JP7181144B2 - Trained model generation method, computer program, welding power supply, and learned model - Google Patents

Description

The present invention relates to a method of generating a trained model, a computer program, a welding power supply, and a trained model.

A consumable-electrode welding power supply includes a welding power supply, a wire feeder, a welding torch, and a gas supply unit that injects a shielding gas around an arc. Patent Literature 1 discloses a technique of picking up an arc sound with a microphone provided on a welding torch and determining the quality of the welding state from the sound pressure level of a specific frequency of the arc sound. Further, Patent Literature 1 discloses a technique of sampling shielding gas flowing sound and determining the injection state of shielding gas based on the sound pressure level of a specific frequency.

JP 2014-113637 A

However, in the welding power supply apparatus according to Patent Document 1, there is insufficient arc sound data that serves as a reference for determining whether the welding state is good or not for the sampled arc sound, resulting in variations in quality determination. There was a technical problem that In addition, when welding conditions such as work material, thickness, groove shape, joint type, work W posture, welding wire material, wire diameter, welding current, and welding voltage change, the problem of judgment variation becomes more pronounced. become.
In order to solve this problem, it is conceivable to analyze a large amount of data indicating the quality of the welding state with respect to the arc sound, but it is generally difficult to efficiently prepare such data.

It is an object of the present invention to provide a generation method, a computer program, a welding power supply, and a learning method capable of efficiently generating and accumulating data for learning and generating a trained model for accurately determining the quality of a welding state from an arc sound. It is to provide a finished model.

A method of generating a trained model according to this aspect is a method of generating a trained model that, when arc sound data is input, outputs pass/fail data indicating the pass/fail of a welding state. Acquire arc sound data obtained by collecting with a microphone, and detect the state of the welded part by imaging the welded part after welding or scanning the welded part with detection light. determining whether the welding state is good or bad based on the obtained detection data, and accumulating learning data in which the obtained arc sound data and the good or bad data indicating the welding state are associated with each other in a storage unit; Based on the learning data accumulated in the storage unit, a trained model is generated by machine learning that outputs the pass/fail data corresponding to the arc sound data when the arc sound data is input.

According to this aspect, it is possible to efficiently prepare the learning data required to generate a learned model for judging the quality of the welding state from the arc sound. The learning data includes arc sound data acquired during welding and good/bad data indicating the quality of the welding state when the arc sound is generated.
Learning data is obtained by picking up the arc sound generated during welding with a microphone. However, it is not possible to know immediately whether the welding state is good or bad when the arc sound is generated.
Therefore, by imaging the welded portion after welding, detection data obtained by detecting the state of the welded portion is obtained. Further, detection data obtained by detecting the state of the welding portion may be acquired by scanning the welding portion with the detection light. Quality data is obtained by judging whether the welding state is good or bad based on the detected data. It is relatively easy to determine whether the welded state is good or bad from the detection data relating to the welded portion, and it is possible to automatically obtain the good or bad data.
The arc sound data thus obtained is associated with the pass/fail data, and stored in the storage unit as learning data. Next, using the learning data accumulated in the storage unit, a learned model for judging the quality of the welding state from the arc sound is generated. A trained model can be generated by subjecting a pre-learning neural network to machine learning. When arc sound data is input to the generated learned model, quality data indicating the quality of the welding state is output, so it is possible to determine the quality of the welding state from the arc sound.

In the method of generating a trained model according to this aspect, by inputting the acquired detection data to a trained model for quality determination that outputs data indicating the quality of a welding state when the detection data is input, Judge the quality of welding.

According to this aspect, learning data is accumulated in the storage unit for each of a plurality of different welding conditions, and a learned model is generated for each of a plurality of welding conditions. Therefore, a learned model suitable for each welding condition can be generated. A trained model can be efficiently generated by machine learning for each welding condition.

In the method of generating a learned model according to this aspect, the learning data for each of a plurality of different welding conditions is accumulated in the storage unit, and based on the learning data of the plurality of welding conditions accumulated in the storage unit to generate a plurality of said learned models different for each of said plurality of welding conditions.

According to this aspect, it is possible to accurately determine the quality of the welded state from the detected data by using the quality determination model, which is a trained neural network.

