JP7780138B2 - AI evaluation method, AI evaluation device, and program for spot welds - Google Patents

Description

The present invention relates to spot welding, in which two overlapping plate materials are sandwiched between electrodes and voltage is applied, and in particular to an AI evaluation method, device, and program for evaluating the weld strength of the spot weld, the material quality, composition, or bonding state of the plate materials, etc.

10 and 11 , spot welding is a method widely used to join metal plates, for example, in the field of metal product manufacturing, such as automobile manufacturing. In spot welding, two metal plate materials 51 and 52 to be joined are arranged with a joint area 53 overlapping each other, and a pair of electrodes 54 and 55 are pressed against the joint area 53 from above and below, and a voltage is applied between the electrodes 54 and 55 from a power source 56.
12 , an electrode rod 54 may be placed in contact with the plate material 51, an electrode 55 may be placed at a position sufficiently distant from the welded portion of the plate material 52, and a current may be applied while the plate materials 51 and 52 are in contact at the welded portion 53. In this manner, current I is passed from one electrode rod 54 to the other electrode rod 55 through the joining region 53, thereby heating and melting both plate materials 51 and 52 in the joining region 53, and then cooling to form a spot weld. In this manner, both plate materials 51 and 52 are joined to each other at the joining region 53 by the spot weld.

However, the weld strength of spot welds can be reduced by various welding defects, such as failed or insufficient welding. Such a reduction in weld strength at spot welds can lead to a reduction in the strength of the entire metal product.

In response to this, Patent Document 1 discloses an electromagnetic induction sensor that includes a transmitter coil that generates a magnetic field, a spot welding electrode, and a rod-shaped member used as a magnetic path to concentrate the magnetic field generated by the transmitter coil on the nugget being inspected, and multiple receiver coils that generate induced voltages in response to the magnetic field that has passed through the nugget, with the transmitter coil wound around the rod-shaped member and the multiple receiver coils wound around the outer periphery of the transmitter coil.

However, this type of electromagnetic induction sensor is only used to evaluate the quality of spot welds, and cannot evaluate the weld strength of spot welds.

Patent No. 3317366

In contrast, one method for evaluating the weld strength of spot welds is to create test specimens and use them for destructive testing. However, in actual metal products, metal plates are often joined using multiple spot welds, making it difficult to evaluate weld strength by destroying the actual metal product. Therefore, if the weld strength of each spot weld could be evaluated without destroying the metal product, it would be possible to calculate the strength of the entire metal product, which would contribute to improving product safety.

Furthermore, with regard to metal sheet materials joined by spot welding, depending on the metal product, the sheet material on the front side may be visible from the outside, but the sheet material on the back side may not be. In such metal products, the sheet material on the back side cannot be seen, making it difficult to confirm whether the sheet materials on both the front and back sides are as designed. Furthermore, while it is possible to ascertain the material of both sheet materials to some extent just by visual inspection, it is impossible to evaluate the condition of each sheet material, i.e., its exact material, composition, or bonding state.

In light of the above, the present invention aims to provide an AI evaluation method for spot welds that non-destructively estimates target variables for evaluating spot welds, such as weld strength and plate condition.

The above object is achieved, according to the present invention, by an AI evaluation method for spot welds, which includes a first step of applying a plurality of excitation signals of different frequencies to an excitation coil arranged adjacent to a spot weld to be evaluated, thereby generating an alternating magnetic field at the spot weld; a second step of measuring a detection signal generated in a detection coil by magnetic flux that is disturbed in accordance with the state of the spot weld; and a third step of evaluating the state of the spot weld based on the detection signal, wherein in the third step, a first objective variable and a second objective variable of the spot weld to be evaluated are estimated based on the detection signal of the second step using a quantitative value of a first objective variable of the evaluation target, which serves as comparison data for the spot weld and which is set in advance, a quantitative value of a second objective variable for classifying the spot weld to be evaluated, and correlation data relating to the correlation between the excitation signal of the first step and the detection signal of the second step, and This is achieved by an AI evaluation method for spot welds, which is configured to obtain a complex amplitude ratio for a one-stage excitation signal as measurement data, and obtain a first objective variable as correlation data from the quantitative values of a first objective variable and a second objective variable in the measurement data and the complex amplitude ratio of the detection signal as comparison data corresponding to the quantitative values of the first objective variable and the second objective variable, define a plurality of types of classification related to the second objective variable, and define an evaluation formula for the correlation data for each classification, then select a classification corresponding to each measurement data related to the correlation data of an unknown actual evaluation target, and estimate the first objective variable of the evaluation target from the measurement data using the first evaluation formula corresponding to the selected classification, wherein the first objective variable of the evaluation target of the spot weld is the weld strength of the spot weld, and the second objective variable of the evaluation target of the spot weld is the state of each of the front and back plates of the spot weld or the spot weld.

