Laser welding control system, laser welding device and machine learning device

Abstract

To provide a laser welding control system that can highly accurately feedback-control a welding condition of laser welding in real time, a laser welding device, and a machine learning device.SOLUTION: This laser welding control system comprises: a band-pass filter 13 through which a beam of a wavelength λ1 for measuring the temperature of a workpiece 2 among beams emitted following radiant heat from a prescribed region 2a including the application position of laser beams applied to the workpiece 2 and set at least in the workpiece 2; a camera 11 that images the beam passed through the band-pass filter 13; temperature distribution image acquisition means 15 for acquiring a temperature distribution image in the prescribed region 2a on the basis of the image acquired through the imaging by the camera 11; and learning means 16 for learning determination of a state of laser welding and / or a welding condition by updating a relationship for determining the state of the laser welding and / or the welding condition on the basis of the temperature distribution image.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a laser welding control system, a laser welding device, and a machine learning device. [Background technology]

[0002] Conventionally, there are laser welding devices that irradiate laser light to weld objects. In recent years, technology has been developed to control laser welding using artificial intelligence (AI) in order to maintain good welding quality.

[0003] For example, a laser welding control system disclosed in Patent Document 1 captures an image of a processing site irradiated with a laser beam during laser welding, compares the captured image with a teaching image for each welding state obtained by machine learning, calculates the degree of agreement between the welding state of the processing site and the welding state of the teaching image, and then feedback-controls the output of the irradiated laser beam based on the calculated degree of agreement and a target degree of agreement based on a target welding state. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 6810973 Summary of the Invention [Problem to be solved by the invention]

[0005] However, in the conventional laser welding control system, machine learning is performed using captured images of the processed area irradiated with a laser beam. These captured images contain many noise components, such as plasma emission generated by laser beam irradiation and scattered light from the irradiated laser beam. Therefore, conventional laser welding control systems have a problem in that it is difficult to perform highly accurate feedback control.

[0006] The present invention has been made to solve the above problems, and aims to provide a laser welding control system, a laser welding device, and a machine learning device that can feedback control welding conditions for laser welding in real time with high accuracy. [Means for solving the problem]

[0007] In order to achieve this object, the laser welding control system described in claim 1 controls laser welding of a welding object by a laser welding device, and emits light due to radiant heat from at least a predetermined area set on the welding object including an irradiation position of the laser light irradiated on the welding object. a separating means for separating the laser beam from the optical path of the laser beam; an optical filter means for passing light of a wavelength for measuring the temperature of the work-piece to be welded; an imaging means for capturing an image of light emitted in association with radiant heat from at least the predetermined region that has passed through the optical filter means; a temperature distribution image acquisition means for acquiring a temperature distribution image in the predetermined region based on an image captured by the imaging means; a learning means for learning to determine the state and / or welding conditions of the laser welding by updating a relationship for determining the state and / or welding conditions of the laser welding based on the temperature distribution image acquired by the temperature distribution image acquisition means; and a control means for controlling laser welding in the laser welding device based on the state and / or welding conditions determined by the learning means.

[0008] The laser welding control system according to claim 2 comprises: a temperature distribution image acquisition means for acquiring a temperature distribution image of the predetermined area based on an image acquired by the imaging means; a learning means for learning to determine a state and / or welding conditions of the laser welding by updating a relationship for determining a state and / or welding conditions of the laser welding based on the temperature distribution image acquired by the temperature distribution image acquisition means; and a control means for controlling laser welding in the laser welding apparatus based on the state and / or welding conditions determined by the learning means.The optical filter means comprises first optical filter means that passes light of a first wavelength, which is used to measure the temperature of the welding object, from among the light emitted due to radiant heat from at least the predetermined region, and second optical filter means that passes light of a second wavelength, which is used to measure the temperature of the welding object, from among the light emitted due to radiant heat from at least the predetermined region; the imaging means comprises first imaging means that images light that is passed through the first optical filter means, from among the light emitted due to radiant heat from at least the predetermined region, and second imaging means that images light that is passed through the second optical filter means, from among the light emitted due to radiant heat from at least the predetermined region; and the temperature distribution image acquisition means acquires a temperature distribution image in the predetermined region based on an image obtained by imaging by the first imaging means and an image obtained by imaging by the second imaging means.

[0009] The laser welding control system according to claim 3 comprises: a temperature distribution image acquisition means for acquiring a temperature distribution image of the predetermined area based on an image acquired by the imaging means; a learning means for learning to determine the state and / or welding conditions of the laser welding by updating a relationship for determining the state and / or welding conditions of the laser welding based on the temperature distribution image acquired by the temperature distribution image acquisition means; and a control means for controlling the laser welding in the laser welding device based on the state and / or welding conditions determined by the learning means, wherein the welding conditions include at least an output balance of the laser beams output from the light sources. A fourth aspect of the present invention provides a laser welding control system according to the third aspect, wherein the welding conditions include at least a pattern of a temporal change in the output balance.

[0010] Claim 5 The laser welding control system according to the present invention comprises: 4 In the laser welding control system described in any one of the above, the learning means learns to determine the state of the laser welding by updating at least a relationship for determining the state of the laser welding based on the temperature distribution image acquired by the temperature distribution image acquisition means, and the control means performs feedback control to adjust the welding conditions of the laser welding in the laser welding device based on the state of the laser welding determined by the learning means so that the state of the laser welding becomes a target state.

[0011] Claim 6 The laser welding control system according to the present invention comprises: 5In the laser welding control system described in any one of the above, the welding conditions include at least one of the output intensity of the laser light output from a light source, the intensity and / or diameter of the laser light irradiated onto the workpiece, the scanning speed of the laser light, the scanning pattern of the laser light, and the flow rate of the assist gas supplied to the periphery of the position irradiated with the laser light.

[0014] Claim 7 The laser welding control system according to the present invention comprises: 6 In the laser welding control system described in any one of the above, the learning means performs learning by deep learning.

[0015] Claim 8 The laser welding control system according to the present invention comprises: 7 In the laser welding control system described in any one of the above, the learning means performs learning by ensemble learning using a plurality of learning models.

[0016] Claim 9 The laser welding device according to the present invention comprises: 8 The laser welding of the welding object is controlled by the laser welding apparatus according to any one of the above. [Effects of the Invention]

[0018] According to the laser welding control system of claim 1, the laser welding of the welding object by the laser welding device is controlled, and light emission due to radiant heat from at least a predetermined area set on the welding object including the irradiation position of the laser light irradiated on the welding object is detected. The light emitted by the separating means is separated from the optical path of the laser light.Among the light beams, light having a wavelength for measuring the temperature of the workpiece passes through the optical filter means. Then, of the light emitted due to radiant heat from at least the predetermined region, the light that has passed through the optical filter means is imaged by the imaging means. Based on the image obtained by the imaging means, a temperature distribution image of the predetermined region is acquired by the temperature distribution image acquisition means. Based on the temperature distribution image acquired by the temperature distribution image acquisition means, the learning means learns to determine the laser welding state and / or welding conditions by updating the relationship for determining the laser welding state and / or welding conditions. Then, based on the laser welding state and / or welding conditions determined by the learning means, the control means controls the laser welding in the laser welding device. In this way, the data used to input the learning means is a temperature distribution image obtained based on the light emitted due to radiation from the predetermined region of the workpiece, imaged in real time. By using the temperature distribution image, seed This allows the learning means to perform learning and the determination of the laser welding state and / or welding conditions without being affected by various noises, thereby providing the effect of enabling feedback control of the laser welding conditions in real time with high accuracy.

