A kind of inverse solution control method, system, device and medium for five-axis galvanometer scanning system

Abstract

The application discloses a kind of anti-solution control methods, systems, equipment and medium for five-axis galvanometer scanning system, mainly related to five-axis galvanometer technical field, to solve the mapping relationship involved in existing approximate mapping scheme is prone to deviation, only rely on offline fitting result or local geometric approximation, not necessarily can guarantee that the control quantity of online output always corresponds target image plane state, in addition, there are multiple tool separation problems, need to be maintained by artificial coordinate alignment or script update incident state Process interconnection.It includes: calculating the error result between the predicted image plane state and the target image plane state;When the error result does not meet the preset output condition, use the prediction control vector to construct a number of candidate control vectors;Calculate the comprehensive score through the error result corresponding to the candidate control vector, output the candidate control vector corresponding to the minimum value of the comprehensive score;When the error result meets the preset output condition, directly output the prediction control vector.

Description

Technical Field

[0001] This application relates to the field of five-axis galvanometer technology, and in particular to an inverse kinematics control method, system, device and medium for five-axis galvanometer scanning systems. Background Technology

[0002] The core focus of a five-axis galvanometer scanning system is not the galvanometer angle itself, but rather the position, exit direction, and stability of the beam on the target image plane after multiple reflections and focusing by the field lens. Deviations in the focal position and incident direction directly affect energy deposition distribution, linewidth consistency, surface morphology, and processing defects. Therefore, a method is needed to achieve stable control over the state of the target image plane.

[0003] In existing engineering practices, a common approach is to first establish an approximate mapping from the target image plane state to the galvanometer drive angle through calibration or offline fitting, and then directly output the drive angle based on this mapping in the online stage. This method is highly efficient when the system structure is relatively stable and the working area is limited, but it is highly dependent on the quality of assembly and calibration. When there is temperature drift, assembly deviation, mirror installation error, or changes in lens parameters, the original mapping relationship is prone to shift. Furthermore, since the inverse mapping may have multiple solutions or approximately multiple solutions under certain operating conditions, relying solely on offline fitting results or local geometric approximations may not guarantee that the online output control quantity always corresponds to the target image plane state.

[0004] Furthermore, existing engineering processes commonly suffer from the problem of multiple tools being separated. Because these steps are not standardized in terms of data structure, coordinate system conventions, and interface forms, engineers often need to manually align coordinates or update the incident state using scripts to maintain process continuity. Summary of the Invention

[0005] This application provides an inverse control method, system, device, and medium for a five-axis galvanometer scanning system to address the problems of existing approximate mapping schemes, such as the ease with which mapping relationships can shift, reliance on offline fitting results or local geometric approximations, and the inability to guarantee that the online output control quantity always corresponds to the target image plane state. In addition, there is the problem of multiple tools being separated, requiring manual coordinate alignment or script updates of the incident state to maintain process continuity.

[0006] In a first aspect, this application provides an inverse kinematics control method for a five-axis galvanometer scanning system, wherein the five-axis galvanometer scanning system consists of four controlled galvanometers and a field mirror, and the method includes: Receive the actual control vector of the controlled galvanometer and send it to each controlled galvanometer channel; obtain the entrance pupil ray state through each controlled galvanometer channel; By using a fixed coordinate alignment relationship, the state of the entrance pupil ray is mapped to the sequential tracing coordinate system to obtain the sequential tracing incident state; using the sequential tracing incident state as the incident state, sequential tracing is performed on the field lens to obtain the actual image plane state. Data samples are constructed based on a preset number of actual control vectors and actual image plane state combinations; using the data samples, forward neural networks and inverse neural networks are trained and obtained. When the target image plane state is obtained, the target image plane state is used as the input of the inverse neural network, and the output is the prediction control vector; the prediction control vector is input into the forward neural network, and the output is the prediction image plane state; the error result between the prediction image plane state and the target image plane state is calculated. When the error result does not meet the preset output conditions, several candidate control vectors are constructed using the predictive control vector; the comprehensive score is calculated based on the error result corresponding to the candidate control vector, and the candidate control vector corresponding to the minimum comprehensive score is output; when the error result meets the preset output conditions, the predictive control vector is output directly.

[0007] In one implementation of this application, the actual control vector of the controlled galvanometer is received and sent to each controlled galvanometer channel; the entrance pupil ray state is obtained through each controlled galvanometer channel, specifically including: Receive the actual control vector of the controlled galvanometer: in, Indicates the first The driving angle of each galvanometer; The actual control vector passes through each controlled galvanometer channel, and using a five-axis galvanometer scanning system, the final outgoing rays from the four mirrors are uniformly converted into entrance pupil rays. in, Indicates the entrance pupil point. Indicates the direction of incidence.