In the above-mentioned configuration, the state of the front and back plate materials related to the spot welded portion is preferably the material quality, composition, structure or bonding state of the plate materials.
The first objective variable may be estimated by selecting an evaluation formula for the first objective variable through classification of the spot welds to be evaluated using the second objective variable.
The first evaluation formula and the evaluation formula corresponding to the second classification preferably use any one of linear analysis, PLS regression analysis, SVM (Support Vector Machine), neural network, and supervised AI analysis methods to define measurement data as explanatory variables.
Preferably, the excitation coil and detection coil are moved along the surface of the spot weld , and measurement operations in the first and second stages are performed at a plurality of measurement points in the measurement target area including the spot weld .
The excitation coil and detection coil are preferably scanned in one or two dimensions along the surface of the spot weld so as to pass near the center of the measurement target area.
Preferably, the third stage of evaluation is performed based on detection signals from a predetermined number of scans near the center of the spot weld among the two-dimensional scans.
The excitation signal is preferably generated at five or more different frequencies, and preferably has a frequency band of about 5 to 100 times from the lowest frequency to the highest frequency.
An AI evaluation device that executes any of the AI evaluation methods described above, preferably
a sensor having a coil;
a measurement unit that generates excitation signals for each frequency to be input to a coil in order to apply an alternating magnetic field to an evaluation object, and processes detection signals output from the coil in response to the generated excitation signals;
a storage unit that stores data relating to correlations between quantitative values of objective variables of analysis and sensors and measurement units for a plurality of evaluation targets;
an estimation unit that estimates a response variable of an evaluation target using data stored in a storage unit based on a detection signal processed by the measurement unit;
It is equipped with
an estimation unit obtains a complex amplitude ratio for a signal from the detection signal as measurement data, defines a plurality of types of classifications based on the quantitative values of the objective variables of the plurality of different evaluation targets stored in the storage unit and the complex amplitude ratios of the detection signals as corresponding comparison data, and defines an evaluation formula for each classification;
The estimation unit selects, for each measurement data, a classification consisting of the state of each plate material on the front and back of the spot weld or the spot weld that corresponds to the measurement data, and estimates the welding strength of the spot weld, which is the objective variable to be evaluated, from the measurement data using an evaluation formula that corresponds to the selected classification.
The program may be one that causes a computer to function as the AI evaluation device for spot welds described above.

The present invention provides an AI evaluation method for spot welds that non-destructively estimates target variables for evaluating spot welds, such as weld strength and the condition of the plate material.

1 is a block diagram showing the overall configuration of an example of an evaluation device for implementing an AI evaluation method for spot welds according to the present invention; FIG. 10 is a diagram illustrating another example of the configuration of the sensor. FIG. 2 is a flow chart showing an evaluation method using the evaluation device of FIG. 1 . FIG. 2 is an explanatory diagram showing an example of an evaluation method using the evaluation device of FIG. 1 . 2 is a schematic explanatory diagram showing an example of a method for measuring a spot weld using the evaluation device of FIG. 1 . FIG. 1. FIG. 4 is a schematic explanatory diagram showing another example of a method for measuring a spot weld using the evaluation device of FIG. FIG. 10 is a schematic explanatory diagram showing yet another example of a method for measuring a spot weld using the evaluation device of FIG. 1 . 1 shows the relationship between predicted and measured values of the welding strength of the spot welds in Example 1, where (A) shows the results at the time of learning and (B) shows the results at the time of verification. 10 shows the relationship between the predicted value and the actually measured value of the welding strength of the spot weld of Example 2, where (A) shows the time of learning and (B) shows the time of verification. FIG. 1 is a schematic explanatory diagram showing a general spot welding. FIG. 1 is a schematic cross-sectional view showing a typical spot welding. FIG. 10 is a schematic explanatory diagram showing a modified example of a general spot welding.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
Fig. 1 shows the configuration of one embodiment of an evaluation device for implementing the AI evaluation method according to the present invention, and Fig. 2 is a diagram showing another example of the configuration of a sensor 10. As shown in Fig. 1, the evaluation device 1 is composed of the sensor 10, a measurement unit 20, and a data processing unit 30.

The sensor 10 comprises an excitation coil 11 and a detection coil 12, and a magnetic path forming section 13. To block external magnetic fields, the sensor 10 is housed in a sensor holder 14 made of, for example, metal, with an open top end, and is supported within the sensor holder 14 by a non-magnetic gap filler (not shown). During measurement, the sensor 10 is positioned upside down so that the open surface 14a of the sensor holder 14 is in contact with the surface of the object to be evaluated.

