A strip steel continuous production line laser welding machine weld quality detection method

Abstract

The present application belongs to the technical field of strip steel production weld detection, and provides a strip steel continuous production line laser welding machine weld quality detection method, which comprises the following steps: in the continuous production process, the trajectory position data of the weld along the width direction of the strip steel is collected in real time to form a historical trajectory sequence, the historical trajectory sequence is analyzed, and the predictability of the weld trajectory fluctuation is evaluated; according to the evaluation result of the predictability, a corresponding compensation mode is dynamically selected. When the trajectory is unpredictable, closed-loop tracking compensation is adopted, so that the sensor can follow the weld center in real time, and the complete cross section can be ensured to be collected all the time; when the trajectory is predictable, feedforward prediction compensation is adopted, the predicted offset is used to correct the profile data, and the complete cross section with the weld center as the reference is also obtained, so that the phenomenon that the complete cross section and the partial cross section appear alternately due to the weld swing is fundamentally eliminated, false positive false reports and false negative missed detections are avoided, and the reliability of the weld quality detection is improved.

Description

Technical Field

[0001] This invention belongs to the field of strip steel production weld inspection technology, specifically a method for inspecting weld quality of laser welding machines in continuous strip steel production lines. Background Technology

[0002] In continuous strip steel production lines (such as pickling lines, cold rolling lines, galvanizing lines, etc.), laser welding machines are used to weld the ends of two coils of strip steel together to form continuous strip steel to ensure uninterrupted production. The quality of the weld directly determines the stability of subsequent rolling, looper operation and surface treatment processes. However, continuous production lines are dynamic operating systems. The strip steel is always in motion during welding and subsequent transport. In actual production, due to factors such as strip steel camber, lag in the centering system response, tension fluctuations, and looper impact, the actual trajectory of the weld seam in the width direction of the strip steel is often not fixed, but exhibits low-frequency oscillation or random drift. In existing detection methods, laser profile sensors are usually fixedly installed at a certain position at the welder exit, with the measurement field of view center aligned with the theoretical weld seam centerline. When the weld seam trajectory oscillates, the cross-sectional data actually collected by the sensor may only cover a part of the weld seam (partial cross-section), or even completely deviate from the weld seam area. Only when the weld seam trajectory just passes through the center of the measurement field of view can the complete weld seam cross-section (complete cross-section) be collected. This phenomenon of alternating "complete cross-sections" and "partial cross-sections" leads to unreliable test data: on the one hand, partial cross-section data cannot truly reflect the overall morphology of the weld, and may misjudge normal welds as having insufficient reinforcement or excessive misalignment, generating false positive alarms and causing unnecessary shutdowns or scrapping; on the other hand, if real defects (such as local porosity or cracks) happen to be located in the weld area that was not collected, they may be missed, forming false negatives, allowing unqualified welds to enter subsequent rolling processes, which in severe cases can cause major production accidents such as strip breakage and damaged rolls. Therefore, the present invention provides a method for inspecting the weld quality of a laser welding machine in a continuous strip steel production line. Summary of the Invention

[0003] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0004] The technical solution adopted by this invention to solve its technical problem is: a method for inspecting the weld quality of a laser welding machine in a continuous strip steel production line, comprising: During continuous production, the trajectory position data of the weld along the width direction of the strip are collected in real time to form a historical trajectory sequence. The historical trajectory sequence is analyzed to evaluate the predictability of weld trajectory fluctuations. Based on predictable assessment results, dynamically select the corresponding compensation mode: If it is determined that it is not predictable, it switches to the first compensation mode and performs closed-loop tracking compensation on the detection data based on the real-time detected weld trajectory position. The first compensation mode includes: using a laser positioning sensor or edge detection algorithm to obtain the weld center position in real time, and dynamically aligning the measurement data through a servo mechanism or digital signal processing; If it is determined to be predictable, switch to the second compensation mode and perform feedforward compensation on the detection data based on the prediction results of the weld trajectory. The second compensation mode includes: using a Kalman filter or a periodic prediction model to predict the weld trajectory and compensating the sensor measurement reference in a feedforward manner; Under the selected compensation mode, the original weld inspection data collected at the laser welding machine exit is compensated to obtain the corrected weld cross-sectional feature data. Based on the corrected weld section feature data, determine whether the weld quality is qualified.

