JP6792880B2 - Laser welding condition judgment device - Google Patents

Description

The present invention relates to a laser welding state determination device suitable for determining a welding state at a laser-welded laser irradiation site of an object.

Conventionally, a laser welding state determining device for determining a welding state of a laser-welded object is known (for example, Patent Documents 1 and 2). The laser welding state determination device described in Patent Document 1 includes a camera that images a laser irradiation portion of an object irradiated with a laser beam, and a determination unit that determines the welding state based on the image captured by the camera. There is. In this device, the determination unit extracts a circular and high-intensity light emitting ring, which is the outer circumference of the keyhole formed in the laser irradiation portion, from the image captured by the camera, and connects the central portion and the peripheral portion of the light emitting ring. Detect the brightness distribution. Then, the degree of penetration at the laser irradiation site is determined based on the comparison result of the brightness distributions.

Further, the laser welding state determination device described in Patent Document 2 includes a camera that captures an image of a laser irradiation portion of an object irradiated with a laser beam, and a determination unit that determines a welding state based on an image captured by the camera. I have. In this device, the determination unit extracts the keyhole from the image captured by the camera, detects the presence or absence of black color in the keyhole, and detects the shape of the black color. Then, when the black area in the keyhole is out of the predetermined range or the length of the molten portion is larger than the predetermined length, it is determined that the welding at the laser irradiation portion is defective.

Japanese Unexamined Patent Publication No. 2010-115680 Japanese Unexamined Patent Publication No. 2003-019584

However, as described above, it is difficult to accurately determine the degree of penetration at the laser irradiation site only by comparing the brightness distribution between the central portion and the peripheral portion of the light emitting ring in the image captured by the camera. Further, as described above, it is difficult to accurately judge the welding quality at the laser irradiation site only by the black shape in the keyhole in the camera image.

The present invention has been made in view of such a point, and an object of the present invention is to provide a laser welding state determination device capable of accurately determining a welding state at a laser irradiation site.

The present invention includes an imaging unit that captures an image of a laser-irradiated portion of an object to be laser-welded and irradiated with a laser beam, an image processing unit that processes an image acquired by imaging by the imaging unit by a predetermined method, and the imaging unit. By comparing the image acquired by imaging in the unit with the learning data showing the relationship between the image data captured by the laser irradiation site obtained by machine learning and the predetermined welding state, the matching with the predetermined welding state is achieved. A first determination unit for determining the degree and a second determination unit for determining the degree of agreement with the predetermined welding state by comparing the processed image acquired by the processing in the image processing unit with the learning data. A third determination unit that determines the degree of agreement with the predetermined welding state based on the numerical parameters acquired by the processing in the image processing unit, the first determination unit, and the second determination unit. A laser welding state determination device including a state determination unit that determines a welding state at the laser irradiation site based on each determination result by the third determination unit.

It is a figure which represented the welding pattern schematically. It is a block diagram of the laser welding system of embodiment. It is a block diagram of the laser welding system of an embodiment. It is a block diagram of the laser welding apparatus included in the laser welding system. This is an example of an image of the welded portion to be welded. It is an internal block diagram of the computer which the laser welding state determination apparatus included in a laser welding system has. It is a detailed block diagram of the main processing part of the computer which the laser welding state determination apparatus has. It is a figure which showed the determination procedure by the laser welding state determination apparatus. It is a flowchart of an example of a routine executed in a laser welding system.

Hereinafter, specific embodiments of the laser welding state determination device according to the present invention will be described with reference to the drawings.

The laser welding state determination device 20 of one embodiment is a device incorporated in the laser welding system 1. The laser welding system 1 is a system in which, for example, a portion where metal plates 2 are butted or overlapped with each other is welded by irradiation with a laser beam. As shown in FIG. 1A, this welding is performed by penetrating the plate material 2 in parallel along the abutted surface in a state where the side surfaces of the two plate materials 2 are abutted against each other (that is, butt penetration). Welding), welding performed non-penetrating the plate material 2 as shown in FIG. 1 (B) (that is, butt non-penetrating welding), and a state in which two plate materials 2 are overlapped in the stacking direction as shown in FIG. 1 (C). Welding performed through the plate material 2 perpendicular to the overlapped surface (that is, overlapped penetration welding), and welding performed non-penetratingly through the plate material 2 as shown in FIG. 1 (D) (that is, Includes any one or more of (lapped non-penetrating welds).

