JP2020157347A - Pulse wave search device, pulse wave search method, pulse arc welding device and pulse arc welding method - Google Patents

Abstract

To automatically search a pulse waveform so as to further reduce a sputter quantity.SOLUTION: A pulse waveform search device, has a generation part for generating waveform information for expressing a pulse waveform, an acquisition part for acquiring a sputter quantity generated when pulse arc welding is executed by using the pulse waveform expressed by the waveform information and a repetitive processing part for preserving mutually corresponding the waveform information and the sputter quantity in a storage part by repetitively executing pulse waveform generation and sputter quantity acquisition in this order. The generation part generates the waveform information expressed by a searched design variable by searching the design variable for expressing the waveform information so as to reduce the sputter quantity from a prediction value by determining the prediction value of the sputter quantity when each piece of the waveform information already preserved in the storage part is set as the design variable when newly generating the waveform information. The repetitive processing part extracts the waveform information which is made to correspond to the minimum sputter quantity from the waveform information in the storage part when a predetermined termination condition is satisfied.SELECTED DRAWING: Figure 1

Description

The present invention relates to a technique for obtaining a pulse waveform in pulse arc welding.

Conventionally, in pulse arc welding, in order to reduce the spatter amount of the welding wire and perform stable welding, it is desired to generate a pulse waveform in which the spatter amount is reduced. The pulse waveform can be specified by, for example, a peak current value, a peak time, a base current value and a base time. For example, in the technique described in Patent Document 1, a range of a peak current value, a peak time, a base current value, and a base time such that the amount of sputtering is reduced is required.

Japanese Patent No. 3300157

However, in the technique described in Patent Document 1, a pulse waveform in which the amount of sputtering is reduced is required by skilled work while conducting a large number of experiments. Therefore, a very large amount of labor (man-hours) is required to obtain a pulse waveform in which the amount of sputtering is reduced.

Further, in the technique described in Patent Document 1, a relatively wide range is required as the range of the peak current value, the peak time, the base current value and the base time so as to reduce the sputtering amount. Here, when actually performing welding, it will be fixed to a specific value selected from the above range. However, since the above range is relatively wide, if the pulse waveform is changed within this range, the amount of sputtering increases or decreases not a little. Skilled work is required to uniquely select a pulse waveform that reduces the amount of sputtering.

The present invention solves the above problems, and is a pulse waveform search device, a pulse waveform search method, a pulse arc welding device, and a pulse capable of reducing the number of steps for searching a pulse waveform so that the amount of spatter is further reduced. It is an object of the present invention to provide an arc welding method.

The first aspect of the present invention is
A device that searches for pulse waveforms used in pulse arc welding.
A generation unit that executes a generation process that generates waveform information representing the pulse waveform, and
An acquisition unit that executes an acquisition process for acquiring the amount of sputtering generated when the pulse arc welding is executed using the pulse waveform represented by the generated waveform information, and an acquisition unit.
A storage unit for storing data and
The generation unit and the acquisition unit are made to repeatedly execute the generation process and the acquisition process in this order, and the waveform information generated by the generation process and the sputtering amount acquired by the acquisition process are mutually exchanged. A repetitive processing unit that repeatedly executes a storage process that is associated and saved in the storage unit,
With
When the waveform information is newly generated in the generation process, the generation unit obtains and obtains a predicted value of the spatter amount when the waveform information already stored in the storage unit is used as a design variable. A design variable representing the waveform information such that the spatter amount is reduced is searched from the predicted value, and the waveform information represented by the searched design variable is generated.
When the predetermined end condition is satisfied, the iterative processing unit ends the execution of the generation process and the acquisition process, and associates the waveform information stored in the storage unit with the smallest sputtering amount. It extracts and outputs the obtained waveform information.

The second aspect of the present invention is
A method for searching for pulse waveforms used in pulse arc welding.
A generation step for generating waveform information representing the pulse waveform, and
An acquisition step of acquiring the amount of sputtering generated when the pulse arc welding is executed using the pulse waveform represented by the generated waveform information, and
The generation step and the acquisition step are repeatedly executed in this order, and the waveform information generated in the generation step and the sputtering amount acquired in the acquisition step are stored in a storage unit in association with each other. Iterative processing steps that repeat processing and execution
With
In the generation step, when the waveform information is newly generated, the predicted value of the spatter amount when the waveform information already stored in the storage unit is used as a design variable is obtained, and the obtained predicted value is obtained. A design variable representing the waveform information such that the spatter amount is reduced is searched for, and waveform information represented by the searched design variable is generated.
When the predetermined end condition is satisfied, the iterative processing step ends the execution of the generation step and the acquisition step, and associates the waveform information stored in the storage unit with the smallest sputtering amount. It extracts and outputs the obtained waveform information.

