Laser processing galvanometer splicing precision closed loop control method and system

Abstract

This invention discloses a closed-loop control method and system for the stitching accuracy of laser processing galvanometers, relating to the field of galvanometer control technology. The method includes acquiring theoretical parameters of the galvanometer system and establishing a theoretical global coordinate system; extracting the local trajectories of the galvanometers in the stitching region of the processed image and performing preliminary spatiotemporal alignment; constructing a set of distance constraints in the stitching region; constructing an overdetermined set of constraint equations to solve for the relative geometric errors of adjacent galvanometers; recursively decomposing the error time series; establishing an error model and predicting future short-term errors; generating feedforward compensation to correct control commands in real time; continuously updating the trajectory, constraints, and model; and introducing a forgetting mechanism to reduce the influence of historical data. This invention achieves high-precision online closed-loop dynamic control of stitching errors, eliminates dependence on complex offline calibration, and enhances the system's accuracy against time-varying disturbances through adaptive weights and predictive feedforward, comprehensively improving the accuracy consistency and stability of large-area laser processing.

Description

Technical Field

[0001] This invention relates to the field of galvanometer control technology, specifically to a closed-loop control method and system for the splicing accuracy of laser processing galvanometers. Background Technology

[0002] As the modern precision manufacturing industry continues to develop towards high speed, high efficiency, and high precision, laser processing, with its advantages of high precision, non-contact operation, and wide material adaptability, has become a core processing method in fields such as electronics, new energy, and aerospace. With laser processing exhibiting a significant trend towards larger sizes, greater precision, and mass production, the limited scanning field of view of a single galvanometer in traditional laser processing technology can no longer meet processing demands. Galvanometer splicing technology has become the key to overcoming these limitations.

[0003] In existing technologies, offline calibration methods are often used to address galvanometer splicing errors. While these methods can reduce system errors to some extent, they rely on the assumption that system parameters remain constant during processing, which has several drawbacks. First, offline calibration cannot compensate for time-varying drift errors caused by factors such as temperature changes and mechanical stress relaxation during processing. Second, offline static calibration is completely ineffective for time-varying motion errors caused by actuator response and vibration in high-speed dynamic machining. Finally, the calibration process is complex and time-consuming, requiring production interruptions and affecting processing speed. Therefore, existing technologies struggle to achieve truly online and real-time accuracy maintenance.

[0004] Furthermore, some technologies attempt to introduce online monitoring, but these methods are mostly limited to post-processing detection or simple feedback correction of the processing results, failing to deeply couple the observation data with the galvanometer control. They also typically lack a strict spatiotemporal synchronization mechanism, failing to consider the end-to-end system latency from image acquisition and data processing to control command activation, resulting in delayed compensation actions and an inability to effectively suppress high-speed processing errors. In addition, existing methods often remain at the level of trajectory following within the field of view of a single galvanometer, lacking relevant processing of observation data from adjacent galvanometers within the stitched area, thus failing to guarantee the consistency of accuracy in the stitched area.

[0005] In summary, current technologies in the field of multi-mirror laser processing have not yet provided a closed-loop control method capable of real-time, online measurement of time-varying errors during processing and precise feedforward compensation based on the dynamic characteristics of these errors. This has become a key bottleneck restricting the development of large-area, high-precision laser processing technology. Summary of the Invention

[0006] To address the aforementioned technical problems, this paper provides a closed-loop control method and system for the splicing accuracy of laser processing galvanometers. This technical solution solves the problem mentioned in the background art of lacking a closed-loop control method that can obtain time-varying errors in real time and online during the processing, and perform accurate feedforward compensation based on the dynamic characteristics of the errors.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A closed-loop control method for the splicing accuracy of laser processing galvanometers includes: The theoretical parameters of the galvanometer system are obtained, and a theoretical global coordinate system is established. The theoretical parameters of the galvanometer system include theoretical geometric parameters, theoretical pose parameters, and response delay parameters. Acquire control commands for the galvanometer in the splicing area, image data of the processing area, and corresponding timestamps during the processing; Based on the control commands of the theoretical global coordinate system and the splicing region galvanometer, the coordinates of the theoretical processing point are generated; Extract the local trajectory formed by galvanometer processing in the stitching region of the image data, wherein the local trajectory includes the geometric features of the processing trajectory; The time offset between the image acquisition time and the effective time of the galvanometer control command is obtained based on the timestamp and response delay parameters, and the theoretical processing point and local trajectory are initially aligned in time. The observed trajectories of local trajectories within the time window are extracted, and corresponding weights are obtained based on geometric consistency and observation stability, followed by preliminary spatial alignment. In the galvanometer stitching area, based on the observation trajectory and theoretical processing trajectory corresponding to the same time window of adjacent galvanometers, the distance constraint of galvanometer stitching is calculated and assigned a corresponding weight to form a distance constraint set. The theoretical processing trajectory is the trajectory formed by theoretical processing points. The distance constraint set includes the distance constraint between the theoretical processing trajectory and the observation trajectory, and the distance constraint between the observation trajectories of adjacent galvanometers. Based on the distance constraint set at different spatial locations, an overdetermined constraint equation set is constructed to solve for the relative geometric error between adjacent galvanometers; Construct an error model and recursively decompose the relative geometric error; Based on the error model and system response delay, the short-term error is predicted and a feedforward compensation amount is generated to apply to the galvanometer control command, thereby correcting the subsequent processing path in real time. During the processing, the local trajectory, distance constraint set, and error model are continuously updated in real time, and a forgetting mechanism is introduced to reduce the impact of historical data on the current prediction.

[0008] Preferably, the observed trajectories of the local trajectories within the extraction time window are weighted based on geometric consistency and observation stability, and preliminary spatial alignment is performed, specifically including: Set a time window; Acquire the local trajectory of the galvanometer in the stitching region within the time window; Extract the observed trajectory from the local trajectory; The weights corresponding to the observation trajectories are obtained based on the degree of geometric consistency and the degree of observation stability. Based on the time offset between the galvanometer control command and the image acquisition time and the effective time of the galvanometer control command, the coordinates of the observation trajectory and the corresponding theoretical processing point are matched, and the correspondence between the image coordinate system where the observation trajectory is located and the theoretical global coordinate system where the theoretical processing point is located is established. The image coordinate system is the original pixel coordinate system inherent to the industrial camera imaging sensor. Based on the correspondence between the image coordinate system where the observation trajectory is located and the theoretical global coordinate system where the theoretical processing point coordinates are located, the spatial transformation matrix is ​​calculated to transform the observation trajectory into the theoretical global coordinate system.

