A heavy frequency target and light beam intelligent coupling method for laser fusion energy

Abstract

The present application is suitable for the field of laser fusion technology, and provides a heavy frequency target and light beam intelligent coupling method for laser fusion energy, which comprises the following steps: in response to the target pellet being injected into the vacuum target chamber, the three-dimensional coordinates and the velocity vector of the target pellet are obtained through a plurality of sets of orthogonal measurement units arranged on the flight path of the target pellet; the machine learning prediction model deployed in the edge computing unit is used to calculate the arrival time and the spatial position of the target pellet reaching the preset striking point within a first time threshold; the transmission light path of the laser light beam is adjusted within a second time threshold through the piezoelectric ceramic light beam deflection device, so that the optical axis of the laser light beam is directed to the predicted spatial position; the laser light is triggered to emit at the predicted arrival time, so that the emitted laser pulse intersects with the target pellet moving to the preset striking point. The whole process from trajectory tracking to light beam alignment can be completed within a millisecond level period, and the micron level spatial alignment and nanosecond level time synchronization of the high-speed moving target pellet under the condition of high frequency operation can be realized.

Description

Technical Field

[0001] This invention relates to the field of laser fusion technology, and in particular to a method for intelligent coupling of a high-repetition-rate target and a laser beam for laser fusion energy. Background Technology

[0002] In inertial confinement fusion research, high-power laser devices focus multiple laser beams onto a miniature fuel target, triggering a fusion reaction through the instantaneously generated extreme high temperature and pressure conditions. One of the core aspects of this process is ensuring that the laser beam and the target precisely intersect within a micrometer-scale space and a nanosecond-scale time, i.e., achieving beam-target coupling. Traditional technical solutions mainly revolve around the design of statically or slowly injected target pellets. This is achieved by using preset target coordinates, combined with a statically calibrated beam transmission path, to calibrate and adjust the beam direction using an open-loop or low-frequency post-feedback method, thereby completing laser target firing at a fixed position.

[0003] However, when operating at the high repetition rate required for future fusion energy devices, the target pellet needs to be injected dynamically at high speed. Coupling schemes based on static or quasi-static targets have delays and limited accuracy in measuring the target pellet's flight trajectory, failing to provide a real-time and complete state vector. Secondly, the lack of millisecond-level online prediction capability for the future spatiotemporal position of the high-speed moving target pellet means the beam control mechanism's response speed is insufficient to complete precise pointing adjustments in an extremely short time. Consequently, the system cannot complete the control process from trajectory perception and future point prediction to rapid beam pointing within the extremely short cycle required for high repetition rate operation.

[0004] In view of this, a method for intelligent coupling of repetition rate target and beam for laser fusion energy is proposed. Summary of the Invention

[0005] This invention provides a method for intelligent coupling of high-repetition-rate targets and beams for laser fusion energy, which solves the problem that the system cannot complete the control process from trajectory perception and future point prediction to rapid beam orientation within an extremely short period of time that meets the requirements of high repetition rate operation.

[0006] This invention provides a method for intelligent coupling of a high-repetition-rate target and a laser beam for laser fusion energy, comprising a measurement unit, an edge computing unit, a beam deflection device, and a laser under coordinated control, including: In response to the injection of the target pellet into the vacuum target chamber, the three-dimensional coordinates and velocity vector of the target pellet are acquired in real time through multiple sets of orthogonal measurement units arranged along the flight path of the target pellet. Based on the three-dimensional coordinates and velocity vector, the machine learning prediction model deployed on the edge computing unit is used to calculate the predicted arrival time and predicted spatial position of the target pellet at the preset impact point within the first time threshold. Based on the predicted spatial position, the transmission optical path of the laser beam is adjusted within a second time threshold by the piezoelectric ceramic beam deflection device so that the optical axis of the laser beam points to the predicted spatial position. The laser is triggered to emit light at the predicted arrival time, so that the emitted laser pulse intersects with the target that has moved to the preset impact point.

[0007] Furthermore, the multiple sets of orthogonal measurement units are multiple sets of orthogonal light curtain velocity measurement units arranged in segments along the flight path of the target; each set of orthogonal light curtain velocity measurement units includes a line laser and a high-speed linear array detector, and calculates the three-dimensional coordinates and velocity vector of the target by measuring the time difference of the light curtain formed by the line laser when the target flies over it.

[0008] Furthermore, the multiple sets of orthogonal light curtain velocity measurement units are specifically three sets, which are respectively arranged at the first distance, the second distance and the third distance from the target injection point, and are used to measure the flight state of the target in segments.

[0009] Furthermore, the machine learning prediction model adopts a hybrid expert network architecture, which includes a gating network and multiple expert networks. The gating network is used to dynamically calculate and output the weighted weights of each expert network based on the currently input target trajectory parameters. The multiple expert networks are respectively used for specialized modeling and prediction of uniform linear motion, disturbed trajectory deviation and small-amplitude vibration motion during target flight.

