KR20240015960A - Welding quality monitoring method and welding quality monitoring system - Google Patents

Abstract

The present invention relates to a welding quality monitoring method and a welding quality monitoring system that can predict and monitor welding quality during the welding process. A spectrum signal output step of sensing the state and outputting a plurality of spectrum signals as a signal in the form of a graph representing the change in light intensity according to the wavelength band, a predetermined number of times during a predetermined time during welding of the base material; In the spectrum signal output step, the plurality of spectrum signals output in chronological order for the predetermined time are calculated, and the distribution of the light intensity according to the wavelength of the welding light according to the time flow during the predetermined time is displayed by color classification. The correlation between the signal processing step derived from the color image map, the welding result of the experimental specimen derived through prior experiments, and the experimental color image map derived from the experimental welding light of the experimental specimen was learned as learning data. Based on a neural network, a welding quality prediction step of predicting the welding quality of the base material from the color image map derived in the signal processing step, and outputting the predicted welding state of the base material predicted in the welding quality prediction step as output data. A result output step may be included.

Description

Welding quality monitoring method and welding quality monitoring system}

The present invention relates to a welding quality monitoring method and a welding quality monitoring system, and more specifically, to a welding quality monitoring method and a welding quality monitoring system that can predict and monitor welding quality during the welding process.

Joining materials by welding has been used in a wide variety of industries for a long time, and in addition to arc welding and gas welding, a wide variety of welding methods have been developed and used, including special welding such as thermite welding, electro slag welding, and electron beam welding.

In recent years, as the need for precision welding of various parts has increased in the automobile, aviation, nuclear, and secondary battery industries, the scope of use of laser welding using high-power industrial lasers is gradually expanding. In addition, in order to improve the efficiency of mass production, automation of welding using welding robots has been continuously developed and utilized for several decades. In order to automate welding, a method that can monitor welds in real time is essential to obtain stable welding quality.

Conventionally, a method of analyzing light emitted from a welded area has been proposed to monitor the quality of welding. This utilizes the principle that the nature of the light emitted from the weld zone varies depending on whether the welding was performed correctly.

However, the conventional welding quality monitoring using light had a problem in that the light emitted from the weld zone was affected by many factors and appeared in many different ways, resulting in low accuracy in predicting and monitoring the welding quality. For example, the light emitted from the weld area during the welding process can appear very different depending on the material and dimensions of the base material being welded, the intensity of the irradiated laser, and the relative moving speed. The difference between a good weld and a bad weld can also vary depending on each wavelength of light. It may appear differently. As such, conventional welding quality monitoring using light has the problem of difficulty in accurately determining whether the weld is good or bad by comprehensively considering various characteristics that may appear over a very wide range during welding.

The present invention is intended to solve various problems including the problems described above. During laser welding, the welding light generated at the welding site is sensed in a dispersed state by wavelength through a spectroscope to determine the quality of welding. Alternatively, the purpose is to provide a welding quality monitoring method and a welding quality monitoring system that can determine defects with high accuracy. However, these tasks are illustrative and do not limit the scope of the present invention.

According to one embodiment of the present invention, a welding quality monitoring method is provided. The welding quality monitoring method is capable of evaluating the welding quality of the base material by monitoring and analyzing the welding light generated at the welding area of the base material in real time during welding of the base material, wherein an image sensor is used to monitor and analyze the welding quality of the base material. The welding light generated momentarily at the welding site is sensed in a dispersed state by wavelength through a spectrometer, and this is a signal in the form of a graph representing the change in the intensity of light according to the wavelength band, and during welding of the base material, a predetermined A spectrum signal output step of outputting a plurality of spectrum signals a predetermined number of times for a period of time; In the spectrum signal output step, the plurality of spectrum signals output in chronological order for the predetermined time are calculated, and the distribution of the light intensity according to the wavelength of the welding light according to the time flow during the predetermined time is displayed by color classification. A signal processing step to derive a color image map; The correlation between the welding results of the experimental specimen derived through prior experiments and the experimental color image map derived from the experimental welding light of the experimental specimen is derived in the signal processing step based on an artificial neural network learned with learning data. A welding quality prediction step of predicting the welding quality of the base material from the color image map; and a result output step of outputting the predicted welding state of the base material predicted in the welding quality prediction step as output data.

According to one embodiment of the present invention, the signal processing step includes: a signal acquisition step of acquiring the plurality of spectrum signals output from the spectrum signal output step in chronological order for the predetermined time; A band setting step of setting a plurality of band areas having different predetermined wavelength band ranges in the plurality of spectrum signals; and calculating a plurality of average values of the light intensity within each band region from each individual spectrum signal acquired in time order, and calculating the plurality of average values for each wavelength of the welding light according to the time order in which the plurality of spectrum signals were acquired. It may include an image drawing output step of outputting the distribution of the average value of the color as the color image map.

According to one embodiment of the present invention, the spectrum signal output step includes: outputting the welding light sensed in a dispersed state by wavelength through the spectrometer at a first time as a first spectrum signal; A second spectrum output step of outputting the welding light sensed in a dispersed state by wavelength through the spectrometer at a second time at a predetermined time interval from the first time as a second spectrum signal; And an n-th spectrum output step of outputting the welding light sensed in a dispersed state by wavelength through the spectrometer at the n-th time having the predetermined time interval from the n-1 time as an n-th spectrum signal. You can.

According to one embodiment of the present invention, the predetermined time is the time from the first time to the nth time, and in the spectrum signal output step, the plurality of spectrum signals are n during the predetermined time. can be printed.

According to an embodiment of the present invention, the signal acquisition step includes converting the first spectrum signal to the nth spectrum signal output from the spectrum signal output step at the predetermined time interval from the first time to the nth time. , and the band setting step includes: a first area setting step of setting a first band area having a predetermined wavelength band range in the plurality of spectrum signals including the first to the nth spectrum signals; A second area setting step of setting a second band area having a predetermined wavelength band range higher than the first band area in the plurality of spectrum signals; and an n-th region setting step of setting an n-th band region having a predetermined wavelength band range higher than the n-1-th band region in the plurality of spectrum signals.

According to one embodiment of the present invention, the image drawing output step is to set the average value of the light intensity in the first band region as a 1-1 average value in the first spectrum signal acquired at the first time. Calculating, calculating the average value of the light intensity in the second band region as a 1-2 average value, and calculating the average value of the light intensity in the n-th band region as a 1-n average value Average value calculation step; In the second spectrum signal acquired at the second time, the average value of the light intensity in the first band region is calculated as a 2-1 average value, and the average value of the light intensity in the second band region is calculated as a 2-1 average value. A second average value calculation step of calculating as a 2-2 average value, and calculating the average value of the light intensity in the n-th band region as a 2-n average value; And in the n-th spectrum signal acquired at the n-th time, the average value of the light intensity in the first band region is calculated as the n-1 average value, and the light intensity in the second band region is calculated as the n-1 average value. An nth average value calculation step of calculating the average value as the n-2th average value, and calculating the average value of the light intensity in the nth band region as the n-nth average value; comprising the first average value calculation step to the first average value. The plurality of average values, including the 1-1st average value to the n-nth average value of the light intensity calculated in the n average value calculation step, are divided into colors according to the light intensity, and the plurality of average values are the first The color image map expressed as a color distribution may be output so that the welding light is distributed by wavelength in chronological order from time to the nth time.