[0005] The beneficial effects of this invention are as follows: This invention dynamically switches compensation modes based on the predictability of weld trajectory fluctuations: when the trajectory is unpredictable, closed-loop tracking compensation is used to enable the sensor to follow the weld center in real time, ensuring that a complete cross-section is always acquired; when the trajectory is predictable, feedforward prediction compensation is used to correct the contour data using the predicted offset, thus obtaining a complete cross-section with the weld center as the reference. This fundamentally eliminates the phenomenon of alternating "complete cross-section" and "partial cross-section" caused by weld oscillation, avoids false positives and false negatives, and improves the reliability of weld quality inspection. This invention automatically matches compensation strategies of different complexities based on the regularity of trajectory fluctuations: when the trajectory exhibits predictable regular fluctuations, low-computing-power feedforward prediction compensation is run, the sensor remains stationary, and data is corrected only through software algorithms; when the trajectory exhibits unpredictable random fluctuations, high-computing-power closed-loop tracking compensation is activated, driving the servo mechanism to track in real time; when the trajectory is extremely stable, it further enters a sleep sub-mode, pausing compensation operations or using fixed parameters and reducing the sampling frequency. This mechanism of allocating computing power on demand enables the system to consume far less computing power on average than the all-time tracking scheme while ensuring detection accuracy, making it particularly suitable for edge computing devices with limited computing power. This invention effectively addresses dynamic disturbances such as strip bending, centering system response lag, tension fluctuations, and looper impact by real-time assessment of weld trajectory predictability and automatic adaptation to compensation modes. The tracking limit protection mechanism in the first compensation mode can promptly identify severe deviations and issue alarms, preventing major accidents such as strip breakage and roll damage caused by missed weld inspections. At the same time, the reduction in false positive alarms significantly reduces unnecessary downtime and scrapping, ensuring the continuous and stable operation of the production line. Attached Figure Description

[0006] The invention will now be further described with reference to the accompanying drawings.

[0007] Figure 1 This is a flowchart of the steps of a laser welding machine weld quality inspection method for a continuous strip steel production line according to the present invention; Figure 2 This is an architecture diagram of a laser welding machine weld quality inspection system for a continuous strip steel production line according to the present invention. Detailed Implementation

