Intelligent laser alignment system

Abstract

The invention discloses an intelligent laser collimation system, which realizes automatic collimation of a laser beam through cooperative control of double reflectors, a high-precision electric actuator and an SAC reinforcement learning algorithm, adopts a cage type integrated structure design of double reflectors and a coaxial connecting rod, realizes rigid constraint and rapid alignment of a light path, and improves the accuracy of laser collimation. A reflector double-axis precise automatic adjusting mechanism based on a miniature electric actuator is developed, an SAC reinforcement learning algorithm is applied to an optical collimation system for the first time, a dynamic reward function design based on light spot offset is proposed, a double-reflector cooperative control model, a light spot offset real-time feedback mechanism and a self-adaptive parameter adjusting strategy are creatively adopted, and the optical collimation system is optimized. Alignment and collimation of laser beams are achieved, a geometric parametric modeling method based on a reflector is developed, related offset can be obtained through theoretical calculation without additional detection through geometric parametric modeling, the adjustment precision is improved to the micron order, and the operation complexity is remarkably reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of laser processing, and particularly relates to an intelligent laser collimation system. Background Art

[0002] In current optical systems, laser beam collimation mainly relies on manually adjusting the mirror mechanism, and fine angle adjustment is achieved through mechanical knobs. Such systems have the following inherent defects: 1. The adjustment accuracy is limited by manual operation, resulting in insufficient adjustment accuracy, usually only reaching millimeter-level accuracy; 2. Repeated trial-and-error adjustments are required, with low efficiency and dependence on the operator's experience, and poor repeatability; 3. There is a lack of real-time feedback mechanism, unable to dynamically compensate for environmental disturbances, and it is difficult to achieve automatic closed-loop control; 4. Whether the beam is collimated depends on direct human eye observation of the light spot, and the result is easily affected by subjective factors. Therefore, those skilled in the art are committed to researching an intelligent optical system that can collimate the beam more objectively, accurately, and intelligently.

[0003] The patent with the publication number CN115203906A discloses a method for constructing an automatic collimation mathematical model for multi-optical paths and multi-optical targets. In the derivation process of the mathematical model, a multi-dimensional vector operation method is adopted. The parameter information of each optical path beam is obtained through a CCD camera and regarded as a unit of a multi-dimensional vector. That is, the parameter information of 8 optical path beams is regarded as an 8-dimensional column vector, which participates in the operations of the collimation target position, pixel offset, step deviation, collimation error, and collimation state summary of all optical paths to construct an automatic collimation mathematical model for multi-optical paths and multi-targets. However, it depends on the multi-channel data acquisition of the CCD camera and a complex mathematical model, with high complexity, and does not solve the problem of high-precision adjustment of a single optical path.

[0004] The patent with the publication number CN109542145B discloses an automatic collimation method for the optical path of a high-power laser device. A near-field CCD camera is used to obtain the original near-field image, and then the obtained original near-field image is subjected to fast Fourier transform to obtain a two-dimensional power spectral density image containing far-field information. The two-dimensional power spectral density image is processed to obtain the ellipticity parameter of the two-dimensional power spectral density image. It is judged whether the circularity rate ρ of the two-dimensional power spectral density image is within the ε neighborhood of 1. The near-field and far-field electric mirror frames are adjusted respectively through near-field closed-loop control and far-field closed-loop control until the circularity rate ρ of the two-dimensional power spectral density image is within the ε neighborhood of 1, thereby completing the automatic collimation of the optical path. It uses Fourier transform to analyze the far-field ellipticity, requires multi-closed-loop control, has high system complexity, limited adjustment speed, and insufficient real-time performance.

