An online optical feedback compensation method for refining welding pose drift of steel

Abstract

The application relates to the field of refined steel welding technology and discloses an online optical feedback compensation method for refined steel welding pose drift. The method first collects optical feedback signals of a welding seam area, then calls historical time sequence data to analyze the correlation among the signals, and accordingly divides the signals into a strong correlation signal combination and a single signal. For the strong correlation signal combination, pose drift identification is carried out; for the single signal, if there is time distribution regularity, regularity features are extracted for feature matching identification, and if there is no time distribution regularity, mutation analysis identification is carried out. The identified pose drift amount is converted into a compensation control signal, which is output to a welding gun pose adjusting mechanism in real time for compensation correction, and a closed-loop feedback adjustment is carried out to ensure the compensation response time. The application realizes online real-time detection and compensation of welding pose drift through a classification processing mode, effectively improves the reliability and robustness of pose detection, and meets the requirements of high-precision welding on pose stability.

Description

Technical Field

[0001] This invention belongs to the field of mechanical manufacturing technology, and more specifically, relates to an online optical feedback compensation method for the welding posture drift of refined steel. Background Technology

[0002] With the rapid development of modern manufacturing, precision welding technology is increasingly widely used in high-end equipment manufacturing fields such as aerospace, automobile manufacturing, and rail transportation. Refined steel, as a high-quality steel, plays a crucial role in the connection of important structural components due to its excellent mechanical properties and welding characteristics. However, during the welding process of refined steel, factors such as uneven workpiece heating, fixture deformation, and welding torch vibration can cause welding posture drift, severely affecting welding quality and accuracy. Traditional posture detection methods suffer from low accuracy and poor real-time performance, failing to meet the demands of high-precision welding. Optical feedback compensation technology, as a non-contact, high-precision measurement and control method, offers a new approach to solving this problem. However, existing technologies still have many shortcomings in applying optical feedback compensation to address posture drift in refined steel welding.

[0003] The shortcomings of existing optical feedback compensation technology in the field of welding posture drift of refined steel are mainly reflected in: the lack of optical posture detection methods for metal welding process, the inability to achieve online real-time compensation of welding posture drift, and the inability of the response speed and accuracy of the compensation system to meet the requirements of high-precision welding. Summary of the Invention

[0004] This invention provides an online optical feedback compensation method for welding posture drift in refined steel. By combining strongly correlated signals and dividing single signals into groups based on the optical feedback signals during the welding process, different drift recognition methods are adopted for different types of optical signals, effectively solving the problem mentioned in the background art that welding posture drift in refined steel cannot be detected and compensated online.

[0005] To achieve the above objectives, the present invention provides an online optical feedback compensation method for welding pose drift of refined steel, comprising the following steps: Optical feedback signals in the weld area during the welding process of refined steel are collected; historical time series data of each optical feedback signal are retrieved to analyze the correlation between the signals, and the optical feedback signals are divided into strongly correlated signal combinations and single signals based on the analysis results; The sampling period is set to 5ms, and time-series data of strongly correlated signal combinations are collected within this period for pose drift identification. By using historical time series data of a single signal to analyze the temporal distribution patterns, and extracting the temporal distribution pattern characteristics when a single signal exhibits a temporal distribution pattern. Collect time-series data of a single signal within the sampling period. When the analysis shows that the time distribution of the single signal has a regularity, the pose drift of the time-series data of the signal within the sampling period is identified based on the extracted time distribution regularity features. When analyzing a single signal that does not exhibit a regular temporal distribution, abrupt changes are analyzed in the time-series data of the single signal during the sampling period, and pose drift is identified based on the analysis results. The identified pose drift is converted into a compensation control signal and output to the welding torch pose adjustment mechanism in real time for compensation and correction. Based on the pose drift recognition results, the compensation control signal is adjusted using closed-loop feedback, and the compensation response time is set to not exceed 10ms to achieve real-time stable control of the welding pose.

[0006] Optionally, the analysis of the correlation between the signals specifically includes: Each optical feedback signal is paired up to form several signal combinations; histograms are plotted using the historical time-series data of two optical feedback signals in each signal combination, and the shape contour of the histogram is extracted and overlapped with the contour of the standard normal distribution curve to calculate the overlap degree of the normal contour of the data distribution corresponding to the optical feedback signal. Mark the central vertical line from the shape outline of the histogram, and divide the central vertical line into several points evenly. Then draw horizontal lines perpendicular to the central vertical line at each division point. These horizontal lines intersect with the curves on both sides of the histogram to form two intersection points on the left and right. Mark the left and right intersection points corresponding to each division point, calculate the distance between the intersection point and the division point, and then compare the left and right distances to obtain the left and right differences corresponding to each division point, thereby calculating the data distribution symmetry corresponding to the optical feedback signal; The normal distribution conformance of each optical feedback signal is calculated by weighted averaging the overlap of the normal profile and the symmetry of the data distribution for each optical feedback signal in each signal combination. This is then compared with a set threshold. If the normal distribution conformance of an optical feedback signal reaches the set threshold, the optical feedback signal is determined to conform to a normal distribution. Based on the judgment results of whether the two optical feedback signals in each signal combination conform to the normal distribution, the appropriate correlation analysis method is selected; the historical time series data of the two optical feedback signals in each signal combination are mapped according to time to obtain a number of paired time series data points, and then the correlation analysis of the number of time series data points is performed according to the appropriate correlation analysis method to obtain the correlation coefficient corresponding to each signal combination; An N×N correlation matrix is ​​created using the correlation coefficients corresponding to each signal combination, and statistical tests are performed based on the correlation matrix to obtain the analysis results on whether each signal combination is correlated.

[0007] Optionally, the process of classifying strongly correlated signal combinations and single signals specifically includes: when the analysis finds that a certain signal combination has a correlation, the signal combination is regarded as a strongly correlated signal combination; when a certain optical feedback signal does not show a correlation in all signal combinations, the optical feedback signal is regarded as a single signal.