As shown in FIG. 2, the laser welding system 1 includes a laser welding device 10. The laser welding apparatus 10 joins the plate materials 2 to each other by irradiating the plate material 2 to be welded in a condensed state while generating a laser beam which is a laser beam as a heat source and locally melting and solidifying the plate material 2. It is a device to do. As shown in FIGS. 2 and 3, the laser welding apparatus 10 includes a laser oscillator 11, a transmission line 12, a condensing optical unit 13, a driving unit 14, a shield gas supply unit 15, and a controller 16. ing.

The laser oscillator 11 is a device that oscillates a laser beam (for example, a laser beam having a first wavelength of 1070 nm). The laser oscillator 11 is configured by using, for example, a CO2 laser or a YAG laser capable of increasing the output. The laser light generated by the laser oscillator 11 is guided to the condensing optical unit 13 through the transmission line 12. In the CO2 laser, the transmission on the transmission line 12 is performed by turning back by the mirror. Further, in the YAG laser, transmission on the transmission line 12 may be performed by transmission by a freely bendable optical fiber in addition to folding back by a mirror.

As shown in FIG. 4, the condensing optical unit 13 includes a laser head 13a, a collimation lens 13b, a dichroic mirror 13c, and a condensing lens 13d. The laser head 13a has a cylindrical housing. The collimation lens 13b, the dichroic mirror 13c, and the condenser lens 13d are arranged in this order on the optical path in the laser head 13a.

The laser beam output from the laser oscillator 11 travels while spreading radially. The collimation lens 13b is a lens that converts a laser beam that travels while spreading radially into a laser beam that travels in parallel. The dichroic mirror 13c is a mirror that transmits the laser light of the first wavelength emitted from the laser oscillator 11 and reflects visible light and the laser light of the second wavelength emitted from the light projecting unit for a camera described later. is there. The condensing lens 13d is a lens that condenses the laser light transmitted through the dichroic mirror 13c to the welding processing point of the plate material 2. In this way, the laser welding apparatus 10 welds the two plate materials 2 by condensing the laser beam of the first wavelength generated by the laser oscillator 11 to an appropriate size and irradiating the plate material 2.

A pedestal 17 is provided below the condensing optical unit 13. The plate material 2 to be welded is placed on the pedestal 17 and fixed to the pedestal 17 using a jig. The condensing optical unit 13 is movable with respect to the pedestal and thus the plate material 2. The drive unit 14 is a motor that moves the condensing optical unit 13 in the horizontal direction and the vertical direction with respect to the pedestal 17. Welding to the plate material 2 by the condensing optical unit 13 proceeds along the surface of the plate material 2 by the driving unit 14 moving the condensing optical unit 13 in the horizontal direction with respect to the pedestal 17. The drive unit 14 may move the pedestal 17 instead of or with the movement of the condensing optical unit 13.

The shield gas supply unit 15 is a portion for blowing the shield gas to the welding processing point of the plate material 2. The spraying of the shield gas is performed in order to prevent oxidation of the welding processing point of the plate material 2 due to the irradiation of the laser beam. The shield gas sprayed by the shield gas supply unit 15 is argon, helium, nitrogen, or the like.

The controller 16 is electrically connected to the drive unit 14 including the condensing optical unit 13. The controller 16 is a control unit (PLC) that controls the output of the laser beam by the condensing optical unit 13 and controls the movement of the condensing optical unit 13 by the driving unit 14.

As shown in FIG. 3, the laser welding system 1 includes a laser welding state determination device 20. The laser welding state determination device 20 is a device that determines the welding state of the plate material 2 laser-welded by the laser welding device 10. In laser welding, violent evaporation occurs at the welded part, and the heat from the laser beam and the evaporation reaction force of the violent evaporation form a molten pool where molten metal accumulates at the processing point, and at the same time, in a deep hole near the center of the molten pool. Welding proceeds with a certain keyhole formed, and deep penetration welding is possible. This keyhole tends to become deeper as the output of the laser beam increases, and becomes shallower as the welding speed increases in the traveling direction.

