Patent Application
20200406392
The present application claims priority under 35 U.S.C. § 119 to Japanese Patent Application No. 2019-121885, filed Jun. 28, 2019. The contents of this application are incorporated herein by reference in their entirety.
The embodiments disclosed herein relate to an evaluation apparatus, an evaluation method, an evaluation system, and a non-transitory computer-readable storage medium.
Some robots known in the art make motions by driving a plurality of joints of the robots. Such robot includes an end effector mounted on the leading end of the robot. The end effector varies depending on the application in which the robot is used, such as machining, welding, and holding of a workpiece. Using the end effector, the robot performs these kinds of work.
JP 2014-198345A discloses an end effector that serves as an inspection device for inspecting a welding spot. Specifically, the inspection device includes a laser head capable of changing a shape defined by a track of laser radiation. While performing laser welding on the workpiece, the inspection device inspects welding spots on the workpiece using the laser head.
According to one aspect of the present disclosure, an evaluation apparatus includes an obtaining circuit, a detecting circuit, a generating circuit, and an evaluating circuit. The obtaining circuit is configured to obtain track information indicating positions of a laser target spot in relation to time. The laser target spot is made by a laser beam on a workpiece for welding. The detecting circuit is configured to detect intensity information indicating intensities of return light in relation to time. The return light is generated by radiating the laser beam onto the workpiece. The generating circuit is configured to generate intensity distribution information indicating the intensities corresponding to the positions by correlating the track information and the intensity information based on time. The evaluating circuit is configured to evaluate a welding state of the welding based on the intensity distribution information.
According to another aspect of the present disclosure, an evaluation method includes obtaining track information indicating positions of a laser target spot in relation to time, the laser target spot being made by a laser beam on a workpiece for welding; detecting intensity information indicating intensities of return light in relation to time, the return light being generated by radiating the laser beam onto the workpiece; correlating the track information and the intensity information based on time to generate intensity distribution information indicating the intensities corresponding to the positions; and evaluating a welding state of the welding based on the intensity distribution information.
According to another aspect of the present disclosure, an evaluation system includes a laser head, a robot, a controller, and an evaluation apparatus. The laser head is configured to change a shape of a track of a laser target spot made by a laser beam on a workpiece for welding. The robot is configured to move the laser head relative to the workpiece. The controller is configured to control a motion of the laser head and a motion of the robot. The evaluation apparatus is configured to evaluate a welding state of the welding. The evaluation apparatus includes an obtaining circuit, a detecting circuit, a generating circuit, and an evaluating circuit. The obtaining circuit is configured to obtain track information indicating positions of the laser target spot in relation to time. The detecting circuit is configured to detect intensity information indicating intensities of return light in relation to time. The return light is generated by radiating the laser beam onto the workpiece. The generating circuit is configured to generate intensity distribution information indicating the intensities corresponding to the positions by correlating the track information and the intensity information based on time. The evaluating circuit is configured to evaluate a welding state of the welding based on the intensity distribution information.
According to the other aspect of the present disclosure, a non-transitory computer-readable storage medium stores a program for causing a computer to execute processing. The processing includes obtaining track information indicating positions of a laser target spot in relation to time, the laser target spot being made by a laser beam on a workpiece for welding; detecting intensity information indicating intensities of return light in relation to time, the return light being generated by radiating the laser beam onto the workpiece; correlating the track information and the intensity information based on time to generate intensity distribution information indicating the intensities corresponding to the positions; and evaluating a welding state of the welding based on the intensity distribution information.
A more complete appreciation of the present disclosure and many of the attendant advantages thereof will be readily obtained as the same becomes better understood by reference to the following detailed description when considered in connection with the accompanying drawings, wherein:
By referring to the accompanying drawings, an evaluation apparatus, an evaluation method, an evaluation system, and an evaluation program according to embodiments of the present disclosure will be described in detail below. It is noted that the following embodiments are provided for example purposes only and are not intended for limiting purposes.
