KR102330429B1 - Deep learning based projection welding defect detection method - Google Patents

Abstract

In the deep learning-based projection welding defect detection method according to an embodiment of the present invention, an artificial neural network is learned using welding information, a detection model is generated based on the learning result, and the welding machine according to projection welding by the welding machine. The step of receiving the acquired input data and the step of determining whether a defect occurs due to projection welding using a detection model with respect to the received input data is included, and the input data may include current data and voltage data.

Description

Deep learning-based projection welding defect detection method {DEEP LEARNING BASED PROJECTION WELDING DEFECT DETECTION METHOD}

The present invention relates to a deep learning-based projection welding defect detection method implemented by a projection welding defect detection system that can be linked to a welding machine and a database. In more detail, it relates to a deep learning-based projection welding defect detection method for deriving whether a defect has occurred due to welding from voltage and current acquired during welding using a defect detection model learned using a deep learning algorithm. .

Projection welding is one of the representative welding methods of resistance welding, and it is a method of welding by applying pressure to the resistance heat generated in the process of transferring current to the contact part of the base material to be joined. 1 is an exemplary view for illustrating a resistance welding method, in which a large current flows through the welding base material as shown in (a) of FIG. It is a method of welding by making a molten state and applying mechanical pressure.

That is, projection welding corresponds to resistance welding such as welding in that it joins the materials to be welded by using the energy generated when a welding current is applied as a heat source in a state in which the materials to be welded are in close contact with each other. In general resistance welding, a current is passed to a welded part on a flat surface using a predetermined electrode, but projection welding is a point in that projection welding processes protrusions on one or both sides of the material to be welded before welding and concentrating current through the protrusions. It is distinct from welding or seam welding. Compared to spot welding using a round bar-shaped electrode, in projection welding, welding current and pressing force are applied to the block-shaped electrode from the top and bottom of the material to be welded with protrusions.

As shown in Fig. 1 (b), there are various types of welding defects. In particular, in the resistance welding including projection welding, determining the welding defects is a very difficult task and the defect determination accuracy is not high.

In the past, the inspector and the inspectee trace the weld line one by one to inspect the weldability, and the inspector visually reads the condition of the welded part and then determines whether there is a defect in the joint. However, the above inspection method has a problem in that the subjective judgment factor of the inspector is strongly reduced, and the inspection reliability is lowered due to the decrease in accuracy.

In addition, data such as current, voltage, pneumatic pressure, and temperature measured from the welding machine are monitored and patterns are analyzed. 2 is an exemplary view showing the process of data monitoring in order to detect possible defects during resistance welding. It is currently being judged. However, the reference value, which is the basis for judging normal and defective, is determined by a person through data collection, and the user has to find out what the normal or bad category of the data is, and it is difficult to use data in a non-linear relationship. .

Accordingly, there is a continuous need to develop a technology for a detection method that allows a user to easily determine whether welding is defective only by data on current and voltage obtainable from a welding machine.

Republic of Korea Patent Publication No. 10-2006-0076562 (published on July 4, 2006)

The present invention is to solve the above-described problems, and the defect detection is possible without the need for individual visual confirmation by the user by automatically deriving whether the welding defect is defective according to the result of learning the recursive neural network. An object of the present invention is to provide a method and a system.

As an embodiment of the present invention, a deep learning-based projection welding defect detection method implemented by a projection welding defect detection system capable of interworking with a welding machine and a database may be provided.

In the deep learning-based projection welding defect detection method according to an embodiment of the present invention, an artificial neural network is learned using welding information, a detection model is generated based on the learning result, and the welding machine according to projection welding by the welding machine. The step of receiving the acquired input data and the step of determining whether a defect occurs due to projection welding using a detection model with respect to the received input data is included, and the input data may include current data and voltage data.

In the deep learning-based projection welding defect detection method according to an embodiment of the present invention, the step of generating the detection model includes the steps of receiving welding information from a database and learning an artificial neural network based on the received welding information. is included, and the welding information may include resistance information and defect information according to the projection welding performed in advance.

In the deep learning-based projection welding defect detection method according to an embodiment of the present invention, the artificial neural network may be a recurrent neural network (RNN).