A computer program according to this aspect is a computer program for generating a learned model that outputs pass/fail data indicating the pass/fail of a welding state when arc sound data is input to the computer, the computer comprising: Acquire arc sound data obtained by collecting arc sound generated during welding with a microphone, and detect the state of the welded part by imaging the welded part after welding or scanning the welded part with detection light. learning data in which the quality of the welding state is determined based on the obtained detection data, and the obtained arc sound data and the quality data indicating the quality of the welding state are associated with each other; is accumulated in a storage unit, and based on the learning data accumulated in the storage unit, when the arc sound data is input, the learned model outputs the pass/fail data corresponding to the arc sound data Execute the process generated by machine learning.

According to this aspect, by causing a computer to execute the computer program, it is possible to efficiently accumulate learning data and generate a learned model for accurately determining the quality of the welding state from the arc sound.

A welding power supply apparatus according to this aspect includes a microphone for collecting arc sound, a trained model for outputting good/bad data indicating the quality of a welding state when arc sound data is input, and a microphone for collecting data. and a notification unit that inputs obtained arc sound data to the learned model and notifies the quality of the welding state indicated by the quality data output from the learned model.

According to this aspect, the welding power supply can determine whether the welding state is good or bad from the arc sound and notify the user. Specifically, the welding power supply acquires arc sound data during welding, inputs the acquired arc sound data to a learned model, and determines the quality of the welding state indicated by the quality data output from the learned model. Notification is made by the notification unit.

The welding power supply according to this aspect has a plurality of learned models, and each learned model is machine-learned for each of a plurality of different welding conditions, and based on the welding conditions, the A selection unit that selects the learned model is provided, and the notification unit notifies the quality of the welding state indicated by the quality data output from the learned model selected by the selection unit.

According to this aspect, the welding power supply selects a learned model corresponding to the welding condition, and uses the selected learned model to determine whether the welding state is good or bad, and notifies the result. By using a learned model suitable for the welding conditions, the welding state can be judged with higher accuracy and the quality can be notified.

The learned model according to this aspect is a learned model that causes a computer to function so as to output pass/fail data indicating the pass/fail of a welding state when arc sound data is input. An input layer to which arc sound data obtained by picking up sound with a microphone is input, an intermediate layer that performs an operation based on a learned weighting factor for the arc sound data input to the input layer, and the arc an output layer that outputs the quality data indicating the quality of the welding state when the sound related to the sound data is emitted, the arc sound data obtained by collecting the arc sound generated during welding with a microphone; The arc sound input to the input layer based on the pass/fail data obtained by detecting the state of the welded portion by imaging the welded portion after welding or scanning the welded portion with detection light. A neural network that learns the weighting coefficient so that the data and the pass/fail data output from the output layer correspond to each other, and the computer receives the arc sound data input to the input layer and the A calculation based on the weighting factor is performed, and the pass/fail data is output.

According to this aspect, when the arc sound data is input to the input layer of the trained model, the quality data indicating the quality of the welding state when the arc sound is generated is output from the output layer. By using the learned model, it becomes possible to determine whether the welding state is good or bad from the arc sound.

According to this aspect, it is possible to efficiently generate and accumulate data for learning, and generate a trained model for accurately determining the quality of the welding state from the arc sound.

1 is a schematic diagram showing a welding power supply according to Embodiment 1; FIG. 1 is a block diagram showing a configuration example of a welding torch according to Embodiment 1; FIG. 1 is a block diagram showing a configuration example of a welding power source according to Embodiment 1; FIG. 3 is a block diagram showing a configuration example of a learning unit according to Embodiment 1; FIG. FIG. 4 is a conceptual diagram showing an example of learning data; 4 is a flow chart showing a method of generating a trained model; 4 is a flow chart showing a method of judging whether a welding state is good or bad and notifying the user. FIG. 10 is a schematic diagram showing a learning system according to Embodiment 2;

Hereinafter, the present invention will be described in detail based on the drawings showing its embodiments. Moreover, at least part of the embodiments described below may be combined arbitrarily.

(Embodiment 1)
FIG. 1 is a schematic diagram showing a welding power source apparatus according to Embodiment 1. FIG. A consumable electrode type welding power source device includes a welding power source 1 , a wire feeder 2 and a welding torch 3 . The welding power supply is semi-automatic.