[0009] Preferably, in the galvanometer stitching area, based on the observation trajectory and theoretical processing trajectory corresponding to adjacent galvanometers within the same time window, the distance constraint for galvanometer stitching is calculated and assigned corresponding weights to form a distance constraint set, specifically including: Within a unified time window, the observation trajectory of the first galvanometer in the stitching area is obtained, and the observation trajectory of the second galvanometer in the stitching area is obtained. The first galvanometer is one of two adjacent galvanometers in the stitching area, and the second galvanometer is the other of two adjacent galvanometers in the stitching area. Within the same time window, match the theoretical processing trajectory formed by the theoretical processing points that correspond to the observed trajectory; The distance between the observation trajectory of the first galvanometer and the theoretical processing trajectory is used as the first distance constraint; The distance between the observation trajectory of the second galvanometer and the theoretical processing trajectory is used as the second distance constraint; The distance between the observation trajectories of the first and second galvanometers is used as the third distance constraint; The distance constraints are weighted according to the corresponding weights of the observed trajectories; The three distance constraints and their corresponding weights are integrated into the distance constraint set.

[0010] Preferably, the construction of an overdetermined set of constraint equations based on distance constraints at different spatial locations, and the solution of the relative geometric error between adjacent galvanometers, specifically includes: The relative geometric error to be determined is introduced into the distance constraint set and transformed into a system of relative geometric error constraint equations; An overdetermined constraint equation system is constructed based on the relative geometric error constraint equation system at different spatial locations; Solve the overdetermined system of equations to obtain the relative geometric error between adjacent galvanometers.

[0011] Preferably, the construction of the error model and the recursive decomposition of the relative geometric error specifically includes: Obtain the online time series of relative geometric error; Slow-changing error models and fast-changing error models are constructed and recursively decomposed based on time series, and the corresponding error components are separated according to the rate of change. Recursive state updates are performed at different frequencies based on the error model and the rate of change of the error components.

[0012] Preferably, the step of predicting short-term errors based on the error model and system response delay, and generating feedforward compensation for the galvanometer control command to correct subsequent processing paths in real time, specifically includes: Obtain the current error value of the error model; Determine the prediction time domain, which covers the duration from the image acquisition time to the effective time of the galvanometer control command; Based on the error model, the current error value is extrapolated to future times to obtain the predicted error value in the prediction time domain; Based on the error prediction value, a feedforward compensation amount matching the time step of the galvanometer control command is generated to correct the galvanometer control command. The corrected galvanometer control command is sent to the galvanometer drive unit to adjust the spatial position of the subsequent machining path in real time.

[0013] Furthermore, a closed-loop control system for the splicing accuracy of laser processing galvanometers is proposed to realize the closed-loop control method for the splicing accuracy of laser processing galvanometers as described above, including: The information module is used to acquire theoretical parameters of the galvanometer system, establish a theoretical global coordinate system, and acquire control commands for the galvanometer in the splicing area, image data of the processing area, and corresponding timestamps during the processing. The constraint module is used to generate the coordinates of the theoretical processing point based on the theoretical global coordinate system and the control commands of the galvanometer in the stitching area, extract the local trajectory formed by the galvanometer processing in the stitching area of ​​the image data, obtain the time offset between the image acquisition time and the effective time of the galvanometer control command based on the timestamp and response delay parameters, perform preliminary time alignment on the theoretical processing point and the local trajectory, extract the observation trajectory of the local trajectory within the time window, obtain the corresponding weights based on geometric consistency and observation stability, and perform preliminary spatial alignment. In the galvanometer stitching area, based on the corresponding observation trajectory and theoretical processing trajectory within the same time window of adjacent galvanometers, the distance constraint of the galvanometer stitching is calculated and assigned corresponding weights to form a distance constraint set. The error module is used to construct an overdetermined set of constraint equations based on distance constraint sets at different spatial locations, solve the relative geometric error between adjacent galvanometers, construct an error model and recursively decompose the relative geometric error, predict the short-term error based on the error model and system response delay, generate a feedforward compensation amount to act on the galvanometer control command, and correct the subsequent processing path in real time. An optimization module is used to continuously update the local trajectory, distance constraint set, and error model in real time during the processing, and to introduce a forgetting mechanism to reduce the impact of historical data on the current prediction.

[0014] Optionally, the constraint module specifically includes: The image unit is used to generate the coordinates of the theoretical processing point and extract the local trajectory formed by the processing of the splicing region galvanometer in the image data based on the theoretical global coordinate system and the control command of the splicing region galvanometer. The alignment unit is used to obtain the time offset between the image acquisition time and the effective time of the galvanometer control command based on the timestamp and response delay parameters, perform preliminary time alignment on the theoretical processing point and local trajectory, extract the observation trajectory of the local trajectory within the time window, obtain the corresponding weights based on geometric consistency and observation stability, and perform preliminary spatial alignment. The distance unit is used to calculate the distance constraints of the galvanometer splicing in the galvanometer splicing area based on the observation trajectory and theoretical processing trajectory corresponding to the adjacent galvanometers within the same time window and assign corresponding weights to form a distance constraint set.

[0015] Optionally, the error module specifically includes: Equation unit, which is used to construct an overdetermined set of constraint equations based on distance constraint sets at different spatial locations, and solve for the relative geometric error between adjacent galvanometers; A modeling unit, which is used to construct an error model and recursively decompose the relative geometric error; The correction unit is used to predict the short-term error based on the error model and system response delay, and generate a feedforward compensation amount to apply to the galvanometer control command to correct the subsequent processing path in real time.