In the first and second aspects, waveform information representing the pulse waveform is generated, pulse arc welding is executed using the pulse waveform represented by the generated waveform information, and the amount of spatter generated during welding execution is acquired. Will be done. The generation of the pulse waveform and the acquisition of the sputtering amount are repeated in this order, and the generated waveform information and the acquired sputtering amount are repeatedly stored in the storage unit in association with each other. When the waveform information is newly generated, the predicted value of the spatter amount when the waveform information already stored in the storage unit is used as a design variable is obtained, and the spatter amount is reduced from the obtained predicted value. Design variables representing various waveform information are searched, and waveform information represented by the searched design variables is generated. When the predetermined end condition is satisfied, the generation of the pulse waveform and the acquisition of the sputtering amount are completed, and the waveform information associated with the minimum sputtering amount among the waveform information stored in the storage unit is extracted and output. Will be done. In this way, since the design variable representing the waveform information that reduces the sputtering amount is searched from the predicted value of the sputtering amount when the waveform information already stored in the storage unit is used as the design variable, the sputtering is performed. It is possible to reduce the man-hours for searching for a pulse waveform whose amount is further reduced.

In the first aspect, for example,
For each generation process, the generation unit obtains a first prediction model composed of a function in which the waveform information is used as the design variable, the sputter amount is used as the response variable, and the response variable is represented by a probability distribution by machine learning. , The predicted value of the spatter amount may be obtained by the first prediction model.

In this aspect, for each generation process, a first prediction model consisting of a function in which waveform information is used as a design variable, the amount of sputtering is used as a response variable, and the response variable is represented by a probability distribution is obtained by machine learning, and the first prediction model is obtained. To obtain the predicted value of the spatter amount. Therefore, according to this aspect, by using the first prediction model obtained by machine learning, it is possible to obtain waveform information having a sputtering amount smaller than the sputtering amount acquired up to that point.

In the first aspect, for example,
The generation unit may obtain a first acquisition function from the first prediction model by Bayesian optimization, and search for a design variable representing the waveform information in which the sputtering amount is reduced based on the first acquisition function. ..

In this aspect, the first acquisition function is obtained from the first prediction model by Bayesian optimization, and a design variable representing waveform information in which the amount of sputtering is reduced is searched for based on the first acquisition function. Therefore, according to this aspect, the number of repetitions of the generation process and the acquisition process can be reduced.

In the first aspect, for example,
The generator further obtains shape information representing the bead shape when the pulse arc welding is executed.
The iterative processing unit
In the storage process, the waveform information, the sputtering amount, and the acquired shape information are associated with each other and stored in the storage unit.
When the predetermined end condition is satisfied, among the waveform information stored in the storage unit, the minimum amount of the sputtering among the waveform information associated with the shape information satisfying the predetermined shape condition is supported. The attached waveform information may be extracted and output.

In this aspect, in the acquisition process, shape information representing the bead shape when pulse arc welding is executed is further acquired. In the storage process, the generated waveform information, the acquired sputtering amount, and the acquired shape information are associated with each other and stored in the storage unit. When the predetermined end condition is satisfied, among the waveform information stored in the storage unit, the waveform information associated with the minimum sputtering amount among the waveform information associated with the shape information satisfying the predetermined shape condition. Is extracted and output. Simply, the waveform information associated with the minimum spatter amount may include waveform information that is not well welded. On the other hand, according to this aspect, since the constraint that the shape information satisfies a predetermined shape condition is added, the waveform information that the shape information does not satisfy the predetermined shape condition, that is, the waveform information that is not welded well is obtained. Can be excluded.

In the first aspect, for example,
For each generation process, the generation unit further machine-learns a second prediction model composed of a function in which the waveform information is used as the design variable, the shape information is used as the response variable, and the response variable is represented by a probability distribution. A second acquisition function composed of a probability that the shape information obtained based on the second prediction model satisfies a predetermined shape condition and the first acquisition function is obtained by Bayesian optimization, and the second acquisition function is obtained. The design variable may be searched based on.

In this embodiment, for each generation process, a second prediction model consisting of a function in which waveform information is used as a design variable, shape information is used as a response variable, and the response variable is represented by a probability distribution is obtained by machine learning, and a second A second acquisition function consisting of a probability that the shape information obtained based on the prediction model satisfies a predetermined shape condition and a first acquisition function is obtained by Bayesian optimization, and a design variable is searched based on the second acquisition function. .. Therefore, according to this aspect, the number of repetitions of the generation process and the acquisition process can be reduced.

A third aspect of the present invention is
The pulse waveform search device of the above aspect and
A welded portion that performs the pulse arc welding using a welding wire of a predetermined material is provided.
The acquisition unit measures and acquires the amount of spatter generated when the pulse arc welding is executed by the welding unit.

