JP2020189305A - Laser machining system, leaning device, and leaning method for leaning device - Google Patents

Abstract

To perform evaluation of laser welding including the cause of abnormality, in good accuracy.SOLUTION: A laser machining system comprises: a laser machining device collecting laser beam on the surface of a machining target to weld the machining target; and a learning device executing learning to the laser machining device. The learning device includes: an input processing unit for inputting a state variable including at least one laser condition relating to the laser machining, and a calculation result calculated on the basis of at least one measurement value measured during execution of the laser machining; and a learning unit for executing learning updating an evaluation model deriving the evaluation of the laser machining on the basis of the state variable.SELECTED DRAWING: Figure 1

Description

The present invention relates to a laser processing system, a learning device and a learning method of a learning device used for evaluating a workpiece manufactured by laser welding.

Laser welding technology irradiates a work piece with laser light emitted from a laser oscillator, melts the work piece with the amount of heat of the laser light, and welds it to another work piece to weld these work pieces. It is a technology for mechanically and electrically connecting. Laser welding technology is generally widespread in a wide range of fields such as home appliances, precision equipment, and automobile parts.

In such laser welding technology, various adjustment items are generally adjusted by trial and error according to the shape and size of each laser oscillator or workpiece, but a processed product of a predetermined quality can be obtained. If this is not possible, there are cases where such trial-and-error adjustments cannot be used. Therefore, as shown in Patent Document 1, a quality judgment is made by comparing the measured value of the internal information of the processing apparatus with the threshold value set in the provisional judgment unit, and the quality with the actual processed product is fed back and provisionally. The threshold value of the determination unit is also updated.

JP-A-2017-164801

However, in the method for evaluating the processing result according to Patent Document 1, the processing result is evaluated after the processing, and it takes time to obtain the processing result. Further, in the method for evaluating the processing result according to Patent Document 1, even if the processing result can be acquired after the processing and it can be detected that the processing result is abnormal, the cause cannot be identified, so that it takes time and effort to deal with the abnormality.

The learning method of the laser processing system, the learning device, and the learning device of the present disclosure aims to enable accurate evaluation of laser welding including the cause of abnormality in view of the above-mentioned conventional problems.

The laser processing system of the present disclosure comprises a laser processing apparatus that welds the workpiece by condensing the laser beam on the surface of the workpiece, and a learning apparatus that executes learning for the laser processing apparatus. In the laser processing system to be configured, the learning device obtains at least one laser condition related to laser processing and a calculation result calculated based on at least one measured value measured during execution of the laser processing. It includes an input processing unit in which the included state variables are input, and a learning unit that executes learning to update an evaluation model for deriving the evaluation of the laser processing based on the state variables.

The learning device of the present disclosure is a learning device that performs learning on a laser processing device that welds the work piece by condensing a laser beam on the surface of the work piece, and is at least one related to laser processing. An input processing unit in which a state variable including a laser condition and a calculation result calculated based on at least one measured value measured during execution of the laser machining is input, and the laser based on the state variable. It is provided with a learning unit that executes learning to update the evaluation model for deriving the evaluation of machining.

The learning method of the learning device of the present disclosure is a learning method of a learning device that executes learning for a laser processing device that welds the work piece by condensing a laser beam on the surface of the work piece. An input step in which a state variable including at least one laser condition related to laser machining and a calculation result calculated based on at least one measured value measured during the execution of the laser machining is input, and the state variable Based on this, it includes a learning step of executing learning to update the evaluation model for deriving the evaluation of the laser machining.

According to the laser processing system, the learning device, and the learning method of the learning device of the present disclosure, it is possible to accurately evaluate the laser welding including the cause of the abnormality.

A block principle diagram showing an outline of the laser machining evaluation system according to the embodiment of the present disclosure. A block principle diagram for explaining an outline of learning steps performed by the learning processing unit of the evaluation system according to the embodiment of the present disclosure. Schematic diagram of the laser welding apparatus according to the embodiment of the present disclosure Schematic diagram for explaining the method of detecting thermal radiation light according to the embodiment of the present disclosure. Schematic diagram for explaining the method of detecting welding defects by measuring the thermal radiation light of the embodiment of the present disclosure. It is a figure for demonstrating an example of the correlation between the heat radiation amount and welding quality of the embodiment of this disclosure, and is the figure which shows the heat radiation light amount-time diagram and the laser output-time diagram side by side. Schematic diagram for explaining the vibration detection method of the embodiment of the present disclosure. Schematic diagram for explaining a method of detecting welding defects by measuring vibration according to the embodiment of the present disclosure. It is a figure for demonstrating an example of the correlation between the vibration and the welding quality of one Embodiment of this disclosure, and is the figure which shows the vibration gain-time diagram and the laser output-time diagram side by side. Schematic diagram for explaining a method of detecting the generation of fume by measuring visible light according to the embodiment of the present disclosure. Schematic diagram for explaining the method of detecting the generation of plasma by the measurement of visible light of one Embodiment of this disclosure. It is a figure for demonstrating an example of the correlation between visible light and welding quality of one Embodiment of this disclosure, and is the figure which shows the visible light amount-time diagram and laser output-time diagram side by side. It is a figure for demonstrating an example of the correlation between visible light and welding quality of one Embodiment of this disclosure, and is the figure which shows the visible light amount-time diagram and laser output-time diagram side by side. It is a figure for demonstrating an example of the correlation between visible light and welding quality of one Embodiment of this disclosure, and is the figure which shows the visible light amount-time diagram and laser output-time diagram side by side. 13A to 13C are schematic views for explaining how to return the OCT signal according to the keyhole shape of the embodiment of the present disclosure, respectively. It is a figure for demonstrating an example of the correlation between the OCT signal strength and the welding quality of one Embodiment of this disclosure, and is the figure which shows the OCT signal strength-time diagram and the laser output-time diagram side by side. Structural diagram of a convolutional neural network applied to the evaluation model in the embodiments of the present disclosure. Functional block diagram of the control unit according to the embodiment of the present disclosure.