[0008] Optionally, the pose drift identification of the time-series data of the strongly correlated signal combination specifically includes: The correlation direction corresponding to the strongly correlated signal combination is obtained based on the correlation coefficient corresponding to the strongly correlated signal combination. By marking several points on a two-dimensional coordinate system with time as the horizontal axis and optical feedback signal data as the vertical axis in the combination of strongly correlated signals, the data change curves corresponding to the two optical feedback signals are formed. Several time points are evenly set on the horizontal axis of the data change curve, and then the tangent slope of each time point is obtained on the data change curve corresponding to the two optical feedback signals. The tangent slopes of the same time point are then grouped into a slope group. The signs of the two tangent slopes in the slope group formed at each time point are extracted and compared with the correlation direction corresponding to the strongly correlated signal combination to identify whether they conform to the correlation direction. Time points that do not conform to the correlation direction are recorded as drift time points, and then the occurrence ratio of drift time points is statistically analyzed. The proportion of occurrences of drift time points is compared with a preset critical proportion. If the proportion of occurrences of drift time points reaches the critical proportion, pose drift is identified in the strongly correlated signal combination, and the amount of pose drift is calculated.

[0009] Optionally, the implementation of the time distribution regularity analysis using historical time series data of a single signal specifically includes: The historical time series data of a single signal is marked with several points in a two-dimensional coordinate system with time as the horizontal axis and optical feedback signal data as the vertical axis to form the data change curve corresponding to the single signal. Fourier transform is used to perform spectral transformation on the data change curve corresponding to a single signal. The peaks in the spectrum are then analyzed to identify periodic components. If periodic components can be identified, it is determined that the single signal has a temporal distribution regularity; otherwise, it is determined that the single signal does not have a temporal distribution regularity.

[0010] Optionally, the specific characteristics of the refining time distribution pattern include: Identify the periodic frequency from the spectrum of a single signal, calculate the period length, and then extract a single complete periodic data curve from the data change curve corresponding to the single signal based on the period length. Inflection points are extracted from a single complete periodic data curve, thereby dividing the period length into a stable segment, an ascending segment, and a descending segment, and simultaneously obtaining the duration of the corresponding segment. The divided stable segment, ascending segment, descending segment, and the duration of the corresponding segment are then used as the characteristics of the time distribution pattern.

[0011] Optionally, the step of identifying pose drift of the signal's temporal data within the sampling period based on the extracted temporal distribution characteristics specifically includes: Extract the sampling time corresponding to each time series data in the time series data of a single signal within the sampling period, and classify them according to the time distribution pattern to form time series datasets corresponding to the stationary segment, rising segment, and falling segment; Extract the maximum and minimum data from the time series dataset corresponding to the stationary segment, and subtract the maximum and minimum data from each other and divide by the duration of the stationary segment to obtain the average change of the stationary segment. The time series dataset corresponding to the rising segment is subjected to rising rate analysis, and the time series dataset corresponding to the falling segment is subjected to falling rate analysis. The average change magnitude of the time series dataset corresponding to the stationary segment, the rise rate of the time series dataset corresponding to the rising segment, and the fall rate of the time series dataset corresponding to the falling segment are compared with the reference values. If the average change magnitude of the time series dataset corresponding to all stationary segments, the rise rate of the time series dataset corresponding to all rising segments, and the fall rate of the time series dataset corresponding to all falling segments all meet the reference values, then the single signal is identified as having no pose drift. Otherwise, the single signal is identified as having pose drift, and the amount of pose drift is calculated.

[0012] Optionally, the abrupt change analysis of the time-series data of a single signal during the sampling period specifically includes: forming adjacent data groups with the time-series data of the single signal during the sampling period and the data at the previous moment; calculating the abrupt change degree of each adjacent data group, specifically by dividing the absolute value of the difference between adjacent data groups by the result of the first data in the adjacent data group and taking the percentage, and comparing it with a reasonable abrupt change degree; if the abrupt change degree of an adjacent data group is greater than the reasonable abrupt change degree, then the adjacent data group is recorded as an abnormal data group; the proportion of abnormal data groups is counted and compared with a preset critical proportion; if the proportion of abnormal data groups reaches the critical proportion, then the single signal is identified as having pose drift and the pose drift amount is calculated.

[0013] Optionally, converting the identified pose drift into a compensation control signal specifically includes: acquiring the components of the identified pose drift in the three coordinate axes, namely the drift Δx in the X-axis direction, the drift Δy in the Y-axis direction, and the drift Δz in the Z-axis direction, and the rotational components around the three coordinate axes, namely the rotational drift around the X-axis. Y-axis rotational drift Z-axis rotational drift Substitute the above six components into the calculation formula of the compensation control signal to obtain the compensation control signal; output the calculated compensation control signal to the welding torch posture adjustment mechanism after digital-to-analog conversion.

[0014] Optionally, the closed-loop feedback adjustment of the compensation control signal based on the pose drift recognition result specifically includes: after the welding torch pose adjustment mechanism performs the compensation operation, the optical feedback signal of the compensated weld area is collected in real time through the optical feedback system, and the compensated optical feedback signal is compared with the reference signal to calculate the compensation error; the compensation error is input to the PID controller for processing, and the output signal of the PID controller is superimposed with the compensation control signal to generate the final control signal, which is output to the welding torch pose adjustment mechanism, thereby forming a closed-loop feedback adjustment to ensure that the compensation response time does not exceed 10ms.

[0015] Compared with the prior art, the present invention has at least the following technical effects: This invention performs correlation analysis on the optical feedback signals during the welding process of refined steel, dividing the signals into strongly correlated signal combinations and single signals, and identifies pose drift based on the characteristics of each. This method improves the pertinence and applicability of drift identification, not only enhancing the timeliness of identification, but also significantly reducing the possibility of false identification, thereby improving the reliability and robustness of welding pose detection.