[0008] In one implementation of this application, the entrance pupil ray state is mapped to a sequential tracing coordinate system through a fixed coordinate alignment relationship to obtain a sequential tracing incident state; using the sequential tracing incident state as the incident state, sequential tracing is performed on the field lens to obtain the actual image plane state, specifically including: Through the formula: The state of the light entering the pupil Convert to trajectory incident state ,in, Indicates the entrance pupil point. Indicates the direction of incidence; by As the incident state, sequential tracing is performed on the field lenses to obtain the image plane intersection point. and image plane direction ; Define relative optical axis The angle of inclination is: Establish the actual image plane state based on the image plane intersection point, image plane direction, and tilt angle: .

[0009] In one implementation of this application, data samples are constructed based on a preset number of actual control vectors and actual image plane state combinations; using the data samples, a forward neural network and an inverse neural network are trained, specifically including: Using the actual control vector in the data sample as input data and the actual image plane state as label, train the first neural network to obtain a forward neural network containing the mapping relationship from the actual control vector to the actual image plane state. Using the actual image plane state in the data sample as input data and the actual control vector as label, a second neural network is trained to obtain an inverse neural network containing the mapping relationship between the actual image plane state and the actual control vector.

[0010] In one implementation of this application, when the target image plane state is obtained, the target image plane state is used as the input to the inverse neural network, and a prediction control vector is output; the prediction control vector is input to the forward neural network, and a predicted image plane state is output; the error between the predicted image plane state and the target image plane state is calculated, specifically including: Obtain the target image plane state ; The target image plane state is used as input to the inverse neural network, and the output is a prediction control vector: in, This represents a reverse neural network. Represents the predictive control vector; Predicted control vector Input forward neural network Output the predicted image plane state: and in, This represents the image plane intersection point corresponding to the predicted image plane state. , This indicates the tilt angle corresponding to the predicted image plane state; Through the formula: Calculate position error and angle error ;in, This indicates the intersection point of the image plane corresponding to the state of the target image plane. , The tilt angle corresponding to the state of the target image plane is indicated, and the error results include position error and angle error.

[0011] In one implementation of this application, when the error result does not meet the preset output conditions, several candidate control vectors are constructed using the predicted control vector. A comprehensive score is calculated based on the error results corresponding to the candidate control vectors, and the candidate control vector corresponding to the minimum comprehensive score is output. Specifically, this includes: When not satisfied At that time, the error result did not meet the preset output conditions; among them, Indicates positional error. Indicates angular error; Using the predictive control vector, k candidate control vectors are constructed within a preset fluctuation range; The candidate image plane state of each candidate control vector is obtained using a forward neural network. Using the candidate image plane state and the target image plane state, calculate the error results corresponding to each candidate control vector; Through the formula: Calculate the overall score J; where, This represents the position error corresponding to the candidate control vector. This represents the angle error corresponding to the candidate control vector; Indicates the weight of the preset angle item; Output the candidate control vector corresponding to the minimum comprehensive score.

[0012] Secondly, this application provides an inverse kinematics control system for a five-axis galvanometer scanning system, the system comprising: The acquisition module receives the actual control vector of the controlled galvanometer and sends it to each controlled galvanometer channel; it obtains the entrance pupil ray state through each controlled galvanometer channel; it maps the entrance pupil ray state to the sequential tracing coordinate system through a fixed coordinate alignment relationship to obtain the sequential tracing incident state; it uses the sequential tracing incident state as the incident state to perform sequential tracing on the field lens to obtain the actual image plane state. The training module is used to construct data samples based on a preset number of actual control vectors and actual image plane state combinations; and to train and obtain forward neural networks and inverse neural networks using the data samples. The error module is used to take the target image plane state as the input of the inverse neural network when the target image plane state is obtained, and output the prediction control vector; input the prediction control vector into the forward neural network and output the prediction image plane state; and calculate the error result between the prediction image plane state and the target image plane state. The output module is used to construct several candidate control vectors using the predictive control vector when the error result does not meet the preset output conditions; calculate the comprehensive score based on the error result corresponding to the candidate control vector, and output the candidate control vector corresponding to the minimum comprehensive score; when the error result meets the preset output conditions, directly output the predictive control vector.