[0008] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0009] Example 1

[0010] One of the core inventive points of this invention is: in response to the shortcomings of the current technology, a method for inspecting the weld quality of a laser welding machine in a continuous strip steel production line is proposed. This method analyzes the predictability of weld trajectory fluctuations in real time and dynamically switches between feedforward prediction compensation and closed-loop tracking compensation modes based on the predictability assessment results. This adaptively optimizes the system's computing power consumption while ensuring the integrity of weld cross-section data acquisition and the reliability of detection. Please see Figure 1 As shown in the embodiment of the present invention, a method for inspecting the weld quality of a laser welding machine in a continuous strip steel production line includes the following steps: Step S1: During continuous production, the trajectory position data of the weld along the width direction of the strip is collected in real time to form a historical trajectory sequence. The historical trajectory sequence is analyzed to evaluate the predictability of weld trajectory fluctuations. The specific process of collecting trajectory location data and forming historical trajectory sequences is as follows: During the continuous production of strip steel, at least one laser profile sensor is installed behind the exit of the laser welding machine. The measurement line of the sensor is perpendicular to the running direction of the strip steel and arranged along the width direction of the strip steel. When the weld seam moves with the strip steel and passes under the sensor, the sensor continuously collects three-dimensional coordinate data of a series of discrete points on the surface of the weld seam along the width direction at a fixed sampling frequency. At each sampling moment, a contour line containing lateral position coordinates and height coordinates is obtained. The lateral coordinates of the weld feature points (such as the weld center or the highest point) in each contour line are extracted and used as the instantaneous trajectory position of the weld in the strip width direction at that moment. As production proceeds, the trajectory positions of multiple consecutive moments are stored in chronological order to form a historical trajectory sequence. The specific process for assessing the predictability of weld trajectory fluctuations based on historical trajectory sequences is as follows: The trajectory position values ​​at adjacent moments in the historical trajectory sequence are differentially calculated to obtain the sequence of changes in trajectory position. By analyzing the distribution characteristics of statistical changes, we can determine whether a trajectory exhibits periodicity or a trend. Specifically: Determine the central tendency of the change: Calculate the absolute median difference of the change series; First, the median of the change series is calculated. Then, the absolute deviation of each change from the median is calculated. The median of these absolute deviations is taken as the absolute median difference. If the absolute median difference is less than the preset fluctuation tolerance parameter (0.2 mm, set based on industry standards), the change is determined to be concentrated near zero. Otherwise, it is determined to be dispersed. Determine the directionality of the trajectory deviation: Count the number of positive, negative and zero numbers in the sequence of changes. If the absolute value of the difference between the number of positive and negative numbers divided by the total number of numbers in the sequence (after removing zero values) is less than the preset balance ratio threshold (0.3), it is determined to have reciprocating oscillation characteristics. If the number of positive numbers exceeds a preset multiple (more than one time) of the number of negative numbers, or if the number of negative numbers exceeds a preset multiple (more than one time) of the number of positive numbers, it is determined to have a unidirectional trend. Otherwise, it is judged as random fluctuation; Determine the continuity of the trajectory pattern: Calculate the maximum length of consecutive identical signs in the sequence, and calculate the proportion of this maximum length to the total length of the sequence, as the maximum length proportion; Simultaneously calculate the frequency of sign changes in the sequence, which is the total number of changes (the number of times that two adjacent changes have different signs) divided by the sequence length minus one, as the sign change frequency. When the central tendency is determined to be concentrated, the offset direction is determined to be oscillating or unidirectional, the maximum length proportion in the regularity does not exceed the preset proportion threshold (one-tenth), and the sign change frequency does not exceed the preset frequency upper limit threshold (two-thirds), the output is predictable. Otherwise, the output is not predictable; It should be noted that the threshold parameters such as "0.3", "one time", "one-tenth", and "two-thirds" in the above judgment are optimized values ​​obtained through statistical analysis and ROC curve optimization based on historical data of weld seam trajectories from multiple continuous strip steel production lines. In practical applications, those skilled in the art can make adaptive adjustments to the above thresholds according to factors such as production line speed, steel grade and specifications, and centering system characteristics.