The magnetic path forming portion 13 is composed of, for example, a bottom portion 13a, a cylindrical portion 13b, and a shaft portion 13c, with the bottom portion 13a supporting the cylindrical portion 13b and the shaft portion 13c. The excitation coil 11 and detection coil 12 are attached to the shaft portion 13c. The magnetic path forming portion 13 is not limited to the shape shown in the figure, and may be composed of each of the above-mentioned portions alone or in combination. In the sensor 10, the detection coil 12 is placed relative to the magnetic path formed by the excitation coil 11 and the magnetic path forming portion 13. Furthermore, during measurement, the spot weld (described below) to be evaluated is placed nearby, so the signal detected by the detection coil 12 can be affected depending on the magnetic permeability of the spot weld, etc. The sensor 10 described above is one example configuration, and other configurations may have similar effects. For example, in Figure 1, the detection coil 12 is disposed on the side of the evaluation object (not shown), but as shown in Figure 2, the positional relationship between the excitation coil 11 and measurement coil 12 may be reversed so that the excitation coil 11 is disposed outside the measurement coil 12, i.e., on the side of the evaluation object 41. The positional relationship between the excitation coil 11 and detection coil 12 can be freely set depending on the evaluation object.

The measurement unit 20 comprises an oscillator 21, a signal processing unit 22, and a control unit 23. The oscillator 21 repeatedly generates a signal of a certain frequency and increases or decreases the frequency of the signal in stages. The signal oscillated by the oscillator 21 is branched into an excitation signal and a reference signal. The excitation signal is transmitted to the excitation coil 11 and output to the signal processing unit 22 as a reference signal. The signal processing unit 22 calculates the temporal change in the detection signal relative to the excitation signal from the detection coil 12 using the reference signal from the oscillator 21. In this case, the signal processing unit 22 has a Fourier transform function and converts the time-axis signal into a frequency-axis signal. The signal processing unit 22 also digitizes the detection signal from the sensor 10 and outputs it to the data processing unit 30. The control unit 23 inputs and outputs data and various control signals to and from the data processing unit 30, and controls the oscillator 21 and the signal processing unit 22.

In this case, as shown in Figure 2, multiple sensors 10 may be used, and the evaluation unit 20 or data processing unit 30 may be provided with a function for calculating data from multiple sensors. Using multiple sensors 10 makes it possible to perform arithmetic operations on the detection signals, such as obtaining the differential signal between two sensors 10, thereby reducing noise and improving the strength of the detection signal level.

In this case, data processing unit 30 estimates the objective variable of the evaluation object from the digital data of the detection signal processed by signal processing unit 22. In the following explanation, the evaluation object is the spot weld, the measurement data that is the detection signal from sensor 10 is the explanatory variable, the first objective variable is the weld strength, and the second objective variable is the material of the spot weld, and the explanation will be given assuming that the weld strength as the first objective variable and the material as the second objective variable are estimated.
The data processing unit 30 is composed of a computer including an input/output interface unit 31 that interfaces with the control unit 23, a storage unit 32 including a main storage unit and an auxiliary storage unit, a calculation unit that performs arithmetic operations such as arithmetic operations, and a control unit that controls the storage unit and the calculation unit, and a data processing program is stored in the auxiliary storage unit.When the data processing program is deployed and executed in the calculation unit, the data processing unit 30 functionally includes a storage unit 33 and an estimation unit 34 shown in Fig. 1. Note that AI evaluation (also called artificial intelligence) refers to the estimation of a target variable from an explanatory variable using a calculation method or algorithm executed in the data processing unit 30.

The memory unit 33 stores data relating to the correlation between the quantitative value of the objective variable for the evaluation object and the estimated value of the objective variable for the evaluation object obtained from the processed detection signal measured using the sensor 10 and the measurement unit 20. The estimation unit 34 applies an alternating magnetic field to the evaluation object using the sensor 10, and estimates the weld strength of the evaluation object using the data stored in the memory unit 33 based on the detection signal processed using the measurement unit 20.

The estimation unit 34 can evaluate the weld strength, for example, according to the following analysis method, using a quantitative value such as the weld strength as the objective variable of the evaluation target that has been obtained in advance. As shown in Figure 3, the weld strength of a spot weld is evaluated as follows.

(Step ST1)
The acquisition of comparison data using pre-prepared test pieces will now be described. The test pieces are prepared as follows: For example, plate materials 51 and 52 made of steel and aluminum plates are spot-welded under various conditions to produce a plurality of test pieces with different materials and weld strengths.

(Step ST2)
In step ST2, a plurality of test pieces having different materials and welding strengths are subjected to an AI evaluation method described below, specifically, comparison data 42 consisting of complex amplitudes and phases in a predetermined frequency range measured by the sensor 10. These plurality of test pieces are measured by a known method such as a tensile test or a cross tensile test, and the welding strength of each test piece is obtained.
The comparison data 42 includes quantitative values of the weld strength of the spot welds, which are related to the first objective variable, measured using a conventional quantitative evaluation method, and data related to the second objective variable, which includes the material and characteristics of the plate material to be spot welded, which require classification. The comparison data 42 is linked to information on the weld strength and material of the spot welds and stored in a database in the storage unit 33.

(Step ST3 (pre-learning))
A first evaluation formula 45 for obtaining a first objective variable and a second evaluation formula 46 for obtaining a second objective variable are created using the quantitative values of the test piece obtained in steps ST1 and ST2 and the obtained comparison data 43. The evaluation formulas 46 are created for each item in the classification 44. Step ST3 is a step also called pre-learning.