According to an embodiment of the present invention, in the image drawing output step, the light intensity indicated by each average value of the plurality of average values can be distinguished based on the difference in color brightness.

According to an embodiment of the present invention, the first average value calculation step uses each band region including the first band region to the nth band region as an independent variable, and the first average value calculated in each band region is used as an independent variable. Using the -1 average value or the 1-n average value as the dependent variable, a correlation is obtained as a result of a regression analysis that estimates the relationship between variables, and the average value of the light intensity is calculated using the obtained correlation. The expected average value of the light intensity of the unused band area may be derived, and the image drawing output step may include the expected average value in the color distribution when outputting the color image map.

According to an embodiment of the present invention, the welding quality prediction step includes storing the experimental color image map derived from the experimental welding light generated during welding of the experimental specimen and the welding result of the experimental specimen as learning data. storage step; A learning step of mechanically learning the correlation between the experimental color image map stored as the learning data and the welding results of the experimental specimen by the artificial neural network; and a determination step of determining welding quality of the base material from the color image map derived in the signal processing step, based on the artificial neural network.

According to an embodiment of the present invention, the information storage step includes deriving a plurality of welding results of the experimental color image map and the corresponding experimental specimen through experimentation in advance and storing them as the learning data, and the learning step. Can learn a plurality of the learning data based on the artificial neural network.

According to one embodiment of the present invention, the welding quality prediction step may further include an optimization step of optimizing the artificial neural network so as to minimize the occurrence of errors in the artificial neural network learned in the learning step.

According to one embodiment of the present invention, the optimization step includes, during welding of the base material, the predicted welding state of the base material predicted from the color image map derived in the signal processing step based on the artificial neural network, and A comparison learning step of comparing the actual welding results of the base material confirmed after completion of welding of the base material and deriving the sum of the errors of the predicted welding state compared to the actual welding results as a loss function; and a correction step of reflecting the optimization result for minimizing the error of the predicted welding state in the artificial neural network of the learning step by considering the loss function derived in the comparative learning step.

According to one embodiment of the present invention, the signal processing step converts the color image map output in the image drawing output step into RGB values according to the CIE 1931 It is a signal in the form of a graph representing the change in the RGB values according to the order, and may further include a signal processing step of outputting the RGB value change graph separately for each RGB region.

According to one embodiment of the present invention, the signal processing step is the average of the R value in the RGB value change graph ( ), average of G values ( ), average of B values ( ), standard deviation of R value (STD(R)), standard deviation of G value (STD(G)), standard deviation of B value (STD(B)), maximum intensity wavelength average ( ) and the standard deviation of the maximum intensity wavelength (STD( )) can be extracted as an input variable.

According to one embodiment of the present invention, the information storage step includes outputting an experimental RGB value change graph from the experimental color image map, extracting experimental input variables from the experimental RGB value change graph, and storing them as the learning data. variable storage step; And an output variable storage step of extracting the welding results of the experimental specimen as a learning output variable divided into a total of four types: unjoined, insufficient penetration, full penetration, and not joined due to a gap, and storing the learning data as the learning data. In this step, the correlation between the experimental input variables and the learning output variables can be mechanically learned by the artificial neural network.

According to one embodiment of the present invention, the output variable storage step processes the largest value among the non-bonding, the lack of penetration, the complete penetration, and the non-bonding due to the gap as 1 and the remaining values as 0. , the learning output variable can be extracted as a combination of numbers 1 and 0.

According to one embodiment of the present invention, the decision step is to extract the input variable from the input variable extraction step based on the artificial neural network that mechanically learned the correlation between the experimental input variable and the learning output variable in the learning step. The welding quality of the base material can be determined by calculating output variables from the input variables.

According to another embodiment of the present invention, a weld quality monitoring system is provided. The welding quality monitoring system is installed on a welding head that performs welding of the base material by irradiating a laser beam to the welding area of the base material, is installed on the optical path of the laser beam, and is generated at the welding area of the base material to the welding head. A spectral sensing unit that specifies the incident welding light to form a spectrum distributed by wavelength and senses the spectrum as an image signal; And a monitoring unit that monitors the predicted welding state of the base material according to the image signal received from the spectral sensing unit, wherein the monitoring unit uses the image signal sensed through the spectral sensing unit to determine a predetermined condition during welding of the base material. A spectrum signal output unit that is applied a predetermined number of times for a period of time and outputs a plurality of spectrum signals as a signal in the form of a graph representing the change in light intensity according to the wavelength band of the image signal; The spectrum signal output unit calculates the plurality of spectrum signals output in chronological order for the predetermined time, and displays the distribution of the light intensity according to the wavelength of the welding light according to the flow by time during the predetermined time, classified by color. A signal processing unit derived from a color image map; Based on an artificial neural network that learned the correlation between the welding results of the experimental specimen, which was derived through a prior experiment, and the experimental color image map derived from the experimental welding light of the experimental specimen, as learning data, the signal processor derived a welding quality prediction unit that predicts welding quality of the base material from the color image map; and an output unit that outputs the predicted welding state of the base material predicted by the welding quality prediction unit as output data.

According to an embodiment of the present invention as described above, during laser welding, the image signal sensed by dispersing the welding light generated at the welding site by wavelength through a spectrometer is used to measure the change in light intensity according to the wavelength band. A color image map derived in the form of a contour map of the distribution of light intensity by calculating the plurality of output spectrum signals is applied to an artificial neural network that learns the correlation between the welding state and the color image map form. Thus, it is possible to determine with high accuracy whether the welding is in good or bad condition in real time.

In addition, rather than using the color image map as the original signal data, the color image map is converted into RGB values according to the CIE 1931 It is a signal in the form of a standardized graph, and by processing it into a graph of changes in RGB values separately represented by each RGB region and applying it to an artificial neural network, the data to be processed can be reduced in terms of the amount of calculation. Accordingly, by allowing users to monitor through a small number of clear and standardized signals, it is possible to implement a welding quality monitoring method and a welding quality monitoring system that can intuitively monitor welding quality and increase convenience. Of course, the scope of the present invention is not limited by this effect.

