Method for identifying welding seam by CO2 welding automatic trolley

Abstract

The invention relates to the technical field of weld joint detection, in particular to a CO2 welding automatic trolley weld joint recognition method. According to the method, by measuring current signals and magnetic conductivity parameters in real time and combining comparison of dynamic change amplitudes, the initial position of a weld joint is accurately recognized, the initial accuracy of weld joint detection is improved, meanwhile, a displacement characteristic curve is extracted, weld joint form data is dynamically adjusted, the continuity and accuracy of weld joint path recognition are ensured, and the path tracking effect is improved; the welding seam boundary and width range are accurately judged in combination with electromagnetic parameter differences, the reliability of welding seam boundary detection is facilitated, irrelevant parameters are removed through frequency selection, edge contours are captured, welding seam surface features are marked, feature point layered clustering and normalization processing are conducted through a Gaussian mixture model, welding seam feature point data are comprehensively integrated, and the accuracy of welding seam boundary detection is improved. The generated feature point model provides an important basis for high efficiency and accuracy of weld joint recognition, and the performance of weld joint detection in the aspects of paths, forms and boundaries is integrally optimized.

Description

Technical Field

[0001] The invention relates to the technical field of weld detection, in particular to a method for CO2 welding automatic trolley to identify welds. Background Art

[0002] The field of weld detection technology aims to support automatic welding equipment to accurately track and control the weld path during the welding process by automatically detecting and identifying key parameters on welded parts, so as to ensure the stability and consistency of welding quality and improve welding efficiency.

[0003] The purpose of the CO2 welding automatic trolley weld recognition method is to accurately identify the weld path on the workpiece through automated means, so that the welding trolley can accurately move and weld along the identified weld trajectory, realize the automation of the welding process, ensure the consistency and accuracy of welding quality, improve welding efficiency, and improve the stability and reliability of welding while reducing production costs.

[0004] The path tracking of traditional methods is not stable enough, resulting in reduced recognition effect when the lighting conditions of the weld path are poor and the welding surface is contaminated. It is difficult to accurately distinguish subtle changes in the left and right boundaries of the weld, resulting in deviations in the determination of the weld width range, affecting the consistency of welding quality. In addition, due to insufficient data processing capabilities, existing methods are unable to effectively detect high-frequency detail changes in welds, resulting in the omission of key defects in quality assessment. The lack of the ability to integrate multi-dimensional data results in limited expression of the global features of the weld, affecting the accurate assessment of the weld morphology. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a method for identifying welds using an automatic CO2 welding trolley.

[0006] In order to achieve the above object, the present invention adopts the following technical scheme: a CO2 welding automatic trolley identification method, comprising the following steps:

[0007] Step 1: Based on the signal output data of the induction sensor of the weld inspection trolley, the current signal and magnetic permeability parameters are measured in real time. The weld starting position is identified by comparing the dynamic change amplitude of each parameter, and the magnetic permeability increment position is detected by the current change difference. The weld starting point position is accurately determined to generate the weld starting position signal;

[0008] Step 2: Based on the weld starting position signal, the weld path is monitored using the real-time current signal output by the sensor, the incremental change of each position on the weld path is recorded, the weld direction data is extracted through the path displacement characteristic curve, and dynamic adjustments are made in combination with the weld morphology changes to obtain the weld path positioning signal;

[0009] Step 3: Based on the weld path positioning signal, the weld boundary is judged by the current intensity and magnetic permeability change data, the left and right boundary areas of the weld are determined, the electromagnetic parameter difference at the boundary position is measured, the weld width range is located in combination with the boundary value, and the weld boundary positioning signal is obtained;

[0010] Step 4: Based on the weld boundary positioning signal, the weld contour is extracted from the low-frequency signal data, irrelevant parameters are eliminated through frequency selection, edge deviation is identified in the high-frequency band, edge contour is captured and weld surface features are marked, small deviation morphological data of the weld is identified, and a frequency domain signal of weld feature points is generated;

[0011] Step 5: Based on the frequency domain signal of the weld feature point and combined with the heat diffusion characteristics, the boundary mapping feature frequency band is called to perform multi-dimensional parameter cross analysis, the weld boundary and temperature gradient parameter combination is mapped to the feature matrix, the comprehensive distribution of each weld feature point is generated, and the weld feature distribution matrix is ​​obtained;

[0012] Step 6: Based on the weld feature distribution matrix, a Gaussian mixture model is used to hierarchically cluster the multidimensional values ​​of the feature points, identify the optimal distribution combination of each feature point, calculate and summarize the normalized results of the values ​​of each point, integrate the model parameters of each point to generate a weld feature point model, and generate a weld recognition model.