A fourth aspect of the present invention is
It is a pulse arc welding method that welds by pulse arc welding.
A generation step for generating waveform information representing the pulse waveform, and
A welding step in which the pulse arc welding is performed using a welding wire of a predetermined material, and
An acquisition step of acquiring the amount of sputtering generated when the pulse arc welding is executed in the welding step using the pulse waveform represented by the generated waveform information, and
The generation step and the acquisition step are repeatedly executed in this order, and the waveform information generated in the generation step and the sputtering amount acquired in the acquisition step are stored in a storage unit in association with each other. Iterative processing steps that repeat processing and execution
With
In the acquisition step, the amount of sputtering generated when the pulse arc welding is executed in the welding step is measured and acquired.
In the generation step, when the waveform information is newly generated, the predicted value of the spatter amount when the waveform information already stored in the storage unit is used as a design variable is obtained, and the obtained predicted value is obtained. A design variable representing the waveform information such that the spatter amount is reduced is searched for, and waveform information represented by the searched design variable is generated.
When the predetermined end condition is satisfied, the iterative processing step ends the execution of the generation step and the acquisition step, and associates the waveform information stored in the storage unit with the smallest sputtering amount. It extracts and outputs the obtained waveform information.

According to the third aspect and the fourth aspect, it is possible to perform pulse arc welding in which the amount of sputtering is further reduced by using an appropriate pulse waveform for the welding wire of a predetermined material.

According to the present invention, it is possible to reduce the man-hours for searching for a pulse waveform that further reduces the amount of sputtering.

It is a block diagram which shows schematic structure example of the pulse arc welding apparatus in this embodiment. It is a figure which shows typically the appearance example of the pulse arc welding apparatus. It is a figure which shows typically an example of the image imaged by a camera. It is a figure which shows typically an example of the waveform information which represents a pulse waveform. 6 is a timing chart schematically showing a pulse waveform represented by the waveform information of FIG. It is a figure which shows and explains the Bayesian optimization of a pulse waveform generation part roughly. It is a figure which shows and explains the Bayesian optimization of a pulse waveform generation part roughly. It is a flowchart which shows outline the operation procedure example of the pulse arc welding apparatus of this embodiment. It is a flowchart which shows the different operation procedure of the pulse arc welding apparatus roughly. It is a figure which shows roughly the process which the search of a pulse waveform progresses.

(Embodiment)
Hereinafter, an embodiment of the present invention will be described with reference to the drawings. In each drawing, the same reference numerals are used for the same components, and detailed description thereof will be omitted as appropriate.

FIG. 1 is a block diagram schematically showing a configuration example of a pulse arc welding apparatus according to the present embodiment. FIG. 2 is a diagram schematically showing an external example of the pulse arc welding apparatus. FIG. 3 is a diagram schematically showing an example of an image captured by a camera.

As shown in FIGS. 1 and 2, the pulse arc welding device 10 includes a pulse waveform search device 100, a welding robot 20, and a camera 30. The welding robot 20 includes a welding power supply 40 and a robot control unit 50. As shown in FIG. 1, the pulse waveform search device 100 includes a display 110, an input unit 120, and a control circuit 130. The control circuit 130 includes a memory 140, a central processing unit (CPU) 150, and peripheral circuits (not shown). The pulse waveform search device 100 shown in FIGS. 1 and 2 may be configured by, for example, a personal computer such as a desktop type, a notebook type, or a tablet type.

The display 110 includes, for example, a liquid crystal display panel. The display 110 is controlled by the CPU 150 and displays, for example, waveform information representing the searched pulse waveform. The display 110 is not limited to the liquid crystal display panel. The display 110 may include other panels such as an organic electroluminescence (EL) panel.

The input unit 120 includes, for example, a mouse or a keyboard. When the input unit 120 is operated by the user, the input unit 120 outputs an operation signal representing the operation content to the control circuit 130. When the display 110 is a touch panel display, the touch panel display may also serve as the input unit 120 instead of the mouse or keyboard.

The memory 140 (corresponding to an example of the storage unit) is composed of, for example, a hard disk, a semiconductor non-volatile memory, or the like. The memory 140 includes, for example, a read-only memory (ROM), a random access memory (RAM), an electrically erasable and rewritable ROM (EEPROM), and the like. For example, the ROM of the memory 140 stores the control program of the present embodiment. The control program includes, for example, a sputtering amount measuring program for measuring the sputtering amount, a pulse waveform generation program for generating a pulse waveform, an iterative processing program for performing iterative processing, and the like. For example, the RAM of the memory 140 stores welding history information 141 (described later).

The CPU 150 functions as a sputtering amount measuring unit 151, a pulse waveform generating unit 152, and a repeating processing unit 153 by operating according to the control program of the present embodiment stored in the memory 140. The functions of each part of the CPU 150 will be described later.

In FIG. 2, the welding robot 20 (corresponding to an example of a welded portion) includes a welding torch 60 electrically connected to a welding power source 40. The welding power supply 40 operates according to a pulse waveform input from the pulse waveform search device 100. The robot control unit 50 controls the welding power supply 40 and the welding torch 60 to continuously weld on the flat plate welding work 70 with a bead-on plate. Sputtering occurs when welding is performed. The pulse waveform search device 100 searches for a pulse waveform in which the amount of sputtering is further reduced with a further reduced man-hour.