As used herein, the term “uniform” means, in a broad sense, exactly uniform or approximately uniform within some tolerance from exactly uniform. Also as used herein, the term “orthogonal” means, in a broad sense, exactly orthogonal or approximately orthogonal within some tolerance from exactly orthogonal. Also as used herein, the term “perpendicular” means, in a broad sense, exactly perpendicular or approximately perpendicular within some tolerance from exactly perpendicular. Also as used herein, the term “parallel” means, in a broad sense, exactly parallel or approximately parallel within some tolerance from exactly parallel. That is, these terms are used with production-related and installation-related tolerances and errors taken into consideration.
By referring to
The head 100 illustrated in
The robot 10 is capable of moving the head 100 along a working line set on the workpiece W. As used herein, the term “working line” refers to an imaginary line set along extension directions in which a working region set on the workpiece W extends. An example (not illustrated) is that a spiral radiation track P is drawn on the workpiece W by moving the head 100 along the working line.
The controller 20 controls motions of the head 100 and motions of the robot 10. Specifically, the controller 20 outputs motion commands to the head 100 and the robot 10, and the head 100 and the robot 10 make motions based on the motion commands output from the controller 20.
Conventional welding state evaluation is performed by: detecting the intensity of an optical feedback of a laser beam returning from the workpiece W (examples of the optical feedback, which is also referred to as return light, include plasma light and infrared light); and analyzing time-series data (waveform data) of the intensity.
In the conventional welding state evaluation, however, the time-series data of the intensity of the optical feedback contains external disturbance noise such as welding fumes, spatter particles, and molten pool vibrations. This makes it necessary to exclude the external disturbance noise from the time-series data so as to obtain values for analysis purposes, and it is difficult to obtain these values accurately. Thus, this conventional evaluation method has room for improvement in evaluation accuracy.
In light of the considerations above, the evaluation method according to this embodiment uses a combination of time-series data of optical feedback intensity and laser track information to generate a spatial intensity distribution regarding an optical feedback 200, which returns from the workpiece W (this intensity distribution will be hereinafter referred to as “intensity distribution information”). Then, the evaluation method performs welding state evaluation based on the generated intensity distribution information.
First, the evaluation method according to this embodiment obtains track information showing a track of a laser target spot on the workpiece W to which a laser beam is radiated (step S01 in
The track information may be obtained, for example, based on a motion command output from the controller 20. Another possible example is to obtain the track information based on an output from an image sensor that takes an image of the workpiece W.
Then, the evaluation method according to this embodiment detects intensity information showing, in a time-series order, intensities of the optical feedback 200 of the radiated laser beam (step S02 in
Next, the evaluation method according to this embodiment generates intensity distribution information based on the track information and the time-series intensity information (step S03 in
Next, the evaluation method according to this embodiment analyzes the generated intensity distribution information to evaluate the welding state, which indicates how the workpiece W is welded by the laser beam. Specifically, an image of the intensity distribution information is subjected to image recognition and classified into a normal welding state or an abnormal welding state. When the welding state has been determined as abnormal, the type of the abnormal welding state is determined.
Thus, the evaluation method according to this embodiment evaluates the welding state of the workpiece W more spatially than when only waveform data, which is time-series data, is analyzed. When the workpiece W has a welding defect such as boring, the welding defect in many cases exists over a plurality of coordinates on the workpiece W. The evaluation method according to this embodiment, which evaluates the welding state spatially using the intensity distribution information, increases the accuracy of evaluation as to whether there is a welding defect, such as boring, on the workpiece W.
In the evaluation method according to this embodiment, it is possible to use machine learning to generate or update an evaluation standard on which the welding state evaluation using the intensity distribution information is based. Using machine learning ensures that an image of the intensity distribution information is classified based on the welding state at a higher level of accuracy, which will be described later.
By referring to
As illustrated in
A non-limiting example of the robot 10 is a six-axis vertical multi-articular robot, and the head 100 is mounted on the leading end portion of the robot 10.
The controller 20 includes a robot controller 21 and a head controller 22. The robot controller 21 controls motions of the robot 10. The head controller 22 is a separate controller separate from the robot controller 21, and controls motions of the head 100. Specifically, the robot controller 21 obtains a setting value set for the head 100 from a terminal device connected to the robot controller 21 via a wire or wirelessly, generates a motion command for the head 100 based on the obtained setting value, and outputs the motion command to the head controller 22.