[0013] As a further solution of the present invention, the specific steps of generating the weld starting position signal are:

[0014] Based on the signal output data of the induction sensor of the weld inspection trolley, the current signal is continuously collected, the peak and valley positions in the current fluctuation are analyzed, different current characteristic points are identified through fluctuation changes, and the position and time values ​​are recorded to generate a current characteristic signal point set;

[0015] Based on the current characteristic signal point set, the weld area is located within the range of current fluctuation and magnetic permeability change, the magnetic permeability increment mutation interval is identified, the magnetic permeability peak value and change trend are calculated, the weld starting position is determined in combination with the current characteristic points, and the weld initial position mark is generated;

[0016] Based on the weld initial position mark, the actual weld starting point is located by further identifying the current change difference and combining the gradually decreasing trend of the magnetic permeability increment, marking the starting point position and the electromagnetic characteristic value, and generating a weld starting position signal;

[0017] The weld starting position signal includes the current intensity, magnetic permeability change value and initial position coordinates of the starting point.

[0018] As a further solution of the present invention, the specific steps of generating the weld path positioning signal are:

[0019] Based on the weld starting position signal, the current signal change value at each position on the weld path is collected in real time, the current increment and direction deviation on the path are measured, the data and position of each measuring point are recorded, and the weld path increment data is generated;

[0020] Based on the weld path incremental data, a displacement curve of current change in the path is calculated, the trend angle and offset value of each position in the curve are extracted, the main direction and turning point of the path are determined, and the weld path trend data is generated;

[0021] Based on the weld path direction data, combined with the continuous change of the path shape, positioning correction is performed, the actual position and direction of the path are recorded, the final positioning information of the weld path is updated, and the weld path positioning signal is obtained;

[0022] The weld path positioning signal includes the displacement angle, path increment characteristics and strike angle of the path.

[0023] As a further solution of the present invention, the specific steps of generating the weld boundary positioning signal are:

[0024] Based on the weld path positioning signal, the boundary position is detected by using the fluctuation of the current intensity, the peak and valley values ​​of the signal intensity are identified, the initial positions of the left and right boundaries of the weld are determined, and the initial signal of the weld boundary is generated;

[0025] Based on the initial signal of the weld boundary, the permeability change is detected within the interval, the difference in permeability between the left and right boundaries is detected, and the specific range of the weld boundary area is determined to obtain the left and right boundary data of the weld;

[0026] Based on the left and right boundary data of the weld, the weld width is measured and the boundary width range is marked by comparing the current intensity and the magnetic permeability change, so as to obtain the weld boundary positioning signal;

[0027] The weld boundary positioning signal includes the left and right boundary coordinates of the weld, the weld width and the boundary electromagnetic characteristic value.

[0028] As a further scheme of the present invention, a process of judging the difference in magnetic permeability within an interval and determining a boundary range, collecting magnetic permeability data point by point, collecting magnetic permeability values ​​in segments within the left and right boundary areas of the weld, establishing a magnetic permeability distribution map within the boundary, determining a magnetic permeability difference threshold, which is used to identify sudden changes in magnetic permeability within the boundary area, screening out areas with significant changes in magnetic permeability by detecting the difference in magnetic permeability between adjacent points, determining the boundary range of the weld, comparing the difference in magnetic permeability between sampling points, determining the position change of the left and right boundaries, and marking the specific range of the weld boundary at the point that meets the threshold.

[0029] As a further solution of the present invention, the specific steps of generating the frequency domain signal of the weld feature point are:

[0030] Based on the weld boundary positioning signal, extract low-frequency signal data, filter and isolate low-frequency components, determine the position and characteristics of the weld contour line, and generate a weld contour signal;

[0031] Based on the weld contour signal, the parameters of the frequency band are screened, irrelevant fluctuations are identified to eliminate useful signals in the high frequency band, small deviations on the weld edge position are captured, and weld edge feature data is obtained;

[0032] Based on the weld edge feature data, each edge feature point is located according to the frequency response value, and the surface deviation point of the weld is marked according to the value fluctuation to generate a weld feature point frequency domain signal;

[0033] The frequency domain signal of the weld characteristic point includes a characteristic frequency band, a deviation amplitude and a frequency response value.

[0034] As a further solution of the present invention, the specific steps of generating the weld feature distribution matrix are:

[0035] Based on the frequency domain signal of the weld feature points, the temperature change parameters of each point are extracted, the temperature difference in the heat diffusion area of ​​the weld boundary is measured, the corresponding relationship between the temperature value and the boundary position is compared, and the heat diffusion boundary parameter set is generated;

[0036] Based on the thermal diffusion boundary parameter set, the temperature data of the weld boundary points are mapped to the spatial coordinates, and the distribution of each feature point in the multidimensional space is determined by the position parameters to generate a multidimensional coordinate matrix of the feature points.

[0037] Based on the multi-dimensional coordinate matrix of feature points, the spatial distribution coordinates of each feature point are arranged, and the thermal diffusion parameters and the position data of the weld boundary points are combined to form a comprehensive distribution feature matrix to obtain the weld feature distribution matrix;

[0038] The weld feature distribution matrix includes heat diffusion values, boundary feature point locations and temperature gradient data.