[0013] In one implementation of this application, the output module includes an output unit. Used when not satisfied At that time, the error result did not meet the preset output conditions; among them, Indicates positional error. Indicates angular error; Using the predictive control vector, k candidate control vectors are constructed within a preset fluctuation range; The candidate image plane state of each candidate control vector is obtained using a forward neural network. Using the candidate image plane state and the target image plane state, calculate the error results corresponding to each candidate control vector; Through the formula: Calculate the overall score J; where, This represents the position error corresponding to the candidate control vector. This represents the angle error corresponding to the candidate control vector; Indicates the weight of the preset angle item; Output the candidate control vector corresponding to the minimum comprehensive score.

[0014] Thirdly, this application provides an inverse kinematics control device for a five-axis galvanometer scanning system, the device comprising: processor; And a memory containing executable code, which, when executed, causes the processor to perform an inverse control method for a five-axis galvanometer scanning system, as described above.

[0015] Fourthly, this application provides a non-volatile computer storage medium storing computer instructions, which, when executed, implement an inverse control method for a five-axis galvanometer scanning system as described above.

[0016] As can be seen from the above technical solutions, this application has the following advantages: This application differs from traditional offline fitting methods in establishing mapping relationships. It constructs a sample dataset of actual control vectors and actual image plane states, training forward and inverse neural networks. In actual operation, the neural network model is used as the core for predicting and verifying control vectors, thus eliminating excessive reliance on the quality of assembly and calibration.

[0017] By employing closed-loop verification and candidate optimization mechanisms, this application addresses the problem of multiple solutions or approximate multiple solutions in inverse mapping. After obtaining the predicted control vector, it does not output it directly but instead inputs it into a forward neural network to obtain the predicted image plane state. This predicted state is then compared with the target image plane state to calculate the error. When the error does not meet the requirements, several candidate control vectors are constructed, and a comprehensive score is calculated. The optimal result is then selected for output. This process effectively filters out the control vector that best matches the target image plane state, avoiding the mismatch between the control quantity and the target state that may occur when relying solely on offline fitting or local geometric approximation.

[0018] This application directly maps the entrance pupil ray state to the sequential tracing coordinate system using a fixed coordinate alignment relationship, eliminating the need for manual coordinate alignment or script updates to the incident state. The unified data structure and coordinate system conventions enable seamless integration between the controlled galvanometer channel, the sequential tracing module, and the neural network training module, eliminating interface barriers between different tools. This not only reduces potential errors from manual operation but also shortens workflow transition time, making the control process of the entire five-axis galvanometer scanning system smoother and more efficient, and reducing the complexity and labor costs of engineering implementation. Attached Figure Description

[0019] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of an inverse kinematics control method for a five-axis galvanometer scanning system provided in an embodiment of this application.

[0021] Figure 2 This is a forward simulation link flowchart provided in an embodiment of this application.

[0022] Figure 3 This is a flowchart of an online inverse closed-loop control provided in an embodiment of this application.

[0023] Figure 4 This is a schematic diagram of the internal structure of an inverse kinematics control system for a five-axis galvanometer scanning system provided in an embodiment of this application.

[0024] Figure 5 This is a schematic diagram of the internal structure of an inverse kinematics control device for a five-axis galvanometer scanning system provided in an embodiment of this application. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Those skilled in the art should understand that the embodiments described below are merely preferred embodiments of this disclosure and do not imply that this disclosure can only be implemented through these preferred embodiments. These preferred embodiments are merely used to explain the technical principles of this disclosure and are not intended to limit the scope of protection of this disclosure. Based on the preferred embodiments provided by this disclosure, all other embodiments obtained by those skilled in the art without creative effort should still fall within the scope of protection of this disclosure.

[0027] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0028] The technical solutions proposed in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0029] The embodiment provides an inverse kinematics control method for a five-axis galvanometer scanning system, such as Figure 1 As shown in the embodiments of this application, the method mainly includes the following steps: Step 110: Receive the actual control vector of the controlled galvanometer and send it to each controlled galvanometer channel; obtain the entrance pupil ray state through each controlled galvanometer channel.

[0030] In some embodiments, this step specifically includes: Receive the actual control vector of the controlled galvanometer: in, Indicates the first The driving angle of each galvanometer; The actual control vector passes through each controlled galvanometer channel, and using a five-axis galvanometer scanning system, the final outgoing rays from the four mirrors are uniformly converted into entrance pupil rays. in, Indicates the entrance pupil point. Indicates the direction of incidence.