[0011] Step S2: Based on the predictable evaluation results, dynamically select the corresponding compensation mode: Step S201: If it is determined that it is not predictable, switch to the first compensation mode and perform closed-loop tracking compensation on the detection data based on the real-time detected weld trajectory position. The first compensation mode includes: using a laser positioning sensor or edge detection algorithm to obtain the weld center position in real time, and dynamically aligning the measurement data through a servo mechanism or digital signal processing; Behind the laser welding machine outlet, on the same mounting beam as the original laser contour sensor, a laser positioning sensor is installed. The measurement line of this sensor is also perpendicular to the strip running direction and arranged along the width direction of the strip. Its installation position makes the measurement point of the sensor near the theoretical position of the weld center. The laser positioning sensor emits a laser beam in real time at a sampling frequency of not less than 1 kilohertz, irradiating the weld area on the surface of the moving strip, and captures the reflected light through a built-in high-speed camera to calculate the position coordinates of the weld center in the width direction of the strip in real time. The weld center position coordinates output in real time by the laser positioning sensor are used as the feedback signal for closed-loop control. They are compared with the preset sensor measurement reference position (i.e. the ideal measurement line position of the laser profile sensor) and the position deviation value between the two is calculated to reflect the current actual trajectory of the weld relative to the fixed measurement line. The servo mechanism is driven based on the position deviation value. Specifically: The servo mechanism includes a high-precision linear module, and a laser profile sensor is fixedly mounted on the moving platform of the module. When the position deviation exceeds the preset tracking dead zone (e.g., 0.1 mm), the servo controller outputs a control signal to drive the linear module to move the laser profile sensor along the width of the strip, so that the sensor's measurement line follows the center position of the actual weld trajectory in real time. The moving speed is proportional to the deviation value. The larger the deviation, the faster the moving speed, thereby achieving rapid tracking of the weld trajectory. The position deviation value is used as the input of digital signal processing. During the movement of the sensor, the actual position offset of the sensor at each sampling moment is recorded, and the offset is superimposed on the lateral coordinate of the original contour data collected by the laser contour sensor. That is, the lateral position of each sampling point is dynamically aligned and corrected. The corrected contour data takes the center of the weld as the origin of the coordinate, thereby ensuring that even if the sensor is moving, the cross-sectional data collected is still an accurate shape based on the center of the weld. It is also equipped with a tracking limit protection mechanism. When the position deviation value continuously exceeds the maximum movable range of the sensor (i.e., exceeds 90% of the linear module stroke) and the duration exceeds the preset time (one second), it is determined that the weld seam trajectory is seriously deviated and cannot be restored to normal detection through tracking compensation. At this time, an alarm signal is output to prompt the operator to check the strip deviation or the alignment system status. Through the aforementioned laser positioning real-time detection, servo mechanism dynamic tracking, digital signal processing alignment correction, and limit protection mechanism, the first compensation mode can always ensure that the measurement line of the laser contour sensor is aligned with the center of the weld when the weld trajectory is unpredictable, thereby obtaining complete and reliable weld cross-sectional data.