[0039] As a further solution of the present invention, the temperature difference in the heat diffusion area of ​​the weld boundary is measured by sensing the real-time temperature data of the sensor, measuring the temperature gradient in the weld boundary area point by point, selecting the heat diffusion area of ​​adjacent characteristic points within the weld boundary, calculating the temperature increment between the characteristic points, performing a difference analysis on the temperature values ​​of each pair of adjacent points, and determining the heat diffusion range and temperature difference trend of the weld boundary.

[0040] As a further solution of the present invention, the specific steps of generating the weld recognition model are:

[0041] Based on the weld feature distribution matrix, a Gaussian mixture model is used to classify the value of each feature point, divide the groups according to the value difference between adjacent points, identify the hierarchical relationship of the feature points, and generate feature point hierarchical group data;

[0042] Based on the feature point hierarchical group data, the numerical characteristics of adjacent feature points are calculated, the points in the hierarchical group are normalized, the combination data is established, and a normalized feature combination is generated;

[0043] Based on the normalized feature combination, the values ​​of each normalized feature point and the spatial coordinates are integrated to form an overall model of the weld boundary and feature points, and a weld recognition model is generated;

[0044] Wherein, the weld recognition model includes feature point positions, normalized parameters and model combination relationships.

[0045] As a further solution of the present invention, the Gaussian mixture model is according to the formula:

[0046]

[0047] Where: x is each feature point in the weld feature distribution matrix, λ is the model parameter set, π k is the weight of the kth Gaussian component, represents the kth improved Gaussian distribution, μ k and∑ k are the mean vector and covariance matrix of the kth distribution, δ is the feature point offset, α is the covariance scaling factor, and β is the matrix translation.

[0048] Compared with the prior art, the advantages and positive effects of the present invention are:

[0049] 1. In the present invention, by real-time measurement of current signals and magnetic permeability parameters, combined with the comparison of dynamic change amplitudes, the starting position of the weld can be accurately identified, the initial accuracy of weld detection can be improved, and the displacement characteristic curve can be extracted at the same time, and the weld morphology data can be dynamically adjusted to ensure the continuity and accuracy of weld path identification and improve the path tracking effect.

[0050] 2. In the present invention, the change data of current intensity and magnetic permeability are combined with the difference of electromagnetic parameters to accurately judge the weld boundary and width range, which is helpful to the reliability of weld boundary detection. By frequency selection, irrelevant parameters are eliminated and edge contours are captured, and weld surface features are marked, which can effectively detect small deviation morphological data and provide a more detailed analysis basis for weld quality assessment.

[0051] 3. In the present invention, hierarchical clustering and normalization of feature points are performed through Gaussian mixture model, which helps to comprehensively integrate weld feature point data. The generated feature point model provides an important basis for the efficiency and accuracy of weld identification, and overall optimizes the performance of weld detection in terms of path, shape, and boundary. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a schematic diagram of the main steps of the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0054] Embodiment 1

[0055] See also Figure 1 The present invention provides a technical solution: a CO2 welding automatic trolley identification method, comprising the following steps:

[0056] Step 1: Based on the signal output data of the induction sensor of the weld inspection trolley, the current signal and magnetic permeability parameters are measured in real time. The weld starting position is identified by comparing the dynamic change amplitude of each parameter, and the magnetic permeability increment position is detected by the current change difference. The weld starting point position is accurately determined to generate the weld starting position signal;

[0057] Step 2: Based on the weld starting position signal, the real-time current signal output by the sensor is used to monitor the weld path, and the incremental changes of each position on the weld path are recorded. The weld direction data is extracted through the path displacement characteristic curve, and dynamic adjustments are made in combination with the weld morphology changes to obtain the weld path positioning signal;

[0058] Step 3: Based on the weld path positioning signal, the weld boundary is judged through the current intensity and magnetic permeability change data, the left and right boundary areas of the weld are determined, the electromagnetic parameter difference at the boundary position is measured, the weld width range is located in combination with the boundary value, and the weld boundary positioning signal is obtained;

[0059] Step 4: Based on the weld boundary positioning signal, the weld contour is extracted from the low-frequency signal data, irrelevant parameters are eliminated through frequency selection, edge deviation is identified in the high-frequency band, edge contour is captured and weld surface features are marked, small deviation morphological data of the weld is identified, and frequency domain signals of weld feature points are generated;

[0060] Step 5: Based on the frequency domain signal of the weld feature points and combined with the heat diffusion characteristics, the boundary mapping feature frequency band is called to perform multi-dimensional parameter cross analysis, the weld boundary and temperature gradient parameter combination is mapped to the feature matrix, the comprehensive distribution of each weld feature point is generated, and the weld feature distribution matrix is ​​obtained;

[0061] Step 6: Based on the weld feature distribution matrix, the Gaussian mixture model is used to perform hierarchical clustering on the multi-dimensional values ​​of the feature points, identify the optimal distribution combination of each feature point, calculate the normalized results of the values ​​of each point and summarize them, integrate the model parameters of each point to generate a weld feature point model, and generate a weld recognition model.