[0016] This invention, after identifying posture drift, converts it into a compensation control signal and outputs it in real time to the welding torch posture adjustment mechanism for compensation and correction, thereby realizing online real-time compensation for welding posture drift. With the help of a fast-response closed-loop feedback adjustment mechanism, the compensation response time is ensured to be no more than 10ms, effectively meeting the requirements of high-precision welding for posture stability.

[0017] This invention distinguishes between two methods for processing a single signal: time distribution regularity analysis and abrupt change analysis. For signals with regular characteristics, feature matching is used for identification, while for signals without regular characteristics, abrupt change threshold is used for identification. This classification method improves the accuracy of pose drift identification and helps ensure the continuous stability of welding quality. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0019] Figure 1 This is a diagram illustrating the implementation steps of the method of the present invention.

[0020] Figure 2 This is a schematic diagram of the correlation analysis and segmentation process of the optical feedback signal in this invention.

[0021] Figure 3 This is a schematic diagram of the pose drift recognition and compensation control signal generation process in this invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. The described embodiments are only possible technical implementations of the present invention and not all possible implementations. Those skilled in the art can obtain other embodiments in conjunction with the embodiments of the present invention without creative effort, and these embodiments are also within the protection scope of the present invention.

[0023] See Figure 1 As shown, this invention proposes an online optical feedback compensation method for welding posture drift of refined steel, which includes the following steps: acquiring optical feedback signals of the weld area during the welding process of refined steel.

[0024] In the embodiments applied to the above scheme, the optical feedback signal of the weld area during the welding process of refined steel includes weld image signal, weld pool spectrum signal, weld temperature field distribution signal, weld geometric deformation signal, etc., wherein the weld image signal includes weld surface texture features, weld edge contour features, etc., the weld pool spectrum signal includes plasma spectrum features, weld pool radiation features, etc., the weld temperature field distribution signal includes temperature gradient distribution features, heat conduction features, etc., and the weld geometric deformation signal includes transverse contraction features, longitudinal stretching features, etc.

[0025] It is important to understand that these optical feedback signals reflect the state information of refined steel welding in different dimensions. The weld image signal directly reflects the surface morphology and geometric features of the weld, the weld pool spectrum signal directly reflects the dynamic characteristics of the weld pool during the welding process, the weld temperature field distribution signal directly reflects the temperature distribution of the heat-affected zone, and the weld geometric deformation signal directly reflects the physical deformation state of the workpiece during the welding process.

[0026] Historical time-series data of each optical feedback signal were retrieved and the correlation between the signals was analyzed. Based on the analysis results, the optical feedback signals were divided into combinations of strongly correlated signals and single signals.

[0027] It should be added that when retrieving historical time-series data of various optical feedback signals for correlation analysis, the breadth of the historical period should be ensured. This is because a broad historical period can cover more welding conditions and process conditions, providing more data samples and thus improving the reliability of statistical analysis. At the same time, the historical period should be limited to the current time. This is because newer data before the current time can better reflect the real-time status of the current refined steel welding. In addition, older data may have problems such as inconsistent formats and different sampling frequencies. Using recent data can ensure data consistency and reliability. Furthermore, limiting the historical period to the current time can reduce the workload of data processing and improve the efficiency of correlation analysis.

[0028] In the preferred implementation of the above scheme, the correlation between each optical feedback signal is analyzed as follows: each optical feedback signal is paired up to form several signal combinations.

[0029] It should be noted that when combining optical feedback signals in pairs, priority should be given to combining optical feedback signals with the same parameters. Similarly, when performing correlation analysis on these signal combinations, priority should also be given to combinations with the same parameters. This is because signal combinations with the same parameters are more likely to exhibit correlation, primarily because optical feedback signals with the same parameters usually reflect the same physical process or welding attribute. For example, surface texture features and edge contour features in weld seam image signals often influence each other during the same welding process, exhibiting high correlation. Similarly, spectral features such as plasma spectral features and molten pool radiation features in weld pool spectral signals typically show correlation during the same molten pool dynamic process. Therefore, prioritizing the analysis of signal combinations with the same parameters improves analysis efficiency. Furthermore, signal combinations with the same parameters are easier to understand and interpret. For instance, analyzing the relationship between surface texture features and edge contour features in weld seam image signals is easier to interpret than analyzing the relationship with weld pool spectral signals, because the former is a fundamental concept in image processing.

[0030] Histograms were plotted using the historical time-series data of the two optical feedback signals in each signal combination. The shape contour of the histogram was extracted and its overlap with the contour of the standard normal distribution curve was calculated to determine the overlap degree of the data distribution normal contour corresponding to the optical feedback signal. The overlap degree of the data distribution normal contour can be specifically calculated by dividing the length of the overlap contour by the length of the standard normal distribution curve contour.

[0031] It is important to know that the standard normal distribution is a special type of normal distribution, where the mean is 0 and the standard deviation is 1.

[0032] It should be added that before plotting the histogram of the historical time-series data after normalizing the two optical feedback signals, the data needs to be converted into a form with a mean of 0 and a standard deviation of 1. This processing method is called standardization, which makes the shape and outline of the plotted histogram close to the standard normal distribution, so that it can be compared with the curve outline of the standard normal distribution at the same scale, thus improving the comparability of the comparison.

[0033] See Figure 2 As shown, mark the central vertical line from the shape outline of the histogram, and divide the central vertical line into several points evenly. Then, draw a horizontal line perpendicular to the central vertical line at each division point. These horizontal lines intersect with the curves on both sides of the histogram to form two intersection points on the left and right.

[0034] Mark the left and right intersection points corresponding to each division point, and calculate the distance between the intersection point and the division point. Then, compare the left and right distances to obtain the left and right difference corresponding to each division point, and use this to calculate the data distribution symmetry of the optical feedback signal. The formula for calculating the data distribution symmetry is as follows: ; In the formula Indicates the first The left and right differences corresponding to the dividing point. Indicates the division point number, Indicates the number of dividing points. This represents the natural constant. From this formula, we can see that the smaller the left-right difference between each dividing point, the greater the symmetry of the data distribution, and the more symmetrical the data distribution.