The camera 30 captures an image 31 (FIG. 3) of the welding work 70 including the periphery of the welding torch 60. The camera 30 outputs the captured image 31 to the pulse waveform search device 100. As shown in FIG. 3, the image 31 captures the light particles 32 due to sputtering.

The sputtering amount measuring unit 151 (corresponding to an example of the acquisition unit) of the CPU 150 measures the sputtering amount based on the image 31. The sputtering amount measuring unit 151 measures the sputtering amount by, for example, cutting the image 31 into a mesh, counting the area to which the light particles 32 belong, and integrating them. For example, the light grain 32a in FIG. 3 corresponds to one mesh, and the light grain 32b corresponds to four meshes. The method for measuring the amount of sputtering is not limited to the method of cutting the image 31 into a mesh. As the sputtering amount, the sputtering amount measuring unit 151 may calculate and integrate the area of the pixels corresponding to the light particles 32 produced by sputtering, or calculate and integrate the number of pixels corresponding to the light particles 32 produced by sputtering. You may.

The sputtering amount measuring unit 151 may acquire an image 31 every T2 for a predetermined time to measure the sputtering amount over a predetermined time T1, and obtain an average value of a plurality of measured values as the sputtering amount. For example, if T1 = 20 [seconds] and T2 = 0.1 [seconds], the average value of 200 measured values can be obtained as the sputtering amount. The spatter amount measuring unit 151 does not measure the arc portion 33 (FIG. 3) shining directly under the welding torch 60 as spatter.

FIG. 4 is a diagram schematically showing an example of waveform information representing a pulse waveform. FIG. 5 is a timing chart schematically showing a pulse waveform represented by the waveform information of FIG. Waveform information representing a pulse waveform will be described with reference to FIGS. 4 and 5.

As shown in FIG. 4, the waveform information 400 representing the pulse waveform that operates the welding power supply 40 of the welding robot 20 has a peak current value Ip [A], a peak time Tp [ms], and a base current value Ib [A]. Includes base time Tb [ms]. The peak current value Ip is the value of the peak current in the pulse waveform. The peak time Tp is the duration of the peak current value Ip. The base current value Ib is the value of the base current in the pulse waveform. The base time Tb is the duration of the base current value Ib. As shown in the waveform numbers PW1 and PW2 in FIG. 5, the pulse waveform can be specified by the waveform information 400 in FIG. The pulse waveforms of waveform numbers PW1 and PW2 show an example of being randomly generated using random numbers.

6 and 7 are diagrams for explaining Bayesian optimization of the pulse waveform generation unit 152. FIG. 8 is a flowchart schematically showing an example of an operation procedure of the pulse arc welding apparatus 10 of the present embodiment. The operation of the pulse arc welding apparatus 10 will be described with reference to FIGS. 6 and 7 with reference to FIG.

In step S1000 of FIG. 8, the pulse waveform generation unit 152 (corresponding to an example of the generation unit) generates an initial value of the pulse waveform. The pulse waveform generation unit 152 may use, for example, a pulse waveform randomly generated using a random number as the initial value of the pulse waveform. For example, the pulse waveform generation unit 152 may use the pulse waveform of waveform number PW1 or waveform number PW2 shown in FIG. 4 as the initial value of the pulse waveform.

In step S1005, the pulse waveform generation unit 152 outputs the generated pulse waveform (for example, the waveform information of the waveform number PW1 in FIG. 4) to the welding power source 40 of the welding robot 20. In step S1010, the welding robot 20 operates the welding power supply 40 according to the output pulse waveform to execute pulse arc welding for a predetermined time T1. Further, the sputtering amount measuring unit 151 measures the sputtering amount generated by pulse arc welding for a predetermined time T1.

In step S1015, the repetitive processing unit 153 stores the waveform information representing the pulse waveform used for welding and the amount of sputtering generated and measured during welding in association with each other as welding history information 141 in the memory 140. .. In step S1020, the repetitive processing unit 153 determines whether or not the amount of sputtering measured in step S1010 and stored in step S1015 is equal to or less than a predetermined value (corresponding to an example of the end condition). If the amount of sputtering exceeds a predetermined value (NO in step S1020), the process proceeds to step S1025.

In step S1025, the pulse waveform generation unit 152 generates a function f (X), which is a prediction model (corresponding to an example of the first prediction model) that stochastically predicts the amount of spatter, by machine learning. The pulse waveform generation unit 152 obtains a predicted value of the sputtering amount by this function f (X). The function f (X) corresponds to the Gaussian process of Bayesian optimization. Schematically, "a general function f (x) returns a scalar f (x) for a certain x, but in the Gaussian process, a normal distribution followed by the value f (X) of f at a certain X ( That is, the mean and variance) are returned. " That is, as schematically shown in FIG. 6, the function f (X), which is a prediction model, follows a normal distribution P [f (X)] rather than a scalar.