[0062] The specific steps to generate the weld start position signal are:

[0063] Based on the signal output data of the induction sensor of the weld inspection trolley, the current signal is continuously collected, the peak and valley positions in the current fluctuation are analyzed, different current characteristic points are identified through fluctuation changes, and the position and time values ​​are recorded to generate a current characteristic signal point set;

[0064] Based on the current characteristic signal point set, the weld area is located within the range of current fluctuation and magnetic permeability change, the magnetic permeability increment mutation interval is identified, the magnetic permeability peak value and change trend are calculated, and the weld starting position is determined in combination with the current characteristic points to generate the weld initial position mark;

[0065] Based on the weld initial position mark, the actual weld starting point is located by further identifying the current change difference and combining the gradual reduction trend of the magnetic permeability increment, marking the starting position and electromagnetic characteristic value, and generating the weld starting position signal;

[0066] Based on the signal output data of the induction sensor of the weld inspection trolley, the Fourier transform algorithm is used to continuously collect the current signal. Specifically, the collected current signal is transformed in the frequency domain and the main frequency component and amplitude change in the signal are extracted. The input data is the original induction signal sequence. The 512-point fast Fourier transform is used for frequency domain conversion, and the peak frequency and the corresponding time position are recorded. The fluctuation in the time domain signal is fitted and analyzed in combination with the signal reconstruction formula, and the peak and trough positions in the fluctuation are identified. A data table containing the time and amplitude characteristics of the current signal is generated, and a current characteristic signal point set is generated;

[0067] Based on the current characteristic signal point set, the wavelet transform algorithm is used to locate the weld area within the range of current fluctuation and magnetic permeability change. Specifically, the current signal and magnetic permeability parameters are continuously sampled and the wavelet coefficients are subjected to layered threshold processing. The number of decomposition layers is set to 5. The position where the wavelet coefficient change rate is higher than the threshold is located as the characteristic point interval. The magnetic permeability peak value and its change trend are calculated by arranging the characteristic points in time series. The magnetic permeability increment mutation interval is determined in combination with the current characteristic value at each time point. The corresponding weld characteristic point area range is marked to generate the weld initial position mark.

[0068] Based on the initial position mark of the weld, the differential signal analysis method is used to gradually identify the current change difference. By constructing a differential sequence and calculating the absolute value of the signal difference value of two consecutive time points, the threshold is set to three times the dynamic standard deviation, and the difference mutation point is marked as the current characteristic point. The actual weld starting point is located in combination with the gradual reduction trend of the magnetic permeability increment. By linearly fitting the incremental trend, the least squares method is used to calculate the fitting curve parameters to determine the inflection point position of the magnetic permeability curve. At the same time, the starting position and its corresponding electromagnetic characteristic value are recorded to generate the weld starting position signal.

[0069] The weld starting position signal includes the current intensity, magnetic permeability change value and initial position coordinates of the starting point.

[0070] The specific steps to generate the weld path positioning signal are:

[0071] Based on the weld starting position signal, the current signal change value at each position on the weld path is collected in real time, the current increment and direction deviation on the path are measured, the data and position of each measurement point are recorded, and the weld path increment data is generated;

[0072] Based on the incremental data of the weld path, the displacement curve of the current change in the path is calculated, the trend angle and offset value of each position in the curve are extracted, the main direction and turning point of the path are determined, and the weld path trend data is generated;

[0073] Based on the weld path direction data and combined with the continuous changes in the path shape, positioning correction is performed, the actual position and direction of the path are recorded, the final positioning information of the weld path is updated, and the weld path positioning signal is obtained;

[0074] Based on the weld starting position signal, the Kalman filter algorithm is used to collect the current signal change value at each position on the weld path in real time. Specifically, the current signal output by the sensor in real time is iteratively processed multiple times, the initial state estimation value and the covariance matrix are set, and the optimal estimation value of the current state is calculated using the prediction update formula. The prediction step includes calculating the current increment and direction offset of the current point on the path according to the state transfer equation. The update step adjusts the covariance according to the error between the sensor observation value and the prediction value and records the corrected measurement point data and position to generate weld path increment data.

[0075] Based on the incremental data of the weld path, a curve fitting algorithm is used to calculate the displacement curve of the current change in the path. Specifically, the fitting model is constructed by using the least squares method, the current change value and spatial coordinates of the path point sequence are set as input variables, and a quadratic curve fitting equation is established and the curve coefficient is calculated. The QR decomposition method is used to solve the linear equation group to reduce the numerical error, and the strike angle and offset value of each position in the fitting curve are extracted. The change rate of the strike angle is calculated by discrete difference and the turning points with a change rate higher than the threshold are marked to generate the weld path direction data.