[0019] According to the laser welding control system of claim 2, The present invention controls laser welding of workpieces using a laser welding device. Light emitted from at least a predetermined region on the workpiece, including the irradiation position of the laser beam irradiated thereon, having a wavelength for measuring the temperature of the workpiece, passes through an optical filter. Then, the light emitted from at least the predetermined region, accompanied by radiant heat, that has passed through the optical filter is captured by an imaging device. A temperature distribution image of the predetermined region is acquired by a temperature distribution image acquisition device based on the image captured by the imaging device. A learning device learns to determine the laser welding state and / or welding conditions by updating a relationship for determining the laser welding state and / or welding conditions based on the temperature distribution image acquired by the temperature distribution image acquisition device. Then, laser welding in the laser welding device is controlled by a control device based on the laser welding state and / or welding conditions determined by the learning device. Thus, data used as input to the learning device is a temperature distribution image obtained based on light emitted from the predetermined region of the workpiece, captured in real time, accompanied by radiation. Using the temperature distribution image, learning by the learning device and determination of the laser welding state and / or welding conditions can be performed without being affected by various noises. Therefore, there is an effect that the welding conditions of the laser welding can be feedback-controlled in real time with high accuracy.Of the light emitted due to radiant heat from at least a predetermined region set on the work-piece, including the irradiation position of the laser light irradiated on the work-piece, light of a first wavelength for measuring the temperature of the work-piece passes through the first optical filter means. Then, of the light emitted due to radiant heat from at least the predetermined region that has passed through the first optical filter means, an image is taken by the first imaging means. Furthermore, of the light emitted due to radiant heat from at least the predetermined region, light of a second wavelength for measuring the temperature of the work-piece passes through the second optical filter means. Then, of the light emitted due to radiant heat from at least the predetermined region that has passed through the second optical filter means, an image is taken by the second imaging means. A temperature distribution image of the predetermined region is acquired by the temperature distribution image acquisition means based on the image acquired by the first imaging means and the image acquired by the second imaging means. In this way, a temperature distribution image of the predetermined region is acquired using light of two wavelengths emitted due to radiant heat from the predetermined region on the work-piece, thereby improving the reliability of the temperature within the predetermined region shown in the temperature distribution image. Therefore, there is an effect that the welding conditions for laser welding can be feedback-controlled with higher accuracy.

[0020] According to the laser welding control system of claim 3, The present invention controls laser welding of workpieces using a laser welding device. Light emitted from at least a predetermined region on the workpiece, including the irradiation position of the laser beam irradiated thereon, having a wavelength for measuring the temperature of the workpiece, passes through an optical filter. Then, the light emitted from at least the predetermined region, accompanied by radiant heat, that has passed through the optical filter is captured by an imaging device. A temperature distribution image of the predetermined region is acquired by a temperature distribution image acquisition device based on the image captured by the imaging device. A learning device learns to determine the laser welding state and / or welding conditions by updating a relationship for determining the laser welding state and / or welding conditions based on the temperature distribution image acquired by the temperature distribution image acquisition device. Then, laser welding in the laser welding device is controlled by a control device based on the laser welding state and / or welding conditions determined by the learning device. Thus, data used as input to the learning device is a temperature distribution image obtained based on light emitted from the predetermined region of the workpiece, captured in real time, accompanied by radiation. Using the temperature distribution image, learning by the learning device and determination of the laser welding state and / or welding conditions can be performed without being affected by various noises. This has the effect of enabling feedback control of the welding conditions for laser welding in real time with high accuracy. Furthermore, a laser welding device controlled by the laser welding control system performs laser welding by irradiating a workpiece with a laser beam formed by combining laser beams output from multiple light sources, and the welding conditions include at least the output balance of the laser beams output from each light source. This allows laser welding to be performed while feedback-controlling the output balance of each laser beam so that the output balance of the laser beams output from the multiple light sources achieves a target laser welding quality based on a determination made by the learning means using a temperature distribution image. This has the effect of enabling laser welding of more stable quality. The laser welding control system of claim 4 achieves the following effect in addition to the effect achieved by the laser welding control system of claim 3. That is, the welding conditions include at least a pattern of temporal change in the output balance of each laser beam output from a plurality of light sources. As a result, laser welding is performed while feedback control of the temporal change in the output balance of each laser beam is performed based on the judgment by the learning means using the temperature distribution image so as to achieve the target laser welding quality. Therefore, there is an effect that laser welding of more stable quality can be achieved.

[0021] Claim 5 According to the laser welding control system described above, claims 1 to 4In addition to the effects of the laser welding control system described in any one of the above, the present invention provides the following effect. That is, the learning means learns to determine the state of laser welding by updating at least the relationship for determining the state of laser welding based on the temperature distribution image acquired by the temperature distribution image acquisition means. Then, based on the state of laser welding determined by the learning means, the control means performs feedback control to adjust the welding conditions of laser welding in the laser welding device so that the state of laser welding becomes a target state. This has the effect of realizing laser welding with high accuracy in real time so that the state of laser welding becomes a target state.

[0022] Claim 6 According to the laser welding control system described above, claims 1 to 5 In addition to the effects of the laser welding control system described in any one of the above, the present invention provides the following effect. That is, the welding conditions include at least one of the output intensity of the laser beam output from the light source, the intensity and / or diameter of the laser beam irradiated onto the workpiece, the scanning speed of the laser beam, the scanning pattern of the laser beam, and the flow rate of the assist gas supplied to the periphery of the position irradiated with the laser beam. Therefore, since the parameters included in the welding conditions can be feedback-controlled in real time with high precision, there is an effect that laser welding of more stable quality can be realized.

[0025] Claim 7 According to the laser welding control system described above, claims 1 to 6 In addition to the effects of the laser welding control system described in any one of the above, the present invention has the following effect: Since the learning means performs learning by deep learning, the accuracy of the judgment by the learning means can be improved. As a result, there is an effect that the welding conditions for laser welding can be feedback-controlled with higher accuracy.

[0026] Claim 8 According to the laser welding control system described above, claims 1 to 7In addition to the effects of the laser welding control system described in any one of the above, the present invention provides the following effect. That is, since the learning means performs learning by ensemble learning using a plurality of learning models, the accuracy of the judgment by the learning means can be improved. As a result, there is an effect that the welding conditions for laser welding can be feedback-controlled with higher accuracy.

[0027] Claim 9 According to the laser welding device described in claim 1 to claim 2, 8 Since the laser welding of the welding object is controlled by the laser welding control system described in any one of the above, it is possible to enjoy the effects of the corresponding laser welding control system. [Brief explanation of the drawings]

[0029] [Figure 1] 1A is a schematic diagram showing the general configuration of a machine learning device according to one embodiment of the present invention, a laser welding control system equipped with the machine learning device and controlling a laser welding device, and a laser welding device controlled by the laser welding control system; and FIG. 1B is a diagram showing the intensity distribution of laser light irradiated onto a workpiece in the laser welding device. [Figure 2] FIG. 1 is a graph showing the wavelength and temperature dependence of blackbody radiation intensity. [Figure 3] FIG. 2 is a schematic diagram illustrating an example of the configuration of a learning means of the machine learning device. [Figure 4] (a) is a diagram showing a schematic example of learning data used for learning by the learning means of the machine learning device, and (b) is a diagram showing a schematic example of output from the learning means when a temperature distribution image of a specified area acquired by a temperature distribution image acquisition means from image data captured in real time by an imaging means of the laser welding control system is input to the learning means. [Figure 5] FIG. 10 is a schematic diagram illustrating a modified example of the laser welding control system of the present invention. [Figure 6]1(a) is a diagram showing an example of the intensity distribution of laser light formed by combining laser light output from two light sources as a modified example of the laser welding device of the present invention, and FIG. 1(b) is a diagram showing the intensity A and diameter B of the main laser light in the laser light shown in FIG. 1(a), the radius C indicating the position where the secondary laser light is formed relative to the main laser light, and the intensity D and diameter E of the secondary laser light. DETAILED DESCRIPTION OF THE INVENTION

[0030] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. Each of the embodiments described below illustrates a preferred specific example of the present invention. Therefore, the numerical values, shapes, materials, components, component placement and connection configurations, etc., shown in the following embodiments are merely examples and are not intended to limit the present invention. Therefore, among the components in the following embodiments, components that are not recited in the independent claims that represent the highest concept of the present invention will be described as optional components. Furthermore, in each drawing, substantially identical components are assigned the same reference numerals, and redundant explanations will be omitted or simplified.

[0031] Fig. 1(a) is a schematic diagram showing the schematic configuration of a machine learning device 18 according to one embodiment of the present invention, a laser welding control system 10 that includes the machine learning device 18 and controls a laser welding device 1, and the laser welding device 1 controlled by the laser welding control system 10. Fig. 1(b) is a diagram showing the intensity distribution of laser light 3 irradiated onto an object to be welded (hereinafter simply referred to as "work") 2 in the laser welding device 1.

[0032] The laser welding device 1 scans and irradiates the workpieces 2, which are, for example, two plate materials fixed to a base 50 in an abutted or overlapped state, with a laser beam 3, thereby welding the abutted or overlapped portions of the workpieces 2. As shown in FIG. 1(a), the laser welding device 1 has at least a light source 20, a transmission path 30, a welding nozzle 40, and a laser welding control system 10.