[0035] The normal distribution conformance of each optical feedback signal is calculated by weighted averaging the overlap of the normal profile and the symmetry of the data distribution for each optical feedback signal in each signal combination. This is then compared with a set threshold. If the normal distribution conformance of an optical feedback signal reaches the set threshold, the optical feedback signal is determined to conform to a normal distribution.

[0036] It should be added that when setting the weight values ​​for the data distribution normal profile overlap and data distribution symmetry in the weighted average, it is considered that profile overlap reflects the similarity between the overall shape of the data distribution and the standard normal distribution, which is usually related to the overall distribution shape of the data. Symmetry reflects the symmetry between the left and right sides of the data distribution, which is usually related to the local details of the data. Overall correlation is greater than local correlation. In addition, profile overlap is more resistant to the influence of outliers and noise in the data. Even if there are a few outliers, the overall distribution of the data can still match the profile of the standard normal distribution, while symmetry may be affected by individual extreme values. In summary, it can be seen that profile overlap can be given a higher weight. For example, the weight values ​​for data distribution normal profile overlap and data distribution symmetry can be 0.6 and 0.4, which satisfies the need to value profile overlap without overemphasizing it.

[0037] It should be further noted that since the values ​​of the normality profile overlap and the symmetry of the data distribution are both between 0 and 1, the normality conformance value obtained after weighted averaging is also between 0 and 1. Therefore, the normality conformance threshold can be set to 0.7, and the purpose of setting the normality conformance threshold is to assist in judging whether the optical feedback signal conforms to the normal distribution.

[0038] It should be explained that the above-mentioned normality distribution assessment of each optical feedback signal in the signal combination is to screen for suitable correlation analysis methods. Commonly used correlation analysis methods include Pearson correlation coefficient and Spearman rank correlation coefficient. However, these two methods have different applicable scenarios. Pearson correlation coefficient is suitable for data that conforms to or is close to a normal distribution, while Spearman rank correlation coefficient is suitable for data that does not conform to a normal distribution. Therefore, it is necessary to assess the conformity of the optical feedback signal to a normal distribution in order to select a more suitable and targeted correlation analysis method and achieve the goal of accurate and reliable correlation analysis results. In addition, the above scheme integrates the overlap of the normal distribution contour and the symmetry of the data distribution in the normality distribution conformity assessment of the optical feedback signal, which can more comprehensively reflect the characteristics of the data distribution and thus more accurately assess whether the data conforms to a normal distribution.

[0039] The appropriate correlation analysis method is selected based on whether the two optical feedback signals in each signal combination conform to a normal distribution. Specifically, if both optical feedback signals in a signal combination conform to a normal distribution, the Pearson correlation coefficient is used as the appropriate correlation analysis method for that signal combination. If neither optical feedback signal conforms to a normal distribution, the Spearman rank correlation coefficient is used. If one optical feedback signal in a signal combination conforms to a normal distribution and the other does not, both the Pearson and Spearman rank correlation coefficients are used for correlation analysis simultaneously. In this case, two correlation analysis results are obtained. The differences between the results can be compared. If the differences are small, the average of the two correlation analysis results is taken as the final correlation analysis result. If the differences are large, the appropriate method can be selected based on actual needs. For example, if the optical feedback signal conforming to a normal distribution is more important in practical applications, the Pearson correlation coefficient can be chosen as the final result; if the optical feedback signal not conforming to a normal distribution is more important in practical applications, the Spearman rank correlation coefficient can be chosen as the final result.

[0040] The historical time-series data of the two optical feedback signals in each signal combination are mapped according to time to obtain a number of paired time-series data points. Then, the correlation analysis of the time-series data points is performed according to the adaptive correlation analysis method to obtain the correlation coefficient corresponding to each signal combination.

[0041] It is important to understand that when performing time-series mapping and pairing of the historical time-series data of two optical feedback signals in various signal combinations, it is necessary to ensure that the timestamps of the paired data are consistent. That is, pairing the data of two optical feedback signals at the same time point. However, the sampling frequencies of the two optical feedback signals may not be consistent. To achieve better pairing, the historical time-series data of the two optical feedback signals can be interpolated and filled in first. The specific operation process is as follows: find the common time range in the datasets of the two optical feedback signals and create a timestamp list containing all time points within this time range. Use the selected interpolation method to fill in the historical time-series data of the two optical feedback signals so that each timestamp has a corresponding data value. This can ensure that there is corresponding data at each time point, thereby avoiding data loss or pairing errors caused by inconsistent timestamps.

[0042] An N×NN×N correlation matrix is ​​created using the correlation coefficients corresponding to each signal combination. Statistical tests are then performed based on the correlation matrix to obtain the analysis results on whether each signal combination is correlated.

[0043] It should be explained that after obtaining the correlation coefficient using correlation analysis, the statistical test is performed to verify whether the obtained correlation coefficient is statistically significant, thereby confirming whether the association between the optical feedback signals is real and reliable.

[0044] It should be added that the statistical test mentioned above involves setting a significance level and checking whether the p-value is below that level. If the p-value is less than the significance level, the correlation is considered significant. For example, assuming a significance level of 0.05, if the p-value is less than 0.05, the correlation is considered significant. If the p-value is greater than or equal to 0.05, the correlation is considered insignificant.

[0045] In a further preferred implementation of the above scheme, the optical feedback signal is divided into strongly correlated signal combinations and single signals, as follows: when the analysis finds that a certain signal combination has a correlation, the signal combination is regarded as a strongly correlated signal combination; when a certain optical feedback signal does not show a correlation in all signal combinations, the optical feedback signal is regarded as a single signal.

[0046] In the example of the above implementation, if the signal combination (A,B) is significantly correlated, then (A,B) is identified as a strongly correlated signal combination; if the optical feedback signal A does not show significant correlation among all signal combinations, then A is identified as a single signal.

[0047] The sampling period is set to 5ms, and time-series data of strongly correlated signal combinations are collected within this period to identify pose drift.