[0010] Furthermore, the method also includes a step of performing edge-cloud collaborative online optimization of the hybrid expert network model: The hybrid expert network model is deployed in the edge computing unit, while an enhanced version of the model containing a larger number of expert networks is deployed in the cloud server. Based on the obtained feedback signals, the prediction performance differences between the model version deployed in the edge computing unit and the enhanced model version deployed in the cloud server are compared and analyzed, and the weight allocation strategy of the gating network is dynamically adjusted; wherein, the feedback signal is the deviation between the actual coupling position and the predicted spatial position after each laser target hit; The optimized gating network parameters of the cloud server are sent to the edge computing unit to update the incremental parameters of the model in the edge computing unit.

[0011] Furthermore, the step of calculating the predicted arrival time of the target pellet at the preset impact point within the first time threshold includes: Based on the real-time acquired target velocity vector, the reference time for the target to reach the preset impact point is calculated through kinematic recursion. The online learning algorithm in the edge computing unit generates a real-time calibration item based on the historical deviation between the predicted time and the actual laser trigger time of the most recent target shootings; and the reference time is corrected using the real-time calibration item to obtain the final predicted arrival time.

[0012] Furthermore, calculating the predicted spatial position of the target pellet reaching the preset impact point within the first time threshold includes: Use the theoretical coordinates of the preset strike point as the base position; A spatial deviation prediction value is calculated and generated by an output branch of the machine learning prediction model dedicated to spatial deviation prediction, based on the real-time trajectory parameters of the target and the real-time environmental disturbance parameters obtained from the system sensors. The spatial deviation prediction value is used to characterize the offset of the target relative to the ideal trajectory in three-dimensional space caused by airflow and vibration. The predicted spatial deviation value is superimposed with the base position to obtain the final predicted spatial position.

[0013] Furthermore, the step of adjusting the transmission optical path of the laser beam within the second time threshold using the piezoelectric ceramic beam deflection device includes: The edge computing unit generates control commands based on the predicted spatial position to drive the high-voltage amplifier circuit to control the pose of the six-degree-of-freedom piezoelectric ceramic platform. The position sensor integrated into the piezoelectric ceramic platform monitors the mirror deflection state in real time, and combines the feedforward signal collected by the vibration sensor for feedback control, so as to stably point the optical axis of the laser beam to the predicted spatial position within the second time threshold.

[0014] Furthermore, the method also includes the step of measuring the actual coupling deviation after the laser pulse is triggered: At the instant the laser pulse intersects with the target, two high-speed gated cameras arranged at orthogonal angles are triggered and the pulse backlight is simultaneously lit to capture transient images containing the target and the laser spot from different angles. The transient image is processed by a binocular vision 3D reconstruction algorithm to reconstruct the spatial contours of the target and the laser spot, and the 3D spatial deviation of the laser spot center relative to the target center is calculated as the deviation between the actual coupling position and the predicted spatial position.

[0015] Furthermore, triggering the laser to emit light at the predicted arrival time includes: The edge computing unit sends a trigger preparation command to the laser during at least one laser system response time before the predicted arrival time, so that the laser enters a stable triggerable state. At the predicted arrival time, a trigger pulse signal with timing jitter lower than the system synchronization accuracy requirement is generated and sent to control the laser in the stable triggerable state to emit light precisely.

[0016] As can be seen from the above technical solutions, the present invention has the following advantages: This invention utilizes multiple sets of orthogonal measurement units deployed along the target's flight path to acquire its three-dimensional coordinates and velocity vector in real time. A machine learning prediction model deployed on an edge computing unit calculates the predicted arrival time and spatial position of the target at a preset impact point within a first time threshold. This drives a piezoelectric ceramic beam deflection device to adjust the optical path to this position within a second time threshold. Finally, at the predicted arrival time, a laser is triggered to emit light, achieving precise convergence of the laser pulse and the moving target at the impact point. This invention overcomes the inherent defects of traditional static coupling schemes in dynamic high-speed scenarios, such as measurement delay, prediction gaps, and adjustment lags, through the synergy of real-time sensing, intelligent prediction, rapid control, and precise triggering. It can complete the entire process from trajectory tracking to beam alignment within milliseconds, thus achieving micrometer-level spatial alignment and nanosecond-level time synchronization of high-speed moving targets under high repetition rate operation conditions. This provides crucial technical support for inertial confinement fusion energy devices to transition from single-shot experiments to continuous operation. Attached Figure Description