[0005] The above-mentioned solutions do not solve the core problems of the laser collimation system such as high precision, low complexity, and intelligent collaborative control. Summary of the Invention

[0006] To solve the technical defects existing in the prior art, the present invention provides an intelligent laser collimation system. It adopts a cage-type integrated structure design of "double mirrors + coaxial connecting rods" to achieve rigid optical path constraint and rapid alignment. It develops a dual-axis precision automatic adjustment mechanism for the mirror based on a micro electric actuator, and for the first time applies the SAC reinforcement learning algorithm to the optical collimation system, proposes a dynamic reward function design based on the spot offset, creatively adopts a cooperative control model of double mirrors, a real-time feedback mechanism of spot offset and an adaptive parameter adjustment strategy to achieve the alignment and collimation of the laser beam, and develops a geometric parametric modeling method based on the mirror, so that relevant offsets can be obtained through theoretical calculation without additional detection based on geometric parametric modeling.

[0007] To achieve the above object, the present invention adopts the following technical solutions: The present invention provides an intelligent laser collimation system, including an optical platform, a laser emitter, an adjustable mirror mount, a first mirror mount, a second mirror mount, a CCD camera, a first coaxial mounting plate, a second coaxial mounting plate, a plurality of coaxially arranged connecting rods and a control module; The laser emitter is mounted on the adjustable mirror mount located on the optical platform. Two coaxially arranged mounting plates on the optical platform are arranged opposite to each other and are connected by a plurality of coaxially arranged connecting rods. The CCD camera is mounted on the first coaxial mounting plate far from the laser emitter. Diaphragms are respectively mounted on the first coaxial mounting plate and the second coaxial mounting plate. The first mirror mount and the second mirror mount are mounted on the optical platform and are sequentially arranged between the adjustable mirror mount and the second coaxial mounting plate. A first mirror and a second mirror are respectively mounted on the first mirror mount and the second mirror mount. The first mirror, the second mirror, the two diaphragms (the aperture of the diaphragm is adjustable from 0.8 mm to 18 mm) and the CCD camera are coaxially arranged; Two adjusting knobs are arranged on the two mirror mounts and are diagonally distributed along the circumferential direction. The ends of the two adjusting knobs are connected to the back of the corresponding mirror, and electric actuators are respectively mounted on the adjusting knobs. Through the cooperation of the electric actuator and the adjusting knob, the mirror can rotate relative to the bottom surface and the side surface for angle adjustment; The collimated laser beam emitted by the laser emitter is reflected twice by the mirrors on the two mirror mounts, passes through the diaphragms on the second coaxial mounting plate and the first coaxial mounting plate, and is incident on the receiving surface of the CCD camera. The control module calculates the theoretical laser spot offset, drives the electric actuator to correspondingly adjust the angles of the two mirrors, and combines the real-time spot image information of the laser beam collected by the CCD camera to finally adjust the center of the laser spot to align with the center of the receiving surface of the CCD camera.

[0008] Further, the control module first calculates the theoretical laser spot offset through geometric modeling to optimize the initial adjustment strategy, and then dynamically corrects the actions of the electric actuator through the SAC deep reinforcement learning algorithm to drive the electric actuator to adjust the mirror angle. The CCD camera is used to collect the real-time spot image information of the laser beam and detect the laser spot offset in real time. Finally, the center of the laser spot is aligned with the center of the receiving surface of the CCD camera, and the laser spot is centered in the collected image.

[0009] Further, the offset of the theoretical laser spot on the target surface of the CCD camera is calculated through the geometric relationships in the following geometric modeling : ; where h 1 is the center distance between the two mirrors, h 2 is the distance from the second mirror to the receiving surface of the CCD camera, the rotation angle of the first mirror is , and the rotation angle of the second mirror is .

[0010] Further, the offset can simultaneously reflect the offset conditions of the laser beam in the horizontal and vertical directions. The rotation angle of the first mirror and the rotation angle of the second mirror can respectively represent the rotation angles of the mirror relative to the bottom surface and the side surface.