(Embodiment)
Hereinafter, embodiments of the present disclosure will be described with reference to FIGS. 1 to 16.

FIG. 1 shows a block principle diagram showing an outline of the laser machining evaluation system 1 according to the embodiment of the present disclosure. The laser machining evaluation system 1 of the present disclosure shown in FIG. 1 has an input processing unit 2 that executes an input step, a learning processing unit 3 that executes a learning step, and an output processing unit 4 that executes an output step as functions. Be prepared.

The input processing unit 2 acquires the physical quantity 15 measured during laser welding (during processing) as a state variable 12 and inputs it to the evaluation model 11 (described later) of the learning processing unit 3.

The learning processing unit 3 includes a trained evaluation model 11 and a data set group 14.

The output processing unit 4 outputs the laser machining evaluation result 13 to the outside.

With such a configuration, the laser machining evaluation system 1 is configured to input the state variable 12 to the trained evaluation model 11 and output the laser machining evaluation result 13. The evaluation model 11 is optimized through learning steps by the learning processing unit 3 using the data set group 14. The laser processing evaluation result 13 is a result of predicting the presence or absence of an abnormality in the laser welding processing when the physical quantity 15 is measured and the cause of the abnormality.

The state variable 12 is a variable based on at least one laser condition related to laser machining and a physical quantity 15 obtained by actual measurement for each laser welding machining. Laser conditions related to laser processing include, for example, laser wavelength, spot diameter (laser intensity distribution profile), laser output profile (laser output, irradiation time, output change), assist gas amount, assist gas flow velocity, assist gas supply position, assist gas. At least one of the supply direction, the position of the dust collecting intake duct, and the dust collecting intake flow velocity. Further, it is preferable to acquire the state variable 12 obtained by converting the physical quantity 15. The information acquired as the physical quantity 15 may include information other than the information necessary for the laser welding processing evaluation. Therefore, it is preferable to obtain the state variable 12 obtained by appropriately converting the physical quantity 15.

The input physical quantity 15 is at least one of the temperature of the processed portion, the amount of heat radiated from the processed portion, the amount of visible light of the processed portion, the vibration amount of the workpiece, and the OCT signal of the processed portion. The physical quantity 15 measured in real time from the start of one laser welding process to the end is input to the evaluation model 11 as a state variable 12. Since a sufficient number of samples is required to capture the tendency of the local value of the physical quantity 15 such as the characteristics of the process such as the curvature of the curve of the laser output profile in the laser welding evaluation, the sampling period (measurement) of the physical quantity 15 is required. The period) is preferably 1/100 or less of the time required to control the output of laser irradiation.

The physical quantity 15 is input to the evaluation model 11 as the state variable 12. The evaluation model 11 is a model including a function that processes the input state variable 12 and converts it into an output (that is, a function that obtains and outputs the laser machining evaluation result 13 based on the state variable 12). By optimizing the function in the processing (learning step) by the learning processing unit 3 described later with reference to FIG. 2, the evaluation model 11 can obtain the highly accurate laser machining evaluation result 13.

The laser processing evaluation result 13 is a result of stepwise evaluation of abnormalities in laser welding processing. The laser machining evaluation result 13 is classified into an n + 1 pattern, which is the sum of one pattern when there is no abnormality in machining and n patterns of the number of patterns causing the abnormality when there is an abnormality in machining. Known failure causes include, for example, spatter generation, fusion penetration, fume generation, plasma generation, laser output fluctuation, spot diameter fluctuation, laser irradiation time fluctuation, and work-induced fluctuation. Can be mentioned.

The laser machining evaluation result 13 is specifically a one-dimensional vector that holds the probability of the above pattern for each n + 1 element. The output processing unit 4 outputs the element having the largest value among the elements, that is, the pattern (defect cause) having the highest probability of occurrence.

The data set group 14 used for learning the evaluation model 11 is a set of two data, an input data and an output data, accumulated. Specifically, the data set group 14 uses the state variable 12 based on the physical quantity 15 measured in one laser welding process as input data, and the laser processing result 16 (when the state variable 12 is measured as output data). (See FIG. 2) is collected as a set for each laser welding process.

The laser processing result 16 is obtained by actual measurement for each welding process and is associated with the physical quantity 15 during the laser welding process. That is, when a certain welding process is performed, the detection of the physical quantity 15 and the determination of the laser processing result 16 are performed as a set, and the set is input to the data set group 14. The laser processing result 16 is judged, for example, as the quality of the processed product by a person or by mechanical means using the prior art, and is input to the data set group 14. The laser machining result 16 is measured by, for example, an apparatus that actually measures the laser machining result, and is input as a signal. Further, the laser processing result 16 may be evaluated by a person in an inspection process, for example, and may be input via an input device such as a keyboard. It is easiest to judge the quality of the welding process based on the quality of the processed product, but the quality of the welding process may be judged based on other criteria.

The laser processing result 16 is a stepwise evaluation of the abnormality of the laser welding processing, and one pattern when there is no abnormality in the processing and n patterns of the number of patterns which cause the abnormality when there is an abnormality in the processing. It is classified into the n + 1 pattern which is the sum of and. Known patterns of abnormal causes include, for example, spatter generation, fusion penetration, fume generation, plasma generation, excess clearance, excessive clearance, insufficient clearance, tool wear, and mold installation error. The cause of the defect is mentioned.