[0016] Optionally, the optimization module specifically includes: The data unit is used to continuously update the local trajectory, distance constraint set, and error model in real time during the processing. The forgetting unit is used to introduce a forgetting mechanism to reduce the impact of historical data on current predictions.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention proposes a closed-loop control method and system for the splicing accuracy of laser processing galvanometers. Through a progressive optimization strategy involving initial spatiotemporal alignment and precision distance constraints, it eliminates the reliance on complex offline calibration and precision assembly. By constructing an online observation and overdetermined constraint optimization model, it achieves high-precision real-time error calculation for galvanometer splicing errors. By performing temporal recursive decomposition and prediction of errors, it improves the dynamic modeling capability, change response speed, and prediction compensation accuracy of errors. By introducing trajectory weight evaluation and a data forgetting mechanism, it enhances the system's adaptive accuracy to time-varying disturbances such as processing noise and thermal drift. Through closed-loop control, it achieves consistency and stability of splicing accuracy during large-area, long-term processing, significantly improving product quality and processing reliability. Attached Figure Description

[0018] Figure 1 This is a flowchart of the closed-loop control method for laser processing galvanometer splicing accuracy proposed in this invention; Figure 2 This is a flowchart of the observation trajectory, corresponding weight acquisition, and preliminary spatial alignment process in this invention; Figure 3 This is a flowchart of the distance constraint set acquisition process in this invention; Figure 4 This is a block diagram of the closed-loop control system for laser processing galvanometer splicing accuracy proposed in this invention. Detailed Implementation

[0019] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0020] Reference Figure 1 - Figure 3 As shown, the closed-loop control method for laser processing galvanometer splicing accuracy in this embodiment of the invention includes: The theoretical parameters of the galvanometer system are obtained, and a theoretical global coordinate system is established. The theoretical parameters of the galvanometer system include theoretical geometric parameters, theoretical pose parameters, and response delay parameters. Specifically, the theoretical parameters of the galvanometer system are obtained by reading the factory configuration file of the galvanometer controller, and the theoretical global coordinate system is established by the theoretical geometric parameters and theoretical pose parameters of the galvanometer system.

[0021] Acquire control commands for the galvanometer in the splicing area, image data of the processing area, and corresponding timestamps during the processing; Specifically, acquiring the control commands of the galvanometer in the splicing area, the image data of the processing area, and the corresponding timestamps during the processing includes: reading the galvanometer control commands for each control cycle from the galvanometer controller in real time during the processing, with each control command carrying a timestamp generated based on a unified time reference; controlling the high-speed camera to acquire image data of the processing area, with each image frame carrying a timestamp generated based on a unified time reference.

[0022] Based on the control commands of the theoretical global coordinate system and the splicing region galvanometer, the coordinates of the theoretical processing point are generated; Extract the local trajectory formed by galvanometer processing in the stitching region of the image data, wherein the local trajectory includes the geometric features of the processing trajectory; Specifically, extracting the local trajectory formed by galvanometer processing in the stitched area of ​​image data includes: performing noise reduction and grayscale enhancement preprocessing on the acquired image data to improve the recognizability of the trajectory boundaries; identifying the preprocessed image to obtain candidate trajectory regions, which include bright spots, bright bands, or feature lines belonging to the laser processing area; extracting the grayscale information of the region corresponding to the candidate trajectory region from the preprocessed grayscale image, and extracting the subpixel-level trajectory center line based on the gradient centroid method; performing connectivity analysis and time sorting on the subpixel-level trajectory center line of the candidate trajectory region to obtain the local trajectory, which includes the local pixel coordinate sequence of trajectory points and lines.

[0023] The time offset between the image acquisition time and the effective time of the galvanometer control command is obtained based on the timestamp and response delay parameters, and the theoretical processing point and local trajectory are initially aligned in time. Specifically, the time offset between the image acquisition time and the effective time of the galvanometer control command is obtained based on the timestamp and response delay parameter. The preliminary time alignment of the theoretical processing points and local trajectories includes: calculating the time difference between image acquisition and galvanometer control command based on the timestamp of the image frame and the timestamp of the control command, and adding the response delay parameter to obtain the time offset between image acquisition and the effective time of the control command; performing linear interpolation on the sequence of theoretical processing points, and resampling the theoretical processing points to the time step aligned with the image frame. The extracted local trajectories are associated with the corresponding image frame timestamps and mapped onto the same time axis, thus completing the time alignment between the theoretical processing points and the local trajectories.

[0024] It should be noted that in this scheme, the high-precision response delay parameters provided by the factory configuration file are used to perform preliminary time alignment between the theoretical processing point and the local trajectory. Under the premise of ensuring basic alignment accuracy, the image acquisition time and the control command activation time are unified to the same time reference, so that the observed local trajectory and the theoretical processing trajectory can be compared on the same time axis, which greatly eliminates time misalignment and improves the accuracy and stability of subsequent geometric consistency judgment, distance constraint construction and relative geometric error solution. The residual micro-time misalignment effect is transformed into a systematic deviation in the spatial domain in the subsequent distance constraint construction, and is adaptively eliminated as a geometric error component to be compensated by distance constraints between adjacent galvanometer observation trajectories, weighted least squares solution and recursive error modeling. Therefore, the final stitching accuracy does not depend on the absolute accuracy of time alignment.

[0025] The observed trajectories of local trajectories within the time window are extracted, and corresponding weights are obtained based on geometric consistency and observation stability, followed by preliminary spatial alignment. Specifically, the observed trajectories of local trajectories within the time window are extracted, corresponding weights are obtained based on geometric consistency and observation stability, and preliminary spatial alignment is performed, including: Set a time window; Acquire the local trajectory of the galvanometer in the stitching region within the time window; Extract the observed trajectory from the local trajectory; The time window is set using a sliding time window method to ensure that the window length covers the time interval from image acquisition to the effective date of the control command, while avoiding crossing the time range where the trajectory geometric features change significantly. During processing, its adaptive adjustment value is linearly selected between upper and lower limits based on the percentage change in trajectory geometric consistency. When the trajectory geometric consistency decreases, the time window length is reduced; when the trajectory geometric consistency tends to stabilize, the time window length is gradually increased to improve the synchronicity, continuity, and responsiveness to changes in the extraction of the observed trajectory.

[0026] Furthermore, the observed trajectory refers to the portion of the local trajectory that satisfies the conditions of continuity, smoothness, and directional consistency. The specific judgment conditions are as follows: Continuity judgment: Calculate the Euclidean distance between adjacent points of the observed trajectory and adjacent points of the corresponding theoretical processing point. If the difference is less than 0.5mm, the continuity is judged to be satisfied; otherwise, it is used as noise filtering. The threshold is a balance value selected based on the conventional laser galvanometer spot diameter and positioning accuracy. In specific implementation, this continuity tolerance can be adaptively fine-tuned within the range of 0.1mm to 1.0mm according to the resolution index of the galvanometer. Smoothness judgment: Calculate the curvature of the observed trajectory and the curvature of the corresponding theoretical trajectory. If the curvature deviation between the two is less than 50 rad / m, it is judged to meet the smoothness requirement. Otherwise, it is used as noise filtering. The threshold is set based on the maximum allowable jerk of the galvanometer motor and the laser processing focal depth tolerance. In specific implementation, it can be selected in the range of 30 rad / m to 100 rad / m according to the allowable curvature change degree of the processing path. Directional consistency judgment: Calculate the direction angle of the observed trajectory segment and the direction angle of the corresponding theoretical trajectory segment. If the deviation of the two direction angles is less than 15°, it is determined that the direction consistency is satisfied; otherwise, it is used as noise filtering. The threshold is set based on the lower limit of the image sensor angle quantization noise and the upper limit of the maximum smooth turning angle of the galvanometer without starting, stopping and decelerating. In specific implementation, it can be selected in the range of 5° to 30° according to the allowable turning angle of the galvanometer jump speed. In this scheme, the three conditions together constitute a multi-level observation trajectory verification mechanism, which can effectively distinguish between the gradual deviation caused by normal processing errors and the severe distortion noise caused by image mis-extraction. The moderately loose tolerance design ensures that the real trajectory data is not mistakenly filtered out, thus providing a sufficient and reliable input basis for the subsequent construction of weighted distance constraints and the solution of overdetermined equations.