1 is a conceptual diagram schematically showing a welding quality monitoring system according to an embodiment of the present invention.
FIG. 2 is a flow chart sequentially showing an embodiment of a welding quality monitoring method using the welding quality monitoring system of FIG. 1.
3 and 4 are flowcharts and conceptual diagrams showing an embodiment of the signal processing step of the welding quality monitoring method of FIG. 2.
FIG. 5 is a flow chart sequentially showing an embodiment of the welding quality prediction step of the welding quality monitoring method of FIG. 2.
FIG. 6 is a flowchart sequentially showing another embodiment of a welding quality monitoring method using the welding quality monitoring system of FIG. 1.
Figure 7 is an image showing the results of an experiment using the welding quality monitoring method of the present invention.
FIGS. 8 and 9 are flowcharts and conceptual diagrams showing another embodiment of the signal processing step of the welding quality monitoring method of FIG. 2.
FIG. 10 is a table showing input variables that can be extracted through the signal processing step of FIG. 8 and output variables according to the input variables.
FIG. 11 is a flowchart sequentially showing another embodiment of the welding quality prediction step of the welding quality monitoring method of FIG. 2.
Figure 12 is a graph showing the results of an experiment using the welding quality monitoring method of the present invention.

Hereinafter, various preferred embodiments of the present invention will be described in detail with reference to the attached drawings.

The embodiments of the present invention are provided to more completely explain the present invention to those skilled in the art, and the following examples may be modified into various other forms, and the scope of the present invention is as follows. It is not limited to examples. Rather, these embodiments are provided to make the present disclosure more faithful and complete and to fully convey the spirit of the present invention to those skilled in the art. Additionally, the thickness and size of each layer in the drawings are exaggerated for convenience and clarity of explanation.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will now be described with reference to drawings that schematically show ideal embodiments of the present invention. In the drawings, variations of the depicted shape may be expected, for example, depending on manufacturing technology and/or tolerances. Accordingly, embodiments of the present invention should not be construed as being limited to the specific shape of the area shown in this specification, but should include, for example, changes in shape resulting from manufacturing.

Figure 1 is a conceptual diagram schematically showing a welding quality monitoring system 100 according to an embodiment of the present invention, and Figure 2 is a schematic diagram of an embodiment of a welding quality monitoring method using the welding quality monitoring system 100 of Figure 1. It is a flowchart shown as follows, and FIGS. 3 and 4 are flowcharts and conceptual diagrams showing an embodiment of the signal processing step (S200) of the welding quality monitoring method of FIG. 2. In addition, Figure 5 is a flowchart sequentially showing an embodiment of the welding quality prediction step (S300) of the welding quality monitoring method of Figure 2, and Figure 6 is another implementation of the welding quality monitoring method using the welding quality monitoring system of Figure 1. It is a flow chart showing examples in order, and Figure 7 is an image showing the results of an experiment using the welding quality monitoring method of the present invention.

First, as shown in FIG. 1, the welding quality monitoring system 100 according to an embodiment of the present invention is a laser welding method that performs welding of the base material (S) by irradiating the base material (S) with the laser beam (1). The device may be a system capable of evaluating the welding quality of the base material (S) by monitoring and analyzing the welding light (2) generated at the welding area (W) of the base material (S) in real time.

For example, as shown in FIG. 1, the laser welding device to which the welding quality monitoring system 100 according to an embodiment of the present invention can be applied includes a light generator 20 that generates a laser beam 1, and It may include a welding head 10 that performs welding of the base material S by irradiating the laser beam 1 generated by the light generating unit 20 to the welding area W of the base material S. In addition, although not shown, the laser welding device includes a support that supports the base material (S) while welding is performed by the welding head (10) and a transport that can move the support relative to the welding head (10). It may include means.

In this laser welding apparatus, when welding is performed by irradiating the laser beam 1 from the welding head 10 toward the welding area W of the base material S, the welding area W of the base material S is performed. At least a partially melted molten pool and keyhole may be formed, spatter may scatter, and welding light 2 may be generated. This welding light 2 may include electromagnetic waves and plasma ranging in the infrared region, visible region, and ultraviolet region.

At this time, the welding device to which the welding quality monitoring system 100 according to an embodiment of the present invention can be applied is not necessarily limited to the laser welding device described above, and may be applied to other types of welding devices. For example, it can be applied to plasma welding, TIG (Tungsten ingot gas welding) welding, MIG (Metal inert gas welding) welding, MAG (Metal active gas welding) welding, and gas welding. In this case, the laser beam (1) It can be replaced by arc or welding flame.

In this way, during welding of the base material (S) in the above-described laser welding device, the welding light (2) generated at the welding area (W) of the base material (S) is monitored and analyzed in real time to ensure the welding quality of the base material (S). The welding quality monitoring system 100 capable of evaluating may largely include a spectral sensing unit 110 and a monitoring unit 120.

As shown in FIG. 1, the spectral sensing unit 110 irradiates the laser beam 1 generated by the light generating unit 20 to the welding site W of the base material S to weld the base material S. In the welding head 10, which is installed on the optical path of the laser beam 1, the welding light 2 generated at the welding site W of the base material S and incident on the welding head 10 is spectralized. Thus, the spectrum 3 is formed in a distributed state by wavelength, and the spectrum 3 can be sensed as an image signal.

For example, the spectral sensing unit 110 is installed within the welding head 10 and reflects the welding light 2 incident on the welding head 10 in a direction different from the optical path of the laser beam 1. ), a spectrometer 112 that specifies the welding light 2 reflected by the mirror 111 into a spectrum 3 distributed by wavelength, and an image that senses the spectrum 3 as the image signal. A sensor 113, a lens barrel 114 with a plurality of lenses installed inside to converge the welding light 2 reflected by the mirror 111, and a spectrometer It may be configured to include an optical cable 115 transmitted to (112).

More specifically, the mirror 111 of the spectral sensing unit 110 is installed to be disposed in the optical path of the laser beam 1 within the welding head 10, and transmits the laser beam from the welding site W of the base material S. The welding light 2 radiated coaxially with the direction in which the beam 1 is irradiated can be received. Accordingly, the spectral sensing unit 110 can receive information according to the shape of the molten pool and keyhole formed at the welding site W of the base material S in a form that is as undistorted as possible.

As the mirror 111, a half mirror that transmits the laser beam 1 and reflects the welding light 2 incident from the base material S to the welding head 10 may be used, and the laser beam It can be installed inclined at a predetermined inclination angle to reflect the welding light (2) incident coaxially with the optical path of (1) in the direction of the barrel 114.