[0031] It should be added that the five-axis galvanometer scanning system is capable of operating in the mechanical coordinate system. The instantaneous attitude of each mirror is updated based on the control vector, and the mirror normal is updated, the mirror plane is constructed, the intersection of the line and the plane is calculated, and the reflection direction is iteratively propagated to obtain the final outgoing light after passing through multiple mirrors.

[0032] By introducing a unified expression of the entrance pupil ray state With the fixed coordinate alignment parameters determined by the assembly and calibration, the system achieves a stable mapping from the mechanical domain output to the sequential tracing input, reducing the risks of intersection failure, tracing interruption, output jump, and repeated manual adaptation between different toolchains, thereby improving the reproducibility and portability of the system.

[0033] More specifically, such as Figure 2 In the mechanical domain simulation area, the galvanometer structure and calibration parameters are set, and the actual control vector is... Input to the five-axis galvanometer scanning system, utilize the multi-mirror geometry solving module within the five-axis galvanometer scanning system, and output the final outgoing ray ( ).

[0034] As an example, in the mechanical coordinate system, the mirror normal vector and rotation axis unit vector at the zero position of each mirror are pre-calibrated. When the drive angle in the control vector is given, the system first updates the mirror normal vector of each mirror. (The text then repeats the process of updating the mirror normal vector.) Taking a single galvanometer as an example, its mirror normal can be expressed as: ; in Indicates rotation around the axis Rotation angle The rotation operator can be expressed using the Rodrigues formula.

[0035] After obtaining the mirror attitude, the system constructs the mirror plane equation based on a known point on the mirror surface, and finds the intersection of the current ray with the mirror plane to obtain the intersection point position; then, it updates the propagation direction according to the law of mirror reflection. By repeatedly executing mirror attitude update, plane construction, intersection calculation, and reflection propagation, the system completes the continuous reflection calculation of the four galvanometers one by one, and finally obtains the outgoing ray after passing through all the galvanometers ( The system does not directly retain the dispersed mirror attitudes, intersections, and intermediate propagation directions, but instead uniformly represents the results of the multi-mirror reflection link as the entrance pupil ray state.

[0036] More specifically, in the digital simulation process of receiving the actual control vector of the controlled galvanometer and distributing it to each controlled galvanometer channel, this application first establishes a mathematical model coupling the mechanical and optical domains based on the structural parameters of the five-axis galvanometer scanning system. Specifically, the control vector is composed of the driving angles of the four controlled galvanometers. In the mechanical coordinate system, combined with the initial normal vectors of each mirror Rotation axis direction and mirror mounting points Calculate the instantaneous normal vector of each mirror at the current driving angle according to the rotation transformation formula: , in, This represents a 3D rotation operator about a given rotation axis. Subsequently, it is based on the updated mirror normal vector. and mirror mounting points Construct corresponding mirror plane For the current propagating light Perform intersection operations between the line and the plane on each frame to obtain the intersection point parameters:

[0037] And further, the reflection points at each level are obtained: ; Then, update the direction of propagation after reflection according to the law of specular reflection:

[0038] And and The incident position and direction of the next-stage reflector are then recursively solved to obtain the final outgoing ray after the continuous action of the four mirrors. Subsequently, the final outgoing ray is uniformly represented as the entrance pupil ray state:

[0039] And by mapping to the sequential tracing coordinate system through a fixed rotation and translation relationship, we obtain:

[0040] This creates a sequential trajectory incident state: Again As the incident condition for sequential tracing of the field lens, the propagation of light in the field lens is mathematically simulated using a sequential optical model of the lens. In one implementation, a sequential optical tracing model can be established using Python's rayoptics to calculate the image plane intersection point, image plane direction, and tilt angle relative to the optical axis, thus forming the actual image plane state. In practical implementation, the aforementioned control issuance, state transfer, simulation scheduling, and data recording can be organized using ROS2 as a communication and scheduling intermediary layer. This establishes a complete digital simulation link from control vector input to image plane state output, serving as the unified physical model foundation for subsequent data sample generation, neural network training, and online closed-loop verification.