The welding state determined by the laser welding state determination device 20 is a state of a keyhole or a molten pool. For example, the presence or absence of penetration of the keyhole and the surface of the base metal without a part of the keyhole being filled with the weld metal. The presence or absence of undercut in which a groove is formed between the welded metal and the surface of the weld metal, the presence or absence of underfill in which the welded surface is dented, unwelded, unpowered laser, etc., are selectively judged from at least two of them. Will be done. The laser welding state determination device 20 includes a camera 21, a computer (PC) 22, and a data server 23.

The camera 21 is an imaging device that images the welded state of the laser-welded plate material 2. As shown in FIG. 4, the camera 21 takes an image coaxially with the laser beam emitted by the condensing optical unit 13. Light from the welded portion of the plate material 2 is focused, reflected, enlarged, and incident on the camera 21. The camera 21 generates an image by photoelectrically converting the incident light (see FIG. 5). It is preferable that the camera 21 generates an image by photoelectrically converting wavelength components of near-infrared light other than visible light together with visible light. The image generated by the camera 21 expresses the brightness of each region obtained by dividing a two-dimensional plane image vertically and horizontally into a predetermined number of pixels with a predetermined number (for example, 256) of gradations.

As shown in FIG. 3, a computer 22 is connected to the camera 21. The image captured by the camera 21 is supplied to the computer 22. The computer 22 is a computer for deep learning having a CPU, a storage, an image processing board, an I / O board, and the like. The storage stores image data. The image processing board processes the captured image of the camera 21. The I / O board processes the input / output of the computer 22. The computer 22 has a driver layer as a software system. Examples of this driver layer include a storage driver, a camera driver, and an I / O board.

The computer 22 has an application layer as a software system. An example of this application layer is the main processing unit 30. The main processing unit 30 includes a parameter setting unit 31, an image processing unit 32, an image processing real-time determination unit 33, an AI processing unit 34, and a comprehensive determination unit 35.

The parameter setting unit 31 is a part for setting parameters used for image processing in the image processing unit 32. The image processing unit 32 is a portion that processes the captured image of the camera 21 read by the camera driver by a predetermined method described later. The image processing real-time determination unit 33 is a portion that determines the welding state of the plate material 2 in real time based on the result of image processing by the image processing unit 32.

The AI processing unit 34 is a portion for determining the welded state of the plate material 2. The AI processing unit 34 has an AI parameter setting unit 34a, an AI learning unit 34b, and an AI real-time determination unit 34c. The AI parameter setting unit 34a is a part for setting parameters used for machine learning for determining the welding state. This AI parameter includes the image size and number of convolution processes, the selection of the convolution filter, the filter selection of the activation process, the filter selection of the thinning process, the size and number of thinning processes, the size and the number of layers of the intermediate layer of the perceptron coupling process. , Dropout ratio, optimizer type selection, convolution processing, activation processing, thinning processing, perceptron coupling processing coefficient, offset initial value, learning rate initial value, and epoch number setting in learning. The AI learning unit 34b is a portion that learns the relationship between the image and the welded state based on the learning image data stored in the data server 23. The AI real-time determination unit 34c determines the welding state in real time based on the result learned by the AI learning unit 34b, the captured image of the camera 21 read by the camera driver, and the image processing result by the image processing unit 32. It is a part.

The comprehensive determination unit 35 is a portion that comprehensively determines the welding state of the plate material 2 based on the determination result by the image processing real-time determination unit 33 and the determination result by the AI real-time determination unit 34c. The comprehensive determination unit 35 can also control the image display of the camera 21 on the monitor of the computer 22, the display of various states including the welding state, and the state display of machine learning for determining the welding state. At the same time, it is possible to output the laser beam by the condensing optical unit 13 and command the position of the laser head 13a by the driving unit 14 to the controller 16.

In the data server 23, the image data showing the image captured by the camera 21 and the welding state are stored as learning data in relation to each other. The learning data stored in the data server 23 is the captured image itself (original image) of the camera 21, the image acquired by image processing on the original image (processed image), and the numerical value acquired by image processing on the original image. Includes parameters. That is, the data of the original image of the camera 21, the data of the processed image, and the numerical parameters are stored in the data server 23 in relation to the welding state.