The pulse waveform generation unit 152 uses the waveform information 400 stored in the welding history information 141 of the memory 140 as the design variable X, the spatter amount as the response variable Y, and the function Y = f (X) which is a prediction model of the spatter amount. Is generated by machine learning. For example, when the welding history information 141 of the memory 140 stores the first to tth design variables X1, X2, ..., Xt and the spatter amounts Y1, Y2, ..., Yt, (X1, The function Y = f (X) is obtained by using Y1), (X2, Y2), ..., (Xt, Yt) as learning data for machine learning. Here, since the design variable X is the waveform information 400, it is a vector value as can be seen from FIG.

In step S1030 of FIG. 8, the pulse waveform generation unit 152 calculates the Bayesian optimization acquisition function (corresponding to an example of the first acquisition function) for all the design variables X. In the present embodiment, the pulse waveform generation unit 152 uses the expected value (EI) of the improvement amount as the acquisition function. In step S1035, the pulse waveform generation unit 152 generates the design variable X when the expected value (EI) of the improvement amount is the maximum value as waveform information representing the next pulse waveform. After that, the process returns to step S1005.

For example, the state in which the sputtering amount exceeds a predetermined value (NO in step S1020) is repeated, and the welding history information 141 of the memory 140 stores the sputtering amount from the first to the tth and the waveform information in association with each other. It is assumed that it has been done. In this case, in step S1025 of FIG. 8, the function ft, which is the t-th prediction model shown in FIG. 6, is determined by the design variables and response variables from the first (X1, Y1) to the t-th (Xt, Yt). (X) is required. In step S1030 of FIG. 8, this function ft (X) is used.

Assuming that the minimum value of the sputtering amounts Y1 to Yt from the first to the tth is Ymin and the function that is the prediction model is f (X), the improvement amount I (X) is defined by the following (Equation 1). The expected value E [I (X)] of the improvement amount I (X) is given by the following (Equation 2).
I (X) = max {0, Ymin-f (X)} (Equation 1)

Then, according to (Equation 2), the expected value E [I (X)] of the improvement amount is calculated for all the design variables X. Here, the upper limit value, the lower limit value, and the step size are set in advance as "all design variables X". That is, the expected value of the improvement amount is calculated for the design variable X for each step width from the upper limit value to the lower limit value. Of the calculated expected value E [I (X)] of the improvement amount, the maximum value is the shaded area in FIG. Therefore, in step S1035 of FIG. 8, a pulse waveform represented by the waveform information which is the design variable Xt + 1 at this time is generated.

In the present embodiment, as described above, the pulse waveform generation unit 152 divides the search range of the design variable X by a grid, and sets the expected value E [I (X)] of the improvement amount at all the grid points. The grid point (that is, the design variable X) at which the expected value E [I (X)] of the improvement amount is maximized is calculated, but the present invention is not limited to this. The pulse waveform generation unit 152 may obtain the design variable X that maximizes the expected value E [I (X)] of the improvement amount by using an optimization method such as particle swarm optimization (PSO), for example.

Then, in the next step S1010 (FIG. 8), pulse arc welding is executed using the pulse waveform represented by the waveform information which is the design variable Xt + 1, and the sputtering amount Yt + 1 shown in FIG. 7 is measured. Further, in step S1025, the t + 1st prediction model shown in FIG. 7 is a set of design variables and response variables from the first (X1, Y1) to the t + 1th (Xt + 1, Yt + 1) as learning data for machine learning. The function Y = ft + 1 (X) is obtained.

On the other hand, in step S1020, if the sputtering amount is equal to or less than a predetermined value (YES in step S1020), the process proceeds to step S1040. In step S1040, the repetition processing unit 153 determines that among the waveform information stored in the welding history information 141 of the memory 140, the waveform information having the smallest sputtering amount (in the present embodiment, it is determined to be equal to or less than a predetermined value in step S1020). The waveform information associated with the spatter amount) is displayed on the display 110 as the searched pulse waveform, and the operation of FIG. 8 ends.

As described above, in the present embodiment, a pulse waveform is obtained by Bayesian optimization. That is, the pulse waveform generation unit 152 obtains a function Y = f (X) which is a prediction model representing a normal distribution, using the waveform information representing the pulse waveform as the design variable X and the measured sputtering amount as the response variable Y. The expected value E [I (X)] of the improvement amount is obtained as an acquisition function, and the design variable X when the expected value of the improvement amount is the maximum value is generated as waveform information representing the next pulse waveform. Therefore, according to the present embodiment, it is possible to automatically search for a pulse waveform in which the amount of sputtering is further reduced. In addition, it does not require a large number of experiments and skilled work, and the man-hours can be reduced as compared with the conventional case.