[0033] The light source 20 is a laser oscillator that oscillates and outputs the laser beam 3, and may be a CO2 laser oscillator, a YAG laser oscillator, a semiconductor laser oscillator, a disk laser oscillator, a fiber laser oscillator, or the like, depending on the type of laser used for welding. In this embodiment, there is one light source 20, and one laser beam 3 is irradiated onto the workpiece 2 via the welding nozzle 40.

[0034] The transmission path 30 transmits the laser light 3 output from the light source 20 to the welding nozzle 40, and is composed of mirrors and lenses when the laser output from the light source 20 is a CO2 laser. Furthermore, when the laser output from the light source 20 is a YAG laser, semiconductor laser, disk laser, or fiber laser, optical fibers may be used in addition to or instead of the mirrors and lenses.

[0035] Welding nozzle 40 is a component that irradiates and scans laser light 3 toward workpiece 2 fixed to base 50. Laser light 3 output from light source 20 and transmitted via transmission path 30 is input to welding nozzle 40. Optical components for shaping laser light 3, such as collimator lens 41, dichroic mirror 42, galvanometer scanner 43, and condenser lens 44, are arranged inside welding nozzle 40. Laser light 3 shaped by these optical components is irradiated from welding nozzle 40 toward workpiece 2.

[0036] The laser light 3 input from the transmission line 30 to the welding nozzle 40 first passes through the collimator lens 41. The collimator lens 41 is a lens for converting the laser light 3 input to the welding nozzle 40 into parallel light.

[0037] The laser light 3, which has been collimated by the collimator lens 41, is then incident on the dichroic mirror 42. The dichroic mirror 42 is a mirror that reflects light of a specific wavelength and transmits other light. In this embodiment, a dichroic mirror 42 that transmits at least the laser light 3 is used. Furthermore, as will be described in detail later, this dichroic mirror 42 reflects light of wavelength λ1, which is used to measure the temperature of the workpiece 2, out of the light emitted from the workpiece 2 in association with radiant heat.

[0038] In this embodiment, an example is shown in which the dichroic mirror 42 reflects light of wavelength λ1 used to measure the temperature of the workpiece 2, out of the light emitted from the workpiece 2 in association with radiant heat, and transmits the laser light 3. However, the role of the dichroic mirror 42 is to separate, out of the light emitted from the workpiece 2 in association with radiant heat, the light of wavelength λ1 used to measure the temperature of the workpiece 2, from the optical path of the laser light 3. Therefore, the dichroic mirror 42 may reflect the laser light 3 to direct the laser light 3 toward the workpiece 2, while transmitting the light of wavelength λ1 used to measure the temperature of the workpiece 2, out of the light emitted from the workpiece 2 in association with radiant heat, thereby separating the light from the optical path of the laser light 3.

[0039] The laser light 3 transmitted through the dichroic mirror 42 is incident on the galvanometer scanner 43. The galvanometer scanner 43 adjusts the irradiation position when the laser light 3 is irradiated toward the workpiece 2, and is composed of at least two reflection mirrors (galvanometer mirrors) not shown.

[0040] One of the two reflecting mirrors is connected to an X-axis displacement motor (not shown). By driving this X-axis displacement motor, the reflection angle of the connected reflecting mirror is changed, and the irradiation position of the laser beam 3 on the workpiece 2 can be displaced in the X-axis direction set on the plane of the workpiece 2.

[0041] The other of the two reflecting mirrors is connected to a Y-axis displacement motor (not shown). By driving this Y-axis displacement motor, the reflection angle of the connected reflecting mirror is changed, and the irradiation position of the laser beam 3 on the workpiece 2 can be displaced in the Y-axis direction, which is perpendicular to the X-axis set on the plane of the workpiece 2.

[0042] Therefore, by driving the X-axis displacement motor and the Y-axis displacement motor, the galvanometer scanner 43 can irradiate the laser light 3 at a desired position on the plane of the workpiece 2. For example, by using the galvanometer scanner 43, it is possible to linearly scan the irradiation position of the laser light 3 on the workpiece 2. Furthermore, by using the galvanometer scanner 43, it is also possible to irradiate the laser light 3 while rotating (revolving) around a certain point.

[0043] It is also possible to omit the galvanometer scanner 43. In this case, the scanning of the laser beam 3 over the workpiece 2 can also be performed by moving the welding nozzle 40 relative to the pedestal 50, rather than using the galvanometer scanner 43. For example, the scanning of the laser beam 3 can be performed by moving the welding nozzle 40 relative to the fixed pedestal 50, or the scanning of the laser beam 3 can be performed by moving the pedestal 50 relative to the fixed welding nozzle 40.

[0044] Furthermore, by moving the positions of base 50 and welding nozzle 40 relatively in a straight line while rotating (revolving) laser beam 3 around a certain point using galvano scanner 43, it is possible to perform scanning in a straight line while rotating the irradiation position of laser beam 3. Furthermore, by moving the positions of base 50 and welding nozzle 40 relatively in a straight line in the X-axis direction, for example, while moving the irradiation position of laser beam 3 in the Y-axis direction using galvano scanner 43, it is also possible to perform scanning while moving the irradiation position of laser beam 3 in a zigzag pattern.

[0045] The laser light 3 that has passed through the galvano scanner 43 is input to the condenser lens 44. The condenser lens 44 is a lens for condensing parallel light at a desired position, and the laser light 3 that has passed through the galvano scanner 43 is condensed by this condenser lens 44 at the irradiation position of the workpiece 2.

[0046] The focusing lens 44 is configured to be able to change the aperture and focal length in cooperation with various optical components. The laser welding apparatus 1 controls the aperture and focal length of the focusing lens 44 in real time according to the laser welding state of the workpiece 2 based on learning by a machine learning device 18 (described later), thereby changing the intensity distribution of the laser beam 3 irradiated onto the workpiece 2, specifically, the intensity A and diameter B of the laser beam 3 shown in FIG. 1(b).

[0047] In addition to irradiating the workpiece 2 with the laser light 3, the welding nozzle 40 is configured to spray an inert gas such as nitrogen gas as an assist gas in order to reduce the oxygen concentration near the irradiation position of the laser light 3 in order to prevent the workpiece 2 from oxidizing during laser welding.

[0048] Specifically, welding nozzle 40 is provided with gas supply path 45 that connects the side surface of welding nozzle 40 with the surface of welding nozzle 40 that faces workpiece 2 (hereinafter referred to as the "bottom surface"). Gas cylinder 60, which stores inert gas, is connected via gas supply valve 61 to the opening of gas supply path 45 provided on the side surface of welding nozzle 40.

[0049] The gas supply valve 61 is a valve that opens and closes the passage of the inert gas supplied from the gas cylinder 60 using an electrical signal, and adjusts the discharge amount (flow rate) of the inert gas by adjusting the degree of opening.

[0050] When the gas supply valve 61 is opened, the inert gas is supplied from the gas cylinder 60 to the gas supply path 45 at a discharge rate corresponding to the degree of opening, and the inert gas is discharged from the opening of the gas supply path 45 provided on the bottom surface of the welding nozzle 40. Then, the discharged inert gas is sprayed onto the workpiece 2 near the position irradiated with the laser light 3.

[0051] In this embodiment, the inert gas (assist gas) is sprayed onto the workpiece 2 from the welding nozzle 40, but it is also possible to provide an outlet connected to the gas cylinder 60 via a gas supply valve 61 in a location separate from the welding nozzle 40, and spray the inert gas from the outlet near the irradiation position of the laser light 3 on the workpiece 2.

[0052] The laser welding control system 10 is a system for controlling laser welding of a workpiece 2 by a laser welding device 1. The laser welding control system 10 is composed of at least a reflecting mirror 14, a bandpass filter 13, a camera 11, and a computer 12.

[0053] In this embodiment, the laser welding control system 10 is described as one of the components of the laser welding apparatus 1, but the laser welding control system 10 may be provided separately from the laser welding apparatus 1. Furthermore, the laser welding control system 10 may be configured as a single device, or may be configured as a combination of multiple separately provided parts and / or devices.