[0041] It should be further explained that in the digital simulation process of this application, a mathematical model coupling the mechanical and optical domains is first established based on the structural parameters of the five-axis galvanometer scanning system. Specifically, the driving angles of the four controlled galvanometers form the control vector. In the mechanical coordinate system, combined with the initial normal vectors, rotation axis directions, and installation position parameters of each mirror, the instantaneous attitude of each mirror at the current driving angle is calculated sequentially according to existing calculation formulas. Then, based on the updated mirror normal vectors and mirror positions, corresponding mirror planes are constructed. The intersection operation of the line and the plane is performed on each mirror for the current propagating ray to obtain the reflection points at each level. The ray propagation direction is updated according to the mirror reflection law. Through a recursive solution process, the final outgoing ray after the continuous action of the four mirrors is obtained. Subsequently, the final outgoing ray is uniformly represented as the entrance pupil ray state and mapped to the sequential tracing coordinate system through a fixed rotation and translation relationship to obtain the sequential tracing incident state. Then, this sequential tracing incident state is used as the incident condition for the sequential tracing of the field mirrors to calculate the image plane intersection point, image plane direction, and tilt angle relative to the optical axis to form the actual image plane state. This established a complete digital simulation link from control vector input to image plane state output.

[0042] Step 120: Map the entrance pupil ray state to the sequential tracing coordinate system through a fixed coordinate alignment relationship to obtain the sequential tracing incident state; use the sequential tracing incident state as the incident state to perform sequential tracing on the field lens to obtain the actual image plane state.

[0043] In some embodiments, this step specifically includes: Through the formula: The state of the light entering the pupil Convert to trajectory incident state ,in, Indicates the entrance pupil point. Indicates the direction of incidence; by As the incident state, sequential tracing is performed on the field lenses to obtain the image plane intersection point. and image plane direction ; Define relative optical axis The angle of inclination is: Establish the actual image plane state based on the image plane intersection point, image plane direction, and tilt angle: .

[0044] It should be further explained that, through the above design, the system uses a unified entrance pupil ray state and image plane state as the core interface that runs through the geometric domain, optical domain and data domain. This not only improves the stability and reproducibility of the link, but also enables subsequent sample generation, model training and online control to be carried out around the same state definition, thus constituting an important structural feature that distinguishes this application from the traditional splicing scheme.

[0045] More specifically, such as Figure 2 The system includes a state interface, coordinate transformation area, and optical domain simulation area. After obtaining the entrance pupil ray state, the system maps the entrance pupil ray state in the mechanical domain to the sequential tracking coordinate system through a fixed coordinate alignment relationship. Specifically, a fixed rotation is applied to the direction vector, and a fixed rotation and translation are applied to the position vector to obtain the sequential tracking incident state. Subsequently, the field lens sequential optical tracking module... As the incident state, sequential tracing is performed on the field lenses to obtain the intersection point of the image plane and the exit direction of the image plane (the actual image plane state).

[0046] Step 130: Construct data samples based on a preset number of actual control vectors and actual image plane state combinations; use the data samples to train and obtain a forward neural network and an inverse neural network.

[0047] In some embodiments, this step may specifically be as follows: Using the actual control vector in the data sample as input data and the actual image plane state as label, train the first neural network to obtain a forward neural network containing the mapping relationship from the actual control vector to the actual image plane state. Using the actual image plane state in the data sample as input data and the actual control vector as label, a second neural network is trained to obtain an inverse neural network containing the mapping relationship between the actual image plane state and the actual control vector.

[0048] It should be further explained that the forward neural network is used to quickly approximate the mapping from the control vector to the image plane state, and can serve as a fast forward proxy for online backcomputation and candidate selection; the inverse neural network is used to output candidate control quantities in real time for the target image plane state. Unlike existing schemes that rely solely on supervised inverse solutions, this application does not directly use the output of the inverse neural network as the execution instruction, but instead introduces closed-loop verification based on the forward neural network.

[0049] Step 140: When the target image plane state is obtained, the target image plane state is used as the input of the inverse neural network to output the prediction control vector; the prediction control vector is input into the forward neural network to output the prediction image plane state; the error result between the prediction image plane state and the target image plane state is calculated.

[0050] In some embodiments, this step may specifically be as follows: Figure 3The state and candidate control quantity generation region and the closed-loop prediction and error verification region are shown below: Obtain the target image plane state ; Using the target image plane state as input to an inverse neural network (IK neural network), the output is a prediction control vector: in, This represents a reverse neural network. Represents the predictive control vector; Predicted control vector Input forward neural network (FK neural network) Output the predicted image plane state: and in, This represents the image plane intersection point corresponding to the predicted image plane state. , This indicates the tilt angle corresponding to the predicted image plane state; The formulas involved in the error calculation module are as follows: Calculate position error and angle error ;in, Indicates the image plane intersection point corresponding to the target image plane state. , This indicates the tilt angle corresponding to the target image plane state. The error results include position error and angle error. It should be noted that the unit for position error is μm, and the unit for angle error is μrad.