The setting value of the head 100 is information that defines the radiation track P in a coordinate system fixed to the head 100. When the robot 10 and the head 100 cooperate to perform laser welding, the position and orientation of the head 100 are subject to change by the robot 10. That is, the coordinate system fixed to the head 100 itself is movable. In this case, the robot controller 21 calculates the position and orientation of the head 100 based on a known posture of the robot 10. Then, based on the position and orientation that have been calculated, the robot controller 21 converts a coordinate point corresponding to the laser target spot specified by the setting value of the head 100 into a coordinate point corresponding to a relative laser target spot that is a laser target spot set with the movement of the robot 10 taken into consideration. Then, the robot controller 21 outputs to the head controller 22 a motion command that includes the coordinate point corresponding to the relative laser target spot.
The robot controller 21 also makes instructions, such as a laser intensity instruction, to the head controller 22.
The head controller 22 receives the motion command for the head 100 from the robot controller 21. Then, based on the motion command, the head controller 22 controls the mechanism (including galvanometer mirrors) of the head 100. Specifically, the head controller 22 is connected to the head 100 via a communication cable 221 and controls the drive shafts of the galvanometer mirrors by making communication with the head 100 via the communication cable 221, such as performing high speed serial communication and making a pulse width modulation (PWM) motor command. The head controller 22 is also connected to the laser oscillator 120 via a communication cable 222. The head controller 22, based on a command from the robot controller 21, transmits an analogue signal to the laser oscillator 120 via the communication cable 222. The analogue signal is a command that causes the laser oscillator 120 to adjust the laser intensity synchronously with a motion command for the head 100. The laser oscillator 120, based on the command from the head controller 22, adjusts the laser intensity and outputs a laser beam to the head 100 via an optical fiber 121.
While in this embodiment the robot controller 21 and the head controller 22 are separate from each other, it is possible to incorporate the functions of the head controller 22 in, for example, the robot controller 21.
The evaluation apparatus 50 generates intensity distribution information, and evaluates the welding state of the workpiece W based on the generated intensity distribution information. The evaluation apparatus 50 is connected to the robot controller 21 and the head 100. The evaluation apparatus 50 is also connected to the learning apparatus 60.
The learning apparatus 60 generates an evaluation standard using machine learning. The evaluation standard is used as a basis of evaluation of the welding state of the workpiece W based on the intensity distribution information.
While in
The laser oscillator 120 outputs a laser beam to the head 100 via the optical fiber 121.
The head 100 changes the laser target spot based on a motion command from the head controller 22. By referring to
As illustrated in
The first incident mirror 101 receives a laser beam entering the head 100 via the optical fiber 121, and reflects the laser beam into the mirror device 102.
The mirror device 102 includes a first galvanometer mirror and a second galvanometer mirror. The first driver 103 makes the first galvanometer mirror swing about X axis. The second driver 104 makes the second galvanometer mirror rotate about Y axis. The drive controller 105 controls motions of the first driver 103 and the second driver 104 based on a motion command input from the head controller 22.
By driving the mirror device 102 using the first driver 103 and the second driver 104 in the above manner, the head 100 changes the laser beam direction. This enables the head 100 to draw, on the workpiece W, a radiation track P having a shape that is based on a motion command.
After the laser beam from the laser oscillator 120 has entered the head 100, the half mirror 106 allows the laser beam to pass through the half mirror 106 while at the same time reflecting the optical feedback 200 into the second incident mirror 107. The second incident mirror 107 reflects the optical feedback 200 into the intensity sensor 108.
The intensity sensor 108 receives the optical feedback 200 and detects the intensity of the optical feedback 200. Then, the intensity sensor 108 outputs the detected intensity to the evaluation apparatus 50.
As described later, the optical feedback 200 comes not only from one point on the laser target spot but also from a vicinity region R, which is a region centered around the laser target spot.
By referring to
An example configuration of the evaluation apparatus 50 will be described first. As illustrated in
The evaluation apparatus 50 includes a computer and various circuits. The computer includes a central processing unit (CPU), a read only memory (ROM), a random access memory (RAM), a hard disk drive (HDD), and input-output ports.
The CPU of the computer reads an evaluation program stored in the ROM and executes the evaluation program to serve the functions of the obtaining part 51a, the detecting part 51b, the generating part 51c, and the evaluating part 51d of the control part 51. It is to be noted that at least one or all of the obtaining part 51a, the detecting part 51b, the generating part 51c, and the evaluating part 51d may be implemented by hardware such as Application Specific Integrated Circuit (ASIC), Graphics Processing Unit (GPU), and Field Programmable Gate Array (FPGA).