In addition, although not shown, a plurality of lenses are disposed on the lens barrel 114, into which the welding light 2 reflected by the mirror 111 is incident, to condense the welding light 2 and then connect it to one end of the optical cable 115. You can join the company as a deputy. At this time, it may be desirable for the lenses disposed in the barrel 114 not to be coated with an ultraviolet ray blocking coating. This is because there is a high possibility that meaningful information contained in the welding light 2 generated during welding is included in the ultraviolet region.

In addition, the other end of the optical cable 115 is disposed close to the spectrometer 112, so that the welding light 2 incident on one end through the barrel 114 is irradiated to the spectroscope 112.

Accordingly, the welding light 2 irradiated by the spectrometer 112 is dispersed by wavelength to form a spectrum 3, and the image sensor 113 is arranged so that this spectrum 3 is distributed on the light-receiving surface. You can.

Here, the spectroscope 112 is shown as using a prism, but it is not necessarily limited to FIG. 1, and a diffraction grating may be used as needed. In addition, the image sensor 113 may be a charge-coupled device (CCD) or a complementary metal-oxide semiconductor (CMOS). In addition, any sensor capable of detecting everything from the infrared region to the ultraviolet region may be selected and used. .

In addition, in the above-described spectral sensing unit 110, the spectrometer 112 is placed remotely separately from the welding head 10 using the optical cable 115 in order to minimize the size of the welding head 10. This may be to prevent interference with the welding head 10 even when the base material S has various shapes, such as when irregularities or protrusions are formed.

Accordingly, when the spectrum 3 is irradiated to the light-receiving surface of the image sensor 113 of the spectral sensing unit 110, the image sensor 113 can output the image signal accordingly. At this time, the output image signal may be a signal containing information on the intensity of each wavelength included in the welding light 2.

Additionally, as shown in FIG. 1, the monitoring unit 120 may monitor the predicted welding state of the base material S according to the image signal received from the spectral sensing unit 110 described above.

For example, the monitoring unit 120 receives the image signal sensed through the spectral sensing unit 110 a predetermined number of times during welding of the base material (S), and transmits the image signal to the wavelength band. It is a signal in the form of a graph representing the change in light intensity according to the change, including a spectrum signal output unit 121 that outputs a plurality of spectrum signals (SS), and a spectrum signal output unit 121 that outputs a plurality of spectrum signals (SS) in chronological order for the predetermined time. A signal derived from a color image map (M) that calculates a plurality of spectral signals (SS) and shows the distribution of the intensity of the light according to the wavelength of the welding light (2) according to the time flow during the predetermined time by color. Based on an artificial neural network that learned the correlation between the processing unit 122, the welding result of the experimental specimen derived through a prior experiment, and the experimental color image map derived from the experimental welding light of the experimental specimen as learning data, A welding quality prediction unit 123 that predicts the welding quality of the base material (S) from the color image map (M) derived from the signal processing unit 122, and predictive welding of the base material (S) predicted by the welding quality prediction unit 123. It may be configured to include an output unit 124 that outputs the status as output data through a user terminal 30 such as a monitor, and the welding quality prediction unit 123 is a test specimen derived through a prior experiment. It may include a database 123a in which the welding results and the experimental color image map (color image map corresponding to the welding result of the experimental specimen) derived from the experimental welding light of the experimental specimen are stored.

Hereinafter, the welding quality monitoring method according to an embodiment of the present invention, which is performed in the monitoring unit 120 of the welding quality monitoring system 100 described above, will be described in detail.

For example, as shown in FIG. 2, the welding quality monitoring method monitors and analyzes in real time the welding light 2 generated at the welding site W of the base material S during welding of the base material S. As a method of evaluating the welding quality of the base material (S), the signal output step (S100), the signal processing step (S200), the welding quality prediction step (S300), and the output step (S400) can be performed in this order. .

First, the signal output step (S100) is a step performed through the spectrum signal output unit 121, where the image sensor 113 transmits the welding light 2 instantaneously generated at the welding site W of the base material S. Receives the image signal sensed and output in a dispersed state by wavelength through the spectrometer 112, and converts the image signal into a signal in the form of a graph representing the change in light intensity according to the wavelength band, of the base material (S). During welding, a plurality of spectrum signals (SS) can be output a predetermined number of times during a predetermined time.

More specifically, as shown in FIG. 4, the spectrum signal output step (S100) transmits the welding light (2) sensed in a dispersed state by wavelength through the spectrometer 112 at the first time (T1) to the first A first spectrum output step (S110) of outputting a spectrum signal (SS-1), and sensing dispersed by wavelength through the spectrometer 112 at a second time with a predetermined time interval from the first time (T1). A second spectrum output step of outputting the welding light 2 as a second spectrum signal (not shown) through the spectrometer 112 at the n-th time (Tn) having the predetermined time interval from the n-1 time. It may include an n-th spectrum output step (S1n0) in which the sensed welding light 2 dispersed by wavelength is output as an n-th spectrum signal (SS-n).

At this time, the predetermined time during which the plurality of spectrum signals SS are output is from the first time T1 to the nth time Tn, and the plurality of spectrum signals SS are output during the predetermined time. n numbers may be output at each predetermined time interval.

The plurality of spectrum signals (SS) output in the spectrum signal output step (S100) are data instantaneously measured by the welding light 2 dispersed by wavelength through the spectrometer 112, for example, at 5,000 per second. As in times, the predetermined time, which is the time from the first time (T1) to the nth time (Tn), and the number of times (n) of the plurality of spectrum signals (SS) vary greatly depending on the user's needs. can be set.

Next, as shown in FIG. 2, the signal processing step (S200) is performed through the signal processing unit 122, and a plurality of spectra are output in chronological order during the predetermined time in the spectrum signal output step (S100). By calculating the signal (SS), a color image map (M) can be derived that shows the distribution of the intensity of the light by wavelength of the welding light (2) according to the time flow during the predetermined time by dividing it into colors.

For example, as shown in FIG. 3, the signal processing step (S200) includes a signal acquisition step (S210) of acquiring a plurality of spectrum signals (SS) output in the spectrum signal output step (S100) in chronological order during the predetermined time. ) and a band setting step (S220) of setting a plurality of band areas (BA) having different predetermined wavelength band ranges in the plurality of spectrum signals (SS), and in each individual spectrum signal acquired in time order, each band Calculate a plurality of average values of the intensity of the light within the area, and color-code the distribution of the plurality of average values for each wavelength of the welding light 2 according to the time order in which the plurality of spectrum signals SS were acquired. The process may proceed in the order of the image drawing output step (S230) of outputting the indicated color image map (M).

More specifically, as shown in FIG. 4, in the signal processing step (S200), the signal acquisition step (S210) includes the first to nth spectrum signals (SS-1) output from the spectrum signal output step (S100). The spectrum signal SS-n can be acquired at the predetermined time interval from the first time T1 to the nth time Tn.