The welding power source 1 includes a first output terminal and a second output terminal for supplying power to the welding torch 3, and a signal terminal for transmitting and receiving signals. One end of the first power cable 41 is connected to the first output terminal of the welding power source 1 , and the other end of the first power cable 41 is connected to the welding torch 3 via the wire feeder 2 . A second output terminal of the welding power source 1 is connected to the work W by a second power cable 42 . The work W is grounded. The welding power source 1 converts the three-phase alternating current of the electric power system P into required welding current and welding voltage, and supplies the welding torch 3 with electric power necessary for arc welding through the first and second power cables 41 and 42 .

The welding power source 1 and the wire feeder 2 are connected by a power transmission line 5 for drive control. The wire feeder 2 and welding torch 3 are also connected by a power transmission line 5 . Welding power supply 1 supplies electric power to wire feeder 2 and welding torch 3 through power transmission line 5 to drive a feed motor of wire feeder 2, a controller 36 (see FIG. 2) of welding torch 3, and the like. supply to

The welding power supply includes a gas cylinder 6 that supplies shielding gas for preventing oxidation of molten metal during arc welding. One end of gas pipe 7 is connected to gas cylinder 6 , and the other end of gas pipe 7 is connected to welding torch 3 via welding power source 1 and wire feeder 2 . Shielding gas from gas cylinder 6 is supplied to welding torch 3 through gas pipe 7 .

One end of the signal line 8 is connected to the signal terminal of the welding power source 1 , and the other end of the signal line 8 is connected to the wire feeder 2 . The wire feeder 2 and welding torch 3 are also connected by a signal line 8 . The welding power source 1 outputs a control signal to the wire feeder 2 through the signal line 8 to control the welding wire feeding speed and the like. The welding power source 1 and the welding torch 3 also transmit and receive various types of information necessary for generating the learned model 154 for judging whether the welding state is good or not, and for judging whether the welding state is good or bad using the learned model 154 .

The wire feeder 2 is a device that feeds a welding wire that functions as a consumable electrode to the welding torch 3 . The welding wire is for example a solid wire. The wire feeder 2 and the welding torch 3 are connected by a torch cable 39, and the welding wire passes through the torch cable 39 and a liner provided inside the welding torch 3 to reach the tip of the welding torch 3. led to. The wire feeding device 2 includes a feeding roller for feeding welding wire, a feeding motor, and the like, and is driven by electric power supplied from the welding power source 1 .
Inside the torch cable 39, a first power cable 41, a gas pipe 7, a liner, a power transmission line 5 and a signal line 8 are arranged. The drive control power supplied from the welding power source 1 to the wire feeder 2 is also supplied to the welding torch 3 through the power transmission line 5 arranged inside the torch cable 39 . Wire feeder 2 can communicate with welding power source 1 and welding torch 3 through signal line 8 . Similarly, welding power source 1 and welding torch 3 can communicate via wire feeder 2 .

The welding torch 3 is made of a conductive material such as a copper alloy, and has a cylindrical contact tip that guides the welding wire to the workpiece W to be welded and supplies the welding current required to generate an arc. The welding wire fed from the wire feeder 2 is delivered so as to protrude from the tip of the contact tip. The first power cable 41 is electrically connected with the contact tip. The contact tip makes contact with a welding wire passing therethrough, and a welding current is supplied to the welding wire.
The welding torch 3 has a hollow cylindrical shape surrounding the contact tip, and has a nozzle for injecting a shielding gas to the workpiece W from an opening at the tip. The shield gas is for preventing oxidation of the workpiece W and the welding wire melted by the arc. The shield gas is, for example, carbon dioxide, a mixed gas of carbon dioxide and argon, an inert gas such as argon, or the like.

<Welding torch>
FIG. 2 is a block diagram showing a configuration example of the welding torch 3 according to the first embodiment. The welding torch 3 includes a communication section 31 , a display section 32 , an operation section 33 , a storage section 34 , a sensor 35 , a control section 36 , a microphone 37 and a notification section 38 .

The communication unit 31 is a circuit that communicates with the wire feeding device 2 or the welding power source 1 . The communication unit 31 modulates the data given from the control unit 36 according to a predetermined communication protocol and transmits the data to the wire feeding device 2 or the welding power source 1 through the signal line 8 . The communication protocol is CAN (Controller Area Network), for example. It also receives data transmitted from the welding power source 1 and the wire feeder 2, demodulates it, and gives the demodulated signal to the control section 36. FIG. Communication between the welding torch 3 and the welding power source 1 is not limited to wired communication, and may be wireless communication.