[0011] Further, a three-dimensional coordinate system is established based on the initial state of the mirror frame. O: (0, 0, 0) is the fixed reference point of the mirror frame base surface, A: (X1, 0, ) and B: (X2, , 0) respectively correspond to the positions after adjusting the upper and lower adjustment knobs. The parameter represents the side length of the mirror. After being adjusted by the electric actuator, the mirror surface AOB forms an angle θ1 with the bottom surface XOY, and the laser beam will exit from the mirror surface at the reflection angle (π / 2 - θ1). In this case, = π / 2 - θ1, = π / 2 - θ1; the mirror surface AOB forms an angle θ2 with the side surface XOZ, and the outgoing light will exit at the reflection angle θ2. In this case, = -π / 4 + θ2, = -π / 4 + θ2.

[0012] Further, the angle θ1 between the plane AOB and the bottom surface XOY is: , and the angle θ2 between the plane AOB and the side surface XOZ: .

[0013] Further, the adjustment ranges of X1 and X2 are [-4 mm, 13 mm]. The influence of X2 on θ1 is much smaller than that of X1 on θ1, and the influence of X1 on θ2 is much smaller than that of X2 on θ2.

[0014] Further, the control module adopts the following optimized initial adjustment strategy: in the initial control stage, the electric actuators on the two mirror mounts are preferentially adjusted to make X1 = 0; ensure that the two mirrors are always perpendicular to the XOY plane to keep the vertical position of the light beam stable.

[0015] Further, during the training process of dynamically correcting the actions of the electric actuators through the SAC deep reinforcement learning algorithm, the collimation evaluation index is the offset distance between the center of the laser spot and the geometric center of the CCD camera target surface , and the reward function is R = -10 , the state space corresponds to the telescopic amounts X1 and X2 of the two mirror mounts, the action space represents the dimension of the output action, corresponding to the forward or backward movement of the electric actuator, and the minimum adjustment step corresponds to the movement amount of 0.1 mm of the electric actuator.

[0016] Further, after loading the trained model, the control module adjusts the electric actuator to adjust the telescopic amount of the mirror mount until the offset between the center of the laser spot and the center of the receiving surface of the CCD camera meets the laser beam collimation requirements.

[0017] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention adopts a cage-type integrated structure design of "double mirrors + coaxial connecting rods" to achieve rigid optical path constraints and rapid alignment, ensuring a high degree of consistency between the theoretical model and the actual physical system, and developing a dual-axis precision automatic adjustment mechanism for mirrors based on micro electric actuators (2-μm resolution); (2) The SAC reinforcement learning algorithm is applied to the optical collimation system for the first time, a dynamic reward function design based on the spot offset is proposed, and a dual-mirror cooperative control model, a real-time feedback mechanism for spot offset and an adaptive parameter adjustment strategy are creatively adopted to achieve the alignment and collimation of the laser beam; (3) A geometric parameterization modeling method based on mirrors is developed to simplify the complex optical path relationship into a computable mathematical expression, form a closed-loop optimization in combination with real-time feedback, and the relevant offset can be obtained through theoretical calculation without additional detection through geometric parameterization modeling; (4) The present invention uses fewer devices and has a simple structure. The mathematical modeling of the offset is easy to calculate and modify, improving the efficiency. The collimation control accuracy depends on the electric actuator. Due to the high accuracy of the electric actuator, the offset of the light adjustment can be almost ignored. Moreover, the mechanical structure and control algorithm adopted by the present invention are easy to debug for non-professionals, and the starting difficulty is low. Brief Description of the Drawings

[0018] The present invention will be further described in detail below in conjunction with the drawings and specific embodiments.