[0054] In addition, in this embodiment, the laser welding control system 10 is described as being provided adjacent to the laser welding apparatus 1 and controlling the laser welding apparatus 1 by sending and receiving signals via wiring or the like, but this is not necessarily limited to this. For example, the laser welding control system 10 may remotely control the laser welding apparatus 1 by sending and receiving signals to and from the laser welding apparatus 1 via a network.

[0055] Laser welding control system 10 may also be configured such that some of its constituent parts / devices are provided in laser welding apparatus 1, and the remaining parts / devices are provided separately from laser welding apparatus 1. For example, reflection mirror 14, bandpass filter 13, and camera 11 may be provided in laser welding apparatus 1, and computer 12 may be provided separately from laser welding apparatus 1. In this case, signals may be transmitted and received between computer 12 and laser welding apparatus 1 via a network.

[0056] Reflection mirror 14 is a mirror for reflecting light emitted due to radiant heat from workpiece 2 output from welding nozzle 40 toward camera 11. When laser light 3 is irradiated onto workpiece 2, the temperature of workpiece 2 rises around the irradiation position of laser light 3, causing light emission due to radiant heat. The light emitted due to radiant heat from at least a predetermined area 2a set on workpiece 2 including the irradiation position of laser light 3 irradiated onto workpiece 2 is incident on welding nozzle 40 and then incident on dichroic mirror 42 via condenser lens 44 and galvanometer scanner 43.

[0057] As described above, dichroic mirror 42 transmits laser light 3 while reflecting light of wavelength λ1 used to measure the temperature of workpiece 2, out of the light emitted from workpiece 2 due to radiant heat. The light reflected by dichroic mirror 42 is output from the inside of welding nozzle 40 to the outside. If dichroic mirror 42 reflects laser light 3 and transmits light of wavelength λ1 used to measure the temperature of workpiece 2, out of the light emitted from workpiece 2 due to radiant heat, the light transmitted by dichroic mirror 42 is output from the inside of welding nozzle 40 to the outside.

[0058] Reflecting mirror 14 reflects the light emitted from the inside of welding nozzle 40 to the outside toward camera 11.

[0059] The bandpass filter 13 corresponds to the optical filter means of the present invention and is a filter that passes only light of a predetermined wavelength. This bandpass filter 13 is provided between the reflecting mirror 14 and the camera 11. Furthermore, the predetermined wavelength of the bandpass filter 13 is set to a wavelength λ1 of light that is used to measure the temperature of the workpiece 2 in the predetermined region 2a from light emitted from the workpiece 2 in association with radiant heat in a temperature distribution image acquisition means 15 described below, which is realized by the computer 12.

[0060] That is, this bandpass filter 13 allows only the light of wavelength λ1 out of the light emitted in response to radiant heat in at least an area including the predetermined area 2a set on the workpiece 2 and reflected by the dichroic mirror 42 and the reflecting mirror 14 to pass through and enter the camera 11. Here, as described above, the dichroic mirror 42 reflects (or transmits) light of wavelength λ1, but in reality, due to the characteristics of the dichroic mirror 42, light of wavelengths other than wavelength λ1 is also included. The bandpass filter 13 is provided for the purpose of allowing only light of wavelength λ1 to enter the camera 11.

[0061] The camera 11 corresponds to the imaging means of the present invention, and captures the light of wavelength λ1 that has passed through this bandpass filter 13. The center of an imaging element (not shown), such as a CCD image sensor or a CMOS image sensor, that constitutes the camera 11 is arranged coaxially with the laser light 3. As a result, light emitted in association with radiant heat from the irradiation position of the laser light 3 on the workpiece 2 is always incident on the center of the angle of view captured by the camera 11. The camera 11 is configured to always capture the same area centered on the irradiation position of the laser light 3 on the workpiece 2. The imaged area includes the predetermined area 2a.

[0062] It is not necessary that the light emitted due to radiant heat from the irradiation position of the laser light 3 on the workpiece 2 always be incident on the center of the area imaged by the camera 11; it is sufficient that the specified area 2a is included in the area imaged by the camera 11 and the irradiation position of the laser light 3 is fixed without moving from any position in the area imaged by the camera 11.

[0063] However, by positioning the irradiation position of the laser beam 3 at the center of the area imaged by the camera 11, it becomes easier to fix the irradiation position of the laser beam 3 to a position within the area imaged by the camera 11. In particular, when driving the galvanometer scanner 43 to change the irradiation position of the laser beam 3 on the workpiece 2, by arranging the center of the imaging element coaxially with the laser beam 3, it becomes extremely easy to fix the irradiation position of the laser beam 3 to a position within the area imaged by the camera 11.

[0064] The image obtained by imaging with this camera 11 is obtained by converting the intensity of light of wavelength λ1 incident on each pixel of the imaging element that constitutes the camera 11 into an electrical signal for each pixel through photoelectric conversion.

[0065] While laser welding is being performed by the laser welding device 1, the camera 11 captures images at predetermined time intervals (for example, 1 second) and outputs the captured image data to the computer 12. In other words, the camera 11 captures images of the state of the workpiece 2 during laser welding in real time.

[0066] The computer 12 is mainly composed of a processing device and a storage device, and realizes various functions by the processing device executing processes based on programs stored in the storage device. In the laser welding control system 10, the computer 12 realizes, as its various functions, a function as a machine learning device 18 and a function as a control means 17.

[0067] The machine learning device 18 learns the state and / or welding conditions of the laser welding performed by the laser welding device 1, and includes at least the temperature distribution image acquiring means 15 and the learning means 16.

[0068] Each time an image is captured by the camera 11, the temperature distribution image acquisition means 15 acquires a temperature distribution image in the predetermined region 2a set on the workpiece 2 based on the image obtained by the image capture. As described above, the image data captured by the camera 11 is obtained by converting the intensity of light of wavelength λ1 that has passed through the bandpass filter 13, among the light emitted in association with radiant heat, in an area including at least the predetermined region 2a set on the workpiece 2, into an electrical signal for each pixel.

[0069] The temperature distribution image acquisition means 15 utilizes the thermal radiation emitted by a black body, i.e., the temperature characteristics of black body radiation. Here, a method for acquiring a temperature distribution image by the temperature distribution image acquisition means 15 will be described with reference to Fig. 2. Fig. 2 is a diagram showing the wavelength and temperature dependence of black body radiation intensity.

[0070] In Figure 2, the vertical axis represents the intensity of blackbody radiation and the horizontal axis represents the wavelength of light radiated from a blackbody, and the blackbody radiation intensity of light of each wavelength is shown for light emitted from a blackbody at blackbody temperatures of 1000K, 2000K, 2500K, and 3000K. As shown in Figure 2, for example, the blackbody radiation intensity of light with wavelength λ1 is uniquely determined depending on the temperature of the blackbody.

[0071] It is known that the relationship between light wavelength λ, temperature T, and intensity I in blackbody radiation can be expressed by Planck's radiation formula shown in the following equation 1. Here, ε is emissivity, h is Planck's constant, c is the speed of light, and k is Boltzmann's constant.

[0072]

number

[0073]

number

[0074] The temperature T can be derived from the intensity of light of wavelength λ1 based on equation 2, but it may also be derived using a lookup table or the like that shows the temperature T relative to the value of intensity I.

[0075] Returning to Figure 1, the learning means 16 learns to determine the state and / or welding conditions of laser welding by updating a relationship (e.g., a calculation model or function) for determining the state and / or welding conditions of laser welding based on the temperature distribution image of the specified region 2a acquired by the temperature distribution image acquisition means 15.

[0076] Here, the configuration of the learning means 16 will be described with reference to Fig. 3. Fig. 3 is a schematic diagram showing an example of the configuration of the learning means 16. A 100 x 100 pixel temperature distribution image acquired by the temperature distribution image acquisition means 15 is input to the learning means 16 shown in Fig. 3. Based on the input temperature distribution image, the learning means 16 outputs the output intensity of the laser beam 3 output from the light source 20 (light source laser beam output intensity), the intensity A (irradiated laser beam intensity A, see Fig. 1(b)) and diameter B (irradiated laser beam diameter B, see Fig. 1(b)) of the laser beam 3 irradiated to the workpiece 2, the scanning speed of the laser beam 3 relative to the workpiece 2, the scanning pattern of the laser beam 3 relative to the workpiece 2, and the flow rate (assist gas flow rate) of an assist gas (inert gas) supplied to the periphery of the position irradiated with the laser beam 3, which are determined as welding conditions suitable for the temperature distribution.