The display unit 32 has a display that displays various information related to welding. The display is, for example, a liquid crystal display panel, and displays welding conditions such as the material and thickness of the work W, the shape of the groove, the type of joint, the orientation of the work W, the material of the welding wire, the wire diameter, the welding current, and the welding voltage. do.

The operation unit 33 includes various switches and buttons for accepting operations by the welder. By operating the operation unit 33, the welder can select between a welding sound learning mode in which the learned model 154 according to the present embodiment is generated and a welding mode in which welding is performed using the learned model 154. . When the operation unit 33 is operated, an operation signal is input to the control unit 36 so that the control unit 36 can recognize the operation state of the operation unit 33 .

The storage unit 34 is non-volatile memory such as EEPROM (Electrically Erasable Programmable ROM) and flash memory. The storage unit 34 stores information such as welding conditions set by operating the operation unit 33 . Welding conditions include, for example, the material and thickness of the workpiece W, the shape of the groove, the type of joint, the posture of the workpiece W, the moving speed of the welding torch 3, the posture and the angle, etc. Recommended welding current, welding voltage, This is information on setting values such as the feeding speed of the welding wire. The welding conditions may be set on the welding power source 1 side.

The sensor 35 is, for example, an imaging unit, and is provided on the welding torch 3 in a posture capable of imaging a welded portion such as a welded bead, a welding mark, or the like. The imaging unit outputs image data obtained by imaging the welding site to the control unit 36 as detection data.
Further, the sensor 35 may be an inspection device that detects the state of the welded portion by scanning the welded portion with a laser (detection light). The inspection device detects the state of the welded portion by laser scanning, and outputs detection data indicating the detection result to the control section 36 .

The microphone 37 is provided in the welding torch 3, picks up the arc sound generated during welding, and outputs the arc sound data to the control unit 36.

The notification unit 38 is, for example, a speaker, and outputs sound indicating whether the welding state is good or bad. Audio output is controlled by the control unit 36 .

The control unit 36 is a computer having a CPU (Central Processing Unit), a processor such as a multi-core CPU, a ROM (Read Only Memory), a RAM (Random Access Memory), an input/output interface, and the like. The control unit 36 executes predetermined processing such as setting welding conditions according to the operation of the operation unit 33 . Control unit 36 executes processing for transmitting welding condition data indicating the set welding conditions to welding power source 1 through communication unit 31 . Further, the control unit 36 executes a process of transmitting the detection data detected by the sensor 35 and the arc sound data collected by the microphone 37 to the welding power source 1 through the communication unit 31 . The arc sound data and detection data may be directly output from the sensor 35 and the microphone 37 to the welding power source 1 .
Furthermore, the control unit 36 can also perform control to display the welding conditions, the information read from the storage unit 34, and the like on the display unit 32 as appropriate. Welding power source 1 determines whether the welding state is good or bad based on arc sound data, as will be described later. and controls the notification unit 38 to notify the quality of the welding state.

<Welding power source>
FIG. 3 is a block diagram showing a configuration example of the welding power source 1 according to Embodiment 1. As shown in FIG. The welding power source 1 includes a main control portion 11 , a power source portion 12 , a communication portion 13 , a storage portion 14 and a learning portion 15 .

The main control unit 11 is a microcomputer having a CPU, and controls the operation of each component that constitutes the welding power source 1 .

The power supply unit 12 is a circuit that supplies electric power to the welding torch 3 for performing arc welding. The power supply unit 12 converts the three-phase AC power input from the power system P into power suitable for arc welding and outputs the power. The power supply unit 12 also converts the three-phase AC power into power suitable for controlling the driving of the wire feeding device 2 and the welding torch 3 and supplies the power to the wire feeding device 2 and the welding torch 3 .

The communication unit 13 is a circuit that communicates with the wire feeding device 2 and the welding torch 3 . For example, it receives arc sound data, detection data, welding condition data, and the like transmitted from the welding torch 3 and the wire feeder 2, and outputs various received data to the main control unit 11, the learning unit 15, and the like. The communication unit 13 also transmits to the welding torch 3 quality data indicating the quality of the welding state determined from the arc sound, as will be described later.

The storage unit 14 is non-volatile memory such as EEPROM and flash memory. The storage unit 14 stores a computer program 14a for generating a learned model 154 for judging the quality of the welding state from the arc sound. The storage unit 14 also accumulates learning data for generating the learned model 154 . Furthermore, the storage unit 14 stores learned models 154 generated for each of a plurality of welding conditions.