Here, the first evaluation formula 45 related to the first objective variable for strength and the second evaluation formula 46 related to the second objective variable for material classification are created using any of the following analytical methods: linear analysis, SVM (Support Vector Machine), PLS regression analysis, neural network, and supervised AI analysis. The same or different methods may be used for the first objective variable and the second objective variable, but a single piece of data obtained in steps ST1 and ST2 is used. The evaluation formula can be easily and accurately defined based on this pre-trained comparison data 43. This completes the preliminary preparation.

(Step ST4)
Next, the actual evaluation process begins. Similar to the test piece in step ST2, measurement data 47 consisting of complex amplitude and phase in a predetermined frequency range is acquired from the actual evaluation target 41, whose weld strength and material are unknown.

(Steps ST5 and ST6)
In step ST5, it is selected whether or not to classify the measurement data 47. If the measurement data 47 is to be classified, the classification is carried out in step ST6. If the measurement data 47 is not to be classified (NO), the process proceeds to step ST7.

(Step ST6)
Specifically, in step ST6, the measurement data 47 is subjected to material classification selection using the second evaluation formula 46 for the second objective variable. For example, the classification of the measurement data 47 of spot welding of plate materials 51, 52 made of two steel plates is determined using classification using the first evaluation formula 46. From the classification results, an actual evaluation target 41 consisting of only steel plates may be selected from the measurement data 47.

(Step ST7)
After classification in step ST6, a first evaluation formula 45 for the first objective variable is selected to determine whether or not to estimate the strength of the spot weld.

(Step ST8)
When estimating the strength of the spot weld in step ST7, in step ST8, the results of classification using first evaluation formula 46 are used to estimate the welding strength of the spot weld, for example, for an evaluation object 41 consisting only of a steel plate, by selecting first evaluation formula 45 for the first objective variable and performing calculations using measurement data 47, and in step ST9, the classification and the estimated value of the welding strength are confirmed, thereby completing the process.

(Step ST10)
In step ST5 above, if the selection is made not to classify the measurement data 47, that is, if the measurement data 47 is only of the spot welds of the plate materials 51, 52 consisting of two steel plates, that is, if the evaluation object 41 is iron, that is, is uniquely determined, classification by calculation of the first evaluation formula 46 is not necessary, and the process proceeds to step ST7 in step ST10, and the welding strength of the spot welds can be estimated in step ST7.

(Step ST11)
If it is not necessary to obtain the strength of the spot weld in step ST7 and it is only desired to obtain the classification of the material used, then there is no need to perform the calculation to estimate the weld strength using first evaluation formula 45 in step ST8. Therefore, it is sufficient to leave the measurement data 47 classified into materials using first evaluation formula 46 in step ST5, proceed to step ST9 in step ST7, and confirm the estimated value of the classification, thereby ending the process.

The comparison data 43 in step ST2 and the evaluation object 41 in step ST4 are the spot welds measured by the sensor 10, and the objective variables include the weld strength and the condition of the front and back plates joined by the spot weld, such as the material, composition, or bonding state.

In step ST2, the comparison data 43 is, for example, the quantitative value of tensile strength when the objective variable is weld strength. Below, we will explain in detail the estimation by the estimation unit 34 when the evaluation object 41 is a spot weld and the objective variable is weld strength.

In addition to the weld strength, the estimation unit 34 can also estimate the condition of the front and back plates of the spot weld. Here, the condition of the plates can be estimated, for example, by the material, composition, or bonding state of the plates.

The above configuration makes it possible to accurately predict the weld strength of a spot weld or the condition of the plate materials on the front and back of the spot weld without destroying the product or other item containing the spot weld. In particular, for plate materials on the back side that cannot be seen from the outside, the material, composition, and bonding state can be accurately predicted without destroying the product or other item containing the spot weld, and it can also be used to determine, for example, whether the plate material is being used correctly as designed.

The estimation unit 34 performs regression analysis and estimation using as parameters the real and imaginary parts of the processed detection signal output from the measurement unit 20, or the complex amplitude ratio, or the first or second derivative value of the amplitude or phase difference related to the frequency of the detection signal, etc. In the following explanation, the real and imaginary parts of the processed detection signal, or the complex amplitude ratio, are used as parameters.
The estimation unit 34 performs PLS regression analysis or AI analysis on the spot welds based on the quantitative values of the weld strength and the detection signals generated by applying an alternating magnetic field using the sensor 10 and processing the detection signals using the measurement unit 20, and generates data to be stored in the recording unit 33. For this reason, the estimation unit 34 may also be called an AI analysis unit.