[0027] The weights corresponding to the observation trajectories are obtained based on the degree of geometric consistency and the degree of observation stability. Among them, the weights corresponding to the observation trajectories, based on the degree of geometric consistency and the degree of observation stability, include: The similarity between the observed trajectory and the theoretical processed trajectory is calculated to obtain the geometric consistency calculation weight, using the following formula: ; In the formula, Calculate weights for geometric consistency. This is the difference between the Euclidean distance between adjacent points on the observed trajectory and the corresponding Euclidean distance between adjacent points in the theoretical processing, used to measure the shape deviation between the observed trajectory and the theoretical trajectory. The normalized standard deviation for geometric consistency is calculated from all values ​​within the current sliding time window. The median absolute deviation is obtained based on the statistical properties of the normal distribution and is used to normalize the shape deviation. The confidence score of the extracted trajectory points is calculated to obtain the observation stability calculation weight, using the following formula: ; In the formula, Calculate weights for observation stability. The confidence score for trajectory edge extraction is the normalized value of the image edge gradient, used to measure the confidence of trajectory point extraction. This represents the positional deviation between the current observation point and the observation point at the same location at the previous moment, used to measure the degree of fluctuation in the observation. To observe the normalized standard deviation of stability, it is calculated from all values ​​within the current sliding time window. The median absolute deviation is obtained by transforming the statistical properties of the normal distribution and is used to normalize the fluctuations; the two are multiplied to obtain the final weight.

[0028] In this scheme, the observation trajectories with different reliability levels are distinguished by calculating the corresponding weights of the observation trajectories. Trajectories with high geometric consistency and good observation stability are given greater weight in the subsequent distance constraint construction and error solution process. This effectively suppresses the influence of noisy trajectories and abnormal observations on the stitching accuracy control and improves the stability and accuracy of the galvanometer stitching area.

[0029] Based on the time offset between the galvanometer control command and the image acquisition time and the effective time of the galvanometer control command, the coordinates of the observation trajectory and the corresponding theoretical processing point are matched, and the correspondence between the image coordinate system where the observation trajectory is located and the theoretical global coordinate system where the theoretical processing point is located is established. The image coordinate system is the original pixel coordinate system inherent to the industrial camera imaging sensor. Based on the correspondence between the image coordinate system where the observation trajectory is located and the theoretical global coordinate system where the theoretical processing point coordinates are located, the spatial transformation matrix is ​​calculated to transform the observation trajectory into the theoretical global coordinate system; It should be noted that, due to the process of galvanometer processing, within the same processing layer Z The axis platform is in a static locked state, meaning the machining points of the adjacent galvanometers in the splicing area are... Z The coordinates are theoretically fixed and consistent. Therefore, the process of calculating the spatial transformation matrix to transform the observation trajectory to the theoretical global coordinate system involves solving the mapping relationship from the two-dimensional observation plane to the two-dimensional theoretical processing plane, rather than three-dimensional reconstruction. The transformed coordinates... Z Axis height and theoretical coordinates Z The shaft heights are consistent; in addition, due to equipment vibration during processing, Z Factors such as minute deformation or thermal drift of the shaft platform lead to actual Z Minor fluctuations in the axis coordinates will cause offsets in the two-dimensional machining trajectory, which will be uniformly incorporated into the relative geometric error model for solution and compensation, without the need for specific adjustments. Z An additional error model was established for the axis, further enhancing the system's adaptability to complex operating conditions and disturbances.

[0030] In this scheme, the spatial alignment here is preliminary spatial alignment. Its core function is to provide a global initial value for subsequent algorithms, significantly reducing the deployment threshold and initialization time of the system, while ensuring a certain level of initial accuracy and converging the pose deviation of the galvanometer to a small range. The residual deviation is incorporated into the overdetermined equations for joint solution when constructing the distance constraint set, and is transformed into part of the relative geometric error parameters. It is adaptively compensated by weighted least squares and recursive error modeling, so that the final stitching accuracy does not depend on the absolute accuracy of the spatial alignment. Through the two-step design of preliminary online alignment and online accuracy optimization, both real-time computation efficiency and closed-loop constraint solution are used to ensure the final accuracy.

[0031] In the galvanometer stitching area, based on the observation trajectory and theoretical processing trajectory corresponding to the same time window of adjacent galvanometers, the distance constraint of galvanometer stitching is calculated and assigned a corresponding weight to form a distance constraint set. The theoretical processing trajectory is the trajectory formed by theoretical processing points. The distance constraint set includes the distance constraint between the theoretical processing trajectory and the observation trajectory, and the distance constraint between the observation trajectories of adjacent galvanometers. Specifically, in the galvanometer stitching area, based on the observation trajectory and theoretical processing trajectory corresponding to adjacent galvanometers within the same time window, the distance constraints for galvanometer stitching are calculated and assigned corresponding weights, forming a set of distance constraints including: Within a unified time window, the observation trajectory of the first galvanometer in the stitching area is obtained, and the observation trajectory of the second galvanometer in the stitching area is obtained. The first galvanometer is one of two adjacent galvanometers in the stitching area, and the second galvanometer is the other of two adjacent galvanometers in the stitching area. Within the same time window, match the theoretical processing trajectory formed by the theoretical processing points that correspond to the observed trajectory; The distance between the observation trajectory of the first galvanometer and the theoretical processing trajectory is used as the first distance constraint; The distance between the observation trajectory of the second galvanometer and the theoretical processing trajectory is used as the second distance constraint; The distance between the observation trajectories of the first and second galvanometers is used as the third distance constraint; In this process, after the aforementioned steps, the connection order of the observed trajectory and the theoretical processing trajectory are consistent and the local directional change trend is consistent. They are consistent in terms of topological structure and local geometric features. The distance constraint calculation prioritizes the distance between the nearest corresponding point pairs. For areas where a strict point pair correspondence cannot be established, the minimum matching distance within the constraint search range is used. The minimum matching distance is the minimum geometric distance between the nearest point pairs between two trajectories within the search range defined by the initial spatiotemporal alignment conditions.