The learning unit 15 is a computer having a processor such as a CPU, a GPU (Graphics Processing Unit) or a multi-core CPU, a ROM, a RAM, an input/output interface, and the like. The learning unit 15 executes a process of accumulating learning data in which arc sound data and detection data are associated with each welding condition in the storage unit 14 .
When sufficient learning data is accumulated, the learning unit 15 generates a learned model 154 for judging the quality of the welding state from the arc sound based on the learning data accumulated in the storage unit 14. Run.
After the learned model 154 is generated, the learning unit 15 uses the learned model 154 to perform a process of determining whether the welding state is good or bad. The main control unit 11 transmits quality data indicating the quality of the welding state determined using the learned model 154 to the welding torch 3 via the communication unit 13, and notifies the welder of the quality of the welding state.

<Study Department>
FIG. 4 is a block diagram showing a configuration example of the learning unit 15 according to the first embodiment, and FIG. 5 is a conceptual diagram showing an example of learning data. The learning unit 15 includes a determiner 151 , an arc sound conversion unit 152 , a learning processing unit 153 and a trained model 154 .

The determiner 151 determines whether the welding state is good or bad based on the detection data acquired from the sensor 35 and outputs the good or bad data indicating the quality of the welding state to the learning processing section 153 . A good welding condition means that the welding current, welding voltage, welding wire feed speed, and short circuit condition during welding are normal. I can judge.
The determiner 151 includes, for example, a pass/fail determination trained model 151a. The pass/fail determination trained model 151a is a trained deep neural network, and has an input layer, an intermediate layer, and an output layer. When the detection data is input to the input layer, quality data indicating the quality of the welding state is output from the output layer. The configuration and type of the neural network are not particularly limited, and deep neural networks such as convolutional neural networks, recurrent neural networks, LSTM (Long Short-Term Memory), etc. , or a deep neural network that combines these. If the detected data is image data, a convolutional neural network should be used, and if the detected data is time-series data, a recursive neural network or the like should be used.
Since a relatively large amount of image data of welded parts associated with the quality of the welded state is accumulated, it is relatively easy to prepare learning data for generating the quality determination trained model 151a. be.
Note that the determiner 151 may acquire monitor data indicating the welding current or welding voltage during welding, and determine whether the welding state is good or bad based on the monitor data. Also, although an example of generating the determiner 151 by supervised learning of a neural network has been described, the determiner 151 may be machine-learned by other known methods such as logistic regression and support vector machine.

The arc sound conversion unit 152 converts arc sound data acquired from the microphone 37 into digital data for generating a learned model. The arc sound conversion unit 152 converts, for example, digital data of a waveform image representing the sound pressure of the arc sound. Needless to say, the data may be converted into time-series digital data. In the welding sound learning mode, arc sound conversion section 152 outputs arc sound data to learning processing section 153 . In the welding mode for judging whether the welding state is good or bad using the learned model 154 , the arc sound conversion unit 152 outputs arc sound data to the learned model 154 .

The learning processing unit 153 performs processing for collecting and accumulating data for learning, and processing for generating a trained model 154 . In the welding sound learning mode, particularly in the stage of collecting and accumulating data for learning, the learning processing unit 153 acquires arc sound data obtained by collecting the arc sound generated during welding with the microphone 37 each time welding is performed, Quality data indicating the quality of the welding state and welding condition data are acquired. Specifically, the learning processing unit 153 acquires arc sound data output from the arc sound conversion unit 152 . Also, the learning processing unit 153 acquires pass/fail data from the determiner 151 . Furthermore, the learning processing unit 153 acquires welding condition data from the main control unit 11 . Then, as shown in FIG. 5, learning data in which the obtained welding condition data, arc sound data, and quality data are associated with each other is stored in the storage unit 14 .
In the welding sound learning mode, when sufficient learning data is accumulated, the learning processing unit 153 reads out the accumulated learning data from the storage unit 14, and based on the read out learning data, learns from the arc sound. A trained model 154 for judging whether the welding state is good or bad is generated.
Furthermore, in the welding mode for judging the quality of the welding state using the learned model 154, the learning processing unit 153 selects a learned model corresponding to the welding condition from among the plurality of learned models 154 stored in the storage unit 14. 154 and read from the storage unit 14 .

When the welding sound learning mode is selected with the operation unit 33, the learning unit 15 executes the following processes. The welding power source 1 may be configured to automatically select the welding sound learning mode when the trained model 154 has not yet been generated.