The storage part 52 corresponds to the RAM and/or the HDD. The RAM and the HDD are capable of storing the track information 52a, the intensity information 52b, the intensity distribution information 52c, and the evaluation model 52d. It is to be noted that the evaluation apparatus 50 may obtain the above-described program and various kinds of information from another computer connected to the evaluation apparatus 50 through a wired or wireless network or from a portable recording medium.
The obtaining part 51a obtains from the robot controller 21 a motion command for the robot 10 and the head 100, and generates the track information 52a based on the obtained motion command.
Specifically, based on the motion command for the robot 10, the obtaining part 51a converts relative laser target spot information included in the motion command for the head 100 into laser target spot information in the coordinate system fixed to the workpiece W. Then, the obtaining part 51a generates the track information 52a by correlating the obtained laser target spot information with, for example, the time at which the motion command was obtained. Then, the obtaining part 51a stores the generated track information 52a in the storage part 52.
From the intensity sensor 108 of the head 100 (see
The generating part 51c generates the intensity distribution information 52c by correlating the track information 52a and the intensity information 52b with each other based on time. As described above, the intensity distribution information 52c is image information made up of pixel values each corresponding to a different coordinate that is included in the coordinate system fixed to the workpiece W and that is assigned a different intensity of the optical feedback 200.
By referring to
There is a time lag between the timing at which a laser beam is radiated to the workpiece W and the timing at which the optical feedback 200 of the laser beam is detected. The generating part 51c, therefore, may take the time lag into consideration when the generating part 51c correlates the track information 52a and the intensity information 52b with each other based on time.
Specifically, as illustrated in
Thus, when the generating part 51c generates the intensity distribution information 52c, the generating part 51c may correlate the track information 52a as of a time point (first time point) with the intensity information 52b as of a time point (second time point) that is behind the first time point. This ensures that more accurate intensity distribution information is obtained.
Referring again to
When the welding state is an abnormal welding state, the evaluating part 51d determines the type of the abnormal welding state. Examples of the type of the abnormal welding state include, but are not limited to, “boring”, “burning-through”, “bead meandering”, and “bead shape irregularity”. “Boring” refers to a state in which there is a hole penetrating the workpiece W. “Burning-through” refers to a state in which a joining piece (base material piece) of the workpiece W is melted away from the welding bead. “Bead meandering” refers to a state in which the welding bead meanders laterally relative to the radiation track P. “Bead shape irregularity” refers to a state in which the shape of the welding bead has a particularly distinctive part (a partial distinctiveness such as in bead width and extra-banking height).
The evaluating part 51d, based on the evaluation model 52d stored in the storage part 52, classifies an image of the intensity distribution information 52c on a welding state basis. Specifically, the evaluating part 51d classifies an image of the intensity distribution information 52c as a normal welding state or an abnormal welding state. Upon determining that the image is an abnormal welding state, the evaluating part 51d classifies the type of the abnormal welding state. A non-limiting example of the evaluation model 52d is a calculation model used in the above-described image recognition processing, that is, a calculation model used to classify an image as a normal welding state or an abnormal welding state. The evaluation model 52d is generated by the learning apparatus 60, described later.
The evaluating part 51d outputs the intensity distribution information 52c and the evaluated welding state to, for example, an external device 150, which is connected to the evaluation apparatus 50. The evaluating part 51d also outputs the intensity distribution information 52c to the learning apparatus 60.
An example configuration of the learning apparatus 60 will be described. The learning apparatus 60 includes a control part 61 and a storage part 62. The control part 61 includes an input receiving part 61a, a correct value assigning part 61b, and a machine learning part 61c (which is a non-limiting example of the machine learning circuit recited in the appended claims). The storage part 62 stores the intensity distribution information 52c and learning information 62a.
The learning apparatus 60 includes a computer and various circuits. The computer includes, for example, CPU, ROM, RAM, HDD, and input/output ports. The CPU of the computer reads an evaluation program stored in the ROM and executes the evaluation program to serve the functions of the input receiving part 61a, the correct value assigning part 61b, and the machine learning part 61c of the control part 61.