[0051] Step 150: When the error result does not meet the preset output conditions, construct several candidate control vectors using the predictive control vector; calculate the comprehensive score based on the error result corresponding to the candidate control vector, and output the candidate control vector corresponding to the minimum comprehensive score; when the error result meets the preset output conditions, directly output the predictive control vector.

[0052] This step can be specifically described as follows: Figure 3 As shown in the control output and safety refinement area: This step may include: Detection module based on preset threshold: ; When not satisfied At that time, the error result did not meet the preset output conditions; among them, Indicates positional error. Indicates angular error; Using the predictive control vector, k candidate control vectors are constructed within a preset fluctuation range; The candidate image plane state of each candidate control vector is obtained using a forward neural network. Using the candidate image plane state and the target image plane state, calculate the error results corresponding to each candidate control vector; Through the formula: Calculate the overall score J; where, This represents the position error corresponding to the candidate control vector. This represents the angle error corresponding to the candidate control vector; Indicates the weight of the preset angle item; Output the candidate control vector corresponding to the minimum comprehensive score.

[0053] When satisfied At that time, the predictive control vector is directly output.

[0054] It should be noted that, under the same test conditions, the closed-loop control accuracy of the proposed solution was statistically analyzed. The embodiments of this application use the 95th percentile value of the error (hereinafter referred to as P95 error, which means that after arranging the errors of all test samples in ascending order, the error value corresponding to the 95th cumulative proportion position, i.e., about 95% of the test sample errors do not exceed this value). The unit of position error is μm, and the unit of angle error is μrad. Starting with the original model and initial refinement, and using the second-round hard-case enhancement model and annealed refinement as the optimized result, the median position error decreased from 3.707 μm to 1.536 μm, and the position error p95 decreased from 10.995 μm to 3.968 μm, representing reductions of 58.6% and 63.9%, respectively. The median angle error decreased from 11.703 μrad to 3.484 μrad, and the angle error p95 decreased from 37.037 μrad to 11.578 μrad, representing reductions of 70.2% and 68.7%, respectively. These results demonstrate that the method proposed in this application not only reduces the average error but also significantly compresses the long-tail error, thereby improving the control accuracy of the target image plane position and exit tilt angle, and enhancing the stability and executability of the online control results.

[0055] In addition, this application Figure 4 This application provides an inverse kinematics control system for a five-axis galvanometer scanning system. For example... Figure 4 As shown in the embodiments of this application, the system mainly includes: The module 210 is used to receive the actual control vector of the controlled galvanometer and send it to each controlled galvanometer channel; the entrance pupil ray state is obtained through each controlled galvanometer channel; the entrance pupil ray state is mapped to the sequential tracing coordinate system through a fixed coordinate alignment relationship to obtain the sequential tracing incident state; the sequential tracing incident state is used as the incident state to perform sequential tracing on the field lens to obtain the actual image plane state. Training module 220 is used to construct data samples based on a preset number of actual control vectors and actual image plane state combinations; and to train and obtain forward neural networks and inverse neural networks using the data samples. Error module 230 is used to, when the target image plane state is obtained, take the target image plane state as the input of the inverse neural network and output the prediction control vector; input the prediction control vector into the forward neural network and output the prediction image plane state; and calculate the error result between the prediction image plane state and the target image plane state. The output module 240 is used to construct several candidate control vectors using the predictive control vector when the error result does not meet the preset output conditions; calculate the comprehensive score based on the error result corresponding to the candidate control vector, and output the candidate control vector corresponding to the minimum comprehensive score; and directly output the predictive control vector when the error result meets the preset output conditions.

[0056] Output module 240 includes an output unit, Used when not satisfied At that time, the error result did not meet the preset output conditions; among them, Indicates positional error. Indicates angular error; Using the predictive control vector, k candidate control vectors are constructed within a preset fluctuation range; The candidate image plane state of each candidate control vector is obtained using a forward neural network. Using the candidate image plane state and the target image plane state, calculate the error results corresponding to each candidate control vector; Through the formula: Calculate the overall score J; where, This represents the position error corresponding to the candidate control vector. This represents the angle error corresponding to the candidate control vector; Indicates the weight of the preset angle item; Output the candidate control vector corresponding to the minimum comprehensive score.

[0057] The above are method embodiments of this application. Based on the same inventive concept, embodiments of this application also provide an inverse kinematics control device for a five-axis galvanometer scanning system. Figure 5 As shown, the device includes: a processor; and a memory storing executable code thereon, which, when executed, causes the processor to perform an inverse control method for a five-axis galvanometer scanning system as described in the above embodiment.