[0032] The distance constraints are weighted according to the corresponding weights of the observed trajectories; The three distance constraints and their corresponding weights are integrated into the distance constraint set.

[0033] The weighting method for the first and second distance constraints is based on the weights corresponding to the observation trajectories. The weighting method for the third distance constraint is based on the square root of the product of the weights of adjacent galvanometer observation trajectories. This weight is only used in the subsequent weighted least squares method to solve the overdetermined constraint equations.

[0034] In this scheme, by constructing three types of complementary constraints—single-mirror observation, theoretical trajectory, and dual-mirror observation trajectory—and combining them with a trajectory weighting mechanism and trajectory distance metric, the scheme not only ensures the consistency of accuracy between the single-mirror region and the stitching region, but also significantly improves the noise resistance accuracy of the constraint system. At the same time, it provides sufficient constraints for solving overdetermined errors, greatly reduces the dependence on the accuracy of initial spatial alignment, reduces equipment debugging costs, and can update constraints in real time during the processing to adapt to real-time pose drift during processing, ensuring stable stitching accuracy over long-term processing.

[0035] Based on the distance constraint set at different spatial locations, an overdetermined constraint equation set is constructed to solve for the relative geometric error between adjacent galvanometers; Specifically, based on the distance constraint sets at different spatial locations, an overdetermined set of constraint equations is constructed to solve for the relative geometric errors between adjacent galvanometers, including: The relative geometric error to be determined is introduced into the distance constraint set and transformed into a system of relative geometric error constraint equations; In this process, the rotational drift, translational drift, and optional scaling of the relative geometric error parameters of adjacent galvanometers to be solved are introduced into the expressions of each distance constraint. Each distance term is rewritten as an error equation with the error parameters to be solved as independent variables, so that the values ​​of the error parameters satisfy the distance constraints, thus obtaining a set of relative geometric error constraint equations.

[0036] In this embodiment, an example of a distance error equation is as follows: ; In the formula, For the first The distance between the nearest points For rotational drift, for Axis translation drift, for Axis translation drift, This is the scaling factor. It is a two-dimensional planar rotation matrix. For the first A two-dimensional column vector of observation points It is a two-dimensional translational drift. For the first If the two-dimensional column vector of a theoretical point is subject to the third distance constraint, it is changed to the two-dimensional column vector of the corresponding adjacent galvanometer observation points.

[0037] An overdetermined constraint equation system is constructed based on the relative geometric error constraint equation system at different spatial locations; Among them, since the number of relative geometric error constraint equations formed by multiple different spatial locations within the time window is much greater than the number of parameters to be determined, an overdetermined constraint equation system is formed. Solve the overdetermined system of equations to obtain the relative geometric error between adjacent galvanometers; Among them, the overdetermined system of equations is solved using the weighted least squares method to obtain the relative geometric error that minimizes the sum of squares of the total residuals; In this scheme, redundant distance constraints are introduced by sampling multiple spatial locations and an overdetermined set of equations is constructed to reduce the sensitivity of the system to single observation errors. Furthermore, the constraint equations with different confidence levels are adaptively weighted based on the weighted least squares method, which effectively improves the accuracy of the relative geometric error between adjacent galvanometers. At the same time, this method can suppress the influence of abnormal observations and local mismatches on the solution results, providing a stable and reliable parameter basis for subsequent error modeling and feedforward compensation.

[0038] Construct an error model and recursively decompose the relative geometric error; Specifically, the error model is constructed and the relative geometric error is recursively decomposed, including: Obtain the online time series of relative geometric error; Slow-changing error models and fast-changing error models are constructed and recursively decomposed based on time series, and the corresponding error components are separated according to the rate of change. The process of constructing slow-changing and fast-changing error models and performing recursive decomposition based on time series data, separating the corresponding error components according to the rate of change, includes: The relative geometric error obtained from the solution is added to the online error time series; It should be noted that the relative geometric error parameters rotational drift, translational drift, and scaling differ in their physical sources and mechanisms of action in actual laser processing. Their origins, time constants, and affected objects are completely different, fundamentally determining that their dynamic characteristics are statistically approximately independent. Specifically: rotational drift mainly originates from changes in the galvanometer mounting angle and thermally induced deformation; its rate of change is constrained by thermal inertia, affecting the trajectory direction. Translational drift mainly originates from structural thermal expansion and micro-displacement of the platform, but its amplitude-frequency characteristics are concentrated in the low-frequency range, affecting the overall trajectory position. Scaling mainly originates from changes in the focal length of the optical system or scanning ratio errors; within the single-layer processing timescale, it can be considered quasi-static, affecting the trajectory scale. Therefore, the correlation between the error components is weak, and they can be approximately considered decoupled.

[0039] Based on the aforementioned decoupling characteristics, this scheme constructs an independent state-space model and Kalman filter for each error component. Under this approximation, the introduced model mismatch error is far smaller than the order of magnitude of current visual observation noise, and its impact on the final stitching accuracy is negligible. Simultaneously, the independent filtering architecture avoids the inversion operation of the high-dimensional covariance matrix in multivariate joint estimation, enabling the entire closed-loop compensation process to be completed in real-time within the control cycle of the galvanometer controller, thus meeting the stringent constraints of computational delay imposed by online processing.