[0048] The aforementioned setting of a sampling period aims to overcome the inherent high false alarm rate problem of pose drift recognition based on a single data point. By specifying a certain data acquisition window, it ensures that each drift detection is based on a set of data rather than isolated values, thereby improving the accuracy and reliability of the recognition. Furthermore, setting a sampling period also helps to: firstly, standardize data acquisition: ensuring that the data acquisition process is carried out at predetermined time intervals, eliminating randomness and disorder, and facilitating system management and automated operation; secondly, ensure data timeliness: regularly collecting the latest data ensures that the optical feedback system can promptly capture changes in pose; and thirdly, optimize resource utilization: rationally arranging the data acquisition frequency avoids the storage and processing burden caused by data redundancy, and prevents information loss due to insufficient data.

[0049] In the example described above, the sampling period can be set based on the sampling interval between the two optical feedback signals in the strongly correlated signal combination. Specifically, if the sampling interval of the optical feedback signals is short, the sampling period can be set to be short to ensure that sufficient data is obtained for analysis within a short time. Conversely, if the sampling interval of the optical feedback signals is long, the sampling period can be extended accordingly to ensure that an appropriate amount of data is collected in each period.

[0050] Among the ways that the above solutions can be implemented, see [link to relevant documentation]. Figure 3 As shown, the process of collecting time-series data of strongly correlated signal combinations for pose drift identification is as follows: the correlation direction corresponding to the strongly correlated signal combination is obtained based on the correlation coefficient corresponding to the strongly correlated signal combination.

[0051] It is important to know that the correlation coefficient obtained from correlation analysis ranges from -1 to 1. A negative correlation coefficient indicates that the correlation direction of a strongly correlated signal combination is negative, while a positive correlation coefficient indicates that the correlation direction of a strongly correlated signal combination is positive.

[0052] By marking several points on a two-dimensional coordinate system with time as the horizontal axis and optical feedback signal data as the vertical axis in the combination of strongly correlated signals, the data change curves corresponding to the two optical feedback signals are formed.

[0053] Several time points are evenly set on the horizontal axis of the data change curve, and then the tangent slope of each time point is obtained on the data change curve corresponding to the two optical feedback signals. The tangent slopes of the same time point are then grouped into a slope group.

[0054] The signs of the two tangent slopes in the slope group formed at each time point are extracted and compared with the correlation direction corresponding to the strongly correlated signal combination to identify whether they conform to the correlation direction. Time points that do not conform to the correlation direction are recorded as drift time points, and then the occurrence ratio of drift time points is statistically analyzed.

[0055] In the example above, when the correlation direction is positive, the signs of the slopes of the two tangents formed at a certain time point are positive and negative, respectively. At this time, it is determined that the time point does not conform to the correlation direction.

[0056] The proportion of occurrences of drift time points is compared with a preset critical proportion. For example, the critical proportion is 0.4. If the proportion of occurrences of drift time points reaches the critical proportion, pose drift is identified in the strongly correlated signal combination, and the pose drift amount is calculated.

[0057] This invention utilizes the correlation direction of signal combinations to determine pose drift. Specifically, if two strongly correlated optical feedback signals maintain the same direction (positive correlation) or opposite direction (negative correlation) for most of the time, then the appearance of opposite trends at certain time points can be considered pose drift. The advantages of this method are: first, strong targeting: using known correlation directions to detect drift provides strong targeting; second, high reliability: based on existing strong correlations, the possibility of false alarms is reduced.

[0058] By using historical time-series data of a single signal to analyze the temporal distribution patterns, the characteristics of the temporal distribution patterns can be extracted when a temporal distribution pattern is found in the analysis of a single signal.

[0059] Specifically, the analysis of the temporal distribution regularity using historical time-series data of a single signal is carried out as follows: several points are marked on a two-dimensional coordinate system with time as the horizontal axis and optical feedback signal data as the vertical axis to form the data change curve corresponding to the single signal.

[0060] Fourier transform is used to perform spectral transformation on the data change curve corresponding to a single signal. The peaks in the spectrum are then analyzed to identify periodic components. If periodic components can be identified, it is determined that the single signal has a temporal distribution regularity; otherwise, it is determined that the single signal does not have a temporal distribution regularity.

[0061] It's important to note that in the spectrum obtained through Fourier transform, each frequency component corresponds to an amplitude value. If a periodic component exists at a specific frequency in the signal, a corresponding peak will appear in the spectrum. This is because periodic components produce significant energy concentrations in the spectrum, i.e., peaks. Therefore, by analyzing the position of these peaks in the spectrum, the presence of periodic components in the signal can be determined. For example, if a peak at 1 Hz is observed in the spectrum, it means that the original signal contains a component with a period of 1 second.

[0062] Furthermore, the temporal distribution pattern is refined as follows: the periodic frequency is identified from the spectrum of a single signal, and the period length is calculated accordingly. Then, based on the period length, a single complete periodic data curve is extracted from the data change curve corresponding to the single signal.

[0063] Inflection points are extracted from a single complete periodic data curve, thereby dividing the period length into a stable segment, an ascending segment, and a descending segment, and simultaneously obtaining the duration of the corresponding segment. The divided stable segment, ascending segment, descending segment, and the duration of the corresponding segment are then used as the characteristics of the time distribution pattern.

[0064] In the above embodiment, it is assumed that the period length of a single signal is 50ms, the duration of the rising segment is 5ms-20ms, the duration of the falling segment is 20ms-35ms, and the duration of the stationary segment is 35ms-5ms.

[0065] It's important to know that inflection points are typically the locations of local maxima or minima in a data curve. Once inflection points are identified, the cycle can be divided into different segments based on these points: a plateau segment refers to the part where the curve slope is close to zero, an ascending segment refers to the part where the curve slope increases, and a descending segment refers to the part where the curve slope decreases.