[0017] Figure 1 This is a system architecture diagram of a special device for a method of intelligent coupling between a high-repetition-rate target and a beam for laser fusion energy, as described in this invention. Figure 2 This is a flowchart illustrating a method for intelligent coupling of a high-repetition-rate target and a laser beam for laser fusion energy, as described in this invention. Figure 3 This is a schematic diagram illustrating the working principle of measuring the flight trajectory of a target in this invention; Figure 4 This is a schematic diagram of the process from receiving trajectory parameters to generating the predicted spatiotemporal location in this invention; Figure 5 This is a schematic diagram illustrating the working principle of the piezoelectric ceramic beam deflection device in this invention. Figure 6 This is a schematic diagram illustrating the working principle of the optical path of the dual-backlight imaging system in this invention. Detailed Implementation

[0018] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “corresponding to,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0019] Example 1 The intelligent coupling method between the high-repetition-rate target and the beam proposed in this invention relies on a dedicated device system integrating real-time sensing, intelligent decision-making, rapid execution, and precise diagnostic functions. For example... Figure 1 As shown, the device system mainly includes a high-repetition-rate target injection module (corresponding to part A in the figure), a target trajectory diagnosis module (corresponding to part B and orthogonal light curtain in the figure), an edge computing and control core (corresponding to FPGA, network and storage in the figure), a beam guidance control module (corresponding to part C and piezoelectric fast reflector in the figure), a laser, a beam-target coupling diagnosis module (corresponding to part D and high-speed camera in the figure), and an edge-cloud collaborative optimization platform (corresponding to edge-cloud collaboration in the figure).

[0020] Specifically, the high-repetition-rate target injection module injects the target pellet into the vacuum target chamber at a high repetition rate. It includes an air gun and a target injection control unit, precisely controlling the injection timing and initial velocity of the target pellet to provide a moving target for subsequent dynamic tracking and coupling. The target pellet flight trajectory diagnosis module measures the state of the moving target pellet in real time and non-contactly. It includes multiple sets of orthogonal light curtain velocity measurement units deployed along the target pellet's flight path. Each light curtain consists of a line laser and a high-speed linear array detector. By measuring the time difference of the obstruction signal generated when the target pellet crosses the light curtain, it calculates and outputs the target pellet's three-dimensional spatial coordinates and velocity vector in real time. The edge computing and control core uses a field-programmable gate array (FPGA) as its core hardware, integrating a high-speed data acquisition interface, dedicated signal processing logic, a lightweight machine learning prediction model engine, control algorithms, and data caching. Its functions are: receiving data from the trajectory diagnosis module in real time; running the prediction model to calculate the predicted arrival time and spatial position of the target pellet in a very short time; generating beam control commands based on the prediction results; and coordinating the synchronization of various parts of the system. The beam guidance and control module is a fast-acting mechanism, its core being a piezoelectric ceramic-driven six-degree-of-freedom fast-reflection mirror system, including a high-voltage drive circuit, a pose monitoring sensor, and a vibration feedforward compensation unit. It receives control commands from the edge computing unit and precisely adjusts the transmission direction of the laser beam in an extremely short time, ensuring the optical axis points in advance to the predicted target position. The laser, as the energy output unit, receives a precise trigger signal from the edge computing unit and accurately emits a high-power laser pulse at the predicted arrival time. The beam-target coupling diagnostic module employs a dual-backlight 3D imaging system, including two high-speed gated cameras arranged at a certain angle, a pulsed backlight, and a long working distance microscope lens. It is synchronously triggered at the moment of laser target impact, capturing transient images of the target and the laser spot, and calculating the spatial deviation of the laser spot center relative to the target center using a binocular vision 3D reconstruction algorithm. The cloud server in the edge-cloud collaborative optimization platform uses historical trajectory data, control commands, and coupled diagnostic results uploaded from the edge side to run optimization algorithms such as reinforcement learning, periodically update or refine the prediction model parameters and control strategies on the edge side, and then distribute the optimized parameters to the edge computing unit to achieve online adaptation and performance improvement of the system.

[0021] like Figure 2 As shown, the steps of the method defined in this invention are described in detail based on this device system: S1. In response to the injection of the target pellet into the vacuum target chamber, the three-dimensional coordinates and velocity vector of the target pellet are acquired in real time through multiple sets of orthogonal measurement units arranged on the flight path of the target pellet. like Figure 3As shown, the multiple sets of orthogonal measurement units are multiple sets of orthogonal light curtain velocity measurement units arranged in segments along the target's flight path. Each set of orthogonal light curtain velocity measurement units includes a line laser and a high-speed linear array detector. By measuring the time difference of the light curtain formed by the line laser as the target flies over it, the three-dimensional coordinates and velocity vector of the target are calculated. Specifically, there are three sets of orthogonal light curtain velocity measurement units, which are respectively arranged at the first, second, and third distances from the target injection point, for segmented measurement of the target's flight state.

