information, it is determined that clustering is not performed (step S105).

[0047] Thereafter, the reception of each measurement result at the measurement position B3, the image conversion of those measurement results, and the acquisition of feature information are performed (steps S101 to S104). At this point, since three pieces of feature information are stored in the storage unit 14, the clustering unit 15 determines to perform clustering and clusters the three pieces of feature information (steps S105, S106). Also in this case, as shown in FIG. 4, since there is a cavity V at the measurement position B2, it is assumed that the two pieces of feature information corresponding to the measurement positions B1 and B3 and the one piece of feature information corresponding to the measurement position B2 are clustered into different clusters. Then, the determination unit 16 determines that there is a defect in the bead 6 because clustering into two clusters has been performed (step S107). And the output unit 17 outputs the determination result (step S109). For example, welding may be interrupted according to the output. In this specific example, the case where the defect is a cavity has been described, but the defect may be other defects such as bead under cracking other than the cavity.

[0048] As described above, according to the defect determination device 1 of the present embodiment, even if it is not possible to prepare a sufficient amount of teacher data for welding defects, it is possible to determine the presence or absence of welding defects by performing clustering, which is unsupervised learning. Further, by specifying the defect position, which is the position of the bead corresponding to the feature information classified into the cluster corresponding to the defect, the position of the defect can also be known. Further, by performing acquisition of feature information, clustering, etc. according to the measurement results received in real time during welding, it is possible to know the presence or absence of defects during welding. For example, when a defect exists, it is also possible to interrupt the welding so as not to perform unnecessary work thereafter.

[0049] Note that, in the defect determination device 1 according to the present embodiment, the dimension of the feature information to be clustered may be reduced. In this case, for example, as shown in FIG. 2B, the defect determination device 1 may further include a dimension reduction unit 18. Then, the dimension reduction unit 18 may reduce the dimension of the feature information acquired by the feature acquisition unit 13 before clustering is performed. Further, the clustering unit 15 may cluster a plurality of pieces of feature information whose dimensions have been reduced by the dimension reduction unit 18. Usually, the larger the amount of information of the feature information, the greater the processing load of clustering and the longer the time required for clustering. Therefore, by reducing the dimension of the feature information by the dimension reduction unit 18, the processing load and processing time of clustering can be reduced.

[0050] The dimensionality reduction unit 18 may have the number of dimensions after reduction set. As this number of dimensions, a number smaller than the number of dimensions of the feature information acquired by the feature acquisition unit 13 is preferably set. Note that the dimensionality reduction unit 18 may reduce the dimension, for example, by principal component analysis (PCA: Principal Component Analysis), by t-SNE (T-distributed Stochastic Neighbor Embedding), by UMAP (Uniform Manifold Approximation and Projection), or by other methods. Note that principal component analysis, t-SNE, and UMAP are already publicly known, and detailed descriptions thereof are omitted.

[0051] Also, when welding and measurement are performed in parallel, in FIG. 4, for example, while the laser beam is sequentially irradiated to a plurality of irradiation positions at a certain measurement position, the irradiation position of the laser beam may be slightly shifted in the x-axis direction. In this case, for example, the plurality of irradiation positions T1 of the laser beam may be arranged in a direction not perpendicular to the x-axis, and the irradiation position R1 may also be a plurality of points arranged in the x-axis direction. Thus, a plurality of measurement results in a direction not perpendicular to the longitudinal direction of the bead 6 may be acquired for each measurement position and converted into an image. In this case, the measurement position may be, for example, a measurement position having a predetermined width (for example, a measurement position where the value of the x-axis is from B1 S to B1 E up to the measurement position, etc.).

[0052] In addition, in the present embodiment, the case where processing is performed on the measurement results obtained from one bead has been described, but this is not necessary. For example, for joints of the same shape, similar welding may be repeated. In this case, since different beads have the same shape, for example, clustering may be performed on the feature information obtained from the first bead and the feature information obtained from the second bead. By doing so, for example, even when the first piece of feature information is obtained for the second bead, by clustering the single piece of feature information and the plurality of pieces of feature information obtained from the first bead, it is also possible to make a determination regarding the first piece of feature information of the second bead. Thus, the bead to be measured by the laser ultrasonic method or the bead to be determined for the presence or absence of a defect may be, for example, one bead, or two or more non-adjacent beads.