(AI evaluation method)
A method for AI evaluation of spot welds using the evaluation device 1 will be described.
in this way,
a first step of applying excitation signals of different frequencies to an excitation coil (11) positioned adjacent to a spot weld to be evaluated to generate an alternating magnetic field at the spot weld;
a second step of measuring a detection signal generated in the detection coil 12 by magnetic flux disturbances according to the state of the spot weld;
and a third step of evaluating the condition of the spot weld based on the detection signal.
In the third stage, correlation data relating to the correlation between the quantitative value of the target variable of the spot weld that is set in advance and the excitation signal of the first stage and the detection signal of the second stage is used to estimate the target variable of the spot weld based on the detection signal of the second stage.

Specifically, in the second stage, the complex amplitude ratio of the detection signal to the excitation signal from the first stage is obtained as measurement data. Among the measurement data, multiple classifications are defined based on the quantitative values of the objective variable and the complex amplitude ratios of the detection signal as corresponding comparison data. An evaluation formula is defined as correlation data for each classification. Furthermore, for each piece of measurement data, a classification corresponding to that measurement data is selected, and the objective variable to be evaluated is estimated from the measurement data using the evaluation formula corresponding to the selected classification.

According to the AI evaluation method of the present invention, first, a plurality of types of classifications are defined for measurement data by comparing quantitative values of objective variables such as the weld strength or plate condition of the evaluation target with corresponding comparison data, and an evaluation formula is defined for each classification. Then, a corresponding classification is selected for each measurement data, and the weld strength or plate condition as the objective variables of the evaluation target are evaluated from the measurement data using the evaluation formula corresponding to the selected classification.
Therefore, by evaluating the weld strength and plate condition at the spot weld, particularly the condition of the plate on the back side that cannot be seen, it is possible to predict the weld strength of the spot weld with high accuracy without destroying the metal product containing the spot weld, and it is also possible to predict expulsion, insufficient welding, and poor welding of the spot weld in accordance with the weld strength.

Furthermore, because measurement data can be used to define classifications and generate correlation data, there is no need to obtain data for defining classifications or data for creating correlation data separately from the measurement data, simplifying processing. Furthermore, before evaluating the measurement data, a classification to which the measurement data should be assigned is selected, and the measurement data is evaluated using an evaluation formula corresponding to the selected classification, allowing for more accurate evaluation of the measurement data.

(Welding strength evaluation method)
Next, a weld strength evaluation method using the weld strength evaluation device 1 will be described.
First, one or more spot welds are prepared, and then the quantitative value of the weld strength of each spot weld is determined by, for example, a known method.

The sensor 10 of the evaluation device 1 is brought into contact with the spot weld to be evaluated and scanned along its surface, performing the following measurement at each measurement point. Specifically, under the control of the control unit 23, the oscillator 21 generates a signal at each frequency while gradually increasing the frequency at intervals of a given frequency (e.g., several kHz) within a specified frequency range, e.g., approximately 3 to 100 times or 5 to 100 times, e.g., approximately 1 kHz to 100 kHz. The signal detected by the detection coil 12 is processed by the signal processing unit 22, converted into a digital signal, and output to the data processing unit 30. The estimation unit 34 determines the correlation between the processed detection signal output from the signal processing unit 22 and the quantitative value of the weld strength, and performs a PLS regression analysis. The results are stored in the memory unit 33. Regression analysis and estimation are performed using only the real and imaginary parts of the detection signal, or the complex amplitude ratio, as parameters.

With the above configuration, the evaluation target is evaluated using excitation signals of multiple, preferably five or more, different frequencies. By using the differences in the propagation characteristics of the excitation signals of various frequencies, the objective variable of the evaluation target at various depths in the spot weld can be estimated. This allows for consistently more accurate evaluation of the evaluation target, regardless of the thickness or material of the plate. That is, based on the positional information of each measurement point and the internal information in the depth direction at each measurement point, the learning effect of the neural network allows for the correlation between predicted values and actual measured values based on the measurement data. Since the measurement data for each frequency varies depending on the thickness and material of the plate used in the spot weld, excitation signals of multiple, preferably five or more, different frequencies are used to reliably detect changes in the amplitude and phase of the measurement data.

As shown in FIG. 5 , the above-described scanning of the spot weld with the sensor 10 is performed in two dimensions, i.e., in the X-Y direction, along the surface of the front-side plate material 51 including the spot weld 57. The scanning range is set to be larger than the size of the spot weld 57 (estimated nugget dimensions). For example, if the diameter of the spot weld 57 is approximately 8 mm, the scanning range is set to approximately 10 mm in each of the X and Y directions. Then, while performing so-called uniaxial scanning in the X direction, measurement is performed at measurement points spaced apart at intervals smaller than the inner diameter of the sensor 10; for example, if the inner diameter of the sensor 10 is approximately 2 mm, measurement is performed at measurement points spaced apart at intervals of approximately 1 mm.
After one scan in the X direction is completed, the sensor 10 is shifted in the Y direction by a predetermined interval and scanned again in the X direction, and the above-mentioned measurement work is carried out for each measurement point at a predetermined interval.