(Other)
(1) In the above embodiment, only the function f (X), which is a prediction model for probabilistically predicting the sputtering amount, is used to obtain a pulse waveform in which the sputtering amount is reduced, but the present invention is limited to this. Absent. The pulse waveform generation unit 152 may further generate a function which is a prediction model for probabilistically predicting shape information representing the shape of the bead, and may also use this function to obtain a pulse waveform in which the amount of spatter is reduced.

FIG. 9 is a flowchart schematically showing different operation procedure examples of the pulse arc welding apparatus 10. FIG. 10 is a diagram schematically showing a process in which the search for the pulse waveform proceeds. The vertical axis of FIG. 10 represents the amount of sputtering, and the horizontal axis represents the average voltage [V] of the pulse waveform.

In the embodiments of FIGS. 9 and 10, the average voltage of the pulse waveform is used as the shape information representing the bead shape. It is known that if the average voltage of the pulse waveform is within a predetermined range, the shape of the bead becomes good and high quality welding can be realized. Therefore, in the embodiment of FIGS. 9 and 10, the correspondence relationship between the waveform information of the pulse waveform and the average voltage is stored in advance in the memory 140.

Steps S1000 to S1010 in FIG. 9 are the same as steps S1000 to S1010 in FIG. 8, respectively. In step S1100 following step S1010, the iterative processing unit 153 uses the correspondence between the waveform information of the pulse waveform stored in the memory 140 and the average voltage from the waveform information representing the pulse waveform output in step S1005. , Find the average voltage of the pulse waveform. The iterative processing unit 153 associates the sputtering amount, the waveform information representing the pulse waveform, and the average voltage of the pulse waveform, and stores the welding history information 141 in the memory 140.

In step S1105, the repetitive processing unit 153 determines whether or not the average voltage stored in step S1015 is within the predetermined range Vr and the amount of sputtering stored in step S1015 is equal to or less than the predetermined value (end condition). (Corresponding to one example) is judged. If the average voltage is outside the predetermined range Vr, or if the sputtering amount exceeds the predetermined value (NO in step S1105), the process proceeds to step S1025. Step S1025 is the same as step S1025 in FIG.

In step S1110, the pulse waveform generation unit 152 stochastically predicts the average voltage with the waveform information 400 stored in the welding history information 141 of the memory 140 as the design variable X and the average voltage as the response variable Z. A function g (X) (corresponding to an example of the second prediction model) is generated by machine learning. The pulse waveform generation unit 152 obtains a predicted value of the average voltage by this function g (X). For example, when the first to tth design variables X1, X2, ..., Xt and the average voltage Z1, Z2, ..., Zt are stored in the welding history information 141 of the memory 140, (X1). , Z1), (X2, Z2), ..., (Xt, Zt) are used as learning data for machine learning, and the function Z = g (X) is obtained.

In step S1115, the pulse waveform generation unit 152 has an acquisition function (second) composed of a probability that the average voltage falls within a predetermined range and an expected value (corresponding to an example of the first acquisition function) of the improvement amount of the spatter amount. (Corresponding to an example of the acquisition function) is calculated for all design variables X. In step S1120, the pulse waveform generation unit 152 generates the design variable X when the acquisition function calculated in step S1115 has the maximum value as waveform information representing the next pulse waveform. After that, the process returns to step S1005.

On the other hand, in step S1105, if the average voltage is within the predetermined range Vr and the sputtering amount is equal to or less than the predetermined value (YES in step S1105), the process proceeds to step S1040. Step S1040 is the same as step S1040 in FIG.

According to the embodiment of FIGS. 9 and 10, as shown in FIG. 10, as the search for the pulse waveform progresses, the average voltage is within the predetermined range Vr and the sputtering amount is further reduced. The waveform can be obtained. Therefore, it is possible to obtain a pulse waveform capable of performing higher quality pulse arc welding while further reducing the amount of sputtering.

In the embodiment of FIGS. 9 and 10, the shape information representing the shape of the bead is not limited to the average voltage of the pulse waveform. The width and height of the bead may be directly used as the shape information indicating the shape of the bead. In this case, a laser sensor that emits a laser beam and receives the reflected wave reflected by the bead may be provided, and the width and height of the bead may be measured by the laser sensor.

Further, in the embodiment of FIGS. 9 and 10, the pulse waveform generation unit 152 obtains the function g (X) which is a prediction model, but it is not necessary to obtain the function g (X). In this case, for example, the pulse waveform generation unit 152 may exclude from the design variable X the waveform information representing the pulse waveform whose average voltage does not fall within the predetermined range Vr (FIG. 10).

Further, for example, the determination step of step S1020 may be replaced with step S1105 while operating according to the operation procedure of FIG. Even in this operation, unless the average voltage falls within the predetermined range Vr (FIG. 10), YES is not obtained in step S1105, so that the operation can be performed in the same manner as in the embodiments of FIGS. 9 and 10.