[0058] Specifically, the server receives the actual control vectors of the controlled galvanometers and distributes them to each controlled galvanometer channel; the entrance pupil ray state is obtained through each controlled galvanometer channel; the entrance pupil ray state is mapped to the sequential tracing coordinate system through a fixed coordinate alignment relationship to obtain the sequential tracing incident state; the sequential tracing incident state is used as the incident state to perform sequential tracing on the field lens to obtain the actual image plane state; data samples are constructed based on a preset number of combinations of actual control vectors and actual image plane states; forward neural networks and inverse neural networks are trained using the data samples; when the target image plane state is obtained, the target image plane state is used as the input of the inverse neural network to output a predicted control vector; the predicted control vector is input into the forward neural network to output a predicted image plane state; the error between the predicted image plane state and the target image plane state is calculated; when the error result does not meet the preset output conditions, several candidate control vectors are constructed using the predicted control vectors; a comprehensive score is calculated based on the error result corresponding to the candidate control vectors, and the candidate control vector corresponding to the minimum comprehensive score is output; when the error result meets the preset output conditions, the predicted control vector is directly output.

[0059] In addition, this application embodiment also provides a non-volatile computer storage medium storing executable instructions, which, when executed, implement the inverse control method for a five-axis galvanometer scanning system as described above.

[0060] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for inverse kinematics control of a five-axis galvanometer scanning system, wherein the five-axis galvanometer scanning system comprises four controlled galvanometers and a field mirror, characterized in that, The method includes: Receive the actual control vector of the controlled galvanometer and send it to each controlled galvanometer channel; obtain the entrance pupil ray state through each controlled galvanometer channel; By using a fixed coordinate alignment relationship, the state of the entrance pupil ray is mapped to the sequential tracing coordinate system to obtain the sequential tracing incident state; using the sequential tracing incident state as the incident state, sequential tracing is performed on the field lens to obtain the actual image plane state. Data samples are constructed based on a preset number of actual control vectors and actual image plane state combinations; using the data samples, forward neural networks and inverse neural networks are trained and obtained. When the target image plane state is obtained, the target image plane state is used as the input of the inverse neural network, and the output is the prediction control vector; the prediction control vector is input into the forward neural network, and the output is the prediction image plane state; the error result between the prediction image plane state and the target image plane state is calculated. When the error result does not meet the preset output conditions, several candidate control vectors are constructed using the predictive control vector; the comprehensive score is calculated based on the error result corresponding to the candidate control vector, and the candidate control vector corresponding to the minimum comprehensive score is output; when the error result meets the preset output conditions, the predictive control vector is output directly.

2. The inverse kinematics control method for a five-axis galvanometer scanning system according to claim 1, characterized in that, Receive the actual control vector of the controlled galvanometer and send it to each controlled galvanometer channel; The state of the entrance pupil light is obtained through each controlled galvanometer channel, specifically including: Receive the actual control vector of the controlled galvanometer: in, Indicates the first The driving angle of each galvanometer; The actual control vector passes through each controlled galvanometer channel, and using a five-axis galvanometer scanning system, the final outgoing rays from the four mirrors are uniformly converted into entrance pupil rays. in, Indicates the entrance pupil point. Indicates the direction of incidence.

3. The inverse kinematics control method for a five-axis galvanometer scanning system according to claim 1, characterized in that, By using a fixed coordinate alignment relationship, the state of the entrance pupil ray is mapped to the sequential tracing coordinate system to obtain the sequential tracing incident state; using the sequential tracing incident state as the incident state, sequential tracing is performed on the field lens to obtain the actual image plane state, specifically including: Through the formula: The state of the light entering the pupil Convert to trajectory incident state ,in, Indicates the entrance pupil point. Indicates the direction of incidence; by As the incident state, sequential tracing is performed on the field lenses to obtain the image plane intersection point. and image plane direction ; Define relative optical axis The angle of inclination is: Establish the actual image plane state based on the image plane intersection point, image plane direction, and tilt angle: 。 4. The inverse kinematics control method for a five-axis galvanometer scanning system according to claim 1, characterized in that, Data samples are constructed based on a preset number of actual control vectors and actual image plane state combinations; Using data samples, forward and inverse neural networks are trained, specifically including: Using the actual control vector in the data sample as input data and the actual image plane state as label, train the first neural network to obtain a forward neural network containing the mapping relationship from the actual control vector to the actual image plane state. Using the actual image plane state in the data sample as input data and the actual control vector as label, a second neural network is trained to obtain an inverse neural network containing the mapping relationship between the actual image plane state and the actual control vector.