[0022] Specifically, each set of orthogonal light curtains consists of two mutually perpendicular light curtains. Each light curtain generates a linear light plane using a linear laser, which is then received by a corresponding high-speed linear array detector. When the target pellet flies through this set of light curtains, it successively blocks the two orthogonal light planes, generating two pulse signals with time intervals. By accurately measuring these two time intervals and combining them with the known spatial positions and spacing of the light curtains, the two-dimensional plane coordinates of the target pellet when it passes through the measurement section can be calculated. Multiple sets of such orthogonal light curtain velocity measurement units, arranged in segments along the flight path, measure the velocity of the target pellet at different locations (different... coordinates) By recording the coordinates and the time of passage through each set of light curtains, the complete three-dimensional spatial coordinates of the target can be obtained in real time through spatial geometry and time difference calculations. and its velocity components in three directions This refers to the velocity vector. The settings of the first, second, and third distances are optimized based on the expected flight velocity range of the target pellet, the overall system response time requirements, and the measurement accuracy requirements. For example, the first distance can be set at approximately 20 cm from the injection point to capture the initial motion state of the target pellet shortly after injection; the second distance can be set at approximately 80 cm to monitor its motion stability during mid-flight; and the third distance can be set at approximately 170 cm, a position close to the preset impact point, to provide the final accurate motion parameters of the target pellet before its arrival. By performing segmented measurements at these three key locations, the system can acquire continuous state information of the target pellet at different flight stages, providing a sufficient data foundation for subsequent accurate predictions.

[0023] S2. Based on three-dimensional coordinates and velocity vectors, using a machine learning prediction model deployed on edge computing units, the predicted arrival time and predicted spatial position of the target pellet at the preset impact point are calculated within the first time threshold. like Figure 4 As shown, the machine learning prediction model adopts a hybrid expert network architecture, which includes a gating network and multiple expert networks. The gating network is used to dynamically calculate and output the weights of each expert network based on the current input target trajectory parameters. The multiple expert networks perform specialized modeling and prediction for uniform linear motion, disturbed trajectory deviation and small-amplitude vibration motion during target flight.

[0024] Specifically, in this network architecture, the gating network receives the real-time trajectory parameters (3D coordinates, velocity vector, and their historical sequence) of the target particle and analyzes them to assess which preset typical mode (smooth, disturbed, or minor vibration) the current flight state is closer to. Based on this assessment, the gating network calculates a set of weight coefficients and assigns them to each expert network. This set of weights determines the proportion of each expert network's opinion in the final prediction result. Each expert network is an independent prediction model, but it is assigned different specializations during the training phase. During training, the expert networks use datasets that focus on different motion characteristics; one expert network focuses on learning the laws of uniform linear motion under ideal conditions; another expert network mainly learns the deviation patterns of the target particle's trajectory under various simulated disturbances; and a third expert network may focus on identifying and predicting high-frequency small-amplitude jitters in the main motion caused by factors such as mechanical vibration. During inference, all expert networks process the same input (current trajectory parameters) in parallel, each outputting a preliminary prediction of future spatiotemporal parameters. The weights generated by the gating network are then used to weight and fuse these initial predictions, resulting in a predicted arrival time and predicted spatial location that integrates multiple possibilities. This architecture allows the model to maintain its grasp of the main motion trends while also responding sensitively to and compensating for various dynamic disturbances, significantly enhancing the robustness and adaptability of the predictions.

[0025] This embodiment also includes a step of performing edge-cloud collaborative online optimization of the hybrid expert network model: 1. Deploy a hybrid expert network model in edge computing units, while deploying an enhanced version of the model containing a larger number of expert networks in cloud servers; 2. Based on the obtained feedback signals, the prediction performance differences between the model version deployed in the edge computing unit and the enhanced model version deployed in the cloud server are compared and analyzed, and the weight allocation strategy of the gating network is dynamically adjusted; where the feedback signal is the deviation between the actual coupling position and the predicted spatial position after each laser target hit; 3. The optimized gating network parameters from the cloud server are sent to the edge computing unit to update the incremental parameters of the model in the edge computing unit.

[0026] Specifically, a streamlined hybrid expert network model is deployed in a resource-constrained edge computing unit to ensure real-time prediction within the first time threshold. Simultaneously, a reinforced version of the model is deployed on a cloud server. This reinforced version is structurally a superset of the edge model; for example, it may contain more expert networks to cover more detailed motion patterns, or employ deeper and wider network layers to enhance representation capabilities. After each laser target strike, the beam-target coupling diagnostic module provides a feedback signal showing the three-dimensional deviation between the actual coupling position and the system's predicted spatial position. The cloud system re-predicts historical trajectory data using the edge model, while simultaneously using the cloud-based reinforced model to predict the same batch of data. By comparing the closeness of the two predictions to the actual coupling positions, the performance difference between the two model versions can be quantified. Based on this difference analysis, the cloud uses algorithms such as reinforcement learning to dynamically adjust the weight allocation strategy of the gating network in the edge model. Here, the strategy refers to a set of rules on how the gating network should intelligently allocate weights to different expert networks when facing various input patterns. After optimization, the cloud extracts and encapsulates the key parameters related to the optimized gating network weight allocation strategy. These updated parameters constitute an incremental parameter package. This parameter package is sent to the edge computing unit via the communication network. Upon receiving it, the edge unit merges or replaces the original model parameters stored locally, thus completing a model update. This incremental update method involves small data transmission volumes and minimal interference with the edge's operational status, ensuring the system's stability and real-time performance during continuous optimization.