Here, as shown in FIG. 5 , the spot weld 57 is scanned by the sensor 10 in the X-Y direction, i.e., multiple lines along the X direction offset from each other by a predetermined distance in the Y direction. In some cases, two scan results passing near the center of the spot weld 57 are extracted from the multiple scan results, and measurement and evaluation are performed with the measurement points spatially expanded. However, this is not limiting, and measurement and evaluation may be performed at multiple measurement points randomly on the spot weld 57. In this way, evaluation is performed based on detection signals from a predetermined number of scans (e.g., one or two scans) of the multiple one-dimensional scans obtained by the two-dimensional scans of the excitation coil 11 and the detection coil 12 at the location closest to the center of the spot weld, thereby ensuring accurate evaluation of the vicinity of the center of the spot weld 57. Alternatively, as shown in FIG. 6 , evaluation may be performed based on measurement data obtained by a single line scan in the X direction passing through the center of the spot weld 57, i.e., a one-dimensional scan.
Furthermore, when evaluating the condition of the plate materials related to the spot welds 57, i.e., the material quality, composition, or bonding condition thereof, the sensor 10 may be configured to scan the area between the spot welds 57, as shown in Fig. 7. This allows measurement data on the plate materials 51, 52 to be obtained without being affected by the spot welds 57, enabling a more accurate evaluation of the plate conditions.

Similarly, for spot weld 57 to be evaluated, sensor 10 is brought into contact with spot weld 57, and under the control of control unit 23, oscillator 21 oscillates signals at each frequency while increasing the frequency stepwise at any interval (e.g., several kHz) within a specified frequency range (e.g., approximately 1 kHz to 100 kHz), and outputs the signals to excitation coil 11. For each signal frequency, signals detected by detection coil 12 are processed by signal processing unit 22 and converted into digital signals, which are then output to data processing unit 30. Estimation unit 34 estimates the weld strength of spot weld 57 to be evaluated based on the processed detection signals output from signal processing unit 22 and the data stored in memory unit 33.
For example, when performing n-level evaluation, the number of data points for the scanning range must be increased depending on the final target accuracy and dimensions such as the width of spot weld 57. The number of data points is preferably determined by taking into consideration spatial imaging of the scanning range and the skin depth of the plate material, and is measured at at least n frequencies.
According to the above configuration, by performing evaluations at multiple measurement locations on spot weld 57 to be evaluated, it is possible to evaluate the weld strength and the condition of the plate material of spot weld 57 as a whole, thereby enabling a more accurate evaluation of the condition of spot weld 57. By scanning excitation coil 11 and detection coil 12 in one or two dimensions, measurement of spot weld 57 can be efficiently performed at multiple measurement locations.

Estimation unit 34 performs AI evaluation and estimation for spot weld 57, which is the evaluation target, using as parameters only the real and imaginary parts of the processed detection signal from measurement unit 20. The AI evaluation is performed by estimation unit 34 using any of the above-mentioned analysis methods, such as linear analysis, SVM, PLS regression analysis, and machine learning such as neural networks and supervised AI analysis methods, in data processing unit 30, using a program that calculates first evaluation formula 45 related to the first objective variable of weld strength and second evaluation formula 46 related to the second objective variable of material, both stored in storage device 32.
A method for estimating the weld strength will be described below.
That is, the estimated value (f') of the weld strength expressed as a complex number is expressed as follows:
Estimated weld strength = f'(Real(e _out /e _in ), Ima(e _out /e _in ))
Here, Real(e _out /e _in ) is the real part of the detection signal (e _out ) for the excitation signal (e _in ), and Ima(e _out /e _in ) is the imaginary part of the detection signal (e _out ) for the excitation signal (e _in ). Here, the detection signal for the excitation signal is expressed as a complex number represented by a real part and an imaginary part, but it may also be expressed by absolute amplitude and phase as long as it represents the same complex quantity.
The estimation unit 34 performs AI evaluation such as PLA regression based on the quantitative values of the welding strength of each plate material and the detection signal obtained by applying an alternating magnetic field using the sensor 10 and processing it using the measurement unit 20, and generates data to be stored in the memory unit 33.

An example of evaluation will be described with reference to FIG.
In this evaluation example, the evaluation device 1 is used to evaluate the welding strength of spot welds.
First, test specimens were created by spot welding one point on two plates for each of the cases of spot welding between the same metals (iron-iron) and dissimilar metals (iron-aluminum). Learning data was then acquired for these spot welds. In this evaluation example, the materials iron and aluminum were used as classification items.
Specifically, the sensor 10 scans a spot weld with an inner diameter of 8 mm in a range of 10 mm in the X and Y directions. At each measurement point in 1 mm increments, the frequency is changed, for example, from 5 kHz to 30 kHz in 5 kHz increments, for six frequencies. Alternatively, the frequency is changed, for example, from 5 kHz to 20 kHz in 1 kHz increments, for six frequencies.
Thereafter, a tensile test is carried out on each test piece to actually measure the welding strength of the spot welds of the test piece.