[0076] Based on the weld path direction data, the iterative closest point algorithm is used in combination with the continuous change of the path morphology for positioning correction. Specifically, the actual position of the path is analyzed with the reference path, the initial pose estimation value is set, and the Euclidean distance between the current path point and the reference point is calculated. The minimum square error of the nearest point pair is used as the objective function to iteratively optimize the path alignment result, update the actual position and direction parameters of the path point, record the corrected point coordinates, update the path data structure, and obtain the weld path positioning signal.

[0077] The weld path positioning signal includes the displacement angle, path increment characteristics and strike angle of the path.

[0078] The specific steps to generate the weld boundary positioning signal are:

[0079] Based on the weld path positioning signal, the boundary position is detected by using the fluctuation of current intensity, the peak and valley values ​​of signal intensity are identified, the initial positions of the left and right boundaries of the weld are determined, and the initial signal of the weld boundary is generated;

[0080] Based on the initial signal of the weld boundary, the permeability change is detected within the interval, the difference in permeability between the left and right boundaries is detected, and the specific range of the weld boundary area is determined to obtain the left and right boundary data of the weld;

[0081] Based on the left and right boundary data of the weld, the weld width is measured and the boundary width range is marked by comparing the current intensity and the magnetic permeability change to obtain the weld boundary positioning signal;

[0082] Based on the weld path positioning signal, the peak-valley detection algorithm is used to detect the boundary position of the current intensity fluctuation. Specifically, the input current signal data is analyzed by sliding window, the window size is set to 50 data points, and the peak and valley values ​​in the signal are identified by the local maximum and minimum search algorithm. The difference threshold of the current intensity change rate in each window is calculated as 10% of the signal amplitude, and the points where the change rate exceeds the threshold are marked as the initial positions of the left and right boundaries. The corresponding time and current value are recorded to generate the initial signal of the weld boundary.

[0083] Based on the initial signal of the weld boundary, the dynamic time warping algorithm is used to detect the permeability change within the interval. Specifically, the dynamic time distance between the permeability signal sequence and the template sequence is calculated, the step size is set to 1, the matching range is 3 time points, and the interval where the permeability difference exceeds the set threshold is determined as the permeability change area of ​​the left and right boundaries. The data points in the left and right boundary change interval are integrated and calculated, and the peak values ​​are extracted. The specific range of the left and right boundaries is marked to obtain the left and right boundary data of the weld.

[0084] Based on the left and right boundary data of the weld, the ratio analysis method is used to measure the weld width by comparing the current intensity and the magnetic permeability change. Specifically, the current intensity difference and the magnetic permeability change rate of the left and right boundary data points are calculated, and the ratio threshold is set to 1.2. The weld width is calculated according to the positions of the left and right boundary points and the corresponding interval range is recorded. In the comparative analysis, the data is normalized to the interval of 0 to 1 by standardization processing, and the width range data of the left and right boundaries are generated to obtain the weld boundary positioning signal.

[0085] The weld boundary positioning signal includes the left and right boundary coordinates of the weld, the weld width and the boundary electromagnetic characteristic value.

[0086] The process of judging the difference in magnetic permeability within the interval and determining the boundary range, collecting magnetic permeability data point by point, collecting magnetic permeability values ​​in segments within the left and right boundary areas of the weld, establishing a magnetic permeability distribution map within the boundary, and determining the magnetic permeability difference threshold to identify magnetic permeability mutations within the boundary area. By detecting the magnetic permeability difference between adjacent points, the area with significant magnetic permeability changes is screened out to determine the boundary range of the weld, compare the magnetic permeability difference between the sampling points, determine the position change of the left and right boundaries, and mark the specific range of the weld boundary at the point that meets the threshold.

[0087] The specific steps to generate the frequency domain signal of weld feature points are:

[0088] Based on the weld boundary positioning signal, extract the low-frequency signal data, filter and isolate the low-frequency components, determine the weld contour position and characteristics, and generate the weld contour signal;

[0089] Based on the weld contour signal, the parameters of the frequency band are screened, irrelevant fluctuations are identified to eliminate useful signals in the high-frequency band, and small deviations in the weld edge position are captured to obtain weld edge feature data;

[0090] Based on the weld edge feature data, each edge feature point is located according to the frequency response value, the surface deviation point of the weld is marked according to the value fluctuation, and the frequency domain signal of the weld feature point is generated;

[0091] Based on the weld boundary positioning signal, a bandpass filtering algorithm is used to extract low-frequency signal data. Specifically, by setting the passband range of the filter to 0 to 50 Hz, the input weld boundary signal is subjected to discrete Fourier transform to decompose the spectral components and shield the high-frequency and ultra-low-frequency parts. The filtered low-frequency signal is restored using inverse Fourier transform, and the low-frequency components are analyzed in time series to locate the signal energy center and mark the key points of the contour line. The time position and amplitude information of the contour line are recorded to generate the weld contour signal.