[0019] Figure 1 It is a structural diagram of the intelligent laser collimation system in the present invention; Figure 2 It is an optical path diagram of the intelligent laser collimation system in the present invention; Figure 3 It is a structural diagram of the first reflecting mirror for angle adjustment in the present invention; Figure 4 It is a three-dimensional coordinate system established based on the initial state of the mirror holder in the present invention; Figure 5 It is an optical path diagram of the outgoing light when the reflecting surface AOB of the present invention forms an angle θ1 with the bottom surface XOY; Figure 6 It is an optical path diagram of the outgoing light when the reflecting surface AOB of the present invention forms an angle θ2 with the side surface XOZ; Figure 7 It is a control diagram of the intelligent laser collimation system in the present invention; Figure 8 It is the variation relationship of θ1 with X1 and X2 in the present invention; Figure 9 It is the variation relationship of θ2 with X1 and X2 in the present invention; Figure 10 It is the specific training process of the training model in the present invention; Figure 11 It is the variation of the reward during the training of the SAC algorithm under 1000 sequences in the present invention; Figure 12 It is the movement amount after loading the model in the present invention and the variation of the telescopic amount of the mirror holder with time; Figure 13 It is the adjusted spot image captured by the coaxial CCD camera in the present invention; Among them, the specific reference numerals are: Optical platform 1, laser emitter 2, adjustable mirror holder 3, first reflecting mirror holder 4, first reflecting mirror 5, first electric actuator 6, second electric actuator 7, second reflecting mirror holder 8, second reflecting mirror 9, third electric actuator 10, fourth electric actuator 11, CCD camera 12, first coaxial mounting plate 13, second coaxial mounting plate 14, connecting rod 15, control module 16. Detailed Description of the Preferred Embodiment

[0020] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0021] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

[0022] An embodiment of the present invention discloses an intelligent laser collimation system. During the optical path debugging and calibration of a laser device, the intelligent laser collimation system ensures that the laser still maintains high-precision collimated output after long-distance propagation by automatically regulating the optical elements in the automatic control system. The system integrates geometric optical modeling, algorithm real-time optimization, and CCD visual feedback, significantly improving the debugging efficiency. As Figure 1 shown, the intelligent laser collimation system includes an optical platform 1, a laser emitter 2, an adjustable mirror mount 3, a first mirror mount 4, a second mirror mount 8, a CCD camera 12, a 30-mm first coaxial mounting plate 13, a 30-mm second coaxial mounting plate 14, a plurality of coaxially arranged connecting rods 15, and a control module 16. The 30 mm refers to the distance between adjacent through holes in the coaxial system; The laser emitter 2 is installed on the adjustable mirror mount 3 located on the optical bench 1. As an alternative light source for the optical path collimation system, the laser emitter 2 uses a small laser in the low-power visible light band, with a divergence angle less than 5 mrad, adjustable light intensity, and is safe and reliable. This design not only avoids overexposure or damage of the CCD camera due to strong light, but also reduces the risk of eye damage, facilitating flexible adjustment of the optical path parameters during debugging. Two coaxially installed plates on the optical bench 1 are arranged opposite to each other and are connected by a plurality of coaxially arranged connecting rods 15. The CCD camera 12 is installed on the first coaxially installed plate 13 far from the laser emitter 2. Diaphragms are installed on both the first coaxially installed plate 13 and the second coaxially installed plate 14. Since the first coaxially installed plate 13 and the second coaxially installed plate 14 are rigidly connected by a plurality of coaxially arranged connecting rods 15 to form a cage structure, this structure can accurately limit the imaging position of the light beam. Because the coaxial system is connected by the coaxial connecting rods 15, two points determine a straight line. If the light beam cannot pass through the diaphragms on these two coaxially installed plates simultaneously, it is regarded as the light beam being uncollimated. Therefore, the angle of the reflector can be adjusted so that the light beam passes through both diaphragms simultaneously and reaches the receiving surface of the CCD camera 12. Here, the diaphragm has the function of accurately limiting the imaging position of the light beam. The first mirror mount 4 and the second mirror mount 8 are installed on the optical bench 1 and are sequentially arranged between the adjustable mirror mount 3 and the second coaxially installed plate 14. The first mirror 5 and the second mirror 9 are respectively installed on the first mirror mount 4 and the second mirror mount 8. The first mirror 5, the second mirror 9, the two diaphragms, and the CCD camera 12 are coaxially arranged; Two adjusting knobs are respectively arranged on the two mirror mounts and are diagonally distributed along the circumferential direction. Electric actuators are respectively installed on the adjusting knobs. Through the cooperation of the electric actuators and the adjusting knobs, the mirrors can rotate relative to the bottom surface and the side surface for angle adjustment; specifically, a first electric actuator 6 and a second electric actuator 7 are installed on the first mirror mount 4, and a third electric actuator 10 and a fourth electric actuator 11 are installed on the second mirror mount 8, replacing the original manual adjusting knobs and realizing the electric control of the mirrors.