In the deep learning-based projection welding defect detection method according to an embodiment of the present invention, the step of receiving the input data obtained from the welding machine includes the step of decomposing the input data into a trend component, a seasonal component and an irregular component according to a time series analysis and inputting at least one of the decomposed trend component, seasonal component, and irregular component to the detection model.

In the deep learning-based projection welding defect detection method according to an embodiment of the present invention, the step of outputting an alarm signal generated according to the determination of whether a defect occurs may be further included.

In the deep learning-based projection welding defect detection method according to an embodiment of the present invention, the step of adjusting the projection current applied from the welding machine according to the determination of whether or not a defect may be further included.

As an embodiment of the present invention, a projection welding defect detection system capable of interworking with a welding machine and a database may be provided.

As an embodiment of the present invention, a computer-readable recording medium in which a program for implementing the above-described method is recorded may be provided.

In the projection welding defect detection system according to an embodiment of the present invention, an artificial neural network is learned using welding information, and a model generator in which a detection model is generated based on the learning result and acquired from a welding machine according to projection welding by a welding machine A control unit for receiving input data and determining whether a defect has occurred due to projection welding using a detection model with respect to the received input data is included, and the input data may include current data and voltage data.

According to the projection welding defect detection method of the present invention, there is an effect that an inspector can reduce the inconvenience of visually checking, and can accurately detect a defect according to the learning result of the deep learning algorithm.

In addition, by using a recursive neural network for data related to voltage or current, which is time-series information, higher defect detection accuracy may appear compared to other deep learning algorithms.

1 is an exemplary diagram illustrating a resistance welding method.
2 is an exemplary diagram illustrating a process of data monitoring in order to detect possible defects during resistance welding.
3 is a flowchart illustrating a projection welding defect detection method according to an embodiment of the present invention.
4 is a flowchart illustrating a step in which a detection model is generated in a method for detecting a defect in projection welding according to an embodiment of the present invention.
5 is a conceptual diagram of a recursive neural network (RNN).
6 is a view showing a defect detection result according to the projection welding defect detection method according to an embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those of ordinary skill in the art can easily implement them. However, the present invention may be embodied in many different forms and is not limited to the embodiments described herein. And in order to clearly explain the present invention in the drawings, parts irrelevant to the description are omitted, and similar reference numerals are attached to similar parts throughout the specification.

Terms used in this specification will be briefly described, and the present invention will be described in detail.

The terms used in the present invention have been selected as currently widely used general terms as possible while considering the functions in the present invention, but these may vary depending on the intention or precedent of a person skilled in the art, the emergence of new technology, and the like. In addition, in a specific case, there is a term arbitrarily selected by the applicant, and in this case, the meaning will be described in detail in the description of the corresponding invention. Therefore, the term used in the present invention should be defined based on the meaning of the term and the overall content of the present invention, rather than the name of a simple term.

When a part "includes" a certain element throughout the specification, this means that other elements may be further included, rather than excluding other elements, unless otherwise stated. In addition, terms such as "...unit" and "module" described in the specification mean a unit that processes at least one function or operation, which may be implemented as hardware or software, or a combination of hardware and software. . In addition, when a part is "connected" to another part throughout the specification, it includes not only a case in which it is "directly connected" but also a case in which it is connected "with another element in the middle".

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

As an embodiment of the present invention, a deep learning-based projection welding defect detection method implemented by a projection welding defect detection system capable of interworking with a welding machine and a database may be provided.

In this specification, a welder refers to a welding device operating in a projection welding method, and the database stores data received from the projection welding defect detection system, or a data management server for transferring the stored data to the detection system. refers to In addition, projection welding is one of the representative welding methods of resistance welding, and refers to a method of welding by applying pressure to resistance heat generated in the process of transferring current to the contact part of the base material to be joined.

3 is a flowchart illustrating a projection welding defect detection method according to an embodiment of the present invention.

Referring to Figure 3, the deep learning-based projection welding defect detection method according to an embodiment of the present invention is a step of learning an artificial neural network using welding information, and generating a detection model based on the learning result (S100), Receiving the input data obtained from the welding machine according to the projection welding by the welding machine (S200) and using the detection model for the received input data, the step of determining whether a defect occurs due to projection welding (S300) is included, The input data may include current data and voltage data.