As shown in FIGS. 7 and 8, the image processing performed on the captured image of the camera 21 includes binarization processing, blob analysis processing, exposure colorization processing, three-dimensionalization processing, rotation processing, and the like. The binarization process is a process of converting an image (original image) captured by the camera 21 into two gradations of white and black at a set threshold value, for example, when the image is represented by shades of 256 gradations. .. The binarization process is executed in a plurality of types corresponding to a plurality of threshold values. The blob analysis process is a method of analyzing a binarized image, and includes the area, circumference length, center of gravity position, and roundness of the blob (specifically, a keyhole or a molten pool) included in the image. It is a process to calculate the flatness, the angle, the position of the quadrangle circumscribing or inscribed in the keyhole or the molten pool, and the length of each side in the vertical and horizontal directions. For image processing, for example, a binarized portion having a threshold value of 128 or more is set in the keyhole, and a binarized portion having a threshold value of 64 or more is set in the molten pool.

In the exposure colorization process, the brightness value (for example, 256 steps) of each pixel in the image captured by the camera 21 is set to a predetermined step (specifically, the number of steps smaller than the total number of gradations of the brightness value, for example, 16 steps). It is a process of determining which of the (arbitrary values) is included, assigning a color to that stage, and displaying the brightness in color. The three-dimensional processing is a process of three-dimensionalizing an image captured by the camera 21 based on the luminance distribution and the luminance gradient distribution.

When the above image processing is executed, the area of the keyhole and the molten pool, the circumference, the position of the center of gravity, the roundness, the flatness, and the angle, the quadrangle circumscribing or inscribed in the keyhole, and the quadrangle inscribed or inscribed in the molten pool are circumscribed or inscribed. The position of the quadrangle and the length of each side in the vertical and horizontal directions, the brightness gradient distribution around the keyhole and the molten pool, the area ratio of each exposed color, and the brightness distribution of the keyhole center of gravity and the center of the molten pool and their brightness distribution. Time changes and the like are calculated as numerical parameters.

The AI learning unit 34b of the AI processing unit 34 reads the image data (including the data of the original image and the data of the processed image) stored in the data server 23, and performs machine learning. Initially, the AI learning unit 34b performs machine learning based on the initial parameters set in the AI parameter setting unit 34a, and thereafter, performs machine learning based on the parameters calculated by the learning. In this machine learning, the coefficient for weighting the brightness value of each pixel of the image, the threshold value, the offset value, etc. are changed to determine the weighting coefficient, etc., which gives the highest probability of finally applying the welded state. Is. Machine learning includes deep learning (DL) in which the weighting layer is multi-layered. DL has a function (optimizer and backpropagation) to check whether the input value is correct from the output value, and automatically optimizes parameters such as weighting coefficient in each layer by gradient descent method etc. Is possible. The DL layer may include a layer that weights locally (for example, only one pixel), or may include a layer that weights globally (for example, all pixels).

The AI learning unit 34b has a convolution layer, an activation layer, a pooling layer, a finalizer, and an optimizer as functional blocks. The convolution layer is a layer that performs image convolution processing, which is a filter function. The activation layer is a look-up table (LUT) that transforms the data output from the convolution layer into a form suitable for processing. The pooling layer is a layer for thinning out the data output from the activation layer. The finalizer is a connecting layer that binds the data output from the thinning layer. The optimizer is a layer that checks whether the input value is correct from the output value.

The AI learning unit 34b learns by executing the processing in each of the above layers based on the relationship between the original image of the camera 21 which is a data set of various welding states stored in the data server 23 and the welding state. As a result, the optimum first parameter is set. Further, the AI learning unit 34b has a relationship between the processed image (specifically, the image after the exposure colorization process) of the camera 21, which is a data set of various welding states stored in the data server 23, and the welding state. By executing the processing in each of the above layers based on the above, the optimum second parameter is set as the learning result. Further, the AI learning unit 34b has a relationship between the processed image (specifically, the image after the three-dimensional processing) of the camera 21, which is a data set of various welding states stored in the data server 23, and the welding state. By executing the processing in each of the above layers based on the above, the optimum third parameter is set as the learning result.

Each learning result of the AI learning unit 34b is supplied to the AI real-time determination unit 34c. The AI real-time determination unit 34c compares the captured image sent from the camera 21 or the processed image from the image processing unit 32 with each learning result of the AI learning unit 31b to check the degree of agreement with various welding states. Then, the welding state of the plate material 2 is determined in real time based on the image captured by the camera 21.