[0066] The system collects time-series data of a single signal within a sampling period. When the analysis shows that the single signal has a regular time distribution pattern, the system identifies the pose drift of the time-series data of the signal within the sampling period based on the extracted time distribution pattern characteristics.

[0067] Preferably, the pose drift identification of the signal’s time series data within the sampling period is performed based on the extracted time distribution pattern characteristics, as described in the following process: extract the sampling time corresponding to each time series data in the time series data of a single signal within the sampling period, and classify them according to the time distribution pattern characteristics to form time series datasets corresponding to the stationary segment, rising segment, and falling segment.

[0068] Extract the maximum and minimum data from the time series dataset corresponding to the stationary segment, and then subtract the maximum and minimum data and divide by the duration of the stationary segment to obtain the average change of the stationary segment.

[0069] The time series dataset corresponding to the rising segment is subjected to rise rate analysis, and the time series dataset corresponding to the falling segment is subjected to fall rate analysis.

[0070] Specifically, when performing rise and fall rate analysis, the rise and fall rates can be calculated by dividing the difference between adjacent data points in the time series dataset corresponding to the rise segment by the mean of the time interval.

[0071] The average change magnitude of the time series dataset corresponding to the stationary segment, the rise rate of the time series dataset corresponding to the rising segment, and the fall rate of the time series dataset corresponding to the falling segment are compared with the reference values. If the average change magnitude of the time series dataset corresponding to all stationary segments, the rise rate of the time series dataset corresponding to all rising segments, and the fall rate of the time series dataset corresponding to all falling segments all meet the reference values, then the single signal is identified as having no pose drift. Otherwise, the single signal is identified as having pose drift, and the amount of pose drift is calculated.

[0072] It should be noted that the reference values ​​include the change amplitude reference value, the rise rate reference value, and the fall rate reference value. The change amplitude reference value can be calculated from the average change amplitude of the steady segment from the time distribution characteristics of a single signal. The rise rate reference value and the fall rate reference value are calculated from the rise rate of the rising segment and the fall rate of the falling segment from the time distribution characteristics of a single signal, respectively.

[0073] This invention analyzes the temporal distribution regularity of a single signal and extracts its characteristics when such regularity is found. This allows for more accurate identification of the behavior patterns of optical feedback signals under normal conditions. Based on these temporal distribution regularity characteristics, pose drift identification is performed on the temporal data of a single signal during the sampling period. By comparing the current signal data with known temporal distribution regularity characteristics, false alarms (false positives) caused by normal data fluctuations are minimized, thereby improving the effectiveness and reliability of the detection results.

[0074] When analyzing a single signal that does not exhibit a regular temporal distribution, abrupt change analysis is performed on the time-series data of the single signal during the sampling period, and pose drift identification is performed based on the analysis results.

[0075] In the specific implementation of the above scheme, the abrupt change analysis of the time series data of a single signal in the sampling period is performed as follows: the time series data of a single signal in the sampling period is combined with the data of the previous moment to form an adjacent data group.

[0076] The mutation rate of each adjacent data group is calculated. The specific calculation formula is to divide the absolute value of the difference between the adjacent data groups by the result of the first data in the adjacent data group and take the percentage. This percentage is then compared with the reasonable mutation rate. For example, the reasonable percentage is 20%. If the mutation rate of an adjacent data group is greater than the reasonable mutation rate, then the adjacent data group is recorded as an abnormal data group.

[0077] It should be added that the reasonable abrupt change rate mentioned above can be calculated by acquiring optical feedback signal data at adjacent time points under the condition that the welding of refined steel is proceeding stably.

[0078] The percentage of abnormal data groups is statistically analyzed and compared with a preset critical percentage. If the percentage of abnormal data groups reaches the critical percentage, a single signal is identified as having pose drift, and the amount of pose drift is calculated.

[0079] It's important to understand that the method of analyzing the abrupt changes in the values ​​of a single signal at adjacent time points to identify pose drift, when analyzing the absence of a regular temporal distribution pattern in a single signal, is based on the fact that changes in optical feedback signals under normal welding conditions are usually slow and continuous. In other words, in most cases, the data differences between adjacent time points are not significant. However, when welding pose drift occurs, sudden and drastic changes may occur. Therefore, by analyzing the differences between data at adjacent time points, abrupt changes that do not conform to normal patterns can be identified, thus determining whether pose drift has occurred. This method is particularly suitable for data without obvious periodicity or regularity.

[0080] This invention performs correlation analysis on the optical feedback signals during the welding process of refined steel, dividing the signals into strongly correlated signal combinations and single signals, and identifies pose drift based on the characteristics of each. This method improves the pertinence and applicability of drift identification, not only enhancing the timeliness of identification, but also significantly reducing the possibility of false identification, thereby improving the reliability and robustness of welding pose detection.

[0081] The identified pose drift is converted into a compensation control signal and output to the welding torch pose adjustment mechanism in real time for compensation and correction.

[0082] In the specific implementation of the above scheme, the process of converting the pose drift into a compensation control signal is as follows: The components of the identified pose drift in the three coordinate axes are obtained, namely the drift Δx in the X-axis direction, the drift Δy in the Y-axis direction, and the drift Δz in the Z-axis direction, as well as the rotational components around the three coordinate axes, namely the rotational drift around the X-axis. Y-axis rotational drift Z-axis rotational drift .

[0083] Substituting the above six components into the calculation formula for the compensation control signal yields the compensation control signal, specifically: = In the formula , , These represent the displacement compensation signals of the welding torch in the X, Y, and Z axes, respectively. , , These represent the rotation compensation signals of the welding torch around the X, Y, and Z axes, respectively. , , , , , These represent the compensation gain coefficients for each direction.

[0084] It should be added that the value of the compensation gain coefficient needs to be set according to the mechanical characteristics and control accuracy requirements of the actual welding system. For example, each compensation gain coefficient can be set to a value between 1.0 and 2.0.