Hereinafter, each step (S100-S300) of the deep learning-based projection welding defect detection method of the present invention will be described.

First, in the projection welding defect detection system, an artificial neural network may be trained using welding information, and a detection model may be generated based on the learning result (S100).

4 is a flowchart illustrating a step (S100) of generating a detection model in the projection welding defect detection method according to an embodiment of the present invention, and in more detail, the detection model is generated (S100) from the database. A step of receiving welding information (S110) and a step of learning an artificial neural network based on the received welding information (S120) may be included. In addition, the welding information may include resistance information and defect information according to the previously performed projection welding.

That is, the projection welding defect detection system may receive welding information stored in a database to be used as data for learning of the artificial neural network.

After receiving the welding information from the database, the artificial neural network is trained ( S120 ). In more detail, learning may be performed using the artificial neural network with respect to the correlation between the resistance information and the defect information.

The resistance information is information on resistance derived based on voltage and current measured during projection welding, and the resistance information may include any one of an instantaneous starting resistance and an average dynamic resistance. The instantaneous starting resistance may represent a value of a resistance calculated from data of current and voltage, and the average dynamic resistance may represent a value obtained by averaging the instantaneous starting resistance per half cycle of current and voltage.

The resistance information may be information obtained according to the projection welding performed in advance by the welding machine. More specifically, the resistance information is obtained by a sensor module provided to measure the current and voltage according to the welding progress on the welding machine. can

Through the artificial neural network, a detection model may be generated according to the learning result on the relationship between the resistance information and the defect information. That is, the detection model generated according to the artificial neural network learning result may be a model for determining whether a defect exists as input data according to projection welding is input.

The artificial neural network (ANN) is a machine learning algorithm that mathematically models the learning method of the brain in which information is stored according to the process of sending and receiving neurons through synapses for connection between neurons. In the artificial neural network, the neurons are nodes , the synapse may be defined as a weight. The artificial neural network is composed of an input layer, at least one hidden layer, and an output layer. In the input layer, each independent variable has a matching node, and in the hidden layer, the node and the previous layer There may be nodes generated by the combination of weights, and the output layer may include nodes generated by combining the neurons and weights of the last hidden layer, and a value output from the node of the output layer may be a dependent variable.

The artificial neural network may include any one of a deep neural network (DNN) and a recurrent neural network (RNN). In addition, the artificial neural network may be generated by combining at least one of a deep neural network (DNN) and a recursive neural network (RNN). In addition, various deep learning algorithms may be applied in addition to the aforementioned artificial neural network. The deep neural network (DNN) may be an artificial neural network composed of a plurality of hidden layers. In addition, unlike the deep neural network (DNN), the recursive neural network (RNN) is a neural network having a cyclic structure as shown in FIG. Because it remembers and reflects in learning, it corresponds to a neural network suitable for analysis of time series data.

In the projection welding defect detection method of the present invention, a recursive neural network (RNN) may use a gradient descent method similar to an artificial neural network, and backpropagation may be used. In addition, since it is learning according to the passage of time, learning based on Back-Propagation Through Time (BPTT) may proceed.

In the artificial neural network learning, an additional data processing procedure may be performed first before learning with respect to resistance information and defect information, which are datasets for learning.

Preferably, the resistance information and the defect information are as one data set, and the resistance values for each predetermined unit time (Ex. 1 ms) obtained from the sensor module of the welding machine during one welding and normal and defective It may be composed of an output variable for determining whether or not That is, when the welding result performed k times is included in the data set, k data sets may be provided. In addition, when the resistance value is obtained for each unit time of 0.1 second for 1 second, 10 resistance values may be included in one data set.

In particular, the output variable may be expressed as (y1, y2) in the form of a pair of data, where y1 corresponds to an output variable for determining whether it is normal, and y2 is an output for determining whether or not it is defective. It can be a variable. That is, when the welding result is normal, the output variable is designated as (1, 0) and learning is performed, and when the welding result is bad, the output variable is designated as (0, 1) and learned. The optimization of the detection model can be performed by configuring the dataset for learning as described above.