[0053] Further, the determination unit 16 may make a prediction according to the degree of variation of the plurality of pieces of feature information included in the cluster corresponding normally. For example, when the variation is larger than a threshold value, it may be predicted that the chip used for welding is worn out. The degree of variation of the plurality of pieces of feature information may be indicated, for example, by a representative value of the distances between the respective pieces of feature information. The representative value may be, for example, an average value, a median value, a maximum value, a minimum value, or the like.

[0054] In addition, in the present embodiment, the case where the measurement results are converted into an image and the feature information obtained from the image is clustered has been described, but this is not necessary. The measurement results themselves may be clustered. In this case, a plurality of measurement results for each measurement position in the longitudinal direction of the bead may be regarded as one set of measurement results. Then, the clustering unit 15 may cluster the plurality of sets of measurement results. Except that the object of clustering changes from feature information to a set of measurement results, it is the same as the above description. Therefore, for example, the defect position, which is the position of the bead corresponding to the cluster set classified into the cluster corresponding to the defect, may be specified. In this case, the defect determination device 1 may not include the image conversion unit 12 or the feature acquisition unit 13.

[0055] Further, in the defect determination device 1, the feature information acquired from the measurement results may be clustered. In this case, a plurality of measurement results for each measurement position in the longitudinal direction of the bead may be regarded as one set of measurement results. Then, the feature acquisition unit 13 may acquire feature information including the feature amounts of the plurality of measurement results for each set of measurement results. That is, one piece of feature information may be acquired from one set of measurement results. The acquisition of this feature information may also be performed in the same manner as the acquisition of feature information from an image. For example, the feature acquisition unit 13 may acquire feature information by inputting each data included in the set of measurement results into a neural network. This neural network may have, for example, a plurality of fully connected layers, or may be other neural networks. Further, this neural network may be, for example, a neural network after learning cut at a predetermined intermediate layer (for example, the Nth layer). And the output of each node of the Nth layer may be used as a feature amount respectively. In this case, except that the feature information is acquired from the set of measurement results instead of the image, it is the same as the above description. Thus, when the feature information is directly acquired from the set of measurement results, the defect determination device 1 may not include the image conversion unit 12.

[0056] Also, in the present embodiment, the case where welding is mainly described when welding is performed by the manipulator 2 having a welding torch has been mainly described, but the manipulator 2 is for moving the laser ultrasonic measurement device 2a and may not perform welding. In this case, the manipulator 2 may not include a welding torch, and the robot control system 100 may not include the welding power source 4.

[0057] In the above-described embodiment, each component may be configured by dedicated hardware, or for components that can be realized by software, they may be realized by executing a program. For example, by a program execution unit such as a CPU reading and executing a software program recorded on a recording medium such as a hard disk or a semiconductor memory, each component can be realized. At the time of its execution, the program execution unit may execute the program while accessing a storage unit or a recording medium. Further, the program may be executed by being downloaded from a server or the like, or may be executed by reading a program recorded on a predetermined recording medium. Also, the computer that executes the program may be singular or plural. That is, centralized processing may be performed, or distributed processing may be performed.

[0058] Further, the present invention is not limited to the above-described embodiments, and various modifications are possible, and it goes without saying that those are also included within the scope of the present invention.

Explanation of Reference Numerals

[0059] 1 Defect determination device, 11 Reception unit, 12 Image conversion unit, 13 Feature acquisition unit, 14 Storage unit, 15 Clustering unit, 16 Determination unit, 17 Output unit, 18 Dimensionality reduction unit

Claims

1. A reception unit that receives measurement results of the laser ultrasonic method regarding the weld bead, An image conversion unit that converts a plurality of measurement results for each measurement position in the longitudinal direction of the bead into an image, A feature acquisition unit that acquires feature information including feature amounts of the image for each image, A clustering unit that clusters a plurality of pieces of feature information, A defect determination device comprising: a determination unit that determines that a defect exists in the bead when the clustering unit clusters into two or more clusters.

2. Further comprising a storage unit, The defect determination device according to claim 1, wherein the feature acquisition unit accumulates in the storage unit by associating a measurement position corresponding to the image and the feature information acquired from the image.

3. When the determination unit determines that a defect exists in the bead, the defect determination device according to claim 2, wherein the defect position, which is the measurement position corresponding to the feature information classified into the cluster corresponding to the defect, is specified using the feature information and the measurement position that are associated and stored in the storage unit.

4. The reception unit receives in real time measurement results regarding the bead formed by the welding when the welding is being performed, The clustering unit starts clustering when a predetermined number of pieces of feature information are acquired, The defect determination device according to any one of claims 1 to 3, wherein the clustering unit and the determination unit each repeatedly perform clustering and determination.

Images