Specifically, first, as shown in FIG. 8, the AI real-time determination unit 34c compares the input original image of the camera 21 with the first parameter from the AI learning unit 34b to match the degree of agreement with various welding states. To judge. Further, the degree of agreement with various welding states is determined by comparing the processed image after the exposure colorization processing based on the input original image of the camera 21 with the second parameter from the AI learning unit 31b. Further, the degree of agreement with various welding states is determined by comparing the processed image after the three-dimensional processing based on the input original image of the camera 21 with the third parameter from the AI learning unit 31b.

Further, the image processing real-time determination unit 33 determines the degree of agreement with various welding states based on the numerical parameters including the processing result of the blob analysis obtained by image processing the input original image of the camera 21. Then, the welding state of the plate material 2 is determined in real time based on the image captured by the camera 21. The parameters used in the image processing real-time determination unit 33 may be set and changed by an operator's input operation to the computer 22.

The learning result (AI state) of the AI learning unit 34b, the determination result of the AI real-time determination unit 34c, and the determination result of the image processing real-time determination unit 33 are supplied to the comprehensive determination unit 35. Then, the comprehensive determination unit 35 comprehensively determines the welding state of the plate material 2 in real time based on the matching degree determination result in the AI real-time determination unit 34c and the matching degree determination result in the image processing real-time determination unit 33. Specifically, by weighting or evaluating the threshold value of each matching degree determination result, the matching degree with the final various welding states is determined, and the welding state having the highest matching degree is determined.

The coefficient used for weighting the matching degree determination result and the numerical value used for the threshold value evaluation in the comprehensive determination unit 35 can be changed and set by an input operation of the operator to the computer 22. That is, the comprehensive determination unit 35 can set the weighting coefficient used for weighting each matching degree determination result and the threshold value used for the threshold evaluation according to the input operation of the operator to the computer 22. The comprehensive determination unit 35 determines the degree of agreement with various welding states according to the above-mentioned weighting coefficient or threshold value set in the main processing unit 30, and determines the welding state having the highest degree of agreement.

The weighting coefficient of the matching degree determination result in the AI real-time determination unit 34c can be appropriately changed in the range of 0 to 100%. Further, the weighting coefficient of the matching degree determination result in the image processing real-time determination unit 33 can be appropriately changed in the range of 0 to 100%. For example, the comprehensive determination unit 35 sets the weighting coefficient of the matching degree determination result in the AI real-time determination unit 34c to 100%, and sets the weighting coefficient of the matching degree determination result in the image processing real-time determination unit 33 to 0%. The welding state may be determined. On the contrary, the weighting coefficient of the matching degree determination result in the AI real-time determination unit 34c is set to 0%, the matching degree determination result in the image processing real-time determination unit 33 is set to 100%, and the welding state is determined. May be good. Further, the comprehensive determination unit 35 may automatically change the above weighting coefficient or the like according to the proficiency level (number of learning data) in the AI learning unit 34b. Further, when it is determined that an error occurs in the weighting coefficient or the like set according to the input operation of the operator, the comprehensive determination unit 35 preferentially uses the standard set by the AI processing unit 34 side over the standard set by the person. It may be that. For example, when comparing the images of the camera 21 in chronological order, there are clearly images with undercuts and underfills, but in order to determine the state of these images as welding NG, the reference set by the AI processing unit 34 side. May be used with priority to determine the welding state.

Further, the comprehensive determination unit 35 can display the state in the middle of calculation by various machine learning in which the AI state is reflected in the original image of the camera 21 on the monitor of the computer 22, and the AI real-time determination unit 34c. It is possible to display the determination result of the above and the determination result of the image processing real-time determination unit 33 together with the original image of the camera 21 and the like. Further, the specific above-mentioned first to third parameter values and various parameter values in the AI state may be displayed.

Next, the operation of the laser welding system 1 will be described.
In the laser welding system 1, when the system power is turned on (step S100 shown in FIG. 9), first, the image data stored in the data server 23 is acquired, and the main processing unit 30 of the computer 22 performs the system teaching process (step S100 shown in FIG. 9). Step S101). Specifically, the system parameter setting process and the above machine learning for the image data stored in the data server 23 are executed. Then, it is determined whether or not the system parameter setting process is completed and the machine learning is completed (step S102). This determination is continued until an affirmative determination is made. When the machine learning is completed, the learning result in the AI learning unit 34b is supplied to the AI real-time determination unit 34c.