FIG. 2 shows a block principle diagram for explaining the outline of the learning steps performed by the learning processing unit 3 of the laser machining evaluation system 1 according to the embodiment of the present disclosure. From the data sets (state variable 12 and laser processing result 16) extracted from the data set group 14, the input data, that is, the state variable 12 based on the physical quantity 15 during laser welding is input to the evaluation model 11. The error 17 between the laser machining evaluation result 13 output from the evaluation model 11 and the laser machining result 16 extracted from the data set group 14 is calculated by the loss function 18. Then, based on this error 17, the weighting coefficient of the evaluation model 11 is updated by the optimization algorithm 19. The update of the weighting coefficient by this series of operations is performed using all the data sets accumulated in the data set group 14. Specifically, the entire data set group 14 is set until the error 17, which is the sum of the differences between the total laser machining result 16 in the total data set group 14 and the laser machining evaluation result 13 estimated by the evaluation model 11, is minimized and converges. Learning is performed by repeatedly updating the weighting coefficient of the evaluation model 11 used. As the loss function 18 suitable for deriving the error 17, the evaluation model 11 classifies each of the above-mentioned abnormal cause patterns and outputs the laser processing evaluation result 13, which is a loss function suitable for the group classification algorithm. It is desirable to use the entropy error. Further, it is desirable to use the steepest descent method or RMSprop as the method used for the optimization algorithm 19.

The learning step can be performed in parallel with the welding process. That is, it is possible to proceed with the optimization of the evaluation model 11 by learning in real time during the welding process. However, the data set group 14 needs the actual laser machining result 16. Therefore, it is preferable to perform the learning step after the laser welding process is completed, which is the timing at which both the input state variable 12 and the laser processing result 16 obtained by actual measurement are obtained.

In order to effectively optimize the evaluation model 11 by the learning step, there is a strong connection (correlation) between the input data (physical quantity 15) and the output data (laser machining result 16) of the data set group 14. It is necessary. In optimizing the laser machining evaluation system 1, it is required to appropriately select a state variable 12 based on a physical quantity 15 that has a strong correlation with abnormalities in laser welding machining as input data. Examples of the physical quantity 15 having an abnormally strong correlation with the laser welding process include the correlation between the amount of thermal radiation emitted from the processed part and the processing quality, the correlation between the amount of visible light and the quality of the processed part, and the correlation between vibration and the processing quality. The correlation between OCT and the quality of the processed part will be described below.

FIG. 3 shows a schematic view of the laser welding apparatus according to the embodiment of the present disclosure. The laser light emitted from the laser oscillator 21 is transmitted to the lens barrel 23 via the laser transmission fiber 22. The laser light 26 emitted from the laser transmission fiber 22 in the lens barrel 23 is formed into parallel light by the collimated lens 24, passes through the first reflection mirror 27 and the second reflection mirror 28, and is transmitted to the condenser lens 25. The light is collected on the surface of the workpiece 20. The laser beam 26 is a fiber laser having a laser wavelength of 1070 nmd and is a continuous wave.

In the embodiment of the present disclosure, a fiber laser having a laser wavelength of 1070 nmd is used, but the optimum laser wavelength may be selected according to the light absorption characteristics of the workpiece 20. For example, when the workpiece 20 is copper or gold, the laser wavelength is preferably as short as 405-450 nm, and when the workpiece 20 is aluminum, the absorption characteristics are good and good welding is possible. The laser wavelength is preferably a wavelength of about 800 nm.

In the embodiment of the present disclosure, a continuous wave is used for the laser beam 26, but a pulse wave laser beam may be used for the laser beam 26. When a continuous wave laser beam is used as the laser beam 26, the amount of heat input to the workpiece 20 can be increased, which is preferable in that laser welding can be performed with high productivity. Further, when a pulsed laser beam is used as the laser beam 26, it is preferable in that the thermal effect during processing can be reduced as compared with a continuous wave.

In the embodiment of the present disclosure, the laser beam emitted by the laser oscillator 21 is guided to the lens barrel 23 by using the laser transmission fiber 22, but the laser beam emitted by the laser oscillator 21 is directed to a mirror or the like. An optical element may be used to guide the lens barrel 23.

In the embodiment of the present disclosure, the fixed point of the workpiece 20 is irradiated with the laser beam 26, but the laser irradiation portion may be scanned to form a continuous welded portion. As a configuration for scanning the laser irradiation unit, a configuration in which the workpiece 20 or the lens barrel 23 is attached to a driving device such as a stage or a robot, or a galvano mirror or the like is used is exemplified.

In the embodiment of the present disclosure, a thermal radiation light sensor 29 for detecting the thermal radiation light generated when the workpiece 20 is irradiated with the laser beam 26 is provided.

In the embodiment of the present disclosure, a laser ultrasonic sensor 31 for detecting vibration generated when the workpiece 20 is irradiated with a laser beam is provided. The vibration generated from the processed portion (the portion of the workpiece 20 irradiated with the laser beam 26, hereinafter also referred to as the “laser irradiated portion”) propagates through the workpiece 20 so as to spread from the processed portion. The workpiece 20 is irradiated with a laser ultrasonic laser 32 for detecting the propagated vibration.

In the embodiment of the present disclosure, the visible light laser 35 (inspection light) emitted from the visible light source 33 is transmitted to the visible light sensor 34 so as to pass near the upper side (the side irradiated with the laser light) of the laser irradiation unit. Is detected by. In the embodiment of the present disclosure, the visible light source 33 that oscillates the visible light laser 35 is used, but a diffusion light source such as a lamp may be used instead of the visible light source 33. When a diffused light source is used, it is possible to measure a wider area. The visible light source 33 and the visible light sensor 34 detect the fumes generated above the laser irradiation unit as will be described later.