[0077] The scanning pattern of the laser beam 3 on the workpiece 2 may be "linear," "rotational," "zigzag," etc. The "linear" scanning pattern is one in which the irradiation position of the laser beam 3 is moved linearly in a fixed direction by moving the welding nozzle 40 linearly relative to the base 50 without driving the galvano scanner 43.

[0078] "Rotation" is a scanning pattern in which the irradiating position of the laser light 3 is moved linearly while rotating by moving the welding nozzle 40 linearly relative to the base 50 while rotating (revolving) the laser light 3 around a certain point using the galvano scanner 43.

[0079] "Zigzag" is a scanning pattern in which the galvanometer scanner 43 is moved back and forth linearly in a predetermined range in a predetermined direction while the welding nozzle 40 is moved linearly relative to the base 50 in a direction perpendicular to the predetermined direction, thereby moving the irradiation position of the laser light 3 in a zigzag pattern.

[0080] In the present invention, the flow rate of the assist gas is a concept that includes the flow velocity of the assist gas, and the learning means 16 may not be the actual flow rate of the assist gas, but may be the flow velocity of the assist gas.

[0081] The learning means 16 is composed of a multilayer neural network 16a having three or more layers, and also has a preprocessing means 16b on the input side of the multilayer neural network 16a.

[0082] The learning means 16 does not directly input the temperature distribution image of the predetermined region 2a acquired by the temperature distribution image acquisition means 15 into the multilayer neural network 16a for learning, but rather preprocesses the temperature distribution image to extract rough characteristics of the temperature distribution image, reduce the amount of data, and convert it into a form suitable for learning in the multilayer neural network 16a. The preprocessing means 16b performs the preprocessing, and includes a convolution layer 16b1, an activation layer 16b2, and a pooling layer 16b3.

[0083] The convolution layer 16b1 is a layer that performs convolution processing on an input temperature distribution image of 100 × 100 pixels by repeatedly performing calculations using, for example, a 3 × 3 pixel filter while shifting it by one pixel at a time. This convolution processing extracts rough characteristics of the temperature distribution image. The image convolved by the convolution layer 16b1 is 98 × 98 pixels.

[0084] The activation layer 16b2 is a layer that converts the output of the convolutional layer 16b1 into a form suitable for learning, and is configured with a lookup table. The pooling layer 16b3 is a layer that reduces the amount of data by thinning the data output from the activation layer 16b2. In the example shown in Figure 3, the image thinned by the pooling layer 16b3 is 49 x 49 pixels. This image S3 is input to the multilayer neural network 16a.

[0085] The multilayer neural network 16a includes an input layer 16a1, one or more hidden layers 16a2, and an output layer 16a3. The input layer 16a1 is a layer that represents the values ​​of each pixel in the input image (the image after the temperature distribution image S1 has been preprocessed by the preprocessing means 16b). The input layer 16a1 is provided with units X1 to X2041 equal to the number of pixels in the input image (49 × 49 pixels = 2401 pixels), and the value of the corresponding pixel is input to each unit. In addition to the values ​​of each pixel in the input image, the input layer 16a1 is also provided with a bias X, which is a unit that stores a bias value.

[0086] Hidden layer 16a2 is an intermediate layer interposed between input layer 16a1 and output layer 16a3. In the example shown in Fig. 3, hidden layer 16a2 is a single layer, and includes 512 units H1 to H512 to which weighted values ​​are added and input from each unit of input layer 16a1, and bias H, which is a unit storing a bias value.

[0087] The value of each unit (excluding bias H) Hn (n = a natural number from 1 to 512) in hidden layer 16a2 is expressed by the following equation: In the following equation, w(m, n) is a weighting coefficient for unit Hn in hidden layer 16a2 in unit Xm (m = a natural number from 1 to 2401) in input layer 16a1. Also, w(b1, n) is a weighting coefficient for unit Hn in hidden layer 16a2 in unit (bias X) in which the bias value (offset value) of input layer 16a1 is stored.

[0088] Hn=w(1,n)×X1+w(2,n)×X2+…+w(2401,n)×X2401+w(b1,n)×bias X The output layer 16a3 is provided with six units A1 to A6 that output welding conditions for laser welding. Specifically, the unit A1 outputs the output intensity (light source laser light output intensity) of the laser beam 3 output from the light source 20. The unit A2 outputs the flow rate (assist gas flow rate) of the assist gas (inert gas) to be supplied to the periphery of the position irradiated with the laser beam 3.

[0089] Unit A3 outputs the scanning speed of the laser beam 3 relative to the workpiece 2. Unit A4 outputs the scanning pattern of the laser beam 3 relative to the workpiece 2. Unit A5 outputs the intensity A of the laser beam 3 irradiated to the workpiece 2 (irradiated laser beam intensity A, see Figure 1(b)). Unit A6 outputs the diameter B of the laser beam 3 irradiated to the workpiece 2 (irradiated laser beam diameter B, see Figure 1(b)).

[0090] The values ​​output from units A1 to A6 are all encoded outputs, and the laser welding conditions are interpreted for the outputs (values) from each of units A1 to A6 by control means 17, which will be described later, and laser welding is controlled.

[0091] The value of each unit Ao (o=a natural number from 1 to 6) in output layer 16a3 is expressed by the following equation: In the following equation, w(o,m) is a weighting coefficient for unit Ao in output layer 16a3 in unit Hn of hidden layer 16a2. Also, w(b2,o) is a weighting coefficient for unit Ao in output layer 16a3 in unit (bias H) in which the bias value (offset value) of hidden layer 16a2 is stored.

[0092] Ao=w(1,o)×H1+w(2,o)×H2+…+w(512,o)×H512+w(b2,o)×bias H Here, the operation of the learning means 16 will be described with reference to Fig. 4. Fig. 4(a) is a diagram schematically showing an example of learning data (teaching data) used for learning by the learning means 16. Fig. 4(b) is a diagram schematically showing an example of output from the learning means 16 when a temperature distribution image of the predetermined region 2a acquired by the temperature distribution image acquisition means 15 from image data captured in real time by the camera 11 is input to the learning means 16.

[0093] The learning means 16 performs learning in advance using learning data such as that shown in Fig. 4(a). The learning data defines, in association with each other, a pattern of the temperature distribution image of the predetermined region 2a of the workpiece 2 and welding conditions for laser welding that are appropriate when the temperature distribution image of the predetermined region 2a of the workpiece 2 has that pattern.

[0094] The welding conditions for laser welding specified in the learning data are the output intensity of the laser light 3 output from the light source 20 (light source laser light output intensity), which is also the output of the learning means 16, the flow rate of the assist gas (inert gas) supplied around the position where the laser light 3 is irradiated, the scanning speed of the laser light 3 relative to the workpiece 2, the scanning pattern of the laser light 3 relative to the workpiece 2, the intensity A of the laser light 3 irradiated to the workpiece 2 (irradiated laser light intensity A, see Figure 1(b)), and the diameter B of the laser light 3 irradiated to the workpiece 2 (irradiated laser light diameter B, see Figure 1(b)).

[0095] A plurality of such learning data are prepared with different patterns of the temperature distribution image, and the prepared plurality of learning data are used to perform learning in the learning means 16. That is, the learning means 16 first sets predetermined initial values ​​for various parameters used in the learning model.

[0096] Then, various parameters used in the learning model of the learning means 16, such as the weighting coefficients w(m,n), w(b1,n), w(n,o), and w(b2,o) in the multilayer neural network 16a, the bias X, and the bias H values, are adjusted so that when a temperature distribution image represented by data is input to the learning means 16, the welding conditions for laser welding associated with the learning data are output from the learning means 16.

[0097] The weighting coefficients are adjusted, for example, by backpropagation, which checks whether the input values ​​are correct from the output values, and are automatically optimized by gradient descent, etc. The more learning data there is in the learning means 16, the higher the accuracy of the output relative to the input.

[0098] While laser welding is being performed by the laser welding device 1, the laser welding control system 10 uses the learning means 16 that has learned in this manner to determine in real time the welding conditions for laser welding that are optimal for the temperature distribution in the specified area 2a of the workpiece 2 at that time.