When various monitoring in the main processing unit 30 is started (step S103) and welding in the laser welding apparatus 10 is started after the above system teaching process is completed (step S104), first, the main processing unit 30 The AI real-time determination unit 34c and the image processing real-time determination unit 33 each determine the welding state of the plate material 2 in real time based on the image captured by the camera 21, and the comprehensive determination unit 35 comprehensively determines the welding state of the plate material 2. Is determined in real time (step S105). Then, the comprehensive determination unit 35 determines whether or not the welded state of the determined plate material 2 is good. (Step S106).

For example, when the butt penetration welding shown in FIG. 1 (A) is required and the determined welding state is through welding, it is determined that the welding state is good, while the determined welding. When the state is non-penetrating welding, it is determined that the welding state is not good. Further, when the overlap non-penetrating welding shown in FIG. 1D is required and the determined welding state is through welding, it is determined that the welding state is not good, while it is determined. When the welded state is non-penetrating welded, it is determined that the welded state is good.

When the comprehensive determination unit 35 determines that the welding condition is not good, the power output of the laser light by the condensing optical unit 13 and the position of the laser head 13a by the drive unit 14 with respect to the controller 16 of the laser welding device 10 ( That is, a command is given to change at least one of the laser beam irradiation positions on the plate material 2 so that a good welding state can be obtained (step S107). When such a command is given, the controller 16 changes the output of the laser beam by the condensing optical unit 13 and the position of the laser head 13a with respect to the pedestal 17 by the driving unit 14 so that a good welding state can be obtained.

On the other hand, when the comprehensive determination unit 35 determines that the welding state is good, it determines whether or not all the welding of the plate material 2 has been completed (step S108). Then, when it is determined that all the welding has not been completed, the determination of the welding state with respect to the welding of the plate material 2 is continued. On the other hand, when it is determined that all welding is completed, various monitoring in the main processing unit 30 is completed (step S109), and the system power is turned off (step S110).

In this way, in the laser welding system 1, the welding state of the plate material 2 can be determined from the image captured by the camera 21. Specifically, the degree of agreement with various welding states is determined using the first to third parameters obtained by machine learning based on the learning data in the data server 23, and the blob analysis obtained by the binarization process is performed. The degree of agreement with various welding states can be determined based on the above, and the welding state of the plate material 2 can be determined in real time by integrating the determination results. Then, when the welding state is not good, the output of the laser beam or the irradiation position can be feedback-controlled so that the welding state is good.

Further, in performing the real-time determination of the welding state of the plate material 2, the degree of coincidence between the original image of the camera 21 and the various welding states by comparison with the first parameter is determined, and the processed image after the exposure colorization processing and the second For determining the degree of agreement with various welding states by comparing with parameters, determining the degree of agreement with various welding states by comparing the processed image after three-dimensional processing with the third parameter, and blob analysis by binarization processing. The degree of agreement with various welding states based on the above is determined, and the degree of agreement with the final various welding states is determined by weighting or evaluating the threshold value. Then, the welded state having the highest degree of coincidence is determined to be the welded state.

The coefficient used for weighting each matching degree determination result and the numerical value used for threshold evaluation can be changed and set by the input operation of the operator to the computer 22, and the proficiency level (number of learning data) in the AI learning unit 34b and the like. It can be changed according to. Therefore, the welded state of the welded portion of the plate material 2 can be accurately determined by changing the weighting coefficient and the numerical value. Then, when the determined welding state is not the desired state, the output of the laser beam and the irradiation position are corrected so that the subsequent welding state becomes the desired state. Therefore, the welding quality can be improved.

Further, when the plate material 2 is welded, the monitor of the computer 22 shows the state in the middle of calculation by various machine learning in which the AI state is reflected in the original image of the camera 21, the judgment result of the AI real-time judgment unit 34c, and the image processing real-time judgment unit. As a result of the determination of 33, the specific above-mentioned first to third parameter values and various parameter values in the AI state can be displayed. Therefore, since the operator can perform the above-mentioned change setting operation such as the weighting coefficient while looking at the monitor, it is possible to ensure the ease of operation by the operator for improving the welding quality.