[0027] In this embodiment, the predicted arrival time of the target pellet at the preset impact point is calculated within the first time threshold, including the following sub-steps: 1. Based on the real-time acquired target velocity vector, the reference time for the target to reach the preset impact point is calculated through kinematic recursion; 2. Using the online learning algorithm in the edge computing unit, a real-time calibration term is generated based on the historical deviation between the predicted time and the actual laser trigger time of the most recent target shootings; and the reference time is corrected using the real-time calibration term to obtain the final predicted arrival time.

[0028] Specifically, the first time threshold is determined based on the overall timing requirements of high repetition rate operation. Its upper limit must ensure sufficient response time for subsequent beam adjustment and laser triggering steps, ensuring that the entire control closed loop can be completed within the total cycle. The preset impact point is a fixed spatial coordinate pre-calibrated within the vacuum target chamber. It is the ideal focal point of all laser beams and the desired location where the target pellet is hit. This point is determined through precise optical and mechanical calibration and serves as the spatial reference benchmark for the entire coupling system. The target pellet velocity vector is provided in real time by the orthogonal light curtain velocimetry unit in step S1, and includes a three-dimensional vector of magnitude and direction. This characterizes the instantaneous motion state of the target at the moment of measurement.

[0029] The specific process of calculating the reference time using kinematic recursion is as follows: The system uses the latest measured three-dimensional coordinates of the target pellet as the starting point and its real-time velocity vector as the initial velocity. It also considers known physical constraints, such as neglecting air resistance and primarily considering the effect of gravitational acceleration in an ultra-high vacuum environment. Based on uniformly accelerated linear motion, the system calculates the theoretical time required for the target pellet to travel from the current measurement point to the preset impact point in its current motion state by solving the kinematic equations. This calculated time is the baseline time, a preliminary prediction based on an ideal physical model and instantaneous measurements.

[0030] Deviation history refers to the sequence of time differences between the predicted arrival time and the actual laser trigger time during several recent consecutive firing attempts, recorded by the system. These differences reflect the systematic errors and random fluctuations in the model prediction. The real-time calibration term is a dynamically generated correction calculated by a lightweight online learning algorithm deployed in the edge computing unit. This algorithm takes the deviation history as input, fits the trend and statistical characteristics of the error in real time, and outputs a microsecond-level time correction value for the current prediction. Specifically, the reference time calculated by kinematic recursion is algebraically added to the real-time calibration term dynamically generated by the online learning algorithm. The purpose of this calibration term is to compensate for time deviations caused by system latency, model approximation errors, and unmodeled perturbations, thereby obtaining a more accurate predicted arrival time that has been corrected by real-time feedback.

[0031] In this embodiment, the predicted spatial position of the target pellet arriving at the preset impact point is calculated within the first time threshold, including the following sub-steps: 1. Use the theoretical coordinates of the preset strike point as the base position; 2. Using an output branch in the machine learning prediction model dedicated to spatial deviation prediction, a spatial deviation prediction value is calculated and generated based on the real-time trajectory parameters of the target and the real-time environmental disturbance parameters obtained from the system sensors. The spatial deviation prediction value is used to characterize the offset of the target relative to the ideal trajectory in three-dimensional space caused by airflow and vibration. 3. The spatial deviation prediction value and the base position are superimposed to obtain the final predicted spatial position.

[0032] Specifically, the base position is the theoretical coordinates of the preset impact point, a fixed point in three-dimensional space, serving as the absolute reference for position prediction. The dedicated output branch for spatial deviation prediction is a specific network structure within the hybrid expert network model. During model training, this branch's training objective is not to directly predict absolute coordinates, but rather to specifically learn the offset between the actual target position and the ideal trajectory position (a straight line extending from the measurement point along the velocity vector direction to the impact point). Its network weights are obtained through training on a large amount of data containing perturbation trajectories. Real-time environmental perturbation parameters are acquired in real-time by auxiliary sensors installed in the target chamber or beam path, which may affect the physical quantities of the target's flight. These parameters include, but are not limited to, residual air pressure fluctuations within the target chamber, temperature gradients, and monitored mechanical vibration spectrum characteristics. These parameters are acquired in real-time and input into the prediction model. The spatial deviation prediction value is a three-dimensional vector calculated by the aforementioned dedicated output branch based on the two sets of inputs. The first set of inputs consists of the target's real-time trajectory parameters, providing information about the moving subject; the second set consists of real-time environmental disturbance parameters, providing information about external disturbances. This branch, through its internally learned complex mapping relationships, predicts the most likely spatial deviation of the target under the current motion state and environmental conditions. This deviation is used to characterize the target's deviation from the ideal trajectory caused by factors such as airflow and vibration.