Hereinafter, in the deep learning-based projection welding defect detection method of the present invention, an artificial neural network more suitable for defect detection among deep neural networks (DNNs) and recursive neural networks (RNNs) according to a predetermined learning result will be described.

In the case of deep neural networks, the learning rate is 0.003, the learning range is 10000 times, the number of hidden layers is 4, the number of nodes in the first hidden layer is 16, the number of nodes in the second hidden layer is 8, and the number of nodes in the third hidden layer is 8. Learning was carried out by setting the number of nodes to 4 and the number of nodes in the fourth hidden layer to 2, and the result of the calculation of the loss function was 2.73E-03(%). According to the learning of the deep neural network, the accuracy in the occurrence of defects was found to be about 50%.

In the case of the recursive neural network, learning was carried out by setting the learning rate to 0.001, the learning range to 10000 times, and the number of nodes of the hidden layer to be 180, and the calculation result of the loss function was 5.45E-05 (%). According to the learning of the recursive neural network, it was confirmed that the accuracy in the occurrence of a defect was 83.3%.

That is, referring to the learning results of the deep neural network (FIG. 6 (a)) and the recursive neural network (FIG. 6 (b)), as shown in FIG. 6, resistance information according to voltage and current, which is time series data, is more It was confirmed that generating a detection model using a recursive neural network showed higher defect detection accuracy.

In the projection welding defect detection system, after the detection model is generated in step S100, input data obtained from the welding machine according to the projection welding by the welding machine may be received (S200).

That is, input data acquired from the welding machine may be transmitted to the detection system in order to determine whether the welding is defective during projection welding by the welding machine. The input data may include current data and voltage data acquired from the sensor module of the welding machine according to the progress of projection welding using the welding machine. In addition, the input data may be converted into resistance data by the acquired current data and voltage data and input to the detection model.

In addition, in the deep learning-based projection welding defect detection method according to an embodiment of the present invention, in the step (S200) of receiving the input data acquired from the welding machine, the input data is analyzed according to a time series of a trend component, a seasonal component, and an irregularity. A step of decomposition into components and a step of inputting at least one of the decomposed trend component, seasonal component, and irregular component to the detection model may be included.

That is, the current data and voltage data (or resistance data converted from them), which are the input data, correspond to time series data whose values can change with the passage of time. It can be broken down into components.

The trend component indicates a component related to the increase or decrease form when time series data increases or decreases, the seasonal component indicates a component that fluctuates according to a predetermined period (Ex. 1 year, 1 quarter, etc.), and the irregular component is It may indicate irregular or erroneous components that are difficult to predict. That is, the input data can be expressed by dividing the increase/decrease pattern of the input data, the variation pattern for each period, and the error according to the time series analysis.

As a time series analysis method for decomposing the input data, well-known techniques may be applied. For example, the time series analysis method may include a moving average method, a SEATS decomposition method, an STL decomposition method, and the like. However, since this is outside the scope of the present invention, detailed description thereof will be omitted.

That is, the input data obtained from the welding machine may be decomposed into a trend component, a seasonal component, and an irregular component, and at least one of the decomposed trend component, seasonal component, and irregular component may be input as a detection model. That is, all of the decomposed trend component, seasonal component, and irregular component may be input to the detection model, but, unlike this, only the trend component may be input to the detection model, or only the seasonal component and irregular component may be input to the detection model, and the trend component and Only irregular components excluding seasonal components may be input to the detection model. In addition, in the detection system, only one or two components among a trend component, a seasonal component, and an irregular component may be specified and input to the detection model.

As described above, when input data is decomposed into trend components, seasonal components, and irregular components and used as an input variable for input into a detection model, when only current data and voltage data are input without decomposition (without separate processing) In comparison, the accuracy of detecting defects according to welding may be further increased.

In addition, by allowing only one or two components to be input to the detection model without using all of the trend component, the seasonal component, and the irregular component, the defect detection efficiency can be increased according to the pattern of the input data.