(2) In the above embodiment, when the sputtering amount becomes a predetermined value or less (the end condition is satisfied) in step S1020 of FIG. 8, the process proceeds to step S1040, and then the operation of FIG. 8 ends. Further, in the embodiment of FIGS. 9 and 10, in step S1105 of FIG. 9, when the average voltage is within the predetermined range Vr and the sputtering amount is equal to or less than the predetermined value (the end condition is satisfied). After the process proceeds to step S1040, the operation of FIG. 9 ends. However, the termination condition is not limited to the condition that the sputtering amount is not more than a predetermined value.

For example, the repetitive processing unit 153 counts the number of times the process returns from step S1035 to step S1005 in FIG. 8, determines whether or not the count value has reached a predetermined number of times (for example, 100 times) in step S1020, and counts. When the value reaches a predetermined number of times, the process may proceed to step S1040. For example, the repetition processing unit 153 counts the number of times the process returns from step S1120 to step S1005 in FIG. 9, determines in step S1105 whether or not the count value has reached a predetermined number of times, and the count value reaches a predetermined number of times. When it reaches, it may proceed to step S1040.

In the operation of FIG. 8 or 9, the operation is not completed unless the sputtering amount is equal to or less than a predetermined value, so that the pulse waveform search time may become excessive. On the other hand, according to this embodiment, since the search ends when the number of searches reaches a predetermined number of times, it is possible to prevent the search time of the pulse waveform from becoming excessive.

(3) In the above embodiment, the acquisition function is not limited to the expected value (EI) of the improvement amount, but may be the probability of improvement (PI) or the lower limit of the reliability limit (LCB). Further, the design variable X that minimizes the function f (X), which is a prediction model of the sputtering amount, may be searched without using the acquisition function. However, it is preferable to obtain the acquisition function by Bayesian optimization because the number of repetitions can be reduced.

(4) In the above embodiment, the waveform information 400 is not limited to the four pieces of information shown in FIG. 4, and may include information that can identify the pulse waveform. For example, the pulse frequency or pulse period may be used instead of the peak time (or base time). In this case, the peak time can be obtained by subtracting, for example, the base time from the reciprocal of the pulse frequency or the pulse period, so that the pulse waveform can be specified.

Further, the average current value may be included instead of the peak current value (or the base current value). In this case, for example, the peak current value can be obtained from the average current value and the other three pieces of information (that is, the base current value, the peak time, and the base time), so that the pulse waveform can be specified.

Further, in the pulse waveforms of FIGS. 4 and 5, the rise time from the base current value to the peak current value and the fall time from the peak current value to the base current value are fixed, but the rise time and the fall time are fixed. Also, as a parameter, it may be included in the waveform information (that is, the design variable). That is, in addition to the peak current value, the peak time, the base current value, and the base time, six pieces of information, the rise time and the fall time, may be included in the waveform information.

Further, when the arc length control generally used in pulse arc welding is used, for example, when the arc length is controlled by width modulation control of the base time, the base time may be excluded from the design variable.

Further, in the above embodiment, the pulse waveform is a square wave, but the pulse waveform is not limited to this, and may be a trapezoidal waveform, a triangular waveform, or a waveform having two pulse peaks including a first pulse and a second pulse.

(5) In the above embodiment, the sputtering amount measuring unit 151 measures the sputtering amount based on the image 31 captured by the camera 30, but is not limited to this. For example, a box-shaped accommodating portion for accommodating spatter scattered from the weld bead 71 (FIG. 2) may be provided, and a weighing scale for measuring the weight of the accommodating portion may be provided. Then, the sputtering amount measuring unit 151 may measure the sputtering amount based on the weight measured by the weighing scale.

(6) In step S1000 of FIGS. 8 and 9, the initial value of the pulse waveform may not be one, but a plurality of (for example, 10) random design variables may be used as the initial value. For example, in FIG. 8, when there are a plurality of initial values (for example, 10 pieces), steps S1005, S1010, and S1015 are executed at the respective initial values, and from a state where there are a plurality of (for example, 10 pieces) welding history information. , Step S1025 and subsequent steps may be executed. For example, in FIG. 9, when there are a plurality of initial values (for example, 10 pieces), steps S1005, S1010, and S1100 are executed at the respective initial values, and from a state where there are a plurality of (for example, 10 pieces) welding history information. , Step S1025 and subsequent steps may be executed.