[0023] When the system works, the collimated laser beam emitted by the laser emitter 2 is reflected twice by the mirrors on the two mirror mounts, passes through the diaphragms on the two coaxially installed plates, and is incident on the receiving surface of the CCD camera 12. The control module 16 calculates the offset of the theoretical laser spot, drives the four electric actuators to finely adjust the angles of the two mirrors respectively, and combines the real-time spot image information of the laser beam collected by the CCD camera 12 to finally adjust the center of the laser spot to be aligned with the center of the receiving surface of the CCD camera 12.

[0024] As Figure 2As shown, the laser beam emitted by the laser emitter 2 irradiates the first mirror 5 at an incident angle β1, and the rotation angle of the first mirror 5 is , after reflection, the laser beam exits at a reflection angle β1 and irradiates the second mirror 9 at an incident angle β2. The center distance between the two mirrors is h 1, and the rotation angle of the second mirror 9 is , and finally the reflected beam irradiates the imaging surface of the CCD camera 12 at a reflection angle β1. According to the Figure 3 geometric relationship shown, the following relational expressions can be easily obtained: + β1 = π / 2; 2β1 = 2β2 + β4; 2β1 + β3 = π / 2; β2 + β4 + = π / 2. Solving the above system of relational expressions gives: β1 = π / 2 - ; β2 = π / 2 - 2 + ; β3 = -π / 2 + 2 ; β4 = 2 - 2 ( > , counterclockwise; < , clockwise). Since the electric actuator can precisely control and , all angle parameters can be determined. Based on these angle values, the offset of the emitted beam on the receiving plane can be calculated through geometric relationships: ; where h 1 is the center distance between the two mirrors, and h 2 is the distance from the second mirror 9 to the receiving surface of the CCD camera 12. Since all components are rigidly mounted on the optical platform 1, h 1 and h 2 are fixed values. Figure 3 The rotation angle of the first mirror 5 and the rotation angle of the second mirror 9 shown in can respectively represent the rotation angles of the mirrors relative to the bottom surface and the side surface. Therefore, the offset

[0025] As shown in Figure 3 , this figure shows the Figure 1 structure of the adjustable mirror mount for mounting the mirror. The mirror mount can adjust the angle of the mirror by adjusting in the up and down two axial directions. Taking the initial state of the mirror mount as the reference, a three-dimensional coordinate system is established, as shown in Figure 4 . O: (0, 0, 0) is the fixed reference point of the base surface of the mirror mount, A: (X1, 0, ), and B:(X2, , 0) respectively correspond to the positions after adjusting the upper and lower adjustment knobs. The parameter represents the side length of the mirror. In this system, = 44mm. X1 and X2 represent the displacement amounts adjusted by the threaded adjustment knobs, and the adjustment ranges are both [-4mm, 13mm]. From the coordinates, the vectors and can be obtained. Therefore, the normal vector of the plane AOB , and the equation of the plane AOB is: . The normal vector of the plane XOY , and the normal vector of the plane XOZ . Therefore, the angle θ1 between the plane AOB and XOY is: , and the angle θ2 between the plane AOB and XOZ is: . As Figure 5 and Figure 6 show, after being adjusted by the electric actuator, the reflecting surface AOB of the mirror forms an angle θ1 with the bottom surface XOY, and the laser beam exits from the reflecting surface at the reflection angle (π / 2 - θ1). In this case, = π / 2 - θ1, = π / 2 - θ1. The reflecting surface AOB of the mirror forms an angle θ2 with the side surface XOZ, and the outgoing light will exit at the reflection angle θ2. In this case, = -π / 4 + θ2, = -π / 4 + θ2.