As a result, the actually measured weld strength (measured value) and part of the above measurement data are used as learning data, and the correlation between the actual measured value of weld strength and the measurement data is learned, for example, using a neural network, and the remaining measurement data is used as verification data to verify the neural network's learning results.

The measurement data obtained in the measurement operation is processed as follows.
That is, the complex amplitude ratio of the detection signal to the excitation signal is obtained as measurement data D, and this measurement data and comparison data are used to implement an analysis method such as a neural network. Here, the complex amplitude ratio refers to the ratio expressed as the (absolute) amplitude ratio and phase difference or the real part and imaginary part of two signals that differ from each other in amplitude and phase.

In addition, the spot welds are classified into, for example, two categories depending on the material of the plate material.
That is, by focusing on the correlation between the actual measured values and the predicted values, the classification of similar measured data is examined and two categories are created: homogeneous metals (iron-iron) and dissimilar metals (iron-aluminum). In this way, the preparation work for the weld strength evaluation is completed.

Next, after the preparation work for the weld strength evaluation, i.e., after classifying each measurement data into its corresponding category, an estimation formula for analytical evaluation, i.e., an evaluation formula, is defined for each category from the measurement data and comparison data. After defining the classification by category and the evaluation formula for each category, analysis using a neural network is performed using the measurement data and comparison data classified into each category to estimate the weld strength (see the measurement on the right side of Figure 4). The estimation formula created at this time becomes the evaluation formula for estimating the weld strength for each classified category.

Next, as a separate analysis, a classification analysis is performed using a neural network on the original measurement data, with the above categories as explanatory variables. In this example, the neural network randomly divides the measurement data into thirds, using two-thirds of the measurement data for learning and the remaining one-third for estimation. This results in the measurement data being reclassified, and in some cases some of the measurement data being moved to a different category. The estimation formula obtained here becomes the "classification formula." In the analysis, these classification formulas are used to classify the measurement data into categories.

Finally, based on the measurement data reclassified into each category in this way, weld strength is estimated using a neural network. In this case, the newly created evaluation formula is used.

In this way, in actual analysis work, the spot welds are first measured, and the acquired measurement data is then classified using a classification formula. The classification formula is then updated based on the revised classification, and the weld strength is calculated using an evaluation formula based on the reclassified measurement data for each category.

Next, a specific example will be described.
(Example)
In the examples, the weld strength of spot welds is evaluated for spot welding of plates made of the same metal (iron-iron) and for spot welding of plates made of different metals (iron-aluminum). The thicknesses of the iron and aluminum plates 51, 52 are both 1 mm.
After the preparation work for evaluating the weld strength of the spot weld described with reference to Figure 4, measurements were performed at 11 points on a line passing near the center of the spot weld (within a range of ±5 mm in 1 mm steps from the center), at frequencies ranging from 5 kHz to 20 kHz, four times the frequency, in 1 kHz steps.

Based on the obtained measurement data, analysis was performed using a neural network with the classified measurement data and comparison data to estimate the weld strength. In this case, 91 sample data were obtained for iron-iron and 360 sample data for iron-aluminum, of which two-thirds were used as learning data and the remaining one-third as verification data.
Then, when classifications were defined for both samples using a neural network with the surface plate being iron, the measurement data could be reliably classified into iron-iron and iron-aluminum classifications with 100% accuracy. After classifying the measurement data for each material of the plate to be spot welded in this way, the neural network was used to evaluate the weld strength for each classification.

As a result, in the case of iron-iron, the correlation coefficient was 0.999 during learning, as shown in Figure 8(A), and 0.918 during verification, as shown in Figure 8(B). In addition, in the case of iron-aluminum, the correlation coefficient was 0.905 during learning, as shown in Figure 9(A), and 0.868 during verification, as shown in Figure 9(B). This makes it possible to predict the weld strength of spot welds for each material type of plate to be spot welded, and it is expected that accuracy can be improved by increasing the amount of measurement data and by continuing measurement work and conducting further learning with the neural network.

The present invention can be implemented in various forms without departing from its spirit. For example, in the above-described embodiment, the definition of the classification and the definition of the evaluation formula for inferring weld strength are both performed using neural analysis, but this is not limited to this, and it is clear that the definitions can also be performed using other analytical methods, such as linear analysis, SVM (Support Vector Machine), PLS regression analysis, and supervised AI analytical methods.

In addition, in the above-described embodiment, the state of the plate material at the spot weld was explained in the case of iron-iron and iron-aluminum, but it is clear that this is not limited to this and can also be applied to estimating the weld strength of spot welds for combinations of iron and other metal plate materials, and even to estimating the composition or bonding state of the plate materials.

Furthermore, when the material to be used for the spot weld has been determined, only the strength can be estimated, and if it is desired to use it for material classification, it is clear that it can be applied alone as needed, such as for material classification only.