<How to generate a learned model>
FIG. 6 is a flow chart showing a method of generating the trained model 154. As shown in FIG. The welding power supply monitors operations by the welder at the operation unit 33, and receives welding conditions (step S11). Welding condition data input to the welding torch 3 is transmitted from the welding torch 3 to the welding power source 1 . Welding power source 1 may receive the welding conditions.

Next, the learning unit 15 of the welding power source 1 acquires arc sound data from the microphone 37 via the arc sound conversion unit 152 (step S12). The learning unit 15 also acquires detection data from the sensor 35 (step S13), and determines whether the welding state is good or bad based on the detection data (step S14). Then, the learning unit 15 associates the welding condition data, the arc sound data, and the pass/fail data, and accumulates the data in the storage unit 14 as learning data (step S15).

Next, the learning unit 15 determines whether or not a predetermined amount of learning data has been accumulated in the storage unit 14 (step S16). If it is determined that the predetermined amount of learning data is not accumulated (step S16: NO), the learning unit 15 returns the process to step S11.

If it is determined that a predetermined amount of learning data is accumulated (step S16: YES), the learning unit 15 reads the learning data accumulated in the storage unit 14, and based on the read learning data, , when arc sound data is input, a learned model 154 that outputs the quality data corresponding to the arc sound data is generated by machine learning (step S17). For example, the weighting factor of the hidden layer can be obtained by minimizing the error function using the gradient descent method.
In step S17, the learning unit 15 generates a different learned model 154 for each of a plurality of welding conditions. For example, the learning unit 15 uses the learning data for the welding condition α to generate learning data for the welding condition α, and uses the learning data for the welding condition β to generate learning data for the welding condition β. Generate. The generated learned model 154 is stored in the storage unit 14 . Incidentally, the generation of the learned model 154 can be performed by determining whether or not sufficient learned models 154 have been accumulated for each welding condition, and generating the learned model 154 based on the prepared data for learning.

The trained model 154 comprises a neural network 154a as shown in FIG. The neural network 154a includes an input layer to which arc sound data is input, an intermediate layer that performs calculations based on learned weighting factors for the arc sound data input to the input layer, and pass/fail data indicating the pass/fail of the welding state. and an output layer that outputs
Arc sound data representing an arc sound is input to the input layer. The arc sound data may be image data representing the sound pressure waveform of the arc sound, or may be time-series data. Image information is input to the input layer via a convolution layer and a convolution layer (not shown).
The output layer has one or more neurons. The output layer comprises, for example, neurons indicating the probability that the weld condition is good and neurons indicating the probability that the weld condition is bad.

The configuration and type of the neural network 154a are not particularly limited, and may be a deep neural network such as a convolutional neural network, a recursive neural network, an LSTM, or a combination of these.
Also, an example of creating a learned model 154 by supervised learning of a neural network has been described, but a pass/fail judgment device for judging the pass/fail of a welding state based on arc sound data and pass/fail data may be a logistic regression or a support vector machine. For example, it may be generated by machine learning by other known methods.

When the welding mode is selected with the operation unit 33, the learning unit 15 executes the following processes. Note that, when the storage unit 14 stores the learned model 154, the welding power source 1 may be configured to automatically select the welding mode.

<Method for Determining Quality of Welding State Using Learned Model 154>
FIG. 7 is a flow chart showing a method for judging whether the welding state is good or bad and notifying the user. The welding power supply monitors operations by the welder at the operation unit 33, and receives welding conditions (step S31). Welding condition data input to the welding torch 3 is transmitted from the welding torch 3 to the welding power source 1 .

Next, learning unit 15 of welding power source 1 selects one learned model 154 corresponding to the welding condition from among a plurality of learned models 154 (step S32). Then, the learning unit 15 acquires arc sound data (step S33), and determines whether the welding state is good or bad using the learned model 154 selected in step S32 (step S34). Specifically, the learning unit 15 inputs the arc sound data to the input layer of the trained model 154, thereby causing the pass/fail data to be output from the output layer.

Next, the main control unit 11 transmits the quality data output from the learned model 154 to the welding torch 3 via the communication unit 13, and notifies the welder of the quality of the welding state by the notification unit 38 (step S35). ).

According to the first embodiment, it is possible to efficiently generate and store data for learning, and generate the trained model 154 for accurately determining the quality of the welding state from the arc sound.