The storage part 62 corresponds to the RAM and/or the HDD. The RAM and the HDD are capable of storing the intensity distribution information 52c and the learning information 62a. It is to be noted that the learning apparatus 60 may obtain the above-described programs and various kinds of information from another computer connected to the learning apparatus 60 through a wired or wireless network or from a portable recording medium.
The input receiving part 61a obtains the intensity distribution information 52c from the evaluation apparatus 50 and stores the intensity distribution information 52c in the storage part 62.
The correct value assigning part 61b obtains the intensity distribution information 52c from the storage part 62; obtains a correct value corresponding to the classified or evaluated welding state corresponding to the intensity distribution information 52c; and assigns the correct value to the intensity distribution information 52c. In this manner, the correct value assigning part 61b selects or generates information serving as the learning information 62a (training data). Then, the correct value assigning part 61b stores the information in the storage part 62.
A non-limiting example of the correct value obtained by the correct value assigning part 61b is an evaluation performed by a human worker who actually saw the intensity distribution information 52c or the welded part of interest. In this case, the correct value assigning part 61b obtains the correct value input by the worker through an input device, such as a keyboard, connected to the learning apparatus 60. Alternatively, the correct value assigning part 61b may obtain the correct value through a network such as the Internet.
While in this embodiment the correct value assigning part 61b has been described as obtaining the intensity distribution information 52c from the evaluation apparatus 50 and generating the learning information 62a, the correct value assigning part 61b may obtain external, correct value data that correlates the intensity distribution information 52c with a correct value.
The machine learning part 61c generates an evaluation model by performing machine learning (specifically, supervised learning) using the learning information 62a stored in the storage part 62. Using the generated evaluation model, the machine learning part 61c updates the evaluation model 52d stored in the storage part 52 of the evaluation apparatus 50. In other words, the machine learning part 61c updates the parameter(s) of the evaluation model 52d stored in the storage part 52.
The evaluation model 52d may be generated using various techniques known in the machine learning art. Examples of the techniques include, but are not limited to, various deep learning techniques such as CNN (Convolutional Neural Network) and RNN (Recurrent Neural Network). Another non-limiting example technique is SVM (Support Vector Machine). It is to be noted that these techniques have been provided for exemplary purposes, and it is possible to select any other learning technique suitable for the information sought to be obtained in generating the evaluation model 52d.
While in this embodiment the learning apparatus 60 has been described as performing supervised learning, the machine learning performed by the learning apparatus 60 may be unsupervised learning. Specifically, the machine learning part 61c may perform, instead of supervised learning using the learning information 62a, unsupervised learning to extract a characteristic(s) of the welding state from the intensity distribution information 52c stored in the storage part 62. The unsupervised learning may be performed using any technique known in the art, examples including, but not limited to, cluster analysis and base analysis.
By referring to
As illustrated in
Under the circumstances, the generating part 51c may correlate the track information 52a and the intensity information 52b with each other in the manner illustrated in
Thus, an intensity of the optical feedback 200 is correlated with the plurality of coordinate points included in the region centered around the laser target spot. This ensures that obtained intensity distribution information 52c is closer to an actual welding state.
In this case, the generating part 51c may generate intensity distribution information for each one round of laser radiation, as indicated by 52c1 to 52cN in
When the evaluating part 51d generates the pieces of intensity distribution information 52c1 to 52cN (each corresponding to a different one round of laser radiation), the evaluating part 51d may evaluate the welding state based on a difference between the pieces of intensity distribution information 52c1 to 52cN (each corresponding to a different closed loop) such that the pieces of intensity distribution information 52c1 to 52cN are arranged in a time-series order.
Specifically, the evaluating part 51d may compare two pieces of intensity distribution information that are next to each other in a time-series order. For example, the evaluating part 51d may compare the intensity distribution information 52c1, which corresponds to the first round of laser radiation, with the intensity distribution information 52c2, which corresponds to the second round of laser radiation. By making this comparison, the evaluating part 51d determines whether there is a region where the decreasing rate of the intensity of the optical feedback 200 is in excess of a threshold. When there is a region where the decreasing rate of the intensity of the optical feedback 200 is in excess of a threshold, the evaluating part 51d determines that the region has the abnormality “boring”. It is to be noted that this region may be made up of a single pixel or a plurality of adjacent pixels.