[0012] Step S202: If it is determined to be predictable, switch to the second compensation mode and perform feedforward compensation on the detection data based on the prediction results of the weld trajectory. The second compensation mode includes: using a Kalman filter or a periodic prediction model to predict the weld trajectory and compensating the sensor measurement reference in a feedforward manner; Extract trajectory fluctuation characteristic parameters from historical trajectory sequences, including the period, amplitude, and phase of the fluctuations; Based on the predictability determination result (reciprocating oscillation characteristics or unidirectional trend), the appropriate prediction model is automatically selected; If the trajectory fluctuations exhibit reciprocating oscillation characteristics, a periodic prediction model is adopted. This model fits a standard oscillation curve based on the trajectory data within the most recent complete oscillation cycle and extrapolates the curve to future times. Specifically, the time history of the weld trajectory deviating from the center position to the maximum offset and then returning to the center position in the previous oscillation cycle is recorded and used as a template. For the current and several future moments, the predicted offset is obtained by looking up a table in the template based on the phase of the current moment in the oscillation cycle. Since the oscillation has a predictable periodicity, the error between the predicted offset and the actual trajectory is controlled within a small range. If the trajectory fluctuations exhibit a unidirectional trend (such as slow drift), a Kalman filter is used for prediction. The Kalman filter establishes a motion model (such as a constant velocity model or a constant acceleration model) describing the changes in trajectory position based on historical trajectory sequences. The filter performs two steps at each sampling time: first, it uses the motion model to predict the trajectory position at the current time based on the trajectory position estimate at the previous time; then, it uses the actual trajectory position observation value collected at the current time to correct the predicted value and obtain the optimal estimate. Through iterative recursion, the Kalman filter can filter out random noise and output a smooth and forward-looking trajectory position prediction value. The predicted value of the next time moment output by the filter is used as the basis for compensation. Secondly, the predicted weld trajectory offset is applied to the raw detection data collected by the laser profile sensor in a feedforward manner. Specifically, the laser profile sensor is fixedly installed on the crossbeam behind the welding machine exit, and its measurement line position remains unchanged. When the sensor collects a profile line, the predicted offset corresponding to that moment is acquired synchronously. This offset is superimposed on the lateral coordinate of each sampling point in the profile line, so that the origin of the profile line coordinates is translated from the fixed sensor measurement reference to the predicted weld center position. This operation is completed in the data post-processing stage without moving the sensor body. Furthermore, the predicted offset is used to pre-compensate the data acquisition at several subsequent sampling times. Since the prediction model can provide the trajectory position in advance for a period of time, the region of interest for data acquisition can be adjusted in advance to ensure that even if the weld trajectory deviates within the prediction range, the acquired contour line can still cover the complete weld section. It should be noted that in the second compensation mode, since the sensor body remains fixed and the servo mechanism does not move, the system's computing power consumption is mainly concentrated on the recursive calculation of the prediction model, which is much lower than the computing power required for servo drive and data alignment in the first compensation mode. Step S203: As a preferred implementation of the second compensation mode, when the fluctuation amplitude of the weld trajectory is less than the preset threshold and shows a stable pattern, the sleep sub-mode is entered to suspend compensation or use fixed compensation parameters to further reduce computing power consumption. During the operation of the second compensation mode, the following two conditions will be continuously monitored: Condition 1: The amplitude of weld trajectory fluctuation is obtained by calculating the difference between the maximum and minimum values ​​of the trajectory position in the historical trajectory sequence; If the fluctuation amplitude is less than the preset amplitude threshold (e.g., 0.1 mm), it indicates that the weld trajectory is basically stable near the sensor measurement reference, and the oscillation or drift is extremely weak. Condition 2: The stability of the trajectory fluctuation pattern. Check whether the predictability conclusion output in step S1 remains "predictable" in multiple consecutive evaluation windows (e.g., ten consecutive historical trajectory sequence windows), and whether the rate of change of characteristic parameters such as the period and amplitude of the trajectory is less than the preset rate of change threshold (e.g., the period change of adjacent windows does not exceed 5%). This indicates that the trajectory fluctuation is not only predictable, but its pattern is consistent over a long period of time and there is no risk of sudden change. When both of the above conditions are met simultaneously, the current operating condition is determined to be in a "super-steady state," requiring no compensation or only extremely low-frequency compensation to meet the detection requirements. At this time, the system enters a sleep sub-mode and performs the following operations: Pause compensation calculation: Stop running the recursive calculation of the Kalman filter or periodic prediction model, no longer output the prediction offset, and the original contour data collected by the laser contour sensor is directly used for subsequent weld quality judgment without lateral coordinate correction. Use a fixed compensation parameter: Alternatively, retain the offset obtained from the last valid prediction as a fixed compensation parameter, apply the same fixed offset correction to all subsequently acquired contour lines, and update this fixed parameter once every long time interval (e.g., every minute) by a low-priority background task. During the update, the prediction model is briefly woken up to perform a calculation, and then goes back to sleep. Reduce sampling frequency: In the sleep sub-mode, the sampling frequency of the laser profile sensor can be reduced from the normal mode to the minimum frequency required to maintain basic detection (e.g., from 1 kHz to 100 Hz), further reducing the computing power overhead of data acquisition and transmission. Wake-up mechanism: Simultaneously set up a wake-up monitoring task to quickly detect at a very low frequency (e.g., once per second) whether the fluctuation amplitude of the current trajectory exceeds the preset amplitude threshold again, or whether the trajectory pattern has changed significantly (e.g., by comparing whether the deviation between the current trajectory position and the fixed compensation parameter exceeds 0.2 mm). Once an anomaly is detected, the system immediately exits the hibernation sub-mode and restores to the complete second compensation mode (predictive compensation) or switches to the first compensation mode based on the predictability assessment results. Through the aforementioned dormant sub-mode, under extremely stable weld seam trajectory conditions, the computational power consumption related to compensation can be reduced to near zero, while still maintaining a rapid response capability to changes in operating conditions, thereby achieving ultimate optimization of computational resources while ensuring detection reliability.