REFERENCE SIGNS LIST 1 Evaluation device 10 Sensor 11 Excitation coil 12 Detection coil 13 Magnetic path forming section 14 Sensor holding section 14a Open surface 20 Measurement section 21 Oscillation section 22 Signal processing section 23 Control section 30 Data processing section 31 Input/output interface section 32 Storage device 33 Storage section 34 Estimation section 41 Evaluation object (spot welded section)
42 Measurement data 43 Comparison data 44 Classification 51, 52 Plate material 53 Joint area 54, 55 Electrode rod 56 Spot welded portion

Claims (11)

a first step of applying excitation signals of different frequencies to an excitation coil positioned adjacent to a spot weld to be evaluated to generate an alternating magnetic field at the spot weld;
a second step of measuring a detection signal generated in a detection coil by a magnetic flux that is disturbed depending on the state of the spot weld;
a third step of evaluating the condition of the spot weld based on the detection signal;
It contains
In the third stage, a first objective variable and/or a second objective variable of the spot weld to be evaluated is estimated based on the detection signal of the second stage using correlation data relating to a correlation between a quantitative value of a first objective variable of the evaluation object that serves as comparison data for the spot weld, which are set in advance, a quantitative value of a second objective variable for classifying the spot weld to be evaluated, and the excitation signal of the first stage and the detection signal of the second stage,
determining, as measurement data, a complex amplitude ratio from the detection signal of the second stage to the excitation signal of the first stage, and defining a first evaluation formula for obtaining the first objective variable as the correlation data from the complex amplitude ratio of the detection signal as comparison data corresponding to the quantitative values of the first objective variable and the second objective variable in the measurement data, and defining a second evaluation formula for obtaining the second objective variable as the correlation data for each of the classifications by defining a plurality of types of classifications related to the second objective variable;
Next, for each piece of measurement data relating to the correlation data of the unknown actual evaluation target, a classification corresponding to the measurement data is selected, and the first objective variable of the evaluation target is estimated from the measurement data using the first evaluation formula corresponding to the selected classification;
a first objective variable of the spot weld to be evaluated is the weld strength of the spot weld,
The AI evaluation method for spot welds, characterized in that a second objective variable of the evaluation target of the spot welds is each of the front and back plate materials related to the spot welds or the state of the spot welds . 2. The AI evaluation method for spot welds according to claim 1 , wherein the state of the front and rear plate materials related to the spot welds or the state of the spot welds is the material, composition, structure or bonding state of the plate materials. 3. The AI evaluation method for spot welds according to claim 1, wherein the first objective variable is estimated by selecting an evaluation formula for the first objective variable through classification of the evaluation target of the spot welds using the second objective variable. 4. The AI evaluation method for spot welds according to claim 1, wherein the first evaluation formula and the second evaluation formula corresponding to the classification define measurement data as explanatory variables using any one of an analysis method of linear analysis, PLS regression analysis, SVM (Support Vector Machine), neural network, and supervised AI analysis method. 5. The AI evaluation method for spot welds described in claim 1, wherein the excitation coil and detection coil are moved along the surface of the spot weld, and measurement operations in the first and second stages are performed at a plurality of measurement points in a measurement target area including the spot weld. 6. The AI evaluation method for spot welds according to claim 5, wherein the excitation coil and detection coil are scanned in one or two dimensions along the surface of the spot weld so as to pass near the center of the measurement target area. 7. The AI evaluation method for spot welds according to claim 6 , wherein a third stage of evaluation is performed based on detection signals from a predetermined number of scans near the center of the spot weld among the two-dimensional scans. 8. The AI evaluation method for spot welds according to claim 1, wherein the excitation signals are generated at five or more mutually different frequencies. 9. The AI evaluation method for spot welds according to claim 8 , wherein the excitation signal has a frequency band that is approximately 5 to 100 times the minimum frequency to the maximum frequency. An AI evaluation device that executes the AI evaluation method according to any one of claims 1 to 9,
a sensor having a coil;
a measurement unit that generates an excitation signal for each frequency to be input to the coil in order to apply an alternating magnetic field to the evaluation object, and processes a detection signal output from the coil in response to the generated excitation signal;
a storage unit configured to store data relating to a correlation between quantitative values of objective variables of a plurality of evaluation targets and the sensor and the measurement unit;
an estimation unit that estimates a response variable of the evaluation target using data stored in the storage unit based on the detection signal processed by the measurement unit;
It is equipped with
the estimation unit obtains a complex amplitude ratio of the detection signal to the excitation signal as measurement data, defines a plurality of types of classifications based on the quantitative values of a plurality of different evaluation targets stored in the storage unit and the complex amplitude ratios of the detection signals as corresponding comparison data, and defines an evaluation formula for each of the classifications;
The AI evaluation device for spot welds, characterized in that the estimation unit selects, for each piece of measurement data, a classification consisting of the front and back plate materials of the spot weld or the state of the spot weld that corresponds to the measurement data, and estimates the welding strength of the spot weld, which is the objective variable to be evaluated, from the measurement data using an evaluation formula that corresponds to the selected classification. A program that causes a computer to function as the AI evaluation device for spot welds according to claim 10.