Next, in the band setting step (S220), a first band having a predetermined wavelength band range is selected from the plurality of spectrum signals (SS) including the first spectrum signal (SS-1) to the n-th spectrum signal (SS-n). A first area setting step of setting an area (BA-1), and a second band area (BA) having a predetermined wavelength band range higher than the first band area (BA-1) in the plurality of spectrum signals (SS) A second area setting step of setting -2), and a third band area (BA-3) having a predetermined wavelength band range higher than the second band area (BA-2) in the plurality of spectrum signals (SS). A third area setting step of setting and an nth area setting of setting the nth band area (BA-n) having a predetermined wavelength band range higher than the n-1th band area in the plurality of spectrum signals (SS). May include steps.

The plurality of band areas (BA) set in the band setting step (S220) are not necessarily limited to the number (4) shown in FIG. 4, and are not limited to the required accuracy of welding quality monitoring or the welding quality monitoring method. Depending on the specifications of the system being operated on, this number can be set to very different numbers.

Next, in the image drawing output step (S230), the average value of the light intensity in the first band area (BA-1) is calculated from the first spectrum signal (SS-1) acquired at the first time (T1). Calculated as a 1-1 average value, the average value of the light intensity in the second band area (BA-2) is calculated as the 1-2 average value, and the light intensity in the third band area (BA-3) A first average value calculation step of calculating the average value of as the 1-3 average value and calculating the average value of the light intensity in the n-th band area (BA-n) as the 1-n average value, and at the second time In the acquired second spectrum signal (not shown), the average value of the light intensity in the first band area (BA-1) is calculated as the 2-1 average value, and in the second band area (BA-2) The average value of the light intensity is calculated as the 2-2 average value, the average value of the light intensity in the third band area (BA-3) is calculated as the 2-3 average value, and the n-th band area (BA-3) is calculated as the 2-3 average value. A second average value calculation step of calculating the average value of the light intensity at -n) as a 2-n average value, and in the nth spectrum signal (SS-n) acquired at the nth time (Tn), the first band region The average value of the light intensity in (BA-1) is calculated as the n-1th average value, and the average value of the light intensity in the second band area (BA-2) is calculated as the n-2th average value, The average value of the light intensity in the third band area (BA-3) is calculated as the n-3th average value, and the average value of the light intensity in the nth band area (BA-n) is calculated as the n-nth average value. It may include an nth average value calculation step.

Accordingly, through the image drawing output step (S230), as shown in FIG. 4, the 1-1 average value of the light intensity calculated in the first average value calculation step to the nth average value calculation step to the The plurality of average values, including the n-nth average value, are divided into colors according to the intensity of the light, and the plurality of average values are welding light (2) according to time order from the first time (T1) to the nth time (Tn). A color image map (M) expressed as a color distribution can be output so that it is distributed by wavelength.

At this time, as shown in FIG. 4, in the image drawing output step (S230), the lower the light intensity, the darker it is, and the higher the light intensity, the brighter it is expressed, so that each of the plurality of average values is determined by the difference in color brightness. The intensity of the light represented by the average value of can be distinguished.

In addition, the image drawing output step (S230) uses a regression equation to derive the expected average value of the light intensity of the band area for which the average value of the light intensity has not been calculated, thereby generating the output color image map (M). Accuracy can be further increased.

For example, the first average value calculation step of the image drawing output step (S230) uses each band region including the first band region (BA-1) to the nth band region (BA-n) as an independent variable, and the Obtaining a correlation (calculating a regression equation) as a result of a regression analysis that estimates the relationship between variables, using the 1-1st average value to the 1-nth average value calculated in each band area as a dependent variable, By using the obtained correlation to derive the expected average value of the light intensity of the band area for which the average value of the light intensity has not been calculated, the image drawing output step (S230) is performed when outputting the color image map (M). The expected average value can be included in the color distribution.

At this time, in the second average value calculation step and the nth average value calculation step, the light intensity of the band region in which the average value of the light intensity is not calculated is calculated using regression analysis like the above-described first average value calculation step. The expected average value can be derived. Therefore, detailed description is omitted.

Subsequently, as shown in FIG. 2, the welding quality prediction step (S300) involves welding results of the experimental specimen derived through prior experimentation, and the experimental color image derived from the experimental welding light of the experimental specimen. This step is performed through the welding quality prediction unit 123 including the database 123a in which the map is stored, and is derived from the welding results of the experimental specimen derived through prior experiments and the experimental welding light of the experimental specimen. Based on the artificial neural network that learned the correlation between the experimental color image maps as learning data, the welding quality of the base material (S) is measured in real time during the welding process from the color image map (M) derived in the signal processing step (S200). It is predictable.

For example, as shown in FIG. 5, the welding quality prediction step (S300) uses the experimental color image map derived from the experimental welding light generated during welding of the experimental specimen and the welding result of the experimental specimen as learning data. An information storage step (S310) of storing information in the database 123a, and a learning step (S320) of mechanically learning the correlation between the experimental color image map stored as the learning data and the welding results of the experimental specimen by the artificial neural network. And based on the artificial neural network, it may include a determination step (S330) of determining the welding quality of the base material (S) from the color image map (M) derived in the signal processing step (S200).

At this time, in the information storage step (S310), the experimental color image map and the welding results of the experimental specimen corresponding thereto are derived through prior experiments and stored as the learning data, and the learning step (S320) is performed in plural numbers. By learning the training data based on the artificial neural network, prediction accuracy using the artificial neural network can be further increased.

Accordingly, finally, through the result output step (S400) performed through the output unit 123, the predicted welding state of the base material (S) predicted in the welding quality prediction step (S300) is monitored using the artificial neural network. It can be output as output data through a user terminal 30 such as .

In addition, as shown in the welding quality monitoring method according to another embodiment of the present invention in FIG. 6, the welding quality prediction step (S300) further includes an optimization step (S330), and in the above-described learning step (S320) The artificial neural network can be optimized to minimize the occurrence of errors in the learned artificial neural network.

Here, the definition of optimization of the artificial neural network made through the optimization step (S330) verifies the accuracy of the prediction using the artificial neural network and calculates the degree of error from the actual result, further improving the prediction accuracy of the artificial neural network. It may refer to a process of correcting the artificial neural network so as to increase it.

For example, in the optimization step (S330), during welding of the base material (S), the predicted welding state of the base material (S) is predicted from the color image map (M) derived in the signal processing step (S200) based on the artificial neural network. And, a comparative learning step in which the actual welding results of the base material (S) confirmed after completion of welding of the base material (S) are compared, and the sum of the errors of the predicted welding state compared to the actual welding results is derived as a loss function (Loss function) In the correction step (S332), the optimization result for minimizing the error of the predicted welding state is reflected in the artificial neural network of the learning step (S320) by considering the loss function derived in the comparison learning step (S331) and (S331), The artificial neural network can be optimized.