Thus, a change over time is used as a subject of evaluation, which increases the level of the accuracy with which the welding state is evaluated. This configuration also ensures that if an abnormality has occurred, it is possible to identify the number of the closed loop (in the time-series order) in which the abnormality occurred.
The evaluating part 51d may also evaluate the welding state based on information obtained by superposing the pieces of intensity distribution information 52c1 to 52cN (each corresponding to a different closed loop).
Specifically, the evaluating part 51d may generate average intensity distribution information by calculating an average of the pieces of intensity distribution information 52c1 to 52cN; and evaluate the welding state based on the average intensity distribution information that has been generated. This configuration reduces external disturbance noise caused by, for example, fumes, increasing the accuracy of welding state evaluation. It is to be noted that the evaluating part 51d may calculate a weight average of the pieces of intensity distribution information 52c1 to 52cN by more heavily weighting a newer piece of intensity distribution information in a time-series order.
The evaluating part 51d may also generate summation intensity distribution information by calculating a sum of the pieces of intensity distribution information 52c1 to 52cN; and evaluate the welding state based on the summation intensity distribution information that has been generated. The summation intensity distribution information is closer to the total amount of heat applied to the weld portions of the workpiece W. Using such intensity distribution information, therefore, increases the accuracy of welding state evaluation.
The evaluating part 51d may also evaluate the welding state based on one or some of the pieces of intensity distribution information 52c1 to 52cN. For example, the evaluating part 51d may evaluate the welding state based solely on the final intensity distribution information (namely, the intensity distribution information 52cN, which corresponds to the N-th round of laser radiation). Thus, the evaluating part 51d may evaluate the welding state based on at least one piece of intensity distribution information, among a plurality of pieces of intensity distribution information corresponding to a plurality of rounds of laser radiation along an identical path.
It is to be noted that the evaluating part 51d may generate an animation image by arranging the pieces of intensity distribution information 52c1 to 52cN in a time-series order to dynamically show a change over time in the intensity distribution; and output the animation image to the external device 150 or some other device.
Thus, when the track information 52a shows more than one round of laser radiation along an identical closed loop, the generating part 51c may generate the intensity distribution information 52c1 to 52cN (each corresponding to a different round of laser radiation). In this case, the evaluating part 51d may evaluate the welding state based on the intensity distribution information 52c1 to 52cN.
As used herein, the term “identical closed loop” refers to a closed loop that is precisely or approximately the same in shape and size. If a laser beam has drawn a plurality of concentric closed loops (such as a concentric double circle), the radiation track P includes a plurality of different closed loops. These different-size closed loops are not regarded as identical closed loops.
While in the above-described embodiment and modification the radiation track P drawn has been described as a closed loop, the radiation track P, which is formed by following an identical path(s), may not necessarily be a closed loop. Other examples include, but are not limited to, a zigzag-shaped track, an S-shaped track, and a C-shaped track.
In this case, the obtaining part 51a of the evaluation apparatus 50 may obtain the image output from the image taking device 300 and generate the track information 52a based on the obtained image.
Thus, the obtaining part 51a may obtain the track information 52a not based on a motion command for the robot 10 and/or the head 100 but based on an image of the workpiece W taken by an image sensor. This makes the track information 52a closer to an actual track drawn on the workpiece W than when the track information 52a is obtained based on a motion command.
The evaluation apparatus 50 according to the above-described embodiment includes the obtaining part 51a, the detecting part 51b, the generating part 51c, and the evaluating part 51d. The obtaining part 51a obtains the track information 52a, which is regarding a laser target spot on the workpiece W to which a laser beam is radiated. The detecting part 51b detects the intensity information 52b, which shows intensities of the optical feedback 200 (of the laser beam from the workpiece W) in a time-series order. The generating part 51c generates the intensity distribution information 52c. The intensity distribution information 52c shows a distribution of the intensities over the workpiece W and correlates the track information 52a and the intensity information 52b with each other based on time. The evaluating part 51d evaluates a welding state (which indicates how the workpiece W is welded by the laser beam) based on the intensity distribution information 52c.