[0099] That is, a temperature distribution image in the predetermined region 2a of the workpiece 2 is obtained by the temperature distribution determination means based on images (images showing, for each pixel, the intensity of light of wavelength λ1 that has passed through the bandpass filter 13 among the light emitted by radiant heat) captured by the camera 11 at predetermined time intervals (for example, 1 second) and that include at least the predetermined region 2a. By inputting the temperature distribution image to the learning means 16 that has performed learning, welding conditions for laser welding that are suited to the temperature distribution image are output from the units A1 to A6 (see FIG. 3), as shown in FIG. 4(b).

[0100] This process from capturing an image by the camera 11 to outputting the welding conditions for laser welding by the learning means 16 is repeated at the above-mentioned predetermined time intervals while laser welding is being performed. In other words, the welding conditions for laser welding that are suitable for the temperature distribution in the predetermined region 2a of the workpiece 2 are determined in real time. Units A1 to A6 of the learning means 16 that indicate the results of this determination are input to the control means 17.

[0101] Continuing the explanation, returning to Fig. 1, the control means 17 feedback-controls the laser welding in the laser welding device 1 based on the welding conditions for laser welding determined by the learning means 16.

[0102] Specifically, control means 17 interprets the welding conditions for laser welding based on the current state of laser welding from the values ​​of units A1 to A6 encoded by learning means 16. Then, control means 17 adds a correction value to each welding condition for laser welding determined by learning means 16 in accordance with other laser welding environments (such as the welding conditions of previous laser welding, room temperature, and displacement of the welding position). Control means 17 feedback-controls each component of laser welding apparatus 1, such as light source 20, welding nozzle 40, and gas supply valve 61, in real time so that laser welding is performed under the welding conditions for laser welding to which the correction value has been added.

[0103] For example, the control means 17 feedback-controls the light source 20 so that the output intensity of the laser light 3 output from the light source 20 (light source laser light output intensity) is the same as that obtained based on the determination of the learning means 16. The control means 17 feedback-controls the gas supply valve 61 so that the flow rate of the assist gas (inert gas) supplied to the periphery of the position where the laser light 3 is irradiated is the same as that obtained based on the determination of the learning means 16.

[0104] The control means 17 feedback-controls the welding nozzle 40 and / or the base 50 so that the scanning speed of the laser beam 3 relative to the workpiece 2 is the same as that obtained based on the determination of the learning means 16. The control means 17 feedback-controls the welding nozzle 40 and / or the base 50 and the galvano scanner 43 so that the scanning pattern of the laser beam 3 relative to the workpiece 2 is the same as that obtained based on the determination of the learning means 16.

[0105] The control means 17 feedback-controls the focusing lens 44 etc. so that the intensity A (irradiated laser light intensity A, see Figure 1(b)) and diameter B (irradiated laser light diameter B, see Figure 1(b)) of the laser light 3 irradiated to the workpiece 2 are obtained based on the judgment of the learning means 16.

[0106] As described above, the machine learning device 18, the laser welding control system 10, and the laser welding device 1 in this embodiment have the following advantages.

[0107] (A) In the laser welding control system 10, light of wavelength λ1 for measuring the temperature of the workpiece 2, among light emitted due to radiant heat from at least a predetermined area 2a set on the workpiece 2 including the irradiation position of the laser light 3 irradiated on the workpiece 2, passes through a bandpass filter 13. Then, among light emitted due to radiant heat from at least the predetermined area 2a, the light that has passed through the bandpass filter 13 is imaged by a camera 11. The image obtained by imaging by the camera 11 is input to a machine learning device 18.

[0108] In the machine learning device 18, a temperature distribution image in the predetermined region 2a is acquired by the temperature distribution image acquisition means 15, and the learning means 16 learns to determine the welding conditions for laser welding by updating the relationship for determining the welding conditions for laser welding based on the temperature distribution image. Then, based on the welding conditions for laser welding determined by the learning means 16, the control means 17 controls the laser welding in the laser welding device 1.

[0109] In this way, the data used to input the learning means 16 is a temperature distribution image obtained based on light emission accompanying radiation from the predetermined region 2a of the workpiece 2, captured in real time. By using the temperature distribution image, learning by the learning means 16 and determination of the welding conditions for laser welding can be performed without being affected by various noise factors, such as plasma light emission generated by irradiation with the laser beam 3, scattered light from the irradiated laser beam 3, sunlight shining on the workpiece 2, indoor lighting, and light from arc discharge in a welding shop. This has the effect of enabling feedback control of the welding conditions for laser welding in real time with high accuracy.

[0110] (A) The welding conditions controlled based on the learning of the learning means 16 are the output intensity of the laser beam 3 output from the light source 20, the intensity A and diameter B of the laser beam 3 irradiated onto the workpiece 2, the scanning speed of the laser beam 3, the scanning pattern of the laser beam 3, and the flow rate of the inert gas (assist gas) supplied to the periphery of the position irradiated with the laser beam 3. Therefore, these parameters included in the welding conditions can be feedback-controlled in real time with high precision, thereby realizing laser welding of more stable quality.

[0111] (c) In the laser welding control system 10, the temperature distribution image acquisition means 15 uses light of one wavelength λ1 as the wavelength of light used to acquire the temperature of the workpiece 2. This requires only one bandpass filter 13 as optical filter means that passes light of wavelength λ1, and at least one camera 11 as imaging means that captures the light of wavelength λ1 that has passed through the bandpass filter 13 among the light emitted by radiant heat from at least the predetermined region 2a, thereby reducing the number of components required to acquire the temperature of the workpiece 2. Therefore, the temperature within the predetermined region shown by the temperature distribution image can be obtained at low cost.

[0112] (d) In the laser welding apparatus 1, the laser welding of the workpiece 2 is controlled by the laser welding control system 10, and therefore the effects of the laser welding control system 10 can be enjoyed.

[0113] The present invention has been described above based on an embodiment, but the present invention is not limited to the above embodiment, and it can be easily inferred that various improvements and modifications are possible within the scope of the present invention.

[0114] For example, in the above embodiment, the temperature distribution image acquiring means 15 acquires the temperature distribution of the predetermined region 2a from the intensity of light of one wavelength λ1 among the light emitted from the predetermined region 2a in association with radiant heat. Alternatively, the temperature distribution image acquiring means 15 may acquire the temperature from the ratio of the intensity of light of a first wavelength λ1 to that of a second wavelength λ2 among the light emitted from the predetermined region 2a in association with radiant heat.

[0115] That is, as shown in Figure 2, in the light radiated from a blackbody, the ratio of the light intensities of the first wavelength λ1 and the second wavelength λ2, which are different wavelengths, varies depending on the temperature of the blackbody. The relationship between the ratio of the light intensities of the first wavelength λ1 and the second wavelength λ2 and the temperature can be expressed by the following equation 3: where I1 is the light intensity of the first wavelength λ1, and I2 is the light intensity of the second wavelength λ2.

[0116]

number

[0117] Here, a configuration for inputting the light intensities I1 and I2 of the first wavelength λ1 and the second wavelength λ2, respectively, to the temperature distribution image acquisition means 15 will be described with reference to Fig. 5. Fig. 5 is a schematic diagram illustrating a modified example of the laser welding control system 10. In Fig. 5, the same components as those in the above embodiment are denoted by the same reference numerals, and the description thereof will be omitted or simplified below.

[0118] The laser welding control system 10 of this modified example has a beam splitter 14a, a reflecting mirror 14b, a first bandpass filter 13a, a second bandpass filter 13b, a first camera 11a, and a second camera 11b instead of the reflecting mirror 14, bandpass filter 13, and camera 11 of the laser welding control system 10 of the above embodiment.

[0119] In this modification, dichroic mirror 42 of welding nozzle 40 reflects (or transmits) light of a first wavelength λ1 and light of a second wavelength λ2 used to measure the temperature of workpiece 2, out of the light emitted from workpiece 2 in association with radiant heat, and the light of first wavelength λ1 and light of second wavelength λ2 are output from inside welding nozzle 40 to the outside. The light output to the outside of welding nozzle 40 is incident on beam splitter 14a. Beam splitter 14a splits the light output from welding nozzle 40 to the outside into two beams of light. Note that a dichroic mirror may be used instead of beam splitter 14a to split the light of first wavelength λ1 and the light of second wavelength λ2.

[0120] One of the beams split by the beam splitter 14a is incident on the first bandpass filter 13a. The first bandpass filter 13a corresponds to the first optical filter means of the present invention and is a filter that passes only light of a first wavelength λ1. The first bandpass filter 13a is provided between the beam splitter 14a and the first camera 11a. The first bandpass filter 13a allows only the light of the first wavelength λ1 to pass through and enter the first camera 11a from the light that is emitted in response to radiant heat in at least an area including the predetermined area 2a set on the workpiece 2 and reflected by the beam splitter 14a.