Therefore, according to the welding quality monitoring method and welding quality monitoring system 100 according to an embodiment of the present invention, during laser welding, the welding light 2 generated at the welding site W of the base material S is transmitted through a spectrometer. The image signal sensed in a dispersed state by wavelength through (112) is output as a plurality of spectrum signals (SS) representing the change in light intensity according to the wavelength band, and the output plurality of spectrum signals (SS) are calculated. By applying the color image map (M) derived from the distribution of light intensity in the form of a contour map to the artificial neural network that learned the correlation between the welding state and the color image map, it can determine whether the welding is good or bad in real time. This can have the effect of being judged with high accuracy.

For example, as shown in the experimental results of FIG. 7, according to the welding quality monitoring method and welding quality monitoring system 100 according to an embodiment of the present invention, the greater the laser output or the faster the speed of laser welding, the more light The intensity of the signal appeared to increase.

More specifically, the horizontal axis of the table showing the experimental results in FIG. 7 represents the output of the laser (1,000 W to 4,000 W), and the vertical axis may represent the speed of laser welding (1.5 mpm to 2.5 mpm).

According to the results of this experiment, as shown in Figure 7, it can be seen that the greater the laser output or the faster the speed of laser welding, the more the intensity of the optical signal increases, and the more clearly the contour of the color image map appears. The size (sharpness) of the optical signal of the color image map (M) depending on the welding conditions (laser output, laser welding speed, etc.) or welding quality (good joint, poor joint, poor penetration, etc.) As it was confirmed that the color distribution (color contour) had a significant difference, it was possible to verify that welding quality could be predicted using a color image map (M) with a specific size and color distribution of the optical signal.

Therefore, according to the welding quality monitoring method and the welding quality monitoring system 100 according to various embodiments of the present invention, during welding of the base material (S), a color image map (M) having the size and color distribution of a specific optical signal is created. By deriving and substituting this into the artificial neural network, the quality of the welded area (W) of the base material (S) can be accurately predicted in real time.

In addition, the welding quality monitoring method of the present invention reduces the amount of calculation by reducing the data to be processed, and allows more accurate monitoring of welding quality through a clear and standardized signal consisting of a small number of data, such as a color image map (color image map) of the raw signal data. Instead of using M) as is, it can be used after processing it into a standardized signal through a certain procedure.

Hereinafter, a method of processing the above-described color image map (M) into a standardized signal and a method of monitoring welding quality using the processed data will be described in more detail.

FIGS. 8 and 9 are flowcharts and conceptual diagrams showing another embodiment of the signal processing step (S200) of the welding quality monitoring method of FIG. 2, and FIG. 10 shows the input that can be extracted through the signal processing step (S200) of FIG. 8. It is a table showing output variables according to variables and input variables, and FIG. 11 is a flowchart sequentially showing another embodiment of the welding quality prediction step (S300) of the welding quality monitoring method of FIG. 2. And, Figure 12 is a graph showing the results of an experiment using a welding quality monitoring method according to another embodiment of the present invention.

First, as shown in FIG. 8, the signal processing step (S200) of the welding quality monitoring method according to another embodiment of the present invention is the signal processing step (S200) of the welding quality monitoring method according to an embodiment of the present invention described above ( After the signal acquisition step (S210), the band setting step (S220), and the image drawing output step (S230) of S200), a signal processing step (S240) and an input variable extraction step (S250) may be further included.

As shown in FIGS. 8 and 9, the signal processing step (S240) converts the color image map (M) output in the image drawing output step (S230) into RGB values by the CIE 1931 It is a signal in the form of a graph showing the change in the RGB values according to the time order in which the spectrum signal (SS) was acquired, and can be output as an RGB value change graph (RG) separately shown for each RGB region.

If the color image map (M), which is the original signal data, is used as is, unnecessary data may exist, which may increase the amount of calculation (increase in processing time). For example, as shown in the color image map (M) of FIG. 9, a characteristic and significant signal exists in the wavelength region of about 350 nm to 750 nm, but there is no characteristic and significant signal in other wavelength regions.

In addition, signals in the wavelength range of about 350 nm to 750 nm, which can be used as meaningful data, were not clear enough to intuitively distinguish the data. In other words, although a clear difference is confirmed in the relevant wavelength range, it can be judged that the data type is not suitable for distinguishing the difference. In addition, the color and shape may vary depending on the standard for drawing (plot) the signal for the color image map (M) (color standard setting and axis setting), the software used, and the display medium (monitor, printed material, etc.). Because it can be expressed, it may be necessary to process the color image map (M), which is the original signal data, using a standardized method or means of expression.

Accordingly. In the signal processing step (S200) of the welding quality monitoring method according to another embodiment of the present invention, rather than using the color image map (M), which is the original signal data, through the signal processing step (S240), the original signal data is It is possible to use specific meaningful features obtained after processing. In addition, through raw signal data processing, signal readability can be improved and user convenience for signal analysis and quality judgment can be increased based on improved readability.

More specifically, in the signal processing step (S240), the color image map (M), which is the original signal data, is first converted to CIE XYZ based on the CIE 1931 2 Degree Standard Observer standard, and then the CIE , and using the finally converted sRBG value, it is a signal in the form of a graph showing the change in the RGB value according to the time order in which the plurality of spectrum signals (SS) were acquired. RGB separately indicated for each RGB region. It was possible to output it as a value change graph (RG).

As shown in FIG. 9, the final output RGB value change graph (RG) was able to clearly express the difference in data compared to the color image map (M). In addition, there is an advantage in terms of the amount of calculation because the data to be processed is reduced, and from the user's perspective, convenience can be increased by making monitoring of welding quality more intuitive through a smaller number of clear and standardized signals.

Using this RGB value change graph (RG), the algorithm of the welding quality monitoring method according to another embodiment of the present invention uses the RGB signal and the maximum intensity wavelength value of the RGB value change graph (RG) to determine welding quality. It can be monitored.

For example, as shown in FIGS. 8 and 10, the signal processing step (S200) follows the signal processing step (S240), and then, through the input variable extraction step (S250), R in the RGB value change graph (RG) The average of the values ( ), average of G values ( ), average of B values ( ), standard deviation of R value (STD(R)), standard deviation of G value (STD(G)), standard deviation of B value (STD(B)), maximum intensity wavelength average ( ) and the standard deviation of the maximum intensity wavelength (STD( )) can be extracted as an input variable.