[0092] Based on the weld contour signal, the fast Fourier transform is used to screen the parameters of the frequency band. Specifically, the energy contribution value of each frequency component is sorted by calculating the spectral density function of the signal, and the frequency band with a contribution value lower than 10% is eliminated as irrelevant fluctuations. The signal with an energy contribution value exceeding the threshold in the high-frequency band is intercepted, and the captured high-frequency band data is subjected to segmented weighted calculation to extract the characteristic value of the weld edge signal, locate the small deviation point on the edge position, and record the spectral characteristics of the deviation to obtain the weld edge characteristic data;

[0093] Based on the weld edge feature data, the frequency response analysis method is used to locate each edge feature point according to the frequency response value. Specifically, by calculating the amplitude-frequency response and phase-frequency response curves of the edge signal, the point with a response value change rate higher than 20% is set as the weld edge feature point, and the weld surface deviation point is marked according to the response value fluctuation range. The spectrum data of the deviation point is segmented and normalized, and its frequency domain distribution characteristics are recorded. A feature point matrix containing frequency, amplitude and response coefficient is generated, and a frequency domain signal of the weld feature point is generated.

[0094] Among them, the frequency domain signal of the weld characteristic point includes characteristic frequency band, deviation amplitude and frequency response value.

[0095] The specific steps to generate the weld feature distribution matrix are:

[0096] Based on the frequency domain signal of the weld feature points, the temperature change parameters of each point are extracted, the temperature difference in the heat diffusion area of ​​the weld boundary is measured, the corresponding relationship between the temperature value and the boundary position is compared, and the heat diffusion boundary parameter set is generated;

[0097] Based on the thermal diffusion boundary parameter set, the temperature data of the weld boundary points are mapped to the spatial coordinates, and the distribution of each feature point in the multidimensional space is determined by the position parameters to generate a multidimensional coordinate matrix of the feature points.

[0098] Based on the multi-dimensional coordinate matrix of feature points, the spatial distribution coordinates of each feature point are arranged, and the thermal diffusion parameters and the position data of the weld boundary points are combined to form a comprehensive distribution feature matrix to obtain the weld feature distribution matrix;

[0099] Based on the frequency domain signal of the characteristic points of the weld, the heat diffusion model is used to extract the temperature change parameters of each point. Specifically, the frequency domain signal data of each characteristic point is subjected to time-frequency analysis, the heat diffusion time constant is set to 10ms, the temperature response of the signal is calculated by piecewise integration, and the temperature gradient is calculated and its corresponding relationship with the boundary position is recorded in combination with the boundary heat diffusion area range of the measurement point. The corresponding relationship data is stored in the key-value pair format to generate the heat diffusion boundary parameter set;

[0100] Based on the thermal diffusion boundary parameter set, a multidimensional space mapping algorithm is used to map the temperature data of the weld boundary points to the spatial coordinates. Specifically, a three-dimensional mapping function is constructed, and the temperature gradient value is used as the input variable of the mapping. Combined with the two-dimensional plane coordinates of the boundary points, the coordinate values ​​of each feature point in the multidimensional space are generated by the interpolation method. The multidimensional coordinates of all feature points are standardized, and the three decimal places of the coordinate value are retained to generate the multidimensional coordinate matrix of the feature points.

[0101] Based on the multi-dimensional coordinate matrix of feature points, the spatial distribution coordinates of each feature point are arranged by a matrix combination algorithm. Specifically, the thermal diffusion parameters and the spatial position data of the weld boundary points are extracted, and the two are combined into one according to the feature point sequence through a matrix merging function. The combined matrix is ​​normalized in rows and columns to avoid the influence of the difference in numerical ranges, and the final comprehensive distribution data is output in a matrix format to obtain the weld feature distribution matrix.

[0102] Among them, the weld feature distribution matrix includes heat diffusion values, boundary feature point locations and temperature gradient data.

[0103] Measure the temperature difference in the heat diffusion area of ​​the weld boundary. Through the real-time temperature data of the sensing sensor, measure the temperature gradient in the weld boundary area point by point. Within the weld boundary, select the heat diffusion area of ​​adjacent feature points, calculate the temperature increment between the feature points, perform difference analysis on the temperature values ​​of each pair of adjacent points, and determine the heat diffusion range and temperature difference trend of the weld boundary.

[0104] The specific steps to generate the weld recognition model are:

[0105] Based on the weld feature distribution matrix, the Gaussian mixture model is used to classify the value of each feature point, divide the groups according to the value difference between adjacent points, identify the hierarchical relationship of the feature points, and generate the feature point hierarchical group data;

[0106] Based on the hierarchical grouping data of feature points, the numerical characteristics of adjacent feature points are calculated, the points in the hierarchical group are normalized, the combined data is established, and the normalized feature combination is generated;

[0107] Based on the normalized feature combination, the values ​​and spatial coordinates of each normalized feature point are integrated to form an overall model of the weld boundary and feature points, and generate a weld recognition model;