In the projection welding defect detection system, using the generated detection model with respect to the input data received in step S200, it may be determined (S300) whether a defect has occurred due to the projection welding. That is, based on current data and voltage data (or resistance data) acquired according to welding by a welding machine, whether a defect has occurred according to the welding may be determined by the detection model. Determination of defect detection can be classified into normal or abnormal, and results can be derived.

In addition, in the deep learning-based projection welding defect detection method according to an embodiment of the present invention, the step of outputting an alarm signal generated according to the determination of whether a defect occurs may be further included. That is, when it is determined that a defect has occurred, an alarm signal may be generated by the detection system and output by an alarm output unit provided in the detection system, or an alarm signal may be output from the welding machine by transmitting the alarm signal to the welding machine. . The alarm signal may be output in various forms. For example, the alarm signal may include a voice signal, a vibration signal, a display signal, and the like, but this is merely exemplary and may be output in any form to notify the user of the occurrence of a welding defect.

In addition, in the deep learning-based projection welding defect detection method according to an embodiment of the present invention, the step of adjusting the projection current applied from the welding machine according to the determination of whether or not a defect may be further included.

The projection current relates to a current applied to a material to be welded (to be welded) during projection welding, and may be data for controlling an operation of a welding machine. That is, as the projection current is adjusted, the welding strength (degree) may be different. In addition, the projection current may coincide with the current data obtained from the welding machine, but may be a different value from the current data obtained from the welding machine because it may be additionally processed separately for operation control of the welding machine.

Preferably, the welding machine includes a sensor module for measuring current and voltage according to the progress of projection welding and a control module for controlling the projection current, and the control module determines that a defect has occurred in the projection welding defect detection system Then, the projection current is adjusted within a preset reference current range, and after the adjustment, the operation of the welding machine can be stopped according to the determination of whether or not a defect has occurred in the defect system for a predetermined reference time after the adjustment.

That is, as the real-time defect detection according to welding proceeds as described above, in the control module that can control the overall operation of the welding machine, the projection current is adjusted within the reference current range to prevent additional defects, and a predetermined reference time After that, if an additional defect occurs, real-time control may be performed so that the operation of the welding machine is temporarily stopped. Accordingly, it is possible to quickly detect defects during welding, stop work immediately upon detection, or prevent further defects.

Additionally, in addition to the detection model, a supplementary detection model may be utilized together with the detection model.

Preferably, the supplementary detection model is a model for supplementally and additionally detecting the occurrence of defects due to projection welding from the converted image data by visualizing the current data and voltage data, and a convolutional neural network (Convolutional Neural Network) for the image data Network, CNN) can be generated according to the learning result. In addition, the image data may be obtained according to a frequency conversion result for the current data and voltage data.

In other words, it is possible to determine with higher accuracy whether a defect has occurred during projection welding by using both the detection result of the complementary detection model for image data and whether the defect has occurred using the above-described detection model. For example, in the projection welding defect detection system, when it is determined that both the detection model and the supplementary detection model are normal, it may be finally confirmed that no defects are generated during welding. Contrary to this, even if the defect occurrence determination result using the detection model is confirmed to be normal, when the complementary detection model result is determined to be defective, re-detection using the present invention may be performed.

In other words, a complementary detection model can be generated according to the convolutional neural network learning result for image data derived by visualizing current data and voltage data, and projection welding is performed complementary to the detection model using the supplementary detection model. It can be determined whether a defect has occurred according to the

As an embodiment of the present invention, a projection welding defect detection system capable of interworking with a welding machine and a database may be provided.

In the projection welding defect detection system according to an embodiment of the present invention, an artificial neural network is learned using welding information, and a model generator in which a detection model is generated based on the learning result and acquired from a welding machine according to projection welding by a welding machine A control unit for receiving input data and determining whether a defect has occurred due to projection welding using a detection model with respect to the received input data is included, and the input data may include current data and voltage data. In addition, the projection welding defect detection system may further include a communication unit. In the communication unit, communication between components in the projection welding defect detection system, as well as communication with other devices such as a database, a welding machine, or other systems may be performed. As the communication method of the communication unit, various wired or wireless communication methods may be used, and the communication method is not limited to a specific communication method.

As an embodiment of the present invention, a computer-readable recording medium in which a program for implementing the above-described method is recorded may be provided.