[0085] The calculated compensation control signal is converted from digital to analog and then output to the welding torch posture adjustment mechanism. The welding torch posture adjustment mechanism includes a position adjustment mechanism and an attitude adjustment mechanism, wherein the position adjustment mechanism responds to the displacement compensation signal. , , The attitude adjustment mechanism responds to the rotation compensation signal to control the position of the welding torch in space. , , Control the spatial orientation of the welding torch.

[0086] Based on the pose drift recognition results, the compensation control signal is adjusted using closed-loop feedback, and the compensation response time is set to not exceed 10ms to achieve real-time stable control of the welding pose.

[0087] In the further implementation of the above scheme, the implementation process of the closed-loop feedback adjustment mechanism is as follows: after the welding torch position adjustment mechanism performs the compensation operation, the optical feedback signal of the compensated weld area is collected in real time through the optical feedback system, and the compensated optical feedback signal is compared with the reference signal to calculate the compensation error.

[0088] The compensation error is input into the PID controller for processing; the proportional coefficient of the PID controller... Integral coefficient Differential coefficients The setting is based on the dynamic characteristics of the welding system; for example, the proportional coefficient. It can be set to 0.8, the integral coefficient. It can be set to 0.2, the differential coefficient. It can be set to 0.1.

[0089] The output signal of the PID controller is superimposed with the compensation control signal to generate the final control signal, which is then output to the welding torch posture adjustment mechanism, thereby forming a closed-loop feedback adjustment to ensure that the compensation response time does not exceed 10ms.

[0090] It is important to understand that setting the compensation response time to no more than 10ms is to meet the requirements of high-precision welding for posture stability. Since posture drift during the welding process of refined steel often occurs at the millisecond level, the compensation system must have the ability to respond quickly. Through a closed-loop feedback adjustment mechanism, the compensation error can be corrected in real time to ensure that the welding posture is always kept within the allowable error range.

[0091] This invention, after identifying posture drift, converts it into a compensation control signal and outputs it in real time to the welding torch posture adjustment mechanism for compensation and correction, thereby realizing online real-time compensation for welding posture drift. With the help of a fast-response closed-loop feedback adjustment mechanism, the compensation response time is ensured to be no more than 10ms, effectively meeting the requirements of high-precision welding for posture stability.

[0092] This invention distinguishes between two methods for processing a single signal: time distribution regularity analysis and abrupt change analysis. For signals with regular characteristics, feature matching is used for identification, while for signals without regular characteristics, abrupt change threshold is used for identification. This classification method improves the accuracy of pose drift identification and helps ensure the continuous stability of welding quality.

[0093] The above are merely preferred embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. An online optical feedback compensation method for the pose drift of refined steel welding, characterized in that, Includes the following steps: Optical feedback signals in the weld area during the welding process of refined steel are collected; historical time series data of each optical feedback signal are retrieved to analyze the correlation between the signals, and the optical feedback signals are divided into strongly correlated signal combinations and single signals based on the analysis results; The sampling period is set to 5ms, and time-series data of strongly correlated signal combinations are collected within this period for pose drift identification. By using historical time series data of a single signal to analyze the temporal distribution patterns, and extracting the temporal distribution pattern characteristics when a single signal exhibits a temporal distribution pattern. Collect time-series data of a single signal within the sampling period. When the analysis shows that the time distribution of the single signal has a regularity, the pose drift of the time-series data of the signal within the sampling period is identified based on the extracted time distribution regularity features. When analyzing a single signal that does not exhibit a regular temporal distribution, abrupt changes are analyzed in the time-series data of the single signal during the sampling period, and pose drift is identified based on the analysis results. The identified pose drift is converted into a compensation control signal and output to the welding torch pose adjustment mechanism in real time for compensation and correction. Based on the pose drift recognition results, the compensation control signal is adjusted using closed-loop feedback, and the compensation response time is set to not exceed 10ms to achieve real-time stable control of the welding pose.

2. The online optical feedback compensation method for welding pose drift of refined steel as described in claim 1, characterized in that, The analysis of the correlation between the signals specifically includes: Each optical feedback signal is paired up to form several signal combinations; histograms are plotted using the historical time-series data of two optical feedback signals in each signal combination, and the shape contour of the histogram is extracted and overlapped with the contour of the standard normal distribution curve to calculate the overlap degree of the normal contour of the data distribution corresponding to the optical feedback signal. Mark the central vertical line from the shape outline of the histogram, and divide the central vertical line into several points evenly. Then draw horizontal lines perpendicular to the central vertical line at each division point. These horizontal lines intersect with the curves on both sides of the histogram to form two intersection points on the left and right. Mark the left and right intersection points corresponding to each division point, calculate the distance between the intersection point and the division point, and then compare the left and right distances to obtain the left and right differences corresponding to each division point, thereby calculating the data distribution symmetry corresponding to the optical feedback signal; The normal distribution conformance of each optical feedback signal is calculated by weighted averaging the overlap of the normal profile and the symmetry of the data distribution for each optical feedback signal in each signal combination. This is then compared with a set threshold. If the normal distribution conformance of an optical feedback signal reaches the set threshold, the optical feedback signal is determined to conform to a normal distribution. Based on the judgment results of whether the two optical feedback signals in each signal combination conform to the normal distribution, the appropriate correlation analysis method is selected; the historical time series data of the two optical feedback signals in each signal combination are mapped according to time to obtain a number of paired time series data points, and then the correlation analysis of the number of time series data points is performed according to the appropriate correlation analysis method to obtain the correlation coefficient corresponding to each signal combination; An N×N correlation matrix is ​​created using the correlation coefficients corresponding to each signal combination, and statistical tests are performed based on the correlation matrix to obtain the analysis results on whether each signal combination is correlated.

3. The online optical feedback compensation method for welding pose drift of refined steel as described in claim 2, characterized in that, The process of classifying strongly correlated signal combinations and single signals specifically includes: when the analysis finds that a certain signal combination has a correlation, the signal combination is regarded as a strongly correlated signal combination; when a certain optical feedback signal does not show a correlation in all signal combinations, the optical feedback signal is regarded as a single signal.