[0013] Step S3: Under the selected compensation mode, the original weld detection data collected at the laser welding machine exit is compensated to obtain the corrected weld cross-sectional feature data. Based on the currently selected compensation mode, obtain the corresponding compensation parameters: In the first compensation mode, the compensation parameter is the actual position offset of the servo mechanism at each sampling time, which is fed back by the laser positioning sensor and recorded by the servo driver; In the second compensation mode, the compensation parameter is the current-time prediction offset output by the Kalman filter or the periodic prediction model. In the dormant sub-mode, the compensation parameter is a fixed offset (the last valid prediction value) or zero; Each original contour line output by the laser contour sensor consists of a series of discrete points. Each point contains a lateral coordinate (relative to the sensor's own measurement coordinate system) and a height coordinate. The compensation parameter (offset) is superimposed onto the lateral coordinates of all points along the contour line to obtain the corrected lateral coordinates, i.e.: Corrected horizontal coordinate = original horizontal coordinate + compensation offset; For the first compensation mode, since the sensor itself has moved with the servo mechanism, the original lateral coordinates have been partially compensated, but the servo position offset still needs to be superimposed to convert it into an absolute coordinate system. For the second compensation mode and the sleep mode, the predicted or fixed offset is directly superimposed. After correction, the origin of the lateral coordinates of all contour lines is unified to the weld center (or a known fixed reference), thereby eliminating the influence of weld trajectory oscillation on the data acquisition position. Since the compensation offset may have slight errors or noise at the edge of the sensor's field of view, the corrected contour line is trimmed: only data points within a preset effective width (e.g., half the maximum width of the weld plus two millimeters) on both sides are retained based on the weld center, and the excess part is discarded. Then, the trimmed contour line is linearly interpolated and resampled at a fixed lateral interval (e.g., 0.1 millimeters) so that all contour lines have the same number of sampling points and lateral coordinate sequence, which facilitates the batch processing of subsequent feature extraction. For each corrected and resampled contour line, calculate the following weld section geometry features: Weld reinforcement height: Within the weld area (identified by the range of transverse coordinates), the difference between the maximum value of the contour height coordinate and the average height of the base material reference plane on both sides. The base material reference plane is obtained by fitting the contour points of a flat area on each side of the weld. Weld width: The lateral distance between two boundary points in the outline where the height coordinate first exceeds a set threshold downwards (recessed) or upwards (excess height) from the base material reference plane; Misalignment: The absolute value of the difference in average height between the base metal area on the left and right sides of the weld center; Weld depression: The difference between the minimum value of the height coordinate of the contour line in the weld area and the average height of the base material reference plane (a negative value indicates depression). The above feature data are organized into a time series according to the sampling time sequence, and each record is accompanied by the following labels: Current compensation mode (first, second, or dormant sub-mode); Timestamp and weld location information (synchronized by the production line encoder).

[0014] Step S4: Based on the corrected weld section feature data, determine whether the weld quality is qualified; First, based on the current steel grade, thickness and subsequent process requirements of the strip, a set of benchmark thresholds are read from the database, including: upper limit of excess height, upper limit of misalignment, allowable deviation range of weld width, and upper limit of indentation depth. These benchmark thresholds are loaded once when the production line switches steel grades and specifications and remain unchanged during the judgment process. The extracted weld reinforcement, misalignment, weld width, and depression depth are compared with their corresponding benchmark thresholds: If the weld reinforcement height does not exceed the upper limit and is not lower than zero (or the preset lower limit), the reinforcement height item is qualified; otherwise, it is unqualified. If the absolute value of the misalignment amount does not exceed the upper limit, the misalignment item is qualified; otherwise, it is unqualified. If the weld width is within the allowable deviation range, the width item is qualified; otherwise, it is unqualified. If the absolute value of the dent depth does not exceed the upper limit, the dent item is qualified; otherwise, it is unqualified. When all four criteria are met, the weld quality is deemed acceptable; if any one criterion is not met, the weld quality is deemed unacceptable. The operator interface displays the judgment results and the first non-conformity (such as "misalignment"). The judgment results are stored in the quality database for traceability.

[0015] This embodiment achieves high reliability, low computing power consumption, and high robustness in laser welding machine weld quality inspection on continuous strip steel production lines through the technical path of "sensing the predictability of weld trajectory - adaptive switching of compensation mode - on-demand allocation of computing power". It effectively solves the contradiction between unreliable inspection data and limited computing power resources in dynamic continuous production.