Thus, the intensity information 52b, which is time-series information, is combined with the track information 52a, which is spatial information. Using this combination ensures that a welding defect associated with shape, such as boring, is evaluated spatially. As a result, the welding state is evaluated at a higher level of accuracy.
The obtaining part 51a may obtain the track information 52a based on a motion command from the robot controller 21 and/or the head controller 22 (which are non-limiting examples of the controller recited in the claims). The head controller 22 controls motions of the head 100, which is capable of change the shape of the radiation track P (the head 100 is a non-limiting example of the laser head recited in the claims). The robot controller 21 controls motions of the robot 10, which moves the head 100 relative to the workpiece W. By obtaining the track information 52a based on a motion command, spatial track information is obtained at a higher level of accuracy and a lower level of processing load.
The obtaining part 51a may obtain the track information 52a based on an image of the workpiece W taken by the image taking device 300 (which is a non-limiting example of the image sensor recited in the claims).
When the track information 52a shows that the radiation track of the laser beam is formed by more than one round of laser radiation along an identical closed loop (which is a non-limiting example of the identical path recited in the appended claims), the generating part 51c may generate the intensity distribution information 52c1 to 52cN (each corresponding to a different round of laser radiation). In this case, the evaluating part 51d may evaluate the welding state based on the pieces of intensity distribution information 52c1 to 52cN (each corresponding to a different closed loop). Thus, when the radiation track of a laser beam is formed by more than one round of laser radiation, the welding state is evaluated based on an intensity distribution corresponding to each one round of laser radiation. That is, a larger number of pieces of information are input for evaluation purposes, resulting in an increased level of evaluation accuracy.
The evaluating part 51d may evaluate the welding state based on an average or a sum of the pieces of intensity distribution information 52c1 to 52cN (each corresponding to a different closed loop, which is a non-limiting example of the identical path recited in the appended claims). This ensures that the welding state is evaluated at a lower level of processing load.
The evaluating part 51d may evaluate the welding state based on a difference between the pieces of intensity distribution information 52c1 to 52cN (each corresponding to a different closed loop, which is a non-limiting example of the identical path recited in the appended claims) such that the pieces of intensity distribution information 52c1 to 52cN are arranged in a time-series order. Thus, a change over time is used as a subject of evaluation, which increases the level of the accuracy with which the welding state is evaluated.
The evaluating part 51d may determine, based on the intensity distribution information 52c, whether the welding state is a normal welding state or an abnormal welding state. When the welding state is an abnormal welding state, the evaluating part 51d may determine the type of the abnormal welding state. Thus, the evaluating part 51d performs an evaluation as to not only whether the welding state is a normal welding state or an abnormal welding state but also as to, when the welding state is the abnormal welding state, the type of the abnormal welding state. This ensures that a user is provided with a more detailed evaluation.
The evaluation method according to the above-described embodiment includes an obtaining step, a detecting step, a generating step, and an evaluating step. In the obtaining step, the track information 52a is obtained. The track information 52a is regarding a laser target spot on the workpiece W to which a laser beam is radiated. In the detecting step, the intensity information 52b is detected. The intensity information 52b shows, in a time-series order, intensities of the optical feedback 200 of the laser beam returning from the workpiece W. In the generating step, the intensity distribution information 52c is generated. The intensity distribution information 52c shows a distribution of the intensities over the workpiece W and correlates the track information 52a and the intensity information 52b with each other based on time. In the evaluating step, the welding state is evaluated based on the intensity distribution information 52c.
Thus, the intensity information 52b, which is time-series information, is combined with the track information 52a, which is spatial information. Using this combination ensures that a welding defect associated with shape, such as boring, is evaluated spatially. As a result, the welding state is evaluated at a higher level of accuracy.