In the embodiment of the present disclosure, an OCT sensor 36 is provided for measuring the depth of a keyhole or the like formed in the laser irradiation portion. The OCT measurement laser 37 emitted from the OCT sensor 36 is reflected by the second reflection mirror 28 and is irradiated to the molten portion 38 of the workpiece 20 via the condenser lens 25. The laser 37 for OCT measurement is reflected by the concave bottom portion (concave bottom surface of the keyhole) of the melting portion 38, is reflected by the second reflection mirror 28 via the condenser lens 25, returns to the OCT sensor 36, and interferes with light. It is calculated as the depth of the concave bottom portion of the molten portion 38 using the phenomenon. That is, the second reflection mirror 28 is formed with a wavelength-selective reflection film that reflects only the OCT measurement laser 37.

FIG. 4 shows a schematic diagram for explaining the thermal radiation light detection method according to the embodiment of the present disclosure. In general, the amount of thermal radiation has a high correlation with the temperature, so the detection of the amount of thermal radiation is used to measure the temperature of the molten portion 38. In the embodiment of the present disclosure, thermal radiation light in a region of about φ5 mm (hereinafter referred to as “detection region”) centered on the laser irradiation position (center position of the laser irradiation portion) is detected. This detection region is wider than the molten portion 38. By reducing the detection region, it is possible to measure the temperature information at a specific position with high accuracy, but since the measurement is affected by the surface condition and the unevenness of the surface, the variation in measurement becomes large. Therefore, the measurement variation can be reduced by setting the region wider than the molten portion 38 as the detection region.

The diameter of the detection region of the thermal radiation light may be equal to or larger than the spot diameter (diameter of the laser irradiation portion) of the laser beam 26 for laser welding.

Further, it is preferable to use light having a wavelength in the infrared region as the measurement heat radiant light, and in the present embodiment of the present disclosure, light having a wavelength of 1300 nm is used as the measurement heat radiant light. When the workpiece 20 is irradiated with laser light, thermal radiation light is emitted from the molten portion 38. The heat radiant light emitted from the processed portion travels in the direction opposite to that of the laser light 26 via the condenser lens 25, is reflected by the first reflection mirror 27, and is incident on the heat radiant light sensor 29. There is. That is, the first reflection mirror 27 is formed with a wavelength-selective reflection film that reflects only thermal radiation light.

Further, even if a bandpass filter that selectively transmits 1300 nm light is arranged on the optical path until the heat radiant light 30 is reflected by the first reflection mirror 27 and then enters the heat radiant light sensor 29. Good. By arranging the bandpass filter, it is possible to prevent light of an unnecessary wavelength from entering the thermal radiation light sensor 29, and it is possible to measure the amount of thermal radiation light with more accuracy.

Further, in the embodiment of the present disclosure, the thermal radiant light having a single wavelength is detected, but two or more wavelengths may be detected by using a plurality of thermal radiant light sensors. By using a plurality of wavelengths of two or more wavelengths, it is possible to detect the amount of heat radiation more accurately than the detection with a single wavelength.

FIG. 5 is a schematic diagram for explaining a method for detecting welding defects by measuring thermal radiation light according to the embodiment of the present disclosure. Welding failure means that the molten portion 38 penetrates the workpiece 20 from the laser irradiation surface (upper surface on the paper surface in FIG. 5) to the back surface (lower surface on the paper surface in FIG. 5). It becomes the melting part 39, and refers to the case where the through melting part 39 is exposed on the back surface, or the case where the spatter 40 is generated from the melting part 38. FIG. 5 shows a state in which the through-melting portion 39 penetrates the workpiece 20 and spatter 40 is generated. In FIG. 5, the code 41 indicates the heat radiated light from the back surface of the workpiece 20, and the code 42 indicates the heat radiated light from the sputtering 40.

FIG. 6 is a diagram for explaining an example of the correlation between the amount of heat radiation and the welding quality according to the embodiment of the present disclosure, and the heat radiation amount-time diagram and the laser output-time diagram are shown up and down. It is a figure which shows side by side.

For example, when the work piece 20 is irradiated with the laser beam 26 having a constant output for a certain period of time as in the laser output profile 43 shown in FIG. 6, the appropriate heat radiation profile 44 at the time of melting is as shown in FIG. Become. Specifically, the amount of thermal radiation increases as the temperature of the laser irradiation unit rises. Then, when the temperature of the laser irradiation unit rises and the balance between the amount of heat input due to the laser irradiation and the heat escape from the laser irradiation unit to the periphery due to the heat conduction of the workpiece becomes equal, the temperature of the laser irradiation unit rises. It is stable (constant) and the amount of heat radiation is also constant. After that, when the laser irradiation is completed, the temperature of the laser irradiation unit decreases, and the amount of thermal radiation decreases as the temperature of the laser irradiation unit decreases.

On the other hand, in the thermal radiation profile 45 at the time of welding defect occurrence, a peak 46 of the thermal radiation amount at the time of occurrence of the fusion defect and a reduction portion 47 of the thermal radiation amount at the time of through welding occur. By detecting such a peak 46 and a reduced portion 47 of the amount of heat radiation, it can be determined that a defect has occurred during laser welding.