By the way, in the above embodiment, it is assumed that the captured image, the parameter, and the welding state determination result of the camera 21 handled by the AI processing unit 34 and the comprehensive determination unit 35 are obtained by one imaging or calculation. However, the present invention is not limited to this. As the image and parameters of the camera 21 handled by the AI processing unit 34 and the comprehensive determination unit 35, and the determination result of the welding state, the average value of the acquired values per unit time may be used. For example, the AI processing unit 34 may determine the degree of agreement with various welding states by using a value obtained by averaging a predetermined number n of data of a time series obtained in the past. Further, the comprehensive determination unit 35 stores the matching degree determination results with various welding states each time in chronological order, and determines the welding state by averaging the matching degree determination results of a predetermined number n. May be good.

Further, in the above embodiment, the data server 23 that stores the learning data is provided in the laser welding state determination device that determines the welding state. However, the present invention is not limited to this. The database for storing the learning data may be an external device connected to the laser welding state determination device 20 via a network, and the learning data may be acquired by the laser welding state determination device 20 via the network.

The present invention is not limited to the above-described embodiments and modifications, and various modifications can be made without departing from the spirit of the present invention.

1: Laser welding system, 10: Laser welding device, 20: Laser welding state determination device, 21: Camera, 22: Computer, 23: Data server, 30: Main processing unit, 34: AI processing unit, 34b: AI learning unit , 34c: AI real-time determination unit, 32: image processing unit, 33: image processing real-time determination unit, 35: comprehensive determination unit.

Claims (12)

An image pickup unit that images the laser irradiation site irradiated with the laser beam in the object to be laser welded,
An image processing unit that processes an image acquired by imaging with the imaging unit by a predetermined method, and an image processing unit.
By comparing the image acquired by the imaging with the imaging unit with the learning data showing the relationship between the image data captured by the laser irradiation site obtained by machine learning and the predetermined welding state, the predetermined welding state is obtained. The first determination unit that determines the degree of coincidence of
A second determination unit that determines the degree of agreement with the predetermined welding state by comparing the processed image acquired by the processing in the image processing unit with the learning data.
A third determination unit that determines the degree of agreement with the predetermined welding state based on the numerical parameters acquired by the processing in the image processing unit.
A state determination unit that determines the welding state at the laser irradiation site based on the determination results of the first determination unit, the second determination unit, and the third determination unit.
A laser welding state determination device. The state determination unit determines the welding state by weighting the determination result by the first determination unit, the determination result by the second determination unit, and the determination result by the third determination unit. The laser welding state determination device according to claim 1. The state determination unit determines the welding state based on the result of comparing each determination result by the first determination unit, the second determination unit, and the third determination unit with the threshold value, respectively. Item 1. The laser welding state determination device according to item 1. The laser according to any one of claims 1 to 3, wherein each determination result by the first determination unit, the second determination unit, and the third determination unit is an average value per unit time. Welding condition judgment device. Used in the image processing unit or the first determination unit, an image acquired by imaging by the imaging unit, the post-processed image used in the second determination unit, or the third determination unit. The laser welding state determination device according to any one of claims 1 to 4, wherein the numerical parameter is an average value per unit time. The laser welding state determination device according to any one of claims 1 to 5, wherein the learning data is acquired via a network. The laser welding state determination device according to any one of claims 1 to 6, wherein the predetermined method includes at least one of a binarization process, an exposure colorization process, and a three-dimensional process. The numerical parameters include the area of the keyhole and the molten pool formed in the laser irradiation site, the peripheral length, the position of the center of gravity, the roundness, the flatness, and the angle, the quadrangle circumscribing or inscribed in the keyhole, and the molten pool. Includes at least one of the circumscribed or inscribed quadrilateral position and side length, the brightness gradient distribution around the keyhole and the molten pool, and the brightness distribution of the keyhole center of gravity and the molten pool center of gravity. The laser welding state determination device according to any one of claims 1 to 7. The invention according to any one of claims 1 to 8, further comprising a display control unit that reflects the learning data in an image acquired by imaging by the imaging unit and displays a state during calculation by the machine learning on a monitor. Laser welding condition judgment device. The laser welding state determination device according to any one of claims 1 to 9, further comprising a laser output changing unit that changes the output power of the laser beam according to the welding state determined by the state determination unit. The processed image used in the second determination unit is an image obtained by the exposure colorization process.
The laser welding state determination device according to any one of claims 1 to 10, wherein the numerical parameter used in the third determination unit is the result of blob processing. The welded state that can be determined by the state determining unit is any one of claims 1 to 11, which includes at least two of penetration, non-penetration, undercut, underfill, unwelded, and laser unpowered. The laser welding state determination device described.