5. The inverse kinematics control method for a five-axis galvanometer scanning system according to claim 1, characterized in that, When the target image plane state is obtained, the target image plane state is used as the input of the inverse neural network, and the output is the prediction control vector; the prediction control vector is input into the forward neural network, and the output is the prediction image plane state. The error between the predicted image plane state and the target image plane state is calculated, specifically including: Obtain the target image plane state ; The target image plane state is used as input to the inverse neural network, and the output is a prediction control vector: in, This represents a reverse neural network. Represents the predictive control vector; Predicted control vector Input forward neural network Output the predicted image plane state: and in, This represents the image plane intersection point corresponding to the predicted image plane state. , This indicates the tilt angle corresponding to the predicted image plane state; Through the formula: Calculate position error and angle error ;in, This indicates the intersection point of the image plane corresponding to the state of the target image plane. , The tilt angle corresponding to the state of the target image plane is indicated, and the error results include position error and angle error.

6. The inverse kinematics control method for a five-axis galvanometer scanning system according to claim 1, characterized in that, When the error result does not meet the preset output conditions, several candidate control vectors are constructed using the predicted control vector. A comprehensive score is calculated based on the error results corresponding to these candidate control vectors, and the candidate control vector corresponding to the minimum comprehensive score is output. Specifically, this includes: When not satisfied At that time, the error result did not meet the preset output conditions; among them, Indicates positional error. Indicates angular error; Using the predictive control vector, k candidate control vectors are constructed within a preset fluctuation range; The candidate image plane state of each candidate control vector is obtained using a forward neural network. Using the candidate image plane state and the target image plane state, calculate the error results corresponding to each candidate control vector; Through the formula: Calculate the overall score J; where, This represents the position error corresponding to the candidate control vector. This represents the angle error corresponding to the candidate control vector; Indicates the weight of the preset angle item; Output the candidate control vector corresponding to the minimum comprehensive score.

7. A reverse engineering control system for a five-axis galvanometer scanning system, characterized in that, The system includes: The acquisition module receives the actual control vector of the controlled galvanometer and sends it to each controlled galvanometer channel; it obtains the entrance pupil ray state through each controlled galvanometer channel; it maps the entrance pupil ray state to the sequential tracing coordinate system through a fixed coordinate alignment relationship to obtain the sequential tracing incident state; it uses the sequential tracing incident state as the incident state to perform sequential tracing on the field lens to obtain the actual image plane state. The training module is used to construct data samples based on a preset number of actual control vectors and actual image plane state combinations; and to train and obtain forward neural networks and inverse neural networks using the data samples. The error module is used to take the target image plane state as the input of the inverse neural network when the target image plane state is obtained, and output the prediction control vector; input the prediction control vector into the forward neural network and output the prediction image plane state; and calculate the error result between the prediction image plane state and the target image plane state. The output module is used to construct several candidate control vectors using the predictive control vector when the error result does not meet the preset output conditions; calculate the comprehensive score based on the error result corresponding to the candidate control vector, and output the candidate control vector corresponding to the minimum comprehensive score; when the error result meets the preset output conditions, directly output the predictive control vector.

8. The inverse kinematics control system for a five-axis galvanometer scanning system according to claim 7, characterized in that, The output module includes output units. Used when not satisfied At that time, the error result did not meet the preset output conditions; among them, Indicates positional error. Indicates angular error; Using the predictive control vector, k candidate control vectors are constructed within a preset fluctuation range; The candidate image plane state of each candidate control vector is obtained using a forward neural network. Using the candidate image plane state and the target image plane state, calculate the error results corresponding to each candidate control vector; Through the formula: Calculate the overall score J; where, This represents the position error corresponding to the candidate control vector. This represents the angle error corresponding to the candidate control vector; Indicates the weight of the preset angle item; Output the candidate control vector corresponding to the minimum comprehensive score.

9. A reverse engineering control device for a five-axis galvanometer scanning system, characterized in that, The device includes: processor; And a memory storing executable code, which, when executed, causes the processor to perform an inverse control method for a five-axis galvanometer scanning system as described in any one of claims 1-6.

10. A non-volatile computer storage medium, characterized in that, It stores computer instructions, which, when executed, implement the inverse control method for a five-axis galvanometer scanning system as described in any one of claims 1-6.

Images