[0026] The control flow of the intelligent laser collimation system is as Figure 7 shown: The control module 16 (industrial control computer) drives the electric actuator by sending control commands through the electric actuator controller to achieve precise fine-tuning of the angles of the two mirrors. At the same time, the CCD camera 12 collects the laser spot image in real time and performs image processing. Finally, the collimated beam output by the laser source is accurately aligned with the center of the coaxial cage system.

[0027] As Figure 8 and Figure 9 show, the figure shows the variation relationships of θ1 and θ2 with X1 and X2. It can be obtained that the angle ranges of θ1 and θ2 are between (73.6039°, 95.1732°), that is, the radian is between (1.2846, 1.6609). From Figure 8 and Figure 9 , it can be known that the influence of X2 on θ1 is much smaller than the influence of X1 on θ1, and the influence of X1 on θ2 is much smaller than the influence of X2 on θ2. Based on the above characteristics, the control module 16 adopts the following optimization strategy, and the training process is as Figure 10 shown: 1. Initial control stage: Prioritize adjusting the second electric actuator 7 and the fourth electric actuator 11 to make X1 = 0; ensure that the two mirrors are always perpendicular to the XOY plane and keep the vertical position of the light beam stable.

[0028] 2. Control algorithm: (1) Adopt the SAC (Soft Actor-Critic) reinforcement learning algorithm; (2) Collimation evaluation index: The offset distance between the center of the laser spot and the geometric center of the CCD target surface ; (3) Reward function: R = -10 (The smaller the offset, the higher the reward); (4) Action space: [-1, 1], corresponding to the forward / backward movement of the electric actuator; (5) Minimum adjustment step size: 0.1 mm.

[0029] Since the SAC algorithm is an off-policy algorithm and the training and update of the algorithm can reuse past training samples, the algorithm adopts the method of experience replay. The specific approach is to set up a replay buffer, store the quadruple data (state, action, reward, next state) sampled from the environment each time into the replay buffer, and then randomly sample several data from the replay buffer for training during training. This makes the samples independent of each other and can improve the sample utilization efficiency.

[0030] The specific process of training using the SAC algorithm is as follows: First, initialize the value network's and with randomized parameters ω1, ω2, and θ, and initialize the target network in the way of with , and . Then initialize the experience replay pool. In each round of the sequence, obtain the initial state s1 of the environment. During the time steps from 1 to T, select the action according to the current policy π, obtain the quadruple and store it in the experience replay pool. During sampling, sample the quadruple from the experience replay pool in the form of a minimum batch. For each tuple, calculate the target Q value using the target network: , where is the reward under the current state-action, is the discount factor, is the predicted value of the target Q network, is the entropy, is the policy selects the action at the next state The logarithmic probability. Then, for the two Critic networks, the loss function is minimized (j = 1, 2) and updated. The action is sampled using the reparameterization trick , and the Actor network is updated through the loss function after mathematical simplification . After that, α and the target networks and are updated. Finally, the loop operation is performed until the time step T. After the final training is completed, the model containing the two adjustment parameter variables corresponding to the first electric actuator 6 and the third electric actuator 10 is loaded, and the control system can control these two electric actuators to achieve precise alignment and collimation of the light beam.