Specifically, when the peak 46 is detected, it can be determined that the sputtering 40 has occurred. As an example of poor welding, as shown in FIG. 5, when spatter 40 is generated, the spatter 40 protruding from the molten portion becomes a source of thermal radiant light different from that of the molten portion, and is emitted from the spatter 40. The synchrotron radiation 42 is incident on the condenser lens 25, reflected by the first reflection mirror 27, and then detected by the synchrotron radiation sensor 29. As a result, as shown in FIG. 6, an increase in the amount of heat radiation light can be seen as compared with the heat radiation profile 44 at the time of proper melting. Further, since the sputtering 40 is scattered at a high speed, the sputter 40 momentarily jumps out of the field of view of the condenser lens 25, so that the increase in the detected thermal radiation becomes the peak shape peak 46. In the embodiment of the present disclosure, when spatter 40 is generated, the thermal radiant light intensity is increased by about 2 to 5 times with respect to the appropriate thermal radiant light intensity at the time of melting, and the peak derived from the spatter is observed. 46 were observed.

On the other hand, as another example of welding failure, when the molten portion penetrates from the laser irradiation surface to the back surface to become the penetrating molten portion 39, and the fused portion is exposed on the back surface of the laser irradiation surface, the thermal synchrotron radiation 30 A part is also ejected from the back surface that penetrates. Therefore, the amount of thermal radiation light incident on the condenser lens 25 and detected by the thermal radiation light sensor 29 is reduced, and as shown in FIG. 6, 10 to 50% of the thermal radiation light intensity at the time of appropriate welding. A reduced portion 47 with reduced thermal radiant light intensity is generated.

Generally, the generation of spatter 40 and the penetration of the molten portion 38 are caused by a rapid increase in the amount of heat input. When the laser beam is irradiated with a constant output as in the laser output profile 43 shown in FIG. 6, when the sputtering 40 occurs and when the melting portion 38 penetrates (when the penetrating melting portion 39 occurs). ), It is highly possible that the laser irradiation portion of the workpiece 20 is not in a normal state. Therefore, when a fluctuation in the amount of heat radiation (peak 46 or reduction portion 47) is detected, it can be inferred that the surface of the workpiece 20 is dirty, scratched, or has foreign matter adhered to it.

Further, when a plurality of workpieces 20 are overlapped, laser light 26 is irradiated from the overlapping direction, and the plurality of workpieces 20 are welded to each other, the plurality of workpieces 20 are overlapped with each other. It is better to make them adhere to each other. This is because if there is a gap between the workpieces 20 due to deformation due to heat during laser welding or distortion of the workpiece 20 before machining, there are problems such as unjoined workpieces 20 and insufficient strength of the joint. Is generated. When there is a gap between the workpieces 20, the molten portion 38 penetrates one of the workpieces 20 and escapes to the gap, and a decrease in thermal synchrotron radiation intensity can be seen. Therefore, in the case of lap welding, it is possible to estimate the occurrence of gaps between the workpieces 20 due to the decrease in thermal radiation light intensity.

FIG. 7 shows a schematic diagram for explaining the vibration detection method according to the embodiment of the present disclosure. Generally, during laser welding, vibration 48 is generated from the molten portion 38 due to melting of the workpiece 20 or movement (flow) of metal generated by the melting. The vibration 48 generated in the molten portion 38 propagates through the workpiece 20 and the atmosphere, and spreads from the molten portion 38. A microphone, an AE sensor, a laser ultrasonic sensor 31, or the like may be used to detect the vibration 48. In the embodiment of the present disclosure, a laser ultrasonic sensor 31 is used, and vibration 48 propagating through the workpiece 20 is detected by using the laser ultrasonic laser 32 oscillated from the laser ultrasonic sensor 31. Further, in the embodiment of the present disclosure, the fixed point of the workpiece 20 is irradiated with the laser beam 26, but the workpiece 20 and the lens barrel 23 are attached to a drive device such as a stage or a robot, or a galvano mirror or the like. The laser welding may be performed by scanning the laser irradiation portion and performing continuous welding. In this case, it is preferable to detect the vibration 48 by the laser ultrasonic sensor 31 because it is easy to cope with continuous welding.

FIG. 8 is a schematic diagram for explaining a method for detecting welding defects by measuring vibration according to the embodiment of the present disclosure. Welding failure means that the molten portion generated in the laser irradiation portion of the workpiece 20 penetrates from the laser irradiation surface to the back surface to become the penetration melting portion 39, and the molten portion is exposed on the back surface of the laser irradiation surface or melts. It refers to the case where spatter 40 is generated from the portion.

FIG. 9 is a diagram for explaining an example of the correlation between vibration and welding quality according to the embodiment of the present disclosure, in which a vibration gain-time diagram and a laser output-time diagram are shown side by side. It is a figure. For example, when the workpiece 20 is irradiated with the laser beam 26 having a constant output for a certain period of time as in the laser output profile 43, the vibration gain profile 50 of the appropriate melting portion 38 can be used to expand and deepen the melting portion 38. Along with this, the vibration gain increases. When the temperature of the laser irradiation part rises and the amount of heat input by laser irradiation and the balance of heat escape from the laser irradiation part to the surroundings due to heat conduction of the work piece become equal, the melting becomes stable and melts. When the size and depth of the portion 38 are stable (constant), the vibration gain is also constant. After that, when the laser irradiation is completed, the size of the molten portion 38 is reduced and the vibration gain is reduced.

On the other hand, in the vibration gain profile 51 when a welding defect occurs, the vibration gain fluctuates when a melting defect occurs. For example, when the sputtering 40 is generated, the peak 52 of the vibration gain derived from the sputtering 40 is generated, and when the through-melt portion 39 is formed on the workpiece 20, the vibration gain is reduced due to the generation of the through-melt portion 39. 53 occurs. Therefore, by detecting the peak 52 of the vibration gain and the reduction portion 53 of the vibration gain, it is possible to determine the occurrence of defects during laser welding.