[0016] Example 2

[0017] Based on the same inventive concept as the laser welding machine weld quality inspection method for a continuous strip steel production line described in the foregoing embodiments, such as... Figure 2 As shown, this application provides a weld quality inspection system for laser welding machines in a continuous strip steel production line, wherein the system specifically includes: Weld trajectory analysis module: During continuous production, the trajectory position data of the weld along the width direction of the strip is collected in real time to form a historical trajectory sequence. The historical trajectory sequence is analyzed to evaluate the predictability of weld trajectory fluctuations. Compensation mode adaptation module: Dynamically selects the corresponding compensation mode based on predictable evaluation results. First compensation unit: If it is determined that it is not predictable, it switches to the first compensation mode and performs closed-loop tracking compensation on the detection data based on the real-time detected weld trajectory position. The first compensation mode includes: using a laser positioning sensor or edge detection algorithm to obtain the weld center position in real time, and dynamically aligning the measurement data through a servo mechanism or digital signal processing; Second compensation unit: If it is determined to be predictable, switch to the second compensation mode and perform feedforward compensation on the detection data based on the prediction results of the weld trajectory; The second compensation mode includes: using a Kalman filter or a periodic prediction model to predict the weld trajectory and compensating the sensor measurement reference in a feedforward manner; Second compensation subunit: As a preferred implementation of the second compensation mode, when the fluctuation amplitude of the weld trajectory is less than the preset threshold and shows a stable pattern, it enters the dormant submode, suspends compensation or adopts fixed compensation parameters to further reduce computing power consumption. Correction data acquisition module: Under the selected compensation mode, the original weld detection data collected at the laser welding machine exit is compensated to obtain the corrected weld cross-sectional feature data; Weld quality assessment module: Based on the corrected weld cross-sectional feature data, determine whether the weld quality is qualified.

[0018] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for inspecting the weld quality of a laser welding machine in a continuous strip steel production line, characterized in that: include: During continuous production, the trajectory position data of the weld along the width direction of the strip is collected in real time to form a historical trajectory sequence. The historical trajectory sequence is analyzed to evaluate the predictability of weld trajectory fluctuations. The process of assessing the predictability of weld trajectory fluctuations includes: The change sequence is obtained by calculating the difference between the trajectory position values ​​of adjacent moments in the historical trajectory sequence; Determine the central tendency of the change: Calculate the absolute median of the change series. If the absolute median is less than the preset fluctuation tolerance parameter, the change is determined to be concentrated near the zero value; otherwise, it is determined to be dispersed. Determine the directionality of trajectory deviation: count the number of positive, negative and zero numbers in the sequence of changes, and determine whether it is a reciprocating oscillation characteristic, a unidirectional trend or random fluctuation based on the ratio of positive and negative numbers; Determine the regularity of the trajectory: calculate the maximum length ratio of consecutive identical signs and the frequency of sign changes in the sequence of changes; When the concentration trend is determined to be concentration, the offset direction is determined to be reciprocating oscillation or unidirectional trend, and the maximum length ratio does not exceed the preset ratio threshold and the sign change frequency does not exceed the preset frequency upper limit threshold, the output is predictable; otherwise, the output is not predictable. Based on predictable assessment results, dynamically select the corresponding compensation mode: If it is determined that it is not predictable, it switches to the first compensation mode and performs closed-loop tracking compensation on the detection data based on the real-time detected weld trajectory position. The first compensation mode includes: using a laser positioning sensor or edge detection algorithm to obtain the weld center position in real time, and dynamically aligning the measurement data through a servo mechanism or digital signal processing; If it is determined to be predictable, switch to the second compensation mode and perform feedforward compensation on the detection data based on the prediction results of the weld trajectory. The second compensation mode includes: using a Kalman filter or a periodic prediction model to predict the weld trajectory and compensating the sensor measurement reference in a feedforward manner; Under the selected compensation mode, the original weld inspection data collected at the laser welding machine exit is compensated to obtain the corrected weld cross-sectional feature data. Based on the corrected weld section feature data, determine whether the weld quality is qualified.