[0031] The specific hyperparameters used in the training are as follows: Among them, the learning rate of the policy network (Actor) = 0.002, the learning rate of the value network (Critic) = 0.002, and the learning rate of the entropy regularization weight (Alpha) = 0.0003, which are used to update the policy network, the value network, and the learning rate weight for controlling entropy regularization respectively. The target entropy = -2. The larger the target entropy, the more tolerant it is to the uncertainty of the environment and the more exploration behaviors are encouraged. On the contrary, the smaller the target entropy, the more the algorithm tends to exploit in the known area. The discount factor is set to 0.99, which is used to balance the importance of the current reward and the future reward. The soft update coefficient is set to 0.005, which is used to control the update speed of the main network and the target network. The sequence step size = 30. Each training sequence contains 30 time steps. After completion, the environment is reset and a new sequence starts. The maximum capacity of the experience replay buffer = 10000. The batch size sampled each time during training = 64. The hidden layer dimension = 256. The state space = [2 1] represents the dimension of the input state. The state space consists of the telescopic amounts of the mirror mounts controlled by the first electric actuator 6 and the third electric actuator 10, and the range is between [-4mm, 13mm]. The initial state is set to [5mm, 5mm]. The electric actuator provides a minimum step size of 2μm within the entire stroke range. The action space = [2 1] represents the dimension of the output action. The ranges of the two actions are both between [-1, 1]. The label -1 represents stepping backward by 0.1mm, and the label 1 represents stepping forward by 0.1mm. The minimum adjustment step size of the algorithm is 0.1mm, and the adjustment accuracy of the electric actuator is 2μm. Each control command output to the electric actuator is within the range of stepping backward 0.1mm and stepping forward 0.1mm. As Figure 11 shown, it is the change of the training reward of the SAC algorithm under 1000 sequences. In Figure 10 , h1 = 25mm, h2 = 75mm. It can be seen from Figure 11 that as the training process progresses, the reward function converges upward continuously and the training is completed within 1000 sequences. AsFigure 12 As shown, after loading the trained model, the algorithm adjusts the first electric actuator 6 and the third electric actuator 10 to gradually adjust the telescopic amount of the mirror mount from the initial state [5mm, 5mm] to [7.289mm, 4.895mm]. The final spot center offset = 0.0295mm (error about 10μm), meeting the collimation requirements. As Figure 13 shown, the adjusted spot image captured by the coaxial CCD camera 12 shows that the laser beam has achieved high-precision aligned and collimated output.

Claims

1. An intelligent laser collimation system, characterized in that, It includes an optical platform, a laser emitter, an adjustable mirror mount, a first mirror mount, a second mirror mount, a CCD camera, a first coaxial mounting plate, a second coaxial mounting plate, a plurality of coaxially arranged connecting rods, and a control module; The laser emitter is installed on the adjustable mirror mount located on the optical platform. Two coaxially mounting plates located on the optical platform are arranged facing each other, and are connected by a plurality of coaxially arranged connecting rods. The CCD camera is installed on the first coaxial mounting plate far from the laser emitter. Diaphragms are respectively installed on the first coaxial mounting plate and the second coaxial mounting plate. The first mirror mount and the second mirror mount are installed on the optical platform, and are sequentially arranged between the adjustable mirror mount and the second coaxial mounting plate. A first mirror and a second mirror are respectively installed on the first mirror mount and the second mirror mount. The first mirror, the second mirror, the two diaphragms, and the CCD camera are coaxially arranged; Two adjusting knobs are arranged on the two mirror mounts and are diagonally distributed along the circumferential direction, and electric actuators are respectively installed on the adjusting knobs. Through the cooperation of the electric actuators and the adjusting knobs, the mirrors can rotate relative to the bottom surface and the side surface for angle adjustment; The collimated laser beam emitted by the laser emitter is reflected twice by the mirrors on the two mirror mounts, passes through the diaphragms on the second coaxial mounting plate and the first coaxial mounting plate, and is incident on the receiving surface of the CCD camera. The control module calculates the theoretical laser spot offset, drives the electric actuators to correspondingly adjust the angles of the two mirrors, and combines the real-time spot image information of the laser beam collected by the CCD camera to finally adjust the center of the laser spot to align with the center of the receiving surface of the CCD camera.