In addition, the learned model 154 can be efficiently generated by performing machine learning for each welding condition.

Furthermore, the welding power supply can use the trained model 154 to determine the quality of the welding state from the arc sound and notify the welder. Specifically, the welding power supply acquires arc sound data during welding, inputs the acquired arc sound data to the learned model 154, and determines the welding state indicated by the pass/fail data output from the learned model 154. The quality can be reported by the reporting unit 38 .

Furthermore, the welding power supply selects a learned model 154 corresponding to the welding condition, and uses the selected learned model 154 to determine whether the welding state is good or bad, and notifies it. By using the learned model 154 suitable for the welding conditions, the welding state can be judged with higher accuracy and notified.

Furthermore, by using the quality determination learned model 151a, which is a trained neural network, it is possible to accurately determine the quality of the welded state from the detected data.

(Embodiment 2)
FIG. 8 is a schematic diagram showing a learning system according to the second embodiment. Welding power supply A according to Embodiment 2 differs from Embodiment 1 in that the learned model 154 is generated by a server. Therefore, the difference will be mainly described below. Since other configurations and effects are the same as those of the embodiment, corresponding portions are denoted by the same reference numerals and detailed description thereof is omitted.

The learning system according to the second embodiment includes a learning processing server 9 and a plurality of welding power supply devices A connected to the learning processing server 9 by wire or wirelessly.

The welding power source 1 of each welding power source A transmits learning data to the learning processing server 9 . The learning processing server 9 accumulates learning data transmitted from each welding power supply A, and generates a learned model 154 for each welding condition in the same manner as in the first embodiment. Then, the learning processing server 9 distributes the generated learned model 154 to each welding power supply A. FIG. Welding power supply A receives and stores learned model 154 distributed from learning processing server 9 . The method of reporting the quality of the welding state using the learned model 154 is the same as in the first embodiment.

According to Embodiment 2, learning data can be collected and accumulated more efficiently by receiving arc sound data from a plurality of welding power supply apparatuses A at the cloud learning processing server 9 . By performing machine learning using a large amount of learning data, it is possible to generate a learned model 154 that can more accurately determine the quality of the welding state from the welding sound.

Further, by distributing the learned model 154 from the learning processing server 9, the learned model 154 can be shared by a plurality of welding power supply units A.

In this embodiment, a semi-automatic welding power supply has been mainly described, but the present invention can also be applied to a welding system using a welding robot.

Further, although the welding power source 1 performs the process of judging whether the welding state is good or bad based on the detection data, the welding torch 3 may be configured to judge the good or bad of the welding state.

Furthermore, although the welding power source 1 side stores the learned model 154 and performs quality judgment using the learned model 154, the processing may be performed on the welding torch 3 side.

Furthermore, in the present embodiment, an example of selecting one learned model 154 corresponding to the selected welding condition has been described. It may be configured to select a plurality of learned models 154 that are close to the conditions. The learning unit 15 may input arc sound data to a plurality of learned models 154 and determine whether the welding state is good or bad based on quality data output from the plurality of learned models 154 .

Furthermore, in the present embodiment, the learned model 154 outputs pass/fail data indicating the pass/fail of the welding state, but the pass/fail data may be subdivided and output. For example, the welding power supply may be provided with a machine-learned model 154 that classifies and outputs data indicating that the welding state is unsatisfactory according to the cause of the failure.
For example, the trained model 154 includes a node that outputs data indicating the probability of being in a defective state due to a first cause of failure, a node that outputs data indicating a probability of being in a defective state due to a second cause of failure, and a node that outputs data indicating the probability of being in a defective state due to a second cause of failure. It is preferable to provide the output layer with a node indicating the probability of being in the state. Needless to say, the learning data includes a defective state due to a first defect cause, a defective state due to a second defect cause, and pass/fail data indicating a good state as teacher data, and the neural network is subjected to machine learning. I wish I could.
Furthermore, when the welding conditions are unsatisfactory, the welding power supply may be configured to output advice corresponding to the cause of the unsatisfactory condition to the notification unit 38 . The advice is information necessary for improving the welding condition, such as how to correct the welding method, causes of defects, and others.

The embodiments disclosed this time are illustrative in all respects and should be considered not restrictive. The scope of the present invention is indicated by the scope of the claims rather than the above-described meaning, and is intended to include all modifications within the scope and meaning equivalent to the scope of the claims.