As an example of welding failure, as shown in FIG. 8, when sputtering 40 occurs, the state of vibration 49 changes because the sputtering 40 protruding from the molten portion disturbs the molten state of the molten portion, and the vibration gain changes. To do. Further, since the sputtering 40 jumps out of the molten portion at high speed, the molten portion returns to a stable state again after the scattering of the sputtering 40, so that the detected increase in vibration gain shows a peak shape. In the embodiment of the present disclosure, only the vibration gain is shown, but the frequency analysis of the obtained vibration may be performed, and when the sputtering 40 occurs, the state of the molten portion changes instantaneously, which is appropriate. A frequency different from the vibration seen in a molten state is detected.

As another example of welding failure, as shown in FIG. 5, the fused portion penetrates from the laser irradiation surface to the back surface to form a penetrating molten portion 39, and when the fused portion is exposed on the back surface of the laser irradiation surface, it vibrates. Since the concave portion at the bottom of the molten portion, which is the main source of the above, is eliminated, the vibration gain is reduced, and the reduced portion 53 of the vibration gain at the time of penetration can be seen.

In general, the generation of spatter 40 and the generation of through-melt portion 39 are caused by a rapid increase in the amount of heat input. When the laser beam is irradiated with a constant output as in the laser output profile 43 shown in FIG. 9, when sputtering 40 or penetration of the molten portion (penetrating molten portion 39) occurs, the laser of the workpiece 20 There is a high possibility that the condition of the irradiated part is not normal. Therefore, when the fluctuation of the vibration gain is measured, it can be inferred that the surface of the workpiece 20 is dirty, scratched, or has foreign matter adhered to it.

Further, in the case of superimposing the plurality of workpieces 20 and irradiating the laser beam 26 from the superposed direction to weld the plurality of workpieces 20, the plurality of workpieces 20 are overlapped with each other. It is better to make them stick together. This is because if there is a gap between the workpieces 20 due to deformation due to heat during laser welding or distortion of the workpiece 20 before machining, there are problems such as unjoined workpieces 20 and insufficient strength of the joint. Is generated. When there is a gap between the workpieces 20, the molten portion 38 penetrates the workpiece 20 and comes out to the gap, and a decrease in vibration gain can be seen. Therefore, in the case of lap welding, it is possible to estimate the occurrence of gaps between the workpieces 20 due to the decrease in vibration gain.

FIG. 10 shows a schematic diagram for explaining a method of detecting the generation of fume by measuring visible light according to the embodiment of the present disclosure. FIG. 11 is a schematic diagram for explaining a method of detecting the generation of plasma by measuring visible light according to the embodiment of the present disclosure. In the embodiment of the present disclosure, a He-Ne laser is used as the visible light source 33. The visible light sensor 34 includes a spectroscopic function using a diffraction grating (not shown) and a sensor (not shown) such as a CCD image sensor that detects the dispersed light.

12A to 12C are diagrams for explaining an example of the correlation between visible light and welding quality according to the embodiment of the present disclosure, respectively, and are a visible light amount-time diagram and a laser output-time diagram. It is a figure which shows and side by side. 12A shows the appropriate melting time, FIG. 12B shows the fume generation, and FIG. 12C shows the visible light amount-time diagram at the time of plasma generation.

In the embodiments of the present disclosure, the first visible light profile 56,58,60 and the second visible light profile 57,59,61 are visible light-time when a specific wavelength is extracted from the spectroscopic light. It is a diagram. The first visible light amount profiles 56, 58, 60 show changes in the light amount (visible light amount) detected by the visible light sensor 34 when a He-Ne laser is used as the visible light source 33.

The light used for the measurement of the second visible light amount profiles 57, 59, 61 (second visible light) may have a visible light wavelength and a wavelength different from that of the He-Ne laser. The second visible light is better as the difference from the wavelength of the He-Ne laser, which is the first visible light, is larger, and visible light having a blue wavelength to a green wavelength is desirable. In the embodiment of the present disclosure, green light of 532 nm was used as the second visible light, and the second visible light amount profile was detected.

As shown in FIG. 12A, the visible light amount of the visible light laser 35 from the visible light source 33 does not change with the passage of time in FIG. 3 at the time of appropriate melting, and the first visible light amount profile 56 at the time of appropriate melting is Shows constant visible light intensity. Similarly, the second visible light profile 57 does not change over time.

On the other hand, as shown in FIG. 10, when the workpiece 20 is irradiated with the laser beam 26 and the fume 54 (vapor) of the workpiece 20 is generated above the molten portion 38, the first visible light amount profile is obtained. 58 has a profile as shown in FIG. 12B. That is, since the fume 54 blocks not only the laser light 26 but also the visible light laser 35, a decrease in the visible light amount is observed in the first visible light amount profile 58. On the other hand, no change is observed in the second visible light amount profile 59.

When the fume 54 is generated, the fume 54 blocks the laser beam 26, so that the amount of heat input by the laser beam 26 to the workpiece 20 is reduced, resulting in insufficient melting (insufficient melting). Therefore, from the decrease in the visible light amount of the first visible light amount profile 58, it is possible to infer the occurrence of welding defects such as unbonded and insufficient joint strength.

Next, as shown in FIG. 11, by irradiating the workpiece 20 with the laser beam 26, the fume (steam) of the workpiece 20 generated upward in the molten portion 38 absorbs the laser beam 26. , It becomes plasma, and plasma 55 may be generated on the molten portion 38. In this case, the plasma 55 shines white. Therefore, as shown in FIG. 12C, an increase in the amount of visible light is observed in both the first visible light amount profile 60 and the second visible light amount profile 61.