The robot system 1 according to the above-described embodiment (which is a non-limiting example of the evaluation system recited in the claims) includes the head 100, the robot 10, the robot controller 21 (which is a non-limiting example of the controller recited in the claims), the head controller 22 (which is a non-limiting example of the controller recited in the claims), and the evaluation apparatus 50. The head 100 is capable of changing the shape of the radiation track P on the workpiece W. The robot 10 moves the head 100 relative to the workpiece W. The head controller 22 controls motions of the head 100, and the robot controller 21 controls motions of the robot 10. The evaluation apparatus 50 evaluates a welding state indicating how the workpiece W is welded by a laser beam. The evaluation apparatus 50 includes the obtaining part 51a, the detecting part 51b, the generating part 51c, and the evaluating part 51d. The track information 52a is regarding a laser target spot on the workpiece W to which the laser beam is radiated. The detecting part 51b detects the intensity information 52b, which shows intensities of the optical feedback 200 (of the laser beam from the workpiece W) in a time-series order. The generating part 51c generates the intensity distribution information 52c. The intensity distribution information 52c shows a distribution of the intensities over the workpiece W and correlates the track information 52a and the intensity information 52b with each other based on time. The evaluating part 51d evaluates the welding state based on the intensity distribution information 52c.
Thus, the intensity information 52b, which is time-series information, is combined with the track information 52a, which is spatial information. Using this combination ensures that a welding defect associated with shape, such as boring, is evaluated spatially. As a result, the welding state is evaluated at a higher level of accuracy.
The robot system 1 according to the above-described embodiment may include the machine learning part 61c. The machine learning part 61c receives the intensity distribution information 52c and training data generated or selected based on the received intensity distribution information 52c and based on a correct value that corresponds to the evaluated welding state and that is assigned to the received intensity distribution information 52c; performs supervised learning using the training data; and generates the evaluation model 52d (which is a non-limiting example of the calculation model recited in the claims) based on the supervised learning. The evaluation model 52d is used by the evaluating part 51d to evaluate the welding state upon receipt of another piece of intensity distribution information 52c. This enables the evaluating part 51d to evaluate the welding state using the evaluation model 52d at a higher level of accuracy.
The machine learning part 61c may receive the intensity distribution information 52c; perform unsupervised learning using the intensity distribution information 52c; and generate a calculation model used by the evaluating part 51d to evaluate the welding state upon receipt of another piece of intensity distribution information 52c.
The evaluating part 51d may evaluate the welding state based on the evaluation model 52d generated or updated by the machine learning part 61c. Thus, the evaluation model 52d may be obtained by machine learning, and using such evaluation model 52d increases the accuracy of welding state evaluation.
The evaluation program according to the above-described embodiment includes an obtaining step, a detecting step, a generating step, and an evaluating step. In the obtaining step, the track information 52a is obtained. The track information 52a is regarding a laser target spot on the workpiece W to which a laser beam is radiated. In the detecting step, the intensity information 52b is detected. The intensity information 52b shows, in a time-series order, intensities of the optical feedback 200 of the laser beam returning from the workpiece W. In the generating step, the intensity distribution information 52c is generated. The intensity distribution information 52c shows a distribution of the intensities over the workpiece W and correlates the track information 52a and the intensity information 52b with each other based on time. In the evaluating step, the welding state is evaluated based on the intensity distribution information 52c.
Thus, the intensity information 52b, which is time-series information, is combined with the track information 52a, which is spatial information. Using this combination ensures that a welding defect associated with shape, such as boring, is evaluated spatially. As a result, the welding state is evaluated at a higher level of accuracy.
In the present disclosure, the term “comprise” and its variations are intended to mean open-ended terms, not excluding any other elements and/or components that are not recited herein. The same applies to the terms “include”, “have”, and their variations.
In the present disclosure, a component suffixed with a term such as “member”, “portion”, “part”, “element”, “body”, and “structure” is intended to mean that there is a single such component or a plurality of such components.
In the present disclosure, ordinal terms such as “first” and “second” are merely used for distinguishing purposes and there is no other intention (such as to connote a particular order) in using ordinal terms. For example, the mere use of “first element” does not connote the existence of “second element”; otherwise, the mere use of “second element” does not connote the existence of “first element”.
In the present disclosure, approximating language such as “approximately”, “about”, and “substantially” may be applied to modify any quantitative representation that could permissibly vary without a significant change in the final result obtained. All of the quantitative representations recited in the present application shall be construed to be modified by approximating language such as “approximately”, “about”, and “substantially”.
Obviously, numerous modifications and variations of the present disclosure are possible in light of the above teachings. It is therefore to be understood that within the scope of the claims, the present disclosure may be practiced otherwise than as specifically described herein.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2019-121885 | Jun 2019 | JP | national |