[0121] The first camera 11a corresponds to the first imaging means of the present invention, and captures light of the first wavelength λ1 that has passed through the first bandpass filter 13a. Similar to the camera 11 of the above embodiment, the first camera 11a always receives light emitted in association with radiant heat from the irradiation position of the laser light 3 on the workpiece 2 at the center of the angle of view captured by the first camera 11a. The first camera 11a is configured to always capture the same area centered on the irradiation position of the laser light 3 on the workpiece 2. The imaged area includes the predetermined area 2a.

[0122] It is not necessary that the light emitted due to radiant heat from the irradiation position of the laser light 3 on the workpiece 2 always be incident on the center of the area imaged by the first camera 11a; as long as the specified area 2a is included in the angle of view, the irradiation position of the laser light 3 may be fixed at any position in the area imaged by the first camera 11a.

[0123] The image obtained by imaging using the first camera 11a is obtained by converting the intensity of light of the first wavelength λ1 incident on each pixel of the imaging element that constitutes the first camera 11a into an electrical signal for each pixel through photoelectric conversion.

[0124] While laser welding is being performed by the laser welding device 1, the first camera 11a takes images at predetermined time intervals (for example, 1 second) and outputs the captured image data to the computer 12. In other words, the first camera 11a takes images of the state of the workpiece 2 during laser welding in real time, and outputs to the temperature distribution image acquisition means 15 the intensity of light of the first wavelength λ1 emitted in association with radiant heat in an area including the predetermined area 2a of the workpiece 2.

[0125] On the other hand, the other light split by beam splitter 14a is reflected by reflecting mirror 14 and passes through second bandpass filter 13b, which corresponds to second optical filter means of the present invention, before being incident on second camera 11b, which corresponds to second imaging means of the present invention. The configurations of second bandpass filter 13b and second camera 11b are similar to those of first bandpass filter 13a and first camera 11a.

[0126] However, the second bandpass filter 13b has a different wavelength of light that passes through it from the first bandpass filter 13a, and passes light of the second wavelength λ2. The second camera 11b captures images of the state of the workpiece 2 during laser welding in real time, in synchronization with the imaging timing of the first camera 11a, and outputs to the temperature distribution image acquisition means 15 the intensity of light of the second wavelength λ2 emitted in association with radiant heat in an area including the predetermined area 2a of the workpiece 2.

[0127] The temperature distribution image acquisition means 15 acquires the temperature distribution of the specified region 2a by the above-mentioned method using the intensity of light of the first wavelength λ1 input from the first camera 11a and the intensity of light of the second wavelength λ2 input from the second camera 11b.

[0128] In this way, the temperature distribution image of the predetermined area 2a on the workpiece 2 is acquired using the light of two wavelengths λ1 and λ2 emitted in association with radiant heat from the predetermined area 2a, thereby improving the reliability of the temperature in the predetermined area 2a shown by the temperature distribution image, thereby enabling feedback control of the welding conditions for laser welding with higher accuracy.

[0129] In the above embodiment and the above modified example, the learning means 16 of the machine learning device 18 determines the welding conditions for laser welding by updating the relationship for determining the welding conditions for laser welding based on the temperature distribution image acquired by the temperature distribution image acquisition means 15. Alternatively, or in addition, the learning means 16 of the machine learning device 18 may determine the state of laser welding by updating the relationship for determining the state of laser welding based on the temperature distribution image acquired by the temperature distribution image acquisition means 15.

[0130] The laser welding state may be, for example, whether the area of ​​the butted or overlapping plate materials serving as the workpiece 2 that has been melted by the irradiation of the laser light 3 is perforated or non-perforated, or it may indicate the weld bead width, or it may indicate the degree of possibility of a welding defect occurring, or the content of a defect that may currently exist. Furthermore, it may also indicate the degree of possibility of each defect occurring for each type of defect. Examples of the defect include the state of underfill, undercut, spatter, etc.

[0131] The control means 17 performs feedback control to adjust the welding conditions of the laser welding in the laser welding apparatus 1 so that the laser welding state becomes the target state, based on the state of the laser welding determined in real time by the learning means 16. This makes it possible to realize laser welding in real time with high accuracy so that the laser welding state becomes the target state.

[0132] In the above embodiment and modified example, a case has been described in which a single light source 20 is used in the laser welding apparatus 1 to irradiate the workpiece 2 with laser light 3 having the intensity distribution shown in Fig. 1(a), and the laser welding control system 10 feedback-controls the intensity A and diameter B of this laser light 3 through machine learning by the machine learning device 18. However, laser welding may be performed by providing a plurality of light sources and combining the laser light output from the plurality of light sources to form the laser light 3 to be irradiated onto the workpiece 2.

[0133] Here, with reference to Fig. 6, an example of laser beam 3 formed by combining laser beams output from two light sources will be described. Fig. 6(a) is a diagram showing an example of the intensity distribution of laser beam 3 formed by combining laser beams output from two light sources. Fig. 6(b) is a diagram showing the intensity A and diameter B of the main laser beam in the laser beam 3 shown in Fig. 6(a), the radius C indicating the position where the secondary laser beam is formed relative to the main laser beam, and the intensity D and diameter E of the secondary laser beam. The laser beam 3 shown in Fig. 6(a) is a multi-beam in which one light source forms a Gaussian beam-shaped main laser beam at the center of the laser beam 3, and the other light source forms a ring beam-shaped secondary laser beam surrounding the periphery of the main laser beam.

[0134] The laser welding control system 10 may have the learning means 16 learn how to determine the output balance of the laser beams output from each of these light sources, and may perform this determination in real time based on the temperature distribution image. Furthermore, the control means 17 may feedback-control the output balance of the laser beams output from each of the light sources based on the state of laser welding determined in real time by the learning means 16 based on the temperature distribution image.

[0135] 6(a) as an example, the learning means 16 may learn the determination of the output intensity of the light source that forms the main laser beam and the output intensity of the light source that forms the secondary laser beam, and may perform this determination in real time based on the temperature distribution image. Furthermore, the control means 17 may feedback-control the output intensity of the light source that forms the main laser beam and the output intensity of the light source that forms the secondary laser beam based on the state of laser welding determined in real time by the learning means 16 based on the temperature distribution image.

[0136] 6(b), the laser welding control system 10 may have the learning means 16 learn the determination of the intensity A and diameter B of the primary laser beam, the radius C indicating the position where the secondary laser beam is formed relative to the primary laser beam, and the intensity D and diameter E of the secondary laser beam, and may make these determinations in real time based on the temperature distribution image. Furthermore, the control means 17 may feedback-control the intensity A and diameter B of the primary laser beam, the radius C indicating the position where the secondary laser beam is formed relative to the primary laser beam, and the intensity D and diameter E of the secondary laser beam, based on the state of laser welding determined by the learning means 16 in real time based on the temperature distribution image.

[0137] As a pattern of the laser beam 3 formed by combining a plurality of light sources, instead of or in addition to the above-described multi-beam, a multi-beam may be used in which one light source forms a main laser beam in the shape of a top hat beam at the center of the laser beam 3, and the other light source forms secondary laser beams in the shape of a ring beam surrounding the periphery of the main laser beam 3. In this case, too, the intensity A and diameter B of the main laser beam in the shape of a top hat beam may be learned by the learning means 16, and these may be determined in real time based on the temperature distribution image.

[0138] Furthermore, the pattern of laser light 3 formed by combining multiple light sources may include a pattern in which single beams formed by each of the multiple light sources are combined, instead of or in addition to the above-mentioned multiple beams. In this case, each single beam may be fixed to either a Gaussian beam or a top-hat beam, or may be a pattern in which a Gaussian beam and a top-hat beam are combined. At least one of the shape (Gaussian beam or top-hat beam), intensity, and diameter of each single beam may be learned by learning means 16, and these may be determined in real time based on the temperature distribution image.