[0040] In this embodiment, any one of the error components of the relative geometric error is modeled as: ; In the formula, For discrete time steps, corresponding to sampling times under a unified time base. for The relative geometric error at time, for The slowly varying error component over time, for The rapidly changing error component at any given time; Define the error state vector: ; In the formula, for The complete state vector of the error at time step [time]. for The slowly varying error component at time, for The rate of change of the slowly varying error at time for The rapidly changing error component at time, for The rapidly changing error of the previous time step; Based on the characteristics of the two components, match the corresponding recursive model; Slowly varying errors correspond to environmental temperature changes and thermal drift, and the changes are gradual. Therefore, a constant velocity recursive model is constructed, and its state transition satisfies: ; In the formula, for The slowly varying error component at time, for The slowly varying error component of the previous time step. for The rate of change of the slowly varying error component at the previous time step. The time step is determined by the control cycle. for Slowly varying noise at any given time; Rapidly changing errors correspond to mechanical vibrations, instantaneous disturbances, and dynamic changes. Therefore, a second-order autoregressive recursive model is constructed, whose state transition satisfies: ; In the formula, for The rapidly changing error component at time, , These are autoregressive coefficients, obtained by parameter estimation and fitting of rapidly changing error time series collected during actual processing. for The rapidly varying error component of the time preceding time. for The rapidly varying error components of the two time points prior to time. for Noise in rapidly changing time processes; The error state transition equation is then: ; In the formula, Here is the state transition matrix. for Time-matter noise; The state transition matrix , ; In the formula, the upper left corner The state transition corresponding to the slowly varying error is shown in the lower right corner. The state transition corresponding to rapidly changing errors; Process noise: ; In the formula, for Time-based process noise, The process noise covariance matrix is ​​used to characterize the dynamic uncertainty of the error. It is initially set by reading and analyzing the statistical characteristics of the system's historical error data, and then adjusted online based on the filter residuals during subsequent operation. The observed values ​​originate from the relative geometric errors obtained by solving the overdetermined constraint equations in the preceding steps. Therefore, the observation model is: ; In the formula, for The observed value at time, For the observation matrix, for Observation noise at any given moment; The observation matrix : ; In the formula, the observation matrix It is a 1×4 row vector used to extract observations from the state vector, i.e., geometric relative error, which is the sum of slow-varying error and fast-varying error; Observation noise: ; In the formula, for Observation noise at any given moment To observe the noise covariance and characterize the measurement error, a preliminary setting is made by reading and analyzing the statistical characteristics of the system's historical error data, and then adjusted online based on the filter residuals during subsequent operation. Based on the Kalman filter, the time series is recursively decomposed to separate the slowly changing error component and the rapidly changing error component according to the rate of error change.

[0041] Recursive state updates with different frequencies are performed based on the error model and the rate of change of error components. Based on the error model and system response delay, the short-term error is predicted and a feedforward compensation amount is generated to apply to the galvanometer control command, thereby correcting the subsequent processing path in real time. Specifically, based on the error model and system response delay, short-term errors are predicted and feedforward compensation is generated to apply to the galvanometer control commands, thereby correcting subsequent processing paths in real time, including: Obtain the current error value of the error model; Determine the prediction time domain; The lower limit of the prediction time domain is the duration between the image acquisition time and the effective time of the galvanometer control command, ensuring that the error model can carry out the error evolution process and effectively predict and compensate for future short-term errors. The upper limit of the prediction time domain is the smaller value between the time window length and the time length of significant change of the rapidly changing error component, ensuring that the prediction is based on effective observation data. Its adaptive adjustment value is linearly selected between the difference between the upper and lower limits according to the percentage change rate of the rapidly changing error component. When the error change rate increases and the observation consistency decreases, the prediction time domain is shortened to reduce the accuracy loss of long-term prediction. When the system tends to be stable and the error change is gradual, the prediction time domain is increased to improve the stability of compensation.

[0042] Based on the error state transition, the current error value is recursively extrapolated to future times to obtain the error prediction value in the prediction time domain; Specifically, according to the error model, the process of extrapolating the current error value to future times to obtain the predicted error value in the prediction time domain includes: obtaining the error state vector at the current time based on the error model, wherein the error state vector includes error components and their rate of change; and updating the error state vector recursively according to the state transition relationship of the error model, gradually extrapolating it to future times in the prediction time domain to obtain the error state at each prediction time.

[0043] Based on the error prediction value, a feedforward compensation amount matching the time step of the galvanometer control command is generated to correct the galvanometer control command. The corrected galvanometer control command is sent to the galvanometer drive unit to adjust the spatial position of the subsequent machining path in real time.

[0044] In this scheme, by constructing an error prediction model that includes system response delay and by performing advance correction on the galvanometer control command based on feedforward compensation, the system can compensate before the error actually occurs, effectively reducing the deviation caused by system time delay and improving the accuracy of the processing path and the continuity and stability of the galvanometer splicing area.

[0045] During the processing, the local trajectory, distance constraint set, and error model are continuously updated in real time, and a forgetting mechanism is introduced to reduce the impact of historical data on the current prediction. Specifically, during the processing, the local trajectory, distance constraint set, and error model are continuously updated in real time, and a forgetting mechanism is introduced to reduce the impact of historical data on the current prediction, including: A time decay factor is introduced for the weight of each distance constraint in the distance constraint set, and its weight is reduced according to the time interval between historical data and the current time. When updating the error model, historical errors are weighted based on time decay weights to achieve dynamic updates of model parameters; The time decay weight is determined by the time decay factor, and the formula is: ; In the formula, For time decay weight, The time decay factor, This represents the time interval between the current moment and the data. Among them, time decay factor The value in the distance constraint is determined based on the geometrically effective time scale of the distance constraint, and its value is the reciprocal of the time window length. Time decay factor The initial value in the error model is the reciprocal of a preset time memory constant. The preset time memory constant value is set based on the thermodynamic time constant and the dominant frequency period of the mechanical vibration of the laser processing system. In this embodiment, the time memory constant of the slowly varying error component is configured to a value within the range of 50s to 200s, preferably 100s, to enhance the historical accumulation smoothing capability for low-frequency drift errors. The time memory constant of the rapidly varying error component is configured to a value within the range of 0.1s to 1s, preferably 0.5s, to improve the real-time tracking response capability for instantaneous distortion errors. Time decay factor... During processing, the system dynamically adjusts based on the rate of change of real-time error. When an increase in the rate of error change is detected, the system improves its performance. To enhance the weight of current observations, the weight is reduced as the rate of error change decreases. To enhance the historical smoothing weight.

[0046] In this scheme, a forgetting mechanism is introduced to perform time decay processing on historical data, making model updates more focused on the observation information at the current moment. This effectively avoids the lagging effect of historical data on the current error and improves the real-time performance and accuracy of error prediction. At the same time, by adaptively adjusting the forgetting factor, the system stability is ensured, the responsiveness to dynamic changes is enhanced, and the overall performance and accuracy of the closed-loop control of galvanometer splicing accuracy are improved.