Further, as described with reference to FIG. 11, when the plasma 55 is generated, the laser beam 26 is absorbed, the amount of heat input by the laser beam 26 to the workpiece 20 decreases, and insufficient melting (insufficient melting). ) Occurs. Therefore, the first visible light amount profile 60 and the second visible light amount profile 61 predict the occurrence of welding defects such as unbonded and poorly joined, and are caused by the influence of the high temperature plasma 55 on the workpiece 20. , It is possible to predict the occurrence of defects such as unnecessary surface melting. Specifically, when the plasma 55 is generated, an increase in the amount of visible light is observed in both the first visible light amount profile 60 and the second visible light amount profile 61, as described with reference to FIG. 12C. Therefore, when both an increase in the visible light amount of the first visible light amount profile 60 and an increase in the visible light amount of the second visible light amount profile 61 are detected, insufficient melting, unbonding, poor bonding, or unnecessary It is possible to infer the occurrence of surface melting.

Generally, in order to prevent the laser beam 26 from being blocked by the fume 54, the fume 54 is removed by wiping the assist gas. Therefore, when the generation of the fume 54 is detected, that is, when the fluctuation of the visible light amount of the first visible light amount profile 60 or the second visible light amount profile 61 is detected, the assist gas type, the assist gas flow velocity, and the assist gas It can be inferred that either the pressure or the spraying position of the assist gas is inappropriate. Further, at this time, generally, a dust collecting intake duct is used to collect dust scattered by blowing the assist gas (not shown). Therefore, when the generation of the fume 54 is detected, it can be predicted that either the position of the dust collecting intake duct or the dust collecting intake flow velocity is inappropriate.

13A to 13C are schematic views for explaining how to return the OCT signal according to the keyhole shape of the embodiment of the present disclosure, respectively. FIG. 14 is a diagram for explaining an example of the correlation between the OCT signal intensity and the welding quality according to the embodiment of the present disclosure, and the OCT signal intensity-time diagram and the laser output-time diagram are up and down. It is a figure which shows side by side in.

FIG. 13A shows a state in which proper welding is performed. In the case of proper welding, the OCT measurement laser 37 from the OCT sensor 36 in FIG. 3 is reflected by the second reflection mirror 28, irradiated to the molten portion 38 through the condenser lens 25, and reflected at the bottom of the keyhole. Then, the reflected light 64 is reflected again by the second reflection mirror 28 through the condenser lens 25, and returns to the OCT sensor 36. At this time, as shown in the appropriate melting OCT signal intensity profile 65 of FIG. 14, the stable OCT signal intensity can be maintained, so that the keyhole depth can be measured.

FIG. 13B shows a case where the keyhole 62 penetrates the workpiece 20. At this time, since the OCT measurement laser 37 passes through the penetrating keyhole 62, the OCT signal intensity is lowered as in the OCT signal intensity 66 at the time of welding failure shown in FIG. FIG. 13C shows the case where the keyhole is bent. At this time, when the OCT measurement laser 37 reflects at the tip of the bent keyhole 63, it does not reflect toward the laser incident surface side of the workpiece 20, but reflects toward the bottom of the keyhole 63. Therefore, the OCT signal strength decreases as shown in FIG. 14 when the welding is defective, as in the OCT signal strength 66. Bending of the keyhole 63 occurs when laser welding is performed by scanning the laser irradiation portion to form a continuous welded portion, and the keyhole 63 shows a bent shape when the scanning speed is excessive. When the scanning speed is excessive, a state in which a predetermined penetration depth cannot be formed or a space remains at the bending tip of the keyhole 63 to generate a blowhole. Therefore, from the decrease in OCT signal strength, it is possible to infer defects such as bending of the keyhole, insufficient penetration of welding, or occurrence of blowhole.

FIG. 15 shows a structural diagram of the convolutional neural network 67 applied to the evaluation model 11 in the embodiment of the present disclosure. First, the state variable 12 is input to the input layer 68, and in the convolution layer 69, for example, in the heat radiation light amount-time diagram of FIG. 6, state variables such as the inclination and peak intensity of the heat radiation profile 45 when a welding defect occurs. The operation of capturing features with 12 local data is repeated at various data positions. As a result, the convolutional layer 69 extracts the characteristics of the entire state variable 12. Next, the pooling layer 70 is processed so that the features extracted by the convolutional layer 69 become prominent. Finally, the fully coupled layer 71 classifies using the features summarized in the pooling layer 70 and outputs the results to the output layer 72.

As described above, the convolutional neural network 67 is a model that is good at grasping and classifying the characteristics of the input data. Therefore, by applying the convolutional neural network 67 to the evaluation model 11 described with reference to FIG. 3, the ability to grasp the characteristics of the input state variable 12 based on the physical quantity 15 is improved, and the laser machining evaluation result 13 which is the output is obtained. It has the effect of greatly improving accuracy.

By the way, although omitted in FIGS. 3-5, 7, 8, 10 and 11, the laser welding apparatus 101 further includes a control unit 112. The control unit 112 has a function as a learning device of the laser machining evaluation system 1.

The control unit 112 will be described with reference to FIGS. 1 and 16. FIG. 16 is a functional block diagram of the control unit 112 according to the embodiment of the present disclosure.

As shown in FIG. 16, the control unit 112 includes a learning unit 104, a storage unit 105, and a calculation unit 106.