[0139] Furthermore, the pattern of laser beam 3 formed by combining multiple light sources may include at least one of a Gaussian single beam formed only by a Gaussian-shaped main laser beam, a top-hat single beam formed only by a top-hat-shaped main laser beam, and a ring beam formed only by a ring-shaped secondary laser beam. In this case, the Gaussian single beam or top-hat single beam may be formed by setting the output intensity of the light source forming the secondary laser beam to zero, or by setting the intensity D of the secondary laser beam to zero. Furthermore, the ring beam may be formed by setting the output intensity of the light source forming the primary laser beam to zero, or by setting the intensity A of the primary laser beam to zero.

[0140] Furthermore, the pattern of the laser light 3 formed by combining a plurality of light sources may be learned by the learning means 16, and these determinations may be made in real time based on the temperature distribution image.

[0141] In this way, laser welding is performed while feedback-controlling the output balance of each laser beam output from the plurality of light sources and / or the pattern of laser beam 3 formed by combining the plurality of light sources so that the target laser welding quality is achieved based on the judgment using the temperature distribution image by learning means 16. Therefore, laser welding of more stable quality can be achieved.

[0142] The welding conditions may also include at least a pattern of temporal change in the output balance of each laser beam output from the multiple light sources. The output balance of each laser beam can be changed over time in a way that is appropriate for the state of the laser welding. Based on the judgment using the temperature distribution image by the learning means 16, feedback control of the temporal change in the output balance of each laser beam is performed to achieve the target laser welding quality. Laser welding can be performed while slowly or instantaneously changing the output balance of each laser beam in a pattern appropriate for each situation, based on the temperature distribution in the predetermined area 2a irradiated with the laser beam 3. This allows for laser welding of more stable quality.

[0143] In the above embodiment and modified example, the welding conditions for laser welding determined by learning by learning means 16 or feedback-controlled by control means 17 based on the determination by learning means 16 include the output intensity of laser beam 3 output from light source 20 (light source laser beam output intensity), the flow rate of assist gas (inert gas) supplied to the vicinity of the position where laser beam 3 is irradiated, the scanning speed of laser beam 3 relative to workpiece 2, the scanning pattern of laser beam 3 relative to workpiece 2, the intensity A of laser beam 3 irradiated to workpiece 2 (irradiated laser beam intensity A, see FIG. 1(b)), and the diameter B of laser beam 3 irradiated to workpiece 2 (irradiated laser beam diameter B, see FIG. 1(b)). However, it is not necessary to include all of these conditions; at least one of them may be included. Furthermore, at least one other welding condition for laser welding may be included. Laser welding control system 10 can precisely feedback-control the parameters included in the welding conditions in real time, thereby achieving laser welding of more stable quality.

[0144] In the above embodiment and modified example, the learning means 16 shown in FIG. 3 has been exemplified as a configuration of the learning means 16. However, this is merely an example, and a main feature of the present invention is that the input to the learning means 16 is a temperature distribution image. In other words, by using a temperature distribution image as input, the learning means 16 can use not only the one shown in FIG. 3 but also other known learning means. For example, learning by the learning means 16 may be performed by deep learning in which the hidden layer 16a2 in the multilayer neural network 16a has two or more layers. This enables feedback control of the welding conditions for laser welding with higher accuracy.

[0145] Furthermore, the learning means 16 may be configured to perform ensemble learning, which includes multiple independent learning models, performs learning on each model, and makes a final decision based on the results of those decisions. This also enables feedback control of the laser welding conditions with higher accuracy. [Explanation of symbols]

[0146] 1. Laser welding equipment 2 Workpiece (object to be welded) 2a Predetermined area 3 Laser light 10 Laser welding control system 13 Bandpass filter (optical filter means) 13a First bandpass filter (first optical filter means) 13b Second bandpass filter (second optical filter means) 11 Camera (imaging means) 11a First camera (first imaging means) 11b Second camera (second imaging means) 15 Temperature distribution image acquisition means 16 Learning Methods 17 Control Means 18 Machine Learning Device 20 light source

Claims

1. A laser welding control system for controlling laser welding of a welding object by a laser welding device, a separating means for separating light emitted due to radiant heat from at least a predetermined region set on the workpiece including an irradiation position of the laser light irradiated on the workpiece from an optical path of the laser light; an optical filter means for passing light of a wavelength for measuring the temperature of the work-piece out of the emitted light separated by the separation means; an imaging means for imaging light emitted by the radiant heat from at least the predetermined region and passing through the optical filter means; a temperature distribution image acquisition means for acquiring a temperature distribution image in the predetermined region based on the image acquired by the imaging means; a learning means for learning to determine the state and / or welding conditions of the laser welding by updating a relationship for determining the state and / or welding conditions of the laser welding based on the temperature distribution image acquired by the temperature distribution image acquisition means; and and control means for controlling laser welding in the laser welding device based on the state and / or welding conditions of the laser welding determined by the learning means.

2. A laser welding control system for controlling laser welding of objects to be welded by a laser welding device, comprising: an optical filter means for passing light of a wavelength for measuring the temperature of the workpiece, among light emitted in association with radiant heat from at least a predetermined region set on the workpiece including an irradiation position of the laser light irradiated on the workpiece; and an imaging means for imaging light emitted by the radiant heat from at least the predetermined region and passing through the optical filter means; a temperature distribution image acquisition means for acquiring a temperature distribution image in the predetermined region based on the image acquired by the imaging means; a learning means for learning to determine the state and / or welding conditions of the laser welding by updating a relationship for determining the state and / or welding conditions of the laser welding based on the temperature distribution image acquired by the temperature distribution image acquisition means; and a control means for controlling laser welding in the laser welding apparatus based on the state and / or welding conditions of the laser welding determined by the learning means, The optical filter means comprises: a first optical filter means for passing light of a first wavelength, which is used to measure the temperature of the work-piece, among light emitted in association with radiant heat from the predetermined region; and and a second optical filter means for passing light of a second wavelength, which is used to measure the temperature of the work-piece, among light emitted in association with radiant heat from at least the predetermined region, and The imaging means a first image capturing means for capturing an image of light emitted by at least the predetermined region due to radiant heat and passing through the first optical filter means; a second image capturing means for capturing an image of light emitted by the radiant heat from at least the predetermined region and passing through the second optical filter means; The laser welding control system is characterized in that the temperature distribution image acquisition means acquires a temperature distribution image in the specified area based on an image obtained by imaging using the first imaging means and an image obtained by imaging using the second imaging means.

3. A laser welding control system for controlling a laser welding device that performs laser welding by irradiating a welding object with a laser beam formed by combining laser beams output from a plurality of light sources, comprising: an optical filter means for passing light of a wavelength for measuring the temperature of the workpiece, among light emitted in association with radiant heat from at least a predetermined region set on the workpiece including an irradiation position of the laser light irradiated on the workpiece; and an imaging means for imaging light emitted by the radiant heat from at least the predetermined region and passing through the optical filter means; a temperature distribution image acquisition means for acquiring a temperature distribution image in the predetermined region based on the image acquired by the imaging means; a learning means for learning to determine the state and / or welding conditions of the laser welding by updating a relationship for determining the state and / or welding conditions of the laser welding based on the temperature distribution image acquired by the temperature distribution image acquisition means; and a control means for controlling laser welding in the laser welding apparatus based on the state and / or welding conditions of the laser welding determined by the learning means, 10. A laser welding control system, wherein the welding conditions include at least an output balance of the laser beams output from the respective light sources.

4. A laser welding control system as described in claim 3, characterized in that the welding conditions include at least a pattern of temporal change in the output balance.

5. the learning means learns to determine the state of laser welding by updating at least a relationship for determining the state of laser welding based on the temperature distribution image acquired by the temperature distribution image acquisition means, 5. A laser welding control system according to claim 1, wherein the control means performs feedback control to adjust the welding conditions of the laser welding in the laser welding device based on the state of the laser welding determined by the learning means so that the state of the laser welding becomes a target state.

6. 6. A laser welding control system according to claim 1, wherein the welding conditions include at least one of an output intensity of the laser light output from a light source, an intensity and / or a diameter of the laser light irradiated onto the workpiece, a scanning speed of the laser light, a scanning pattern of the laser light, and a flow rate of an assist gas supplied to the vicinity of the position irradiated with the laser light.

7. 7. The laser welding control system according to claim 1, wherein the learning means performs learning by deep learning.

8. 8. The laser welding control system according to claim 1, wherein the learning means performs learning by ensemble learning using a plurality of learning models.

9. A laser welding device in which laser welding of objects to be welded is controlled by the laser welding control system according to any one of claims 1 to 8.

Images