[0047] Reference Figure 4 As shown, further, combining the above-mentioned closed-loop control method for laser processing galvanometer splicing accuracy, a closed-loop control system for laser processing galvanometer splicing accuracy is proposed, including: The information module is used to acquire theoretical parameters of the galvanometer system, establish a theoretical global coordinate system, and acquire control commands for the galvanometer in the splicing area, image data of the processing area, and corresponding timestamps during the processing. The constraint module is used to generate the coordinates of the theoretical processing point based on the theoretical global coordinate system and the control commands of the galvanometer in the stitching area, extract the local trajectory formed by the galvanometer processing in the stitching area of ​​the image data, obtain the time offset between the image acquisition time and the effective time of the galvanometer control command based on the timestamp and response delay parameters, perform preliminary time alignment on the theoretical processing point and the local trajectory, extract the observation trajectory of the local trajectory within the time window, obtain the corresponding weights based on geometric consistency and observation stability, and perform preliminary spatial alignment. In the galvanometer stitching area, based on the corresponding observation trajectory and theoretical processing trajectory within the same time window of adjacent galvanometers, the distance constraint of the galvanometer stitching is calculated and assigned corresponding weights to form a distance constraint set. The error module is used to construct an overdetermined set of constraint equations based on distance constraint sets at different spatial locations, solve the relative geometric error between adjacent galvanometers, construct an error model and recursively decompose the relative geometric error, predict the short-term error based on the error model and system response delay, generate a feedforward compensation amount to act on the galvanometer control command, and correct the subsequent processing path in real time. An optimization module is used to continuously update the local trajectory, distance constraint set, and error model in real time during the processing, and to introduce a forgetting mechanism to reduce the impact of historical data on the current prediction.

[0048] The constraint module includes: The image unit is used to generate the coordinates of the theoretical processing point and extract the local trajectory formed by the processing of the splicing region galvanometer in the image data based on the theoretical global coordinate system and the control command of the splicing region galvanometer. The alignment unit is used to obtain the time offset between the image acquisition time and the effective time of the galvanometer control command based on the timestamp and response delay parameters, perform preliminary time alignment on the theoretical processing point and local trajectory, extract the observation trajectory of the local trajectory within the time window, obtain the corresponding weights based on geometric consistency and observation stability, and perform preliminary spatial alignment. The distance unit is used to calculate the distance constraints of the galvanometer splicing in the galvanometer splicing area based on the observation trajectory and theoretical processing trajectory corresponding to the adjacent galvanometers within the same time window and assign corresponding weights to form a distance constraint set.

[0049] The error module includes: Equation unit, which is used to construct an overdetermined set of constraint equations based on distance constraint sets at different spatial locations, and solve for the relative geometric error between adjacent galvanometers; A modeling unit, which is used to construct an error model and recursively decompose the relative geometric error; The correction unit is used to predict the short-term error based on the error model and system response delay, and generate a feedforward compensation amount to apply to the galvanometer control command to correct the subsequent processing path in real time.

[0050] The optimization modules include: The data unit is used to continuously update the local trajectory, distance constraint set, and error model in real time during the processing. The forgetting unit is used to introduce a forgetting mechanism to reduce the impact of historical data on current predictions.

[0051] In summary, the advantages of this invention are: it eliminates the need for costly and time-consuming offline calibration before processing, improving the practicality and deployment efficiency of the project and eliminating reliance on complex offline calibration; it achieves high-precision online real-time compensation for laser processing galvanometer splicing, dynamically correcting time-varying errors and ensuring the accuracy of the splicing area; it improves the dynamic modeling capability and change response speed of errors; it enhances the system's accuracy in predicting processing noise, environmental interference, and errors; it ensures the consistency of splicing accuracy during large-area, long-term processing, significantly improving the overall stability of laser processing and meeting the core requirements of high quality and high reliability in the precision manufacturing field.

[0052] 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 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 claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A closed-loop control method for the splicing accuracy of laser processing galvanometers, characterized in that, include: The theoretical parameters of the galvanometer system are obtained, and a theoretical global coordinate system is established. The theoretical parameters of the galvanometer system include theoretical geometric parameters, theoretical pose parameters, and response delay parameters. Acquire control commands for the galvanometer in the splicing area, image data of the processing area, and corresponding timestamps during the processing; Based on the control commands of the theoretical global coordinate system and the splicing region galvanometer, the coordinates of the theoretical processing point are generated; Extract the local trajectory formed by galvanometer processing in the stitching region of the image data, wherein the local trajectory includes the geometric features of the processing trajectory; The time offset between the image acquisition time and the effective time of the galvanometer control command is obtained based on the timestamp and response delay parameters, and the theoretical processing point and local trajectory are initially aligned in time. The observed trajectories of local trajectories within the time window are extracted, and corresponding weights are obtained based on geometric consistency and observation stability, followed by preliminary spatial alignment. In the galvanometer stitching area, based on the observation trajectory and theoretical processing trajectory corresponding to the same time window of adjacent galvanometers, the distance constraint of galvanometer stitching is calculated and assigned a corresponding weight to form a distance constraint set. The theoretical processing trajectory is the trajectory formed by theoretical processing points. The distance constraint set includes the distance constraint between the theoretical processing trajectory and the observation trajectory, and the distance constraint between the observation trajectories of adjacent galvanometers. Based on the distance constraint set at different spatial locations, an overdetermined constraint equation set is constructed to solve for the relative geometric error between adjacent galvanometers; Construct an error model and recursively decompose the relative geometric error; Based on the error model and system response delay, the short-term error is predicted and a feedforward compensation amount is generated to apply to the galvanometer control command, thereby correcting the subsequent processing path in real time. During the processing, the local trajectory, distance constraint set, and error model are continuously updated in real time, and a forgetting mechanism is introduced to reduce the impact of historical data on the current prediction.

2. The closed-loop control method for laser processing galvanometer splicing accuracy according to claim 1, characterized in that, The observed trajectories of local trajectories within the extraction time window are weighted based on geometric consistency and observation stability, and preliminary spatial alignment is performed, specifically including: Set a time window; Acquire the local trajectory of the galvanometer in the stitching region within the time window; Extract the observed trajectory from the local trajectory; The weights corresponding to the observation trajectories are obtained based on the degree of geometric consistency and the degree of observation stability. Based on the time offset between the galvanometer control command and the image acquisition time and the effective time of the galvanometer control command, the coordinates of the observation trajectory and the corresponding theoretical processing point are matched, and the correspondence between the image coordinate system where the observation trajectory is located and the theoretical global coordinate system where the theoretical processing point is located is established. The image coordinate system is the original pixel coordinate system inherent to the industrial camera imaging sensor. Based on the correspondence between the image coordinate system where the observation trajectory is located and the theoretical global coordinate system where the theoretical processing point coordinates are located, the spatial transformation matrix is ​​calculated to transform the observation trajectory into the theoretical global coordinate system.