[0033] Finally, the fixed base position (theoretical coordinates) and the predicted dynamic spatial deviation value (offset vector) generated by the model are added together according to their dimensions. In this way, the final prediction result includes both the absolute coordinate reference of the system and intelligent compensation for the current dynamic disturbance, thereby achieving accurate adaptive prediction of the future position of the high-speed moving target.

[0034] S3. Based on the predicted spatial position, the transmission optical path of the laser beam is adjusted within the second time threshold by using a piezoelectric ceramic beam deflection device, so that the optical axis of the laser beam points to the predicted spatial position. The piezoelectric ceramic beam deflection device, a six-degree-of-freedom piezoelectric ceramic platform, performs this control. Utilizing the inverse piezoelectric effect of piezoelectric ceramics, it generates precise displacements at the micrometer or even nanometer scale under a high-voltage electric field, thereby altering the spatial orientation of the reflecting mirror. For example... Figure 5As shown, one specific embodiment of the piezoelectric ceramic beam deflection device includes a piezoelectric adjustment frame and a manual five-dimensional frame. A plane mirror M1 is mounted on the piezoelectric adjustment frame, serving as the core fast-reflection mirror actuator, achieving rapid deflection with nanometer-level precision via a piezoelectric ceramic actuator. A focusing mirror M2 is mounted on the manual five-dimensional frame, primarily used for pre-alignment of the beam transmission path during the initial system setup and adjustment phase. In actual operation, control commands generated by the edge computing unit drive the high-voltage amplifier circuit to control the piezoelectric adjustment frame, causing the plane mirror M1 to produce the required micro-displacement.

[0035] The determination of the second time threshold is related to the overall system's repetition rate cycle. After deducting the time consumed by trajectory measurement and prediction calculation, the remaining time window is the time limit within which the beam adjustment action must be completed. The transmission optical path refers to the optical path of the laser emitted from the laser, passing through a series of optical elements such as mirrors and lenses, and finally reaching the center impact point of the target chamber. Adjusting this optical path essentially involves changing the angle of at least one mirror (fast-reflecting mirror) in the path to alter the final direction of the emitted beam.

[0036] The principle behind aligning the optical axis with the predicted spatial position lies in the geometric transformation of spatial coordinates. Specifically, the edge computing unit calculates the mirror deflection angle (pitch, yaw) required to align the beam's focal point with the predicted spatial position based on the current position of the fast-reflecting mirror. By controlling the piezoelectric ceramic platform to generate corresponding micro-displacements, precise adjustment of the mirror angle can be achieved, thereby guiding the optical axis of the beam in real time to point in advance to the target's impending location, preparing for precise convergence.

[0037] In this embodiment, adjusting the transmission optical path of the laser beam within a second time threshold using a piezoelectric ceramic beam deflection device includes the following steps: 1. The edge computing unit generates control commands based on the predicted spatial location, driving the high-voltage amplifier circuit to control the pose of the six-degree-of-freedom piezoelectric ceramic platform; 2. The position sensor integrated into the piezoelectric ceramic platform monitors the mirror deflection status in real time, and combines the feedforward signal collected by the vibration sensor for feedback control, so as to stabilize the optical axis of the laser beam to the predicted spatial position within the second time threshold.

[0038] The edge computing unit generates corresponding digital control commands based on the calculated required mirror deflection. These commands are sent to a high-voltage amplifier circuit via a high-speed interface. The high-voltage amplifier circuit receives the low-voltage digital or analog command signal and amplifies it to a high voltage sufficient to drive the piezoelectric ceramic platform. The amplified high-voltage signal is then applied to each piezoelectric ceramic actuator of the six-degree-of-freedom piezoelectric ceramic platform. The pose of the six-degree-of-freedom piezoelectric ceramic platform refers to its three translational degrees of freedom in space. The state is defined by three rotational degrees of freedom. By independently controlling the extension and retraction of each actuator, the overall platform pose required to achieve the deflection of the target beam can be precisely combined.

[0039] An integrated pose sensor within the piezoelectric adjustable frame monitors the actual deflection state of the plane mirror M1 in real time. The control unit compares the feedback signal with the target command, calculates the error, and corrects the output in real time through a control algorithm. Furthermore, the system integrates a vibration sensor to collect environmental vibration signals. These vibration signals are pre-injected into the control algorithm as feedforward signals, actively compensating for mirror jitter caused by vibration and significantly improving the stability accuracy of the beam after pointing to the predicted spatial position.