10 Pulse arc welding device 20 Welding robot 100 Pulse waveform search device 140 Memory 141 Welding history information 151 Spatter amount measurement unit 152 Pulse waveform generation unit 153 Repeat processing unit

Claims (8)

A device that searches for pulse waveforms used in pulse arc welding.
A generation unit that executes a generation process that generates waveform information representing the pulse waveform, and
An acquisition unit that executes an acquisition process for acquiring the amount of sputtering generated when the pulse arc welding is executed using the pulse waveform represented by the generated waveform information, and an acquisition unit.
A storage unit for storing data and
The generation unit and the acquisition unit are made to repeatedly execute the generation process and the acquisition process in this order, and the waveform information generated by the generation process and the sputtering amount acquired by the acquisition process are mutually exchanged. A repetitive processing unit that repeatedly executes a storage process that is associated and saved in the storage unit,
With
When the waveform information is newly generated in the generation process, the generation unit obtains and obtains a predicted value of the spatter amount when the waveform information already stored in the storage unit is used as a design variable. A design variable representing the waveform information such that the spatter amount is reduced is searched from the predicted value, and the waveform information represented by the searched design variable is generated.
When the predetermined end condition is satisfied, the iterative processing unit ends the execution of the generation process and the acquisition process, and associates the waveform information stored in the storage unit with the smallest sputtering amount. Extract and output the obtained waveform information,
Pulse waveform search device. For each generation process, the generation unit obtains a first prediction model composed of a function in which the waveform information is used as the design variable, the sputter amount is used as the response variable, and the response variable is represented by a probability distribution by machine learning. , The predicted value of the spatter amount is obtained by the first prediction model.
The pulse waveform search device according to claim 1. The generation unit obtains a first acquisition function from the first prediction model by Bayesian optimization, and searches for a design variable representing the waveform information in which the sputtering amount is reduced based on the first acquisition function.
The pulse waveform search device according to claim 2. The generator further obtains shape information representing the bead shape when the pulse arc welding is executed.
The iterative processing unit
In the storage process, the waveform information, the sputtering amount, and the acquired shape information are associated with each other and stored in the storage unit.
When the predetermined end condition is satisfied, among the waveform information stored in the storage unit, the minimum amount of the sputtering among the waveform information associated with the shape information satisfying the predetermined shape condition is supported. Extracts and outputs the attached waveform information,
The pulse waveform search device according to claim 1. For each generation process, the generation unit further machine-learns a second prediction model composed of a function in which the waveform information is used as the design variable, the shape information is used as the response variable, and the response variable is represented by a probability distribution. A second acquisition function composed of a probability that the shape information obtained based on the second prediction model satisfies a predetermined shape condition and the first acquisition function is obtained by Bayesian optimization, and the second acquisition function is obtained. Search for the design variable based on
The pulse waveform search device according to claim 4. A method for searching for pulse waveforms used in pulse arc welding.
A generation step for generating waveform information representing the pulse waveform, and
An acquisition step of acquiring the amount of sputtering generated when the pulse arc welding is executed using the pulse waveform represented by the generated waveform information, and
The generation step and the acquisition step are repeatedly executed in this order, and the waveform information generated in the generation step and the sputtering amount acquired in the acquisition step are stored in a storage unit in association with each other. Iterative processing steps that repeat processing and execution
With
In the generation step, when the waveform information is newly generated, the predicted value of the spatter amount when the waveform information already stored in the storage unit is used as a design variable is obtained, and the obtained predicted value is obtained. A design variable representing the waveform information such that the spatter amount is reduced is searched for, and waveform information represented by the searched design variable is generated.
When the predetermined end condition is satisfied, the iterative processing step ends the execution of the generation step and the acquisition step, and associates the waveform information stored in the storage unit with the smallest sputtering amount. Extract and output the obtained waveform information,
Pulse waveform search method. The pulse waveform search device according to any one of claims 1 to 5.
A welded portion that performs the pulse arc welding using a welding wire of a predetermined material is provided.
The acquisition unit measures and acquires the amount of sputtering generated when the pulse arc welding is executed by the welding unit.
Pulse arc welding equipment. It is a pulse arc welding method that welds by pulse arc welding.
A generation step for generating waveform information representing the pulse waveform, and
A welding step in which the pulse arc welding is performed using a welding wire of a predetermined material, and
An acquisition step of acquiring the amount of sputtering generated when the pulse arc welding is executed in the welding step using the pulse waveform represented by the generated waveform information, and
The generation step and the acquisition step are repeatedly executed in this order, and the waveform information generated in the generation step and the sputtering amount acquired in the acquisition step are stored in a storage unit in association with each other. Iterative processing steps that repeat processing and execution
With
In the acquisition step, the amount of sputtering generated when the pulse arc welding is executed in the welding step is measured and acquired.
In the generation step, when the waveform information is newly generated, the predicted value of the spatter amount when the waveform information already stored in the storage unit is used as a design variable is obtained, and the obtained predicted value is obtained. A design variable representing the waveform information such that the spatter amount is reduced is searched for, and waveform information represented by the searched design variable is generated.
When the predetermined end condition is satisfied, the iterative processing step ends the execution of the generation step and the acquisition step, and associates the waveform information stored in the storage unit with the smallest sputtering amount. Extract and output the obtained waveform information,
Pulse arc welding method.

Images