That is, the above-described method can be written as a program that can be executed on a computer, and can be implemented in a general-purpose digital computer that operates the program using a computer-readable medium. In addition, the structure of the data used in the above-described method may be recorded in a computer-readable medium through various means. A recording medium for recording an executable computer program or code for performing various methods of the present invention should not be construed as including temporary objects such as carrier waves or signals. The computer-readable medium may include a storage medium such as a magnetic storage medium (eg, a ROM, a floppy disk, a hard disk, etc.) and an optically readable medium (eg, a CD-ROM, a DVD, etc.).

The above description of the present invention is for illustration, and those of ordinary skill in the art to which the present invention pertains can understand that it can be easily modified into other specific forms without changing the technical spirit or essential features of the present invention. will be. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. For example, each component described as a single type may be implemented in a distributed manner, and likewise components described as distributed may be implemented in a combined form.

Although the preferred embodiment of the present invention has been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements by those skilled in the art using the basic concept of the present invention as defined in the following claims are also provided. is within the scope of the That is, the scope of the present invention is indicated by the following claims rather than the above detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalents should be interpreted as being included in the scope of the present invention. do.

S100: An artificial neural network is learned using welding information, and a detection model is generated based on the learning result
S200: Receive input data acquired from the projection welding machine by the welding machine
S300: Determining whether or not a defect occurs due to projection welding using a detection model for the received input data
S110: Receive welding information from the database
S120: artificial neural network learning based on welding information

Claims (8)

In a deep learning-based projection welding defect detection method implemented by a projection welding defect detection system that can be linked with a welding machine and a database,
learning a recursive neural network using welding information, and generating a first detection model based on a learning result of the recursive neural network;
receiving input data acquired by the welding machine according to projection welding by the welding machine;
determining whether a defect has occurred due to the projection welding using the first detection model with respect to the received input data;
generating image data through visualization of the received input data;
learning a convolutional neural network using the image data, and generating a second detection model based on a learning result of the convolutional neural network; and
Complementarily determining whether a defect has occurred due to the projection welding using the second detection model;
The input data includes current data and voltage data, a deep learning-based projection welding defect detection method. The method of claim 1, wherein in the step of generating the first detection model,
receiving the welding information from the database; and
a step of learning the recursive neural network based on the received welding information;
The welding information includes a deep learning-based projection welding defect detection method that includes resistance information and defect information according to the projection welding performed in advance. 3. The method of claim 2,
The resistance information and the defect information,
As one data set, it is composed of resistance values for each predetermined unit time obtained from the sensor module of the welding machine during one welding and an output variable for determining whether it is normal or defective, a deep learning-based Projection welding defect detection method. The method of claim 3, wherein the receiving of the input data obtained from the welding machine comprises:
decomposing the input data into a trend component, a seasonal component, and an irregular component according to time series analysis; and
Deep learning-based projection welding defect detection method comprising the step of inputting at least one of the decomposed trend component, seasonal component, and irregular component to the detection model. The method of claim 1,
Deep learning-based projection welding defect detection method further comprising the step of outputting an alarm signal generated according to the determination of whether the defect has occurred. The method of claim 1,
Deep learning-based projection welding defect detection method further comprising the step of adjusting the projection current applied from the welding machine according to the determination of the occurrence of the defect. 4. The method of claim 3,
The output variable is
It is represented in the form of a pair of (y1, y2) data, where y1 is an output variable for determining whether it is normal, and y2 is an output variable for determining whether or not it is defective. Deep learning-based projection welding defect detection Way. In a projection welding defect detection system that can be linked with a welding machine and a database,
a model generator for learning an artificial neural network using welding information and generating a first detection model based on the learning result; and
A control unit that receives the input data obtained from the welding machine according to the projection welding by the welding machine, and determines whether a defect occurs due to the projection welding using the first detection model with respect to the received input data,
The control unit is
Image data is generated through visualization of the received input data, a convolutional neural network is trained using the image data, a second detection model is generated based on a learning result of the convolutional neural network, and the second detection Using a model to determine whether a defect has occurred due to the projection welding,
A projection welding defect detection system in which the input data includes current data and voltage data.

Images