[0040] In this embodiment, the step of measuring the actual coupling deviation after laser pulse triggering is as follows: 1. At the instant the laser pulse intersects with the target, two high-speed gated cameras arranged at orthogonal angles are triggered and the pulse backlight is lit simultaneously to capture transient images containing the target and the laser spot from different angles. 2. Transient images are processed using a binocular vision 3D reconstruction algorithm to reconstruct the spatial contours of the target and laser spot, and the 3D spatial deviation of the laser spot center relative to the target center is calculated as the deviation between the actual coupling position and the predicted spatial position.

[0041] Two high-speed gated cameras are arranged at orthogonal angles on both sides of the target chamber, and are equipped with pulsed backlights. For example... Figure 6 As shown, the system includes backlight 1 and backlight 2. The light, after collimation, illuminates the target uniformly from both sides through an illumination window. Two high-speed gated cameras (camera 1 and camera 2) are aligned with the center of the target chamber at approximately orthogonal angles. When the excitation laser intersects with the target, the pulsed backlight and cameras are simultaneously triggered, capturing clear transient images from two different angles. Due to the extremely short exposure time, the image freezes the physical state at the moment of impact, effectively avoiding motion blur. After acquiring two synchronized transient images, a binocular vision 3D reconstruction algorithm is used to identify and segment the contours of the target and the laser spot in the two images. Then, by matching the corresponding pixels of the same object in the two viewpoints and combining the pre-calibrated intrinsic and extrinsic parameters of the cameras, the actual coordinates of the target surface point cloud and the laser spot center in 3D space are calculated. Through point cloud fitting, the geometric center of the target can be determined. Subsequently, the algorithm calculates the difference vector between the 3D coordinates of the laser spot center and the 3D coordinates of the target center. This is the three-dimensional spatial deviation. This deviation vector reflects the spatial deviation between the actual point of laser impact and the ideal center of the target.

[0042] S4. Trigger the laser to emit light at the predicted arrival time, so that the emitted laser pulse intersects with the target that has moved to the preset impact point.

[0043] To ensure that the laser can be reliably and accurately triggered at the predicted arrival time, this embodiment adopts a hierarchical triggering control strategy, which specifically includes the following sub-steps: 1. The edge computing unit sends a trigger preparation command to the laser at least once before the predicted arrival time, within the laser system response time, so that the laser enters a stable triggerable state. 2. At the predicted arrival time, generate and send a trigger pulse signal with timing jitter lower than the system synchronization accuracy requirement to control the laser in a stable and triggerable state to accurately emit light.

[0044] Specifically, a trigger preparation command is issued within a defined time window prior to the predicted arrival time. This window covers at least the complete time required for the laser to reach a stable operating point from receiving the command, i.e., the laser system response time. This command puts the laser's front-end system into a ready state in advance, thereby eliminating the time uncertainty and delay caused by the step response of its internal electro-optical devices. When the system's global clock reaches the predicted arrival time, the edge computing unit generates a trigger pulse signal with an extremely steep rising edge. At this time, the laser is in a stable triggerable state, and its internal system is in a quasi-steady state, awaiting the final command. After the trigger pulse signal arrives, the switching devices inside the laser are driven, thereby controlling the release of stored energy in the form of laser pulses. In this way, the laser pulse emission time is synchronized with the predicted arrival time with high precision, ensuring that the laser pulse and the target that has just moved to the preset impact point achieve optimal spatiotemporal intersection.

[0045] It is understood that those skilled in the art can combine various implementation methods in the above embodiments under the guidance of the above examples to obtain technical solutions with multiple implementation methods.

[0046] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent coupling of a high-repetition-rate target and a laser beam for laser fusion energy, characterized in that, It is controlled collaboratively by a measurement unit, an edge computing unit, a beam deflection device, and a laser, including: In response to the injection of the target pellet into the vacuum target chamber, the three-dimensional coordinates and velocity vector of the target pellet are acquired in real time through multiple sets of orthogonal measurement units arranged along the flight path of the target pellet. Based on the three-dimensional coordinates and velocity vector, a machine learning prediction model deployed in the edge computing unit is used to calculate the predicted arrival time and predicted spatial position of the target pellet at the preset impact point within a first time threshold. The machine learning prediction model adopts a hybrid expert network model, which includes a gating network and multiple expert networks. The gating network is used to dynamically calculate and output the weighted weights of each expert network based on the currently input target pellet trajectory parameters. The multiple expert networks are respectively used for specialized modeling and prediction of uniform linear motion, disturbed trajectory deviation, and small-amplitude vibration motion during target pellet flight. Based on the predicted spatial position, the transmission optical path of the laser beam is adjusted within a second time threshold by using a piezoelectric ceramic beam deflection device, so that the optical axis of the laser beam points to the predicted spatial position. The laser is triggered to emit light at the predicted arrival time, so that the emitted laser pulse intersects with the target that has moved to the preset impact point.