4. The online optical feedback compensation method for welding pose drift of refined steel as described in claim 2, characterized in that, The specific steps for pose drift identification using time-series data of strongly correlated signal combinations include: The correlation direction corresponding to the strongly correlated signal combination is obtained based on the correlation coefficient corresponding to the strongly correlated signal combination. By marking several points on a two-dimensional coordinate system with time as the horizontal axis and optical feedback signal data as the vertical axis in the combination of strongly correlated signals, the data change curves corresponding to the two optical feedback signals are formed. Several time points are evenly set on the horizontal axis of the data change curve, and then the tangent slope of each time point is obtained on the data change curve corresponding to the two optical feedback signals. The tangent slopes of the same time point are then grouped into a slope group. The signs of the two tangent slopes in the slope group formed at each time point are extracted and compared with the correlation direction corresponding to the strongly correlated signal combination to identify whether they conform to the correlation direction. Time points that do not conform to the correlation direction are recorded as drift time points, and then the occurrence ratio of drift time points is statistically analyzed. The proportion of occurrences of drift time points is compared with a preset critical proportion. If the proportion of occurrences of drift time points reaches the critical proportion, pose drift is identified in the strongly correlated signal combination, and the amount of pose drift is calculated.

5. The online optical feedback compensation method for welding posture drift of refined steel as described in claim 1, characterized in that, The specific implementation of the analysis of temporal distribution patterns using historical time-series data of a single signal includes: The historical time series data of a single signal is marked with several points in a two-dimensional coordinate system with time as the horizontal axis and optical feedback signal data as the vertical axis to form the data change curve corresponding to the single signal. Fourier transform is used to perform spectral transformation on the data change curve corresponding to a single signal. The peaks in the spectrum are then analyzed to identify periodic components. If periodic components can be identified, it is determined that the single signal has a temporal distribution regularity; otherwise, it is determined that the single signal does not have a temporal distribution regularity.

6. The online optical feedback compensation method for welding pose drift of refined steel as described in claim 5, characterized in that, The specific characteristics of the extraction time distribution pattern include: Identify the periodic frequency from the spectrum of a single signal, calculate the period length, and then extract a single complete periodic data curve from the data change curve corresponding to the single signal based on the period length. Inflection points are extracted from a single complete periodic data curve, thereby dividing the period length into a stable segment, an ascending segment, and a descending segment, and simultaneously obtaining the duration of the corresponding segment. The divided stable segment, ascending segment, descending segment, and the duration of the corresponding segment are then used as the characteristics of the time distribution pattern.

7. The online optical feedback compensation method for welding pose drift of refined steel as described in claim 6, characterized in that, The specific steps for identifying pose drift of the signal's temporal data within the sampling period based on the extracted temporal distribution characteristics include: Extract the sampling time corresponding to each time series data in the time series data of a single signal within the sampling period, and classify them according to the time distribution pattern to form time series datasets corresponding to the stationary segment, rising segment, and falling segment; Extract the maximum and minimum data from the time series dataset corresponding to the stationary segment, and subtract the maximum and minimum data from each other and divide by the duration of the stationary segment to obtain the average change of the stationary segment. The time series dataset corresponding to the rising segment is subjected to rising rate analysis, and the time series dataset corresponding to the falling segment is subjected to falling rate analysis. The average change magnitude of the time series dataset corresponding to the stationary segment, the rise rate of the time series dataset corresponding to the rising segment, and the fall rate of the time series dataset corresponding to the falling segment are compared with the reference values. If the average change magnitude of the time series dataset corresponding to all stationary segments, the rise rate of the time series dataset corresponding to all rising segments, and the fall rate of the time series dataset corresponding to all falling segments all meet the reference values, then the single signal is identified as having no pose drift. Otherwise, the single signal is identified as having pose drift, and the amount of pose drift is calculated.

8. The online optical feedback compensation method for welding pose drift of refined steel as described in claim 1, characterized in that, The abrupt change analysis of the time-series data of a single signal during the sampling period specifically includes: forming adjacent data groups by combining the time-series data of the single signal during the sampling period with the data from the previous moment; calculating the abrupt change degree of each adjacent data group, specifically by dividing the absolute value of the difference between adjacent data groups by the result of the first data in the adjacent data group and taking the percentage, and comparing it with the reasonable abrupt change degree; if the abrupt change degree of an adjacent data group is greater than the reasonable abrupt change degree, then the adjacent data group is recorded as an abnormal data group; the proportion of abnormal data groups is counted and compared with a preset critical proportion; if the proportion of abnormal data groups reaches the critical proportion, then the single signal is identified as having pose drift and the pose drift amount is calculated.

9. The online optical feedback compensation method for welding pose drift of refined steel as described in claim 1, characterized in that, The specific steps of converting the identified pose drift into a compensation control signal include: acquiring the components of the identified pose drift in the three coordinate axes, namely the drift Δx in the X-axis direction, the drift Δy in the Y-axis direction, and the drift Δz in the Z-axis direction, as well as the rotational components around the three coordinate axes, namely the rotational drift around the X-axis. Y-axis rotational drift Z-axis rotational drift Substitute the above six components into the calculation formula of the compensation control signal to obtain the compensation control signal; output the calculated compensation control signal to the welding torch posture adjustment mechanism after digital-to-analog conversion.

10. The online optical feedback compensation method for welding pose drift of refined steel as described in claim 9, characterized in that, The closed-loop feedback adjustment of the compensation control signal based on the pose drift recognition result specifically includes: after the welding torch pose adjustment mechanism performs the compensation operation, the optical feedback signal of the compensated weld area is collected in real time through the optical feedback system, and the compensated optical feedback signal is compared with the reference signal to calculate the compensation error; the compensation error is input to the PID controller for processing, and the output signal of the PID controller is superimposed with the compensation control signal to generate the final control signal, which is output to the welding torch pose adjustment mechanism, thereby forming a closed-loop feedback adjustment to ensure that the compensation response time does not exceed 10ms.

Images