2. The intelligent laser collimation system according to claim 1, wherein The control module first calculates the theoretical laser spot offset through geometric modeling to optimize the initial adjustment strategy, and then dynamically corrects the actions of the electric actuators through the SAC deep reinforcement learning algorithm, drives the electric actuators to adjust the mirror angles. The CCD camera is used to collect the real-time spot image information of the laser beam and detect the laser spot offset in real time. Finally, the center of the laser spot aligns with the center of the receiving surface of the CCD camera, and the laser spot is located at the center of the collected image.

3. The intelligent laser collimation system according to claim 2, wherein, Calculate the offset of the theoretical laser spot on the target surface of the CCD camera through the geometric relationships in the following geometric modeling : ； Among them, h 1 is the center distance between the two reflectors, h 2 is the distance from the second reflector to the receiving surface of the CCD camera, and the rotation angle of the first reflector is , and the rotation angle of the second reflector is .

4. The intelligent laser collimation system according to claim 3, wherein Offset can simultaneously reflect the offset conditions of the laser beam in the transverse and longitudinal directions, and the rotation angle of the first mirror and the rotation angle of the second mirror can respectively represent the rotation angles of the mirrors relative to the bottom surface and the side surface.

5. The intelligent laser collimation system according to claim 4, characterized in that, A three-dimensional coordinate system is established with the initial state of the mirror frame as the reference. O: (0, 0, 0) is the fixed reference point of the mirror frame base surface. A: (X1, 0, ), and B: (X2, , 0) respectively correspond to the positions after adjusting the upper and lower two adjustment knobs. The parameter represents the side length of the mirror. After being adjusted by the electric actuator, the mirror surface AOB forms an angle θ1 with the bottom surface XOY, and the laser beam will be emitted from the mirror surface at the reflection angle (π / 2 - θ1). In this case, = π / 2 - θ1, = π / 2 - θ1; the mirror surface AOB forms an angle θ2 with the side surface XOZ, and the outgoing light will be emitted at the reflection angle θ2. In this case, = -π / 4 + θ2, = -π / 4 + θ2.

6. The intelligent laser collimation system according to claim 5, characterized in that, The included angle θ1 between the plane AOB and the bottom surface XOY is: , and the included angle θ2 between the plane AOB and the side surface XOZ: .

7. The intelligent laser collimation system according to claim 6, wherein, The adjustment ranges of X1 and X2 are [-4mm, 13mm].

8. The intelligent laser collimation system according to claim 7, wherein The control module adopts the following initial adjustment optimization strategy: In the initial control stage, the electric actuators on the two mirror mounts are preferentially adjusted to make X1 = 0; ensure that the two mirrors are always perpendicular to the XOY plane and keep the vertical position of the beam stable.

9. The intelligent laser collimation system according to claim 8, characterized in that, During the training process of dynamically correcting the actions of the electric actuator through the SAC deep reinforcement learning algorithm, the collimation evaluation index is the offset distance between the center of the laser spot and the geometric center of the CCD camera target surface , and the reward function is R = -10 , the state space corresponds to the telescopic amounts X1 and X2 of the two mirror mounts, the action space represents the dimension of the output action, corresponding to the forward or backward movement of the electric actuator, and the minimum adjustment step corresponds to a movement of 0.1 mm of the electric actuator.

10. The intelligent laser collimation system according to claim 9, wherein, After loading the trained model, the control module adjusts the electric actuators to adjust the telescopic amount of the mirror mounts until the offset between the center of the laser spot and the center of the receiving surface of the CCD camera meets the laser beam collimation requirements.

Images