[0108] Based on the weld feature distribution matrix, the Gaussian mixture model is used to classify the value of each feature point. Specifically, the probability density modeling of the feature distribution matrix data is performed, the initial number of components of the mixed distribution is set to 3, and the expectation maximization algorithm is used for parameter estimation. The initial mean and covariance matrix are calculated by the k-means clustering algorithm, and the convergence threshold is set to 10^-4. The numerical differences between adjacent points are divided into groups according to the Gaussian distribution probability, and the hierarchical relationship of the feature points is marked on the divided group data to generate feature point hierarchical group data;

[0109] Based on the hierarchical group data of feature points, the normalization method is used to calculate the numerical characteristics of adjacent feature points. Specifically, the minimum and maximum values ​​of the feature points in the hierarchical group are calculated, a linear normalization formula is constructed, and the normalization range is set to 0 to 1. The values ​​of the points in each group are processed according to the normalization formula to generate normalized point set data, and the weighted average method is used to combine the normalized point values ​​with the feature center data of the group, mark the combined feature points and store the results to generate a normalized feature combination;

[0110] Based on the normalized feature combination, the coordinate integration algorithm is used to integrate the values ​​and spatial coordinates of each normalized feature point. Specifically, the normalized feature points are traversed, and the spatial coordinate index of the point is matched with its value to construct a three-dimensional feature point model. The spatial coordinate value of each feature point and its normalized value are set to be arranged in sequence as a matrix unit. The generated matrix data is sparsely processed to improve the calculation efficiency, and the overall spatial representation of the boundary points and feature points is formed to generate a weld recognition model.

[0111] Among them, the weld recognition model includes feature point positions, normalized parameters and model combination relationships.

[0112] Gaussian mixture model, according to the formula:

[0113]

[0114] Where: x is each feature point in the weld feature distribution matrix, λ is the model parameter set, π k is the weight of the kth Gaussian component, represents the kth improved Gaussian distribution, μ k and∑ k are the mean vector and covariance matrix of the kth distribution, δ is the feature point offset, α is the covariance scaling factor, and β is the matrix translation;

[0115] Execution process: First, each feature point value x in the weld feature distribution matrix is ​​initially processed, and the basic model of each group is initialized according to the overall model parameter set λ, including the group mean μ k , covariance matrix Σ k and weight coefficient π k , mean μ k Represents the average distribution position of feature points, covariance matrix Σ k It represents the correlation and distribution shape between the points in the group, and the weight coefficient π k represents the proportion of the group in the overall model, and then introduces an offset δ in the mean adjustment step to adjust the mean μ k Fine-tune to correct the position offset of feature points caused by external factors to ensure that the distribution is more consistent with the actual data, and then apply the scaling factor α to the covariance matrix Σ k Zoom in and out to control the concentration and dispersion of the group distribution, increase the adaptability of the model to different degrees of variation in the data set, introduce a translation β to correct the reference position of the covariance matrix, so that the distribution can better adapt to the asymmetric data form, and finally iteratively calculate the weight coefficient π k The value of is used to ensure that each group maintains a normalized distribution in the overall model. The whole process ultimately generates hierarchical group data of feature points to achieve accurate classification of feature points.

[0116] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A CO2 welding automatic trolley identification method, characterized in that: The following steps are involved: Step 1: Based on the signal output data of the induction sensor of the weld inspection trolley, the current signal and magnetic permeability parameters are measured in real time. The weld starting position is identified by comparing the dynamic change amplitude of each parameter, and the magnetic permeability increment position is detected by the current change difference. The weld starting point position is accurately determined to generate the weld starting position signal; Step 2: Based on the weld starting position signal, the weld path is monitored using the real-time current signal output by the sensor, the incremental change of each position on the weld path is recorded, the weld direction data is extracted through the path displacement characteristic curve, and dynamic adjustments are made in combination with the weld morphology changes to obtain the weld path positioning signal; Step 3: Based on the weld path positioning signal, the weld boundary is judged by the current intensity and magnetic permeability change data, the left and right boundary areas of the weld are determined, the electromagnetic parameter difference at the boundary position is measured, the weld width range is located in combination with the boundary value, and the weld boundary positioning signal is obtained; Step 4: Based on the weld boundary positioning signal, the weld contour is extracted from the low-frequency signal data, irrelevant parameters are eliminated through frequency selection, edge deviation is identified in the high-frequency band, edge contour is captured and weld surface features are marked, small deviation morphological data of the weld is identified, and a frequency domain signal of weld feature points is generated; Step 5: Based on the frequency domain signal of the weld feature point and combined with the heat diffusion characteristics, the boundary mapping feature frequency band is called to perform multi-dimensional parameter cross analysis, the weld boundary and temperature gradient parameter combination is mapped to the feature matrix, the comprehensive distribution of each weld feature point is generated, and the weld feature distribution matrix is ​​obtained; Step 6: Based on the weld feature distribution matrix, a Gaussian mixture model is used to hierarchically cluster the multidimensional values ​​of the feature points, identify the optimal distribution combination of each feature point, calculate and summarize the normalized results of the values ​​of each point, integrate the model parameters of each point to generate a weld feature point model, and generate a weld recognition model.