1 welding power source, 2 wire feeder, 3 welding torch, 11 main control unit, 14a computer program, 15 learning unit, 151 determiner, 151a learned model for quality judgment, 152 arc sound conversion unit, 153 learning processing unit, 153a selection unit, 154 trained model, 154a neural network, 34 storage unit, 35 sensor, 36 control unit, 37 microphone, 38 notification unit

Claims (6)

A method for generating a trained model that outputs pass/fail data indicating the pass/fail of a welding state when arc sound data is input, comprising:
Acquire arc sound data obtained by collecting the arc sound generated during welding with a microphone,
Acquiring detection data obtained by detecting the state of the welded part by imaging the welded part after welding or scanning the welded part with detection light,
Determining whether the welding state is good or bad based on the acquired detection data,
accumulating learning data in which the acquired arc sound data and the pass/fail data indicating the pass/fail of the welding state are associated with each other in a storage unit for each of a plurality of different welding conditions ;
When the arc sound data is input based on the learning data of the plurality of welding conditions accumulated in the storage unit, the plurality of welding outputs the pass/fail data corresponding to the arc sound data. A trained model generation method that generates multiple trained models that differ for each condition by machine learning. 2. The quality of the welding state is determined by inputting the acquired detection data to a learned model for quality determination that outputs data indicating the quality of the welding state when the detection data is input. How to generate a trained model for . A computer program for generating a learned model that outputs quality data indicating the quality of a welding state when arc sound data is input to a computer,
to the computer;
Acquire arc sound data obtained by collecting the arc sound generated during welding with a microphone,
Acquiring detection data obtained by detecting the state of the welded part by imaging the welded part after welding or scanning the welded part with detection light,
Determining whether the welding state is good or bad based on the acquired detection data,
accumulating learning data in which the acquired arc sound data and the pass/fail data indicating the pass/fail of the welding state are associated with each other in a storage unit for each of a plurality of different welding conditions ;
When the arc sound data is input based on the learning data of the plurality of welding conditions accumulated in the storage unit, the plurality of welding outputs the pass/fail data corresponding to the arc sound data. A computer program for executing a process of generating a plurality of trained models that differ for each condition by machine learning. a microphone for picking up the arc sound;
a trained model that outputs pass/fail data indicating the pass/fail of a welding state when arc sound data is input;
A welding power source apparatus comprising: an informing unit for inputting arc sound data obtained by sampling with the microphone into the learned model, and informing the quality of the welding state indicated by the quality data output from the learned model. There is
There are a plurality of said learned models, and each learned model is machine-learned for each of a plurality of different welding conditions,
a selection unit that selects the learned model corresponding to the welding condition based on the welding condition;
The notification unit notifies the quality of the welding state indicated by the quality data output from the learned model selected by the selection unit.
Welding power supply . A method for generating a trained model that outputs pass/fail data indicating the pass/fail of a welding state when arc sound data is input, comprising:
Acquire arc sound data obtained by collecting the arc sound generated during welding with a microphone,
Acquiring detection data obtained by detecting the state of the welded part by imaging the welded part after welding or scanning the welded part with detection light,
Determining whether the welding state is good or bad by inputting the acquired detection data into a learned model for determining whether the welding state is good or bad when the detection data is input, which outputs data indicating the quality of the welding state,
accumulating learning data in which the obtained arc sound data and the pass/fail data indicating the pass/fail of the welding state are associated with each other in a storage unit;
Based on the learning data accumulated in the storage unit, a trained model is generated by machine learning that outputs the pass/fail data corresponding to the arc sound data when the arc sound data is input.
How to generate a trained model. A computer program for generating a learned model that outputs quality data indicating the quality of a welding state when arc sound data is input to a computer,
to the computer;
Acquire arc sound data obtained by collecting the arc sound generated during welding with a microphone,
Acquiring detection data obtained by detecting the state of the welded part by imaging the welded part after welding or scanning the welded part with detection light,
Determining whether the welding state is good or bad by inputting the acquired detection data into a learned model for determining whether the welding state is good or bad when the detection data is input, which outputs data indicating the quality of the welding state,
accumulating learning data in which the obtained arc sound data and the pass/fail data indicating the pass/fail of the welding state are associated with each other in a storage unit;
Based on the learning data accumulated in the storage unit, a trained model is generated by machine learning that outputs the pass/fail data corresponding to the arc sound data when the arc sound data is input.
A computer program for executing a process.

Images