At the time of welding, the calculation unit 106 acquires information (physical quantity 15) acquired from each of the thermal radiation sensor 29, the visible light sensor 34, and the OCT sensor 36 as a state variable 12. That is, the calculation unit 106 has a function as the input processing unit 2 shown in FIG. The calculation unit 106 outputs the laser machining evaluation result 13 using the evaluation model 11. That is, the calculation unit 106 has a function as an output processing unit 4. After the laser welding of the workpiece 20, the state variable 12 input to the calculation unit 106 at the time of laser welding and the laser processing result 16 are sequentially stored in the storage unit 105 as a data set. That is, the storage unit 105 has a function as a data set group 14. The learning unit 104 executes learning of the evaluation model 11 based on all the data sets (data set group 14) accumulated in the storage unit 105, and feeds back the learned evaluation model 11 to the calculation unit 106. That is, the learning unit 104, the storage unit 105, and the calculation unit 106 cooperate to function as the learning processing unit 3.

According to the present disclosure, by updating the evaluation model while accurately evaluating the welding quality using the evaluation model, it is possible to predict welding abnormalities regardless of skill level. Therefore, it is expected that the number of defects will be reduced by early response to abnormalities and the productivity will be improved by reducing the downtime of the equipment.

1 Laser processing evaluation system 2 Input processing unit 3 Learning processing unit 4 Output processing unit 11 Evaluation model 12 State variable 13 Laser processing evaluation result 14 Data set group 15 Physical quantity 16 Laser processing result 17 Error 18 Loss function 19 Optimization algorithm 20 Processing Object 21 Laser oscillator 22 Fiber for laser transmission 23 Lens tube 24 Collimating lens 25 Condensing lens 26 Laser light 27 First reflection mirror 28 Second reflection mirror 29 Thermal radiation sensor 30 Thermal radiation 31 Laser ultrasonic sensor 32 Laser ultrasonic Laser 33 Visible light source 34 Visible light sensor 35 Visible light laser (inspection light)
36 OCT sensor 37 Laser for OCT measurement 38 Melting part 39 Penetrating melting part 40 Spatter 41 Thermal radiation from the back surface of the workpiece 20 42 Thermal radiation from the sputtering 40 Laser output profile 44 Thermal radiation profile at the time of appropriate melting 45 Thermal radiation profile when melting failure occurs 46 Peak of thermal radiation profile derived from spatter 47 Decrease in the amount of thermal radiation light when penetrating and melting 48 Vibration from the molten part 49 Vibration when spatter occurs 50 Vibration gain profile during proper welding 51 Vibration gain profile when welding failure occurs 52 Radiation gain peak due to spatter 53 Decrease in vibration gain during penetration welding 54 Fume 55 Plasma 56 First visible light amount profile at appropriate melting 57 First at appropriate melting Second visible light amount profile 58 First visible light amount profile at the time of fume generation 59 Second visible light amount profile at the time of fume generation 60 First visible light amount profile at the time of plasma generation 61 Second visible light amount profile at the time of plasma generation 62 Keyhole 63 Bending keyhole 64 Reflected light from the tip of the keyhole 65 OCT signal strength during proper welding 66 OCT signal strength when welding failure occurs 67 Folded neural network 68 Input layer 69 Folded layer 70 Pooling layer 71 Fully coupled Layer 72 Output layer 101 Laser welding device 104 Learning unit 105 Storage unit 106 Calculation unit 112 Control unit

Claims (9)

It is a laser processing system consisting of a laser processing device that welds the work piece by condensing the laser light on the surface of the work piece, and a learning device that executes learning for the laser processing device. hand,
The learning device is
An input processing unit in which a state variable including at least one laser condition related to laser machining and a calculation result calculated based on at least one measured value measured during execution of the laser machining is input.
A learning unit that executes learning to update an evaluation model for deriving an evaluation of the laser machining based on the state variable is provided.
Laser processing system. At least one measured value measured during the execution of the laser machining is the temperature of the machining part, the amount of heat radiated light from the machining section, the amount of visible light of the machining section, the vibration amount of the workpiece, and the OCT signal of the machining section. At least one of
The laser processing system according to claim 1. The laser processing device is
A thermal radiation sensor that measures the thermal radiation emitted from the processing unit during the execution of the laser machining is included.
The measured value is the amount of heat radiation from the processed portion.
The laser processing system according to claim 2. The laser processing device is
Includes a visible light sensor that detects visible light emitted from a visible light source installed near the machined portion during the execution of the laser machining.
The measured value is the amount of visible light of the visible light.
The laser processing system according to claim 2 or 3. The laser processing device is
Includes a vibration sensor that detects vibration propagating through the workpiece.
The measured value is the amount of vibration of the work piece.
The laser processing system according to any one of claims 2 to 4. The laser processing device is
Includes an OCT sensor that measures the depth of keyholes formed in the machined area
The measured value is an OCT signal of the processed portion.
The laser processing system according to any one of claims 2 to 5. At least one laser condition related to the laser processing includes laser wavelength, spot diameter (laser intensity distribution profile), laser output profile (laser output, irradiation time, output change), assist gas amount, assist gas flow velocity, assist gas supply position, and so on. At least one of the assist gas supply direction, the dust collection intake duct position, and the dust collection intake flow velocity.
The laser processing system according to any one of claims 1 to 6. A learning device that executes learning for a laser processing device that welds the work piece by condensing the laser light on the surface of the work piece.
An input processing unit in which a state variable including at least one laser condition related to laser machining and a calculation result calculated based on at least one measured value measured during execution of the laser machining is input.
A learning unit that executes learning to update an evaluation model for deriving an evaluation of the laser machining based on the state variable is provided.
Learning device. It is a learning method of a learning device that executes learning for a laser processing device that welds the work piece by condensing the laser light on the surface of the work piece.
An input step in which a state variable including at least one laser condition related to laser machining and a calculation result calculated based on at least one measured value measured during execution of the laser machining is input.
It comprises a learning step of performing learning to update an evaluation model that derives an evaluation of the laser machining based on the state variables.
Learning method of learning device.

Images