3. The closed-loop control method for laser processing galvanometer splicing accuracy according to claim 1, characterized in that, In the galvanometer stitching area, based on the observation trajectory and theoretical processing trajectory corresponding to adjacent galvanometers within the same time window, the distance constraint for galvanometer stitching is calculated and assigned corresponding weights to form a distance constraint set, specifically including: Within a unified time window, the observation trajectory of the first galvanometer in the stitching area is obtained, and the observation trajectory of the second galvanometer in the stitching area is obtained. The first galvanometer is one of two adjacent galvanometers in the stitching area, and the second galvanometer is the other of two adjacent galvanometers in the stitching area. Within the same time window, match the theoretical processing trajectory formed by the theoretical processing points that correspond to the observed trajectory; The distance between the observation trajectory of the first galvanometer and the theoretical processing trajectory is used as the first distance constraint; The distance between the observation trajectory of the second galvanometer and the theoretical processing trajectory is used as the second distance constraint; The distance between the observation trajectories of the first and second galvanometers is used as the third distance constraint; The distance constraints are weighted according to the corresponding weights of the observed trajectories; The three distance constraints and their corresponding weights are integrated into the distance constraint set.

4. The closed-loop control method for laser processing galvanometer splicing accuracy according to claim 1, characterized in that, The set of distance constraints based on different spatial locations is used to construct an overdetermined set of constraint equations to solve for the relative geometric error between adjacent galvanometers, specifically including: The relative geometric error to be determined is introduced into the distance constraint set and transformed into a system of relative geometric error constraint equations; An overdetermined constraint equation system is constructed based on the relative geometric error constraint equation system at different spatial locations; Solve the overdetermined system of equations to obtain the relative geometric error between adjacent galvanometers.

5. The closed-loop control method for laser processing galvanometer splicing accuracy according to claim 1, characterized in that, The construction of the error model and the recursive decomposition of the relative geometric error specifically include: Obtain the online time series of relative geometric error; Slow-changing error models and fast-changing error models are constructed and recursively decomposed based on time series, and the corresponding error components are separated according to the rate of change. Recursive state updates are performed at different frequencies based on the error model and the rate of change of the error components.

6. The closed-loop control method for laser processing galvanometer splicing accuracy according to claim 1, characterized in that, The method of predicting short-term errors based on the error model and system response delay, and generating feedforward compensation for the galvanometer control commands to correct subsequent processing paths in real time, specifically includes: Obtain the current error value of the error model; Determine the prediction time domain, which covers the duration from the image acquisition time to the effective time of the galvanometer control command; Based on the error model, the current error value is extrapolated to future times to obtain the predicted error value in the prediction time domain; Based on the error prediction value, a feedforward compensation amount matching the time step of the galvanometer control command is generated to correct the galvanometer control command. The corrected galvanometer control command is sent to the galvanometer drive unit to adjust the spatial position of the subsequent machining path in real time.

7. A closed-loop control system for the splicing accuracy of laser processing galvanometers, used to implement the closed-loop control method for the splicing accuracy of laser processing galvanometers as described in any one of claims 1-6, characterized in that, include: The information module is used to acquire theoretical parameters of the galvanometer system, establish a theoretical global coordinate system, and acquire control commands for the galvanometer in the splicing area, image data of the processing area, and corresponding timestamps during the processing. The constraint module is used to generate the coordinates of the theoretical processing point based on the theoretical global coordinate system and the control commands of the galvanometer in the stitching area, extract the local trajectory formed by the galvanometer processing in the stitching area of ​​the image data, obtain the time offset between the image acquisition time and the effective time of the galvanometer control command based on the timestamp and response delay parameters, perform preliminary time alignment on the theoretical processing point and the local trajectory, extract the observation trajectory of the local trajectory within the time window, obtain the corresponding weights based on geometric consistency and observation stability, and perform preliminary spatial alignment. In the galvanometer stitching area, based on the corresponding observation trajectory and theoretical processing trajectory within the same time window of adjacent galvanometers, the distance constraint of the galvanometer stitching is calculated and assigned corresponding weights to form a distance constraint set. The error module is used to construct an overdetermined set of constraint equations based on distance constraint sets at different spatial locations, solve the relative geometric error between adjacent galvanometers, construct an error model and recursively decompose the relative geometric error, predict the short-term error based on the error model and system response delay, generate a feedforward compensation amount to act on the galvanometer control command, and correct the subsequent processing path in real time. An optimization module is used to continuously update the local trajectory, distance constraint set, and error model in real time during the processing, and to introduce a forgetting mechanism to reduce the impact of historical data on the current prediction.

8. The closed-loop control system for laser processing galvanometer splicing accuracy according to claim 7, characterized in that, The constraint module specifically includes: The image unit is used to generate the coordinates of the theoretical processing point and extract the local trajectory formed by the processing of the splicing region galvanometer in the image data based on the theoretical global coordinate system and the control command of the splicing region galvanometer. The alignment unit is used to obtain the time offset between the image acquisition time and the effective time of the galvanometer control command based on the timestamp and response delay parameters, perform preliminary time alignment on the theoretical processing point and local trajectory, extract the observation trajectory of the local trajectory within the time window, obtain the corresponding weights based on geometric consistency and observation stability, and perform preliminary spatial alignment. The distance unit is used to calculate the distance constraints of the galvanometer splicing in the galvanometer splicing area based on the observation trajectory and theoretical processing trajectory corresponding to the adjacent galvanometers within the same time window and assign corresponding weights to form a distance constraint set.

9. The closed-loop control system for laser processing galvanometer splicing accuracy according to claim 7, characterized in that, The error module specifically includes: Equation unit, which is used to construct an overdetermined set of constraint equations based on distance constraint sets at different spatial locations, and solve for the relative geometric error between adjacent galvanometers; A modeling unit, which is used to construct an error model and recursively decompose the relative geometric error; The correction unit is used to predict the short-term error based on the error model and system response delay, and generate a feedforward compensation amount to apply to the galvanometer control command to correct the subsequent processing path in real time.

10. The closed-loop control system for laser processing galvanometer splicing accuracy according to claim 7, characterized in that, The optimization module specifically includes: The data unit is used to continuously update the local trajectory, distance constraint set, and error model in real time during the processing. The forgetting unit is used to introduce a forgetting mechanism to reduce the impact of historical data on current predictions.

Images