Subsequently, as shown in FIG. 11, the information storage step (S310) of the welding quality prediction step (S300) outputs an experimental RGB value change graph from the experimental color image map, and experimental input is obtained from the experimental RGB value change graph. An input variable storage step (S311) in which variables are extracted and stored as the learning data, and the welding results of the experimental specimen are extracted as output variables for learning divided into a total of four categories: unjoined, insufficient penetration, full penetration, and unjoined due to gap. It includes an output variable storage step (S312) of storing the learning data as the learning data, and the learning step (S320) can mechanically learn the correlation between the experimental input variables and the learning output variables by the artificial neural network.

At this time, as shown in FIG. 10, in the output variable storage step (S312), among the non-bonding, the lack of penetration, the complete penetration, and the non-bonding due to the gap, the largest value is set to 1 and the remaining values are set to 0. By processing, the learning output variable can be extracted as a combination of numbers 1 and 0.

Accordingly, the welding quality prediction step (S300) is, finally, through the judgment step (S330), and in the learning step (S320), the artificial Based on a neural network, the welding quality of the base material (S) can be determined by calculating output variables from the input variables extracted from the input variable extraction step (S250).

Hereinafter, experimental examples to aid understanding of the present invention will be described. The following experimental examples are presented as examples to aid understanding of the present invention, and of course, the present invention is not limited to the following experimental examples.

As shown in Figure 10, as an input variable, in the RGB value change graph (RG), the average of R values ( ), average of G values ( ), average of B values ( ), standard deviation of R value (STD(R)), standard deviation of G value (STD(G)), standard deviation of B value (STD(B)), maximum intensity wavelength average ( ) and the standard deviation of the maximum intensity wavelength (STD( )) Eight input variables were calculated. In addition, a total of four output variables were applied: non-bonding, insufficient penetration, complete penetration (quality satisfaction), and non-bonding due to a gap.

At this time, in the case of the output variable, the largest value among the four nodes output from the artificial neural network model can be treated as 1, and the remaining values can be treated as 0. That is, the outputs of 1000, 0100, 0010, and 0001 may indicate non-joint, insufficient penetration, complete penetration, and non-connection due to gap in that order. A total of 5,400 pieces of data were generated, of which 4,590 pieces of data (85%) were used for learning, and 810 pieces of data (15%) were used for verification.

The final model was estimated through learning a total of 5,000 times, and Figure 12 shows the resulting learning results. As shown in Figure 12, it was confirmed that the accuracy and loss function converged after about 1,000 learning sessions. In this way, through learning of the artificial neural network model, it was finally found to have an accuracy of 97.67% for the training data and an accuracy of 91.38% for the verification data.

Therefore, according to the welding quality monitoring method according to another embodiment of the present invention, rather than using the color image map (M), which is the original signal data, as is, the color image map (M) is converted to RGB values according to the CIE 1931 XYZ color system. Convert and process this into an RGB value change graph (RG), which is a signal in the form of a standardized graph representing the change in the RGB values according to the time order in which the plurality of spectrum signals (SS) were acquired, and is separately represented by each RGB region. , by applying it to the artificial neural network, the data to be processed can be reduced in terms of computational volume. Therefore, by allowing the user to monitor through a small number of clear and standardized signals, it can have the effect of making welding quality monitoring intuitive and increasing convenience.

The present invention has been described with reference to the embodiments shown in the drawings, but these are merely exemplary, and those skilled in the art will understand that various modifications and equivalent other embodiments are possible therefrom. Therefore, the true scope of technical protection of the present invention should be determined by the technical spirit of the attached patent claims.

1: Laser beam
2: Welding enthusiast
3: Spectrum
10: welding head
20: Light generation unit
30: user terminal
100: Welding quality monitoring system
110: Spectral sensing unit
111: mirror
112: Spectroscopy
113: image sensor
114: barrel
115: optical cable
120: Monitoring unit
121: Spectrum signal output unit
122: signal processing unit
123: Welding quality prediction unit
123a: database
124: output unit
S: Base material
W: Welding area
SS: Multiple spectral signals
BA: multiple band areas
AG: average value change graph
RG: RGB value change graph
M: color image map

Claims (18)