2. The intelligent coupling method for high-repetition-rate targets and beams for laser fusion energy according to claim 1, characterized in that, The multiple sets of orthogonal measurement units are multiple sets of orthogonal light curtain velocity measurement units arranged in segments along the flight path of the target; each set of orthogonal light curtain velocity measurement units includes a line laser and a high-speed linear array detector. By measuring the time difference when the target passes over and blocks the light curtain formed by the line laser, the three-dimensional coordinates and velocity vector of the target are calculated.

3. The intelligent coupling method for high-repetition-rate targets and beams for laser fusion energy according to claim 2, characterized in that, The multiple sets of orthogonal light curtain velocity measurement units are specifically three sets, which are respectively arranged at the first distance, the second distance and the third distance from the target injection point, and are used to measure the flight state of the target in segments.

4. The intelligent coupling method for high-repetition-rate targets and beams for laser fusion energy according to claim 1, characterized in that, The method further includes a step of performing edge-cloud collaborative online optimization of the hybrid expert network model: The hybrid expert network model is deployed in the edge computing unit, while an enhanced version of the model containing a larger number of expert networks is deployed in the cloud server. Based on the obtained feedback signals, the prediction performance differences between the model version deployed in the edge computing unit and the enhanced model version deployed in the cloud server are compared and analyzed, and the weight allocation strategy of the gating network is dynamically adjusted; wherein, the feedback signal is the deviation between the actual coupling position and the predicted spatial position after each laser target hit; The optimized gating network parameters of the cloud server are sent to the edge computing unit to update the incremental parameters of the model in the edge computing unit.

5. The intelligent coupling method for high-repetition-rate targets and beams for laser fusion energy according to claim 4, characterized in that, The step of calculating the predicted arrival time of the target pellet at the preset impact point within the first time threshold includes: Based on the real-time acquired target velocity vector, the reference time for the target to reach the preset impact point is calculated through kinematic recursion. The online learning algorithm in the edge computing unit generates a real-time calibration item based on the historical deviation between the predicted time and the actual laser trigger time of the most recent target shootings; and the reference time is corrected using the real-time calibration item to obtain the final predicted arrival time.

6. The intelligent coupling method for high-repetition-rate targets and beams for laser fusion energy according to claim 5, characterized in that, The step of calculating the predicted spatial position of the target pellet reaching the preset impact point within the first time threshold includes: Use the theoretical coordinates of the preset strike point as the base position; A spatial deviation prediction value is calculated and generated by an output branch of the machine learning prediction model dedicated to spatial deviation prediction, based on the real-time trajectory parameters of the target and the real-time environmental disturbance parameters obtained from the system sensors. The spatial deviation prediction value is used to characterize the offset of the target relative to the ideal trajectory in three-dimensional space caused by airflow and vibration. The predicted spatial deviation value is superimposed with the base position to obtain the final predicted spatial position.

7. The intelligent coupling method for high-repetition-rate targets and beams for laser fusion energy according to claim 1, characterized in that, The adjustment of the laser beam transmission path by the piezoelectric ceramic beam deflection device within the second time threshold includes: The edge computing unit generates control commands based on the predicted spatial position to drive the high-voltage amplifier circuit to control the pose of the six-degree-of-freedom piezoelectric ceramic platform. The position sensor integrated into the piezoelectric ceramic platform monitors the mirror deflection state in real time, and combines the feedforward signal collected by the vibration sensor for feedback control, so as to stably point the optical axis of the laser beam to the predicted spatial position within the second time threshold.

8. The intelligent coupling method for high-repetition-rate targets and beams for laser fusion energy according to claim 7, characterized in that, The method also includes the step of measuring the actual coupling deviation after laser pulse triggering: At the instant the laser pulse intersects with the target, two high-speed gated cameras arranged at orthogonal angles are triggered and the pulse backlight is simultaneously lit to capture transient images containing the target and the laser spot from different angles. The transient image is processed by a binocular vision 3D reconstruction algorithm to reconstruct the spatial contours of the target and the laser spot, and the 3D spatial deviation of the laser spot center relative to the target center is calculated as the deviation between the actual coupling position and the predicted spatial position.

9. The intelligent coupling method for high-repetition-rate targets and beams for laser fusion energy according to claim 1, characterized in that, The step of triggering the laser to emit light at the predicted arrival time includes: The edge computing unit sends a trigger preparation command to the laser during at least one laser system response time before the predicted arrival time, so that the laser enters a stable triggerable state. At the predicted arrival time, a trigger pulse signal with timing jitter lower than the system synchronization accuracy requirement is generated and sent to control the laser in the stable triggerable state to emit light precisely.

Images