2. The method for inspecting weld quality of a laser welding machine in a continuous strip steel production line according to claim 1, characterized in that: In the second compensation mode, when the fluctuation amplitude of the weld trajectory is less than the preset threshold and shows a stable pattern, the system enters the sleep sub-mode, suspends compensation or uses fixed compensation parameters to further reduce computing power consumption.

3. The method for inspecting weld quality of a laser welding machine in a continuous strip steel production line according to claim 1, characterized in that: The deviation directionality of the determined trajectory is specifically as follows: If the absolute value of the difference between the number of positive numbers and the number of negative numbers, divided by the total number of numbers after removing zero values ​​from the sequence, is less than the preset balance ratio threshold, it is determined to have reciprocating oscillation characteristics. If the number of positive numbers exceeds a preset multiple of the number of negative numbers, or if the number of negative numbers exceeds a preset multiple of the number of positive numbers, it is determined to have a unidirectional trend. Otherwise, it is judged as random fluctuation.

4. The method for inspecting weld quality of a laser welding machine in a continuous strip steel production line according to claim 1, characterized in that: The first compensation model specifically includes: Install a laser positioning sensor to obtain the position coordinates of the weld center in the strip width direction in real time; The coordinates of the weld center position are compared with the preset sensor measurement reference position to obtain the position deviation value; The servo mechanism is driven by the position deviation value, so that the laser contour sensor moves along the width of the strip and tracks the center of the weld line in real time. The actual position offset of the sensor is superimposed on the horizontal coordinate of the original contour data to achieve dynamic alignment correction.

5. The method for inspecting weld quality of a laser welding machine in a continuous strip steel production line according to claim 1, characterized in that: The second compensation model specifically includes: Extract trajectory fluctuation feature parameters from historical trajectory sequences; The prediction model is selected based on the predictability determination result: if it is a reciprocating oscillation characteristic, a periodic prediction model is adopted, and the predicted offset is obtained by looking up a table based on the oscillation period template; if it is a unidirectional trend, a Kalman filter is used for prediction. The predicted offset is superimposed onto the lateral coordinates of the original contour data acquired by the laser contour sensor in a feedforward manner, so that the origin of the contour line is translated to the predicted weld center position.

6. The method for inspecting weld quality of a laser welding machine in a continuous strip steel production line according to claim 2, characterized in that: The hibernation sub-mode includes at least one of the following operations: Pause the recursive calculation of the prediction model and do not correct the lateral coordinates of the original contour data; The last valid predicted offset is retained as a fixed compensation parameter, and the same fixed offset correction is applied to all subsequent contour lines. Reduce the sampling frequency of the laser profile sensor; Configure a wake-up monitoring task to exit the hibernation sub-mode when the trajectory fluctuation amplitude is detected to exceed the preset threshold again or the trajectory pattern changes.

7. The method for inspecting weld quality of a laser welding machine in a continuous strip steel production line according to claim 1, characterized in that: The process of compensating the original weld inspection data to obtain the corrected weld section feature data includes: Obtain the corresponding compensation parameters based on the current compensation mode; The compensation parameters are superimposed onto the lateral coordinates of each sampling point in the original contour line to obtain the corrected lateral coordinates. The corrected outline is trimmed, retaining data points within a preset effective width range based on the weld center; The cropped outline is resampled at fixed horizontal intervals; Extract at least one geometric feature from the resampled contour line, including weld reinforcement, weld width, misalignment, and weld depression.

8. The method for inspecting weld quality of a laser welding machine in a continuous strip steel production line according to claim 1, characterized in that: The process of determining whether a weld is of acceptable quality includes: Based on the strip steel type, thickness, and subsequent process requirements, read the preset benchmark threshold. The extracted weld section feature data is compared with the corresponding benchmark threshold. If all features meet the requirements, the weld is deemed qualified; otherwise, it is deemed unqualified.

9. The method for inspecting the weld quality of a laser welding machine in a continuous strip steel production line according to claim 1, characterized in that: The first compensation mode is equipped with a tracking limit protection mechanism. When the position deviation value continuously exceeds the maximum movable range of the laser profile sensor and the duration exceeds a preset time, an alarm signal is output.

Images