2. The CO2 welding automatic trolley weld identification method according to claim 1 is characterized in that: The weld starting position signal includes the current intensity, magnetic permeability change value and initial position coordinates of the starting point.

3. The CO2 welding automatic trolley weld identification method according to claim 1 is characterized in that: The weld path positioning signal includes a displacement angle of the path, a path increment feature and a strike angle.

4. The CO2 welding automatic trolley weld identification method according to claim 1 is characterized in that: The specific steps of generating the weld boundary positioning signal are: Based on the weld path positioning signal, the boundary position is detected by using the fluctuation of the current intensity, the peak and valley values ​​of the signal intensity are identified, the initial positions of the left and right boundaries of the weld are determined, and the initial signal of the weld boundary is generated; Based on the initial signal of the weld boundary, the permeability change is detected within the interval, the difference in permeability between the left and right boundaries is detected, and the specific range of the weld boundary area is determined to obtain the left and right boundary data of the weld; Based on the left and right boundary data of the weld, the weld width is measured and the boundary width range is marked by comparing the current intensity and the magnetic permeability change, so as to obtain the weld boundary positioning signal; The weld boundary positioning signal includes the left and right boundary coordinates of the weld, the weld width and the boundary electromagnetic characteristic value.

5. The CO2 welding automatic trolley weld identification method according to claim 4 is characterized in that: The process of judging the difference in magnetic permeability within the interval and determining the boundary range, collecting magnetic permeability data point by point, collecting magnetic permeability values ​​in segments within the left and right boundary areas of the weld, establishing a magnetic permeability distribution map within the boundary, and determining the magnetic permeability difference threshold to identify magnetic permeability mutations within the boundary area. By detecting the magnetic permeability difference between adjacent points, the area with significant magnetic permeability changes is screened out to determine the boundary range of the weld, compare the magnetic permeability difference between the sampling points, determine the position change of the left and right boundaries, and mark the specific range of the weld boundary at the point that meets the threshold.

6. The CO2 welding automatic trolley weld identification method according to claim 1 is characterized in that: The frequency domain signal of the weld characteristic point includes a characteristic frequency band, a deviation amplitude and a frequency response value.

7. The CO2 welding automatic trolley weld identification method according to claim 1 is characterized in that: The specific steps of generating the weld feature distribution matrix are: Based on the frequency domain signal of the weld feature points, the temperature change parameters of each point are extracted, the temperature difference in the heat diffusion area of ​​the weld boundary is measured, the corresponding relationship between the temperature value and the boundary position is compared, and the heat diffusion boundary parameter set is generated; Based on the thermal diffusion boundary parameter set, the temperature data of the weld boundary points are mapped to the spatial coordinates, and the distribution of each feature point in the multidimensional space is determined by the position parameters to generate a multidimensional coordinate matrix of the feature points. Based on the multi-dimensional coordinate matrix of feature points, the spatial distribution coordinates of each feature point are arranged, and the thermal diffusion parameters and the position data of the weld boundary points are combined to form a comprehensive distribution feature matrix to obtain the weld feature distribution matrix; The weld feature distribution matrix includes heat diffusion values, boundary feature point locations and temperature gradient data.

8. The CO2 welding automatic trolley weld identification method according to claim 7 is characterized in that: The method measures the temperature difference in the heat diffusion area of ​​the weld boundary, measures the temperature gradient in the weld boundary area point by point through the real-time temperature data of the sensing sensor, selects the heat diffusion area of ​​adjacent characteristic points in the weld boundary, calculates the temperature increment between the characteristic points, performs difference analysis on the temperature values ​​of each pair of adjacent points, and determines the heat diffusion range and temperature difference trend of the weld boundary.

9. The CO2 welding automatic trolley weld identification method according to claim 1 is characterized in that: The specific steps of generating the weld recognition model are: Based on the weld feature distribution matrix, a Gaussian mixture model is used to classify the value of each feature point, divide the groups according to the value difference between adjacent points, identify the hierarchical relationship of the feature points, and generate feature point hierarchical group data; Based on the feature point hierarchical group data, the numerical characteristics of adjacent feature points are calculated, the points in the hierarchical group are normalized, the combination data is established, and a normalized feature combination is generated; Based on the normalized feature combination, the values ​​of each normalized feature point and the spatial coordinates are integrated to form an overall model of the weld boundary and feature points, and a weld recognition model is generated; Wherein, the weld recognition model includes feature point positions, normalized parameters and model combination relationships.

10. The CO2 welding automatic trolley welding seam identification method according to claim 1 is characterized in that: The Gaussian mixture model is based on the formula: Where: x is each feature point in the weld feature distribution matrix, λ is the model parameter set, π k is the weight of the kth Gaussian component, represents the kth improved Gaussian distribution, μ k and∑ k are the mean vector and covariance matrix of the kth distribution, δ is the feature point offset, α is the covariance scaling factor, and β is the matrix translation.

Images