In the welding quality monitoring method that can evaluate the welding quality of the base material by monitoring and analyzing the welding light generated at the welding site of the base material in real time during welding of the base material,
An image sensor senses the welding light momentarily generated at the welding site of the base material in a dispersed state by wavelength through a spectrometer, and this is a signal in the form of a graph representing the change in light intensity according to the wavelength band, wherein the base material A spectrum signal output step of outputting a plurality of spectrum signals a predetermined number of times during a predetermined time during welding;
In the spectrum signal output step, the plurality of spectrum signals output in chronological order for the predetermined time are calculated, and the distribution of the light intensity according to the wavelength of the welding light according to the time flow during the predetermined time is displayed by color classification. A signal processing step to derive a color image map;
The correlation between the welding results of the experimental specimen derived through prior experiments and the experimental color image map derived from the experimental welding light of the experimental specimen is derived in the signal processing step based on an artificial neural network learned with learning data. A welding quality prediction step of predicting the welding quality of the base material from the color image map; and
A result output step of outputting the predicted welding state of the base material predicted in the welding quality prediction step as output data;
Including, welding quality monitoring method. According to claim 1,
The signal processing step is,
a signal acquisition step of acquiring the plurality of spectrum signals output from the spectrum signal output step in chronological order for the predetermined time;
A band setting step of setting a plurality of band areas having different predetermined wavelength band ranges in the plurality of spectrum signals; and
From each individual spectrum signal acquired in time order, a plurality of average values of the light intensity within each band region are calculated, and this is calculated as the plurality of average values for each wavelength of the welding light according to the time order in which the plurality of spectrum signals were acquired. An image drawing output step of outputting the color image map showing the distribution of the average value by color classification;
Including, welding quality monitoring method. According to claim 2,
The spectrum signal output step is,
A first spectrum output step of outputting the welding light sensed in a dispersed state by wavelength through the spectrometer at a first time as a first spectrum signal;
A second spectrum output step of outputting the welding light sensed in a dispersed state by wavelength through the spectrometer at a second time at a predetermined time interval from the first time as a second spectrum signal; and
an n-th spectrum output step of outputting the welding light sensed in a dispersed state by wavelength through the spectrometer at an n-th time having the predetermined time interval from the n-1 time as an n-th spectrum signal;
Including, welding quality monitoring method. According to claim 3,
The predetermined time is,
It is the time from the first time to the nth time,
In the spectrum signal output step, the plurality of spectrum signals are,
A welding quality monitoring method in which n items are output during the predetermined time. According to claim 3,
The signal acquisition step is,
Acquiring the first spectrum signal to the nth spectrum signal output from the spectrum signal output step at the predetermined time interval from the first time to the nth time,
The band setting step is,
A first area setting step of setting a first band area having a predetermined wavelength band range in the plurality of spectrum signals including the first spectrum signal to the nth spectrum signal;
A second area setting step of setting a second band area having a predetermined wavelength band range higher than the first band area in the plurality of spectrum signals; and
An n-th region setting step of setting an n-th band region having a predetermined wavelength band range higher than the n-1-th band region in the plurality of spectrum signals;
Including, welding quality monitoring method. According to claim 5,
The image drawing output step is,
In the first spectrum signal acquired at the first time, the average value of the light intensity in the first band region is calculated as a 1-1 average value, and the average value of the light intensity in the second band region is calculated as a 1-1 average value. A first average value calculation step of calculating as a 1-2 average value, and calculating the average value of the light intensity in the n-th band region as a 1-n average value;
In the second spectrum signal acquired at the second time, the average value of the light intensity in the first band region is calculated as a 2-1 average value, and the average value of the light intensity in the second band region is calculated as a 2-1 average value. A second average value calculation step of calculating as a 2-2 average value, and calculating the average value of the light intensity in the n-th band region as a 2-n average value; and
In the n-th spectrum signal acquired at the n-th time, the average value of the light intensity in the first band region is calculated as the n-1 average value, and the average value of the light intensity in the second band region is calculated as the n-1th average value. An nth average value calculation step of calculating as the n-2th average value, and calculating the average value of the light intensity in the nth band region as the nnth average value,
The plurality of average values, including the 1-1st average value to the nnth average value of the light intensity calculated in the first average value calculation step to the nth average value calculation step, are divided into colors according to the light intensity. , A welding quality monitoring method that outputs the color image map expressed as a color distribution so that the plurality of average values are distributed by wavelength of the welding light in chronological order from the first time to the nth time. According to claim 6,
The image drawing output step is,
A welding quality monitoring method that distinguishes the light intensity represented by each average value of the plurality of average values based on the difference in color brightness. According to claim 6,
The first average value calculation step is,
Each band region including the first band region to the nth band region is an independent variable, and the 1-1st average value to the 1-nth average value calculated in each band region is a dependent variable, Obtaining a correlation as a result of a regression analysis that estimates the relationship between the two, and using the obtained correlation to derive an expected average value of the light intensity in a band area for which the average value of the light intensity has not been calculated,
The image drawing output step is,
Welding quality monitoring method including the expected average value in the color distribution when outputting the color image map. According to claim 2,
The welding quality prediction step is,
An information storage step of storing the experimental color image map derived from the experimental welding light generated during welding of the experimental specimen and the welding result of the experimental specimen as learning data;
A learning step of mechanically learning the correlation between the experimental color image map stored as the learning data and the welding results of the experimental specimen by the artificial neural network; and
A determination step of determining welding quality of the base material from the color image map derived in the signal processing step, based on the artificial neural network;
Including, welding quality monitoring method. According to clause 9,
The information storage step is,
The experimental color image map and the corresponding welding results of the experimental specimen are derived through prior experiments and stored in plural numbers as the learning data,
The learning step is,
A welding quality monitoring method that learns a plurality of the learning data based on the artificial neural network. According to clause 9,
The welding quality prediction step is,
An optimization step of optimizing the artificial neural network so as to minimize the occurrence of errors in the artificial neural network learned in the learning step;
A method for monitoring welding quality, further comprising: According to claim 11,
The optimization step is,
During welding of the base material, the predicted welding state of the base material predicted from the color image map derived in the signal processing step based on the artificial neural network is compared with the actual welding result of the base material confirmed after welding of the base material is completed. Thus, a comparative learning step of deriving the sum of the errors of the predicted welding state compared to the actual welding result as a loss function; and
A correction step of reflecting the optimization result for minimizing the error of the predicted welding state in the artificial neural network of the learning step by considering the loss function derived in the comparative learning step;
Including, welding quality monitoring method. According to clause 9,
The signal processing step is,
The color image map output in the image drawing output step is converted into RGB values according to the CIE 1931 , a signal processing step of outputting a graph of RGB value changes separately represented by each RGB region;
A method for monitoring welding quality, further comprising: According to claim 13,
The signal processing step is,
In the RGB value change graph, the average of R values ( ), average of G values ( ), average of B values ( ), standard deviation of R value (STD(R)), standard deviation of G value (STD(G)), standard deviation of B value (STD(B)), maximum intensity wavelength average ( ) and the standard deviation of the maximum intensity wavelength (STD( )) an input variable extraction step of extracting at least one of the following as an input variable;
A method for monitoring welding quality, further comprising: According to claim 14,
The information storage step is,
An input variable storage step of outputting an experimental RGB value change graph from the experimental color image map, extracting experimental input variables from the experimental RGB value change graph, and storing them as the learning data; and
An output variable storage step of extracting the welding results of the experimental specimen as learning output variables divided into a total of four categories: non-bonding, insufficient penetration, full penetration, and non-bonding due to a gap, and storing them as the learning data,
The learning step is,
A welding quality monitoring method that mechanically learns the correlation between the experimental input variables and the learning output variables by the artificial neural network. According to claim 15,
The output variable storage step is,
Among the non-bonding, the lack of penetration, the complete penetration, and the non-bonding due to the gap, the largest value is treated as 1 and the remaining values are treated as 0, and the learning output variable is extracted as a combination of numbers of 1 and 0, How to monitor weld quality. According to claim 15,
The judgment step is,
In the learning step, based on the artificial neural network that mechanically learns the correlation between the experimental input variables and the learning output variables, output variables are calculated from the input variables extracted from the input variable extraction step to weld the base material. Welding quality monitoring method to determine quality. A welding head that performs welding of a base material by irradiating a laser beam to the welding area of the base material is installed on the optical path of the laser beam and specifies the welding light generated at the welding area of the base material and incident on the welding head. a spectral sensing unit that forms a spectrum distributed by wavelength and senses the spectrum as an image signal; and
It includes a monitoring unit that monitors the predicted welding state of the base material according to the image signal received from the spectral sensing unit,
The monitoring unit,
The image signal sensed through the spectral sensing unit is applied a predetermined number of times during welding of the base material, and the image signal is a signal in the form of a graph showing the change in light intensity according to the wavelength band, a spectrum signal output unit that outputs a plurality of spectrum signals;
The spectrum signal output unit calculates the plurality of spectrum signals output in chronological order for the predetermined time, and displays the distribution of the light intensity according to the wavelength of the welding light according to the flow by time during the predetermined time, classified by color. A signal processing unit derived from a color image map;
Based on an artificial neural network that learned the correlation between the welding results of the experimental specimen, which was derived through a prior experiment, and the experimental color image map derived from the experimental welding light of the experimental specimen, as learning data, the signal processor derived a welding quality prediction unit that predicts welding quality of the base material from the color image map; and
an output unit that outputs the predicted welding state of the base material predicted by the welding quality prediction unit as output data;
Including, a welding quality monitoring system.

Images