A method and system for low response time control of a laser galvanometer

Abstract

The application relates to the field of laser processing and discloses a control method and system for low response time of a laser galvanometer, which realizes real-time collection and time alignment of galvanometer angle switching and mechanical inertia parameters, fuses the parameters into accurate motion description, filters and denoises the parameters, estimates smooth inertia variation, extracts dynamic characteristics, removes noises, predicts risks, supports online parameter self-adaptation and real-time correction to improve stability, predicts overshoot oscillation risks under fast response, adjusts parameters to generate an optimized electromagnetic driving force sequence if the risks exceed a risk threshold, combines real-time time sequence simulation to verify positioning accuracy and determine a correction coefficient, realizes real-time update of driving force matching logic, such as driving enhancement when inertia increases, generates an accurate stop control signal, executes angle adjustment and monitors oscillation, records processing quality feedback if the oscillation amplitude is lower than an amplitude threshold, and iteratively optimizes simulation parameters and driving force strategies based on the feedback. The application can improve the accuracy of a laser galvanometer during large-angle switching.

Description

Technical Field

[0001] This invention relates to the field of laser processing, and in particular to a control method and system for low response time of laser galvanometers. Background Technology

[0002] Laser galvanometers are core actuators in precision manufacturing, laser cutting, and marking, and their response speed and stopping accuracy directly affect processing quality. With the increasing demands for high-speed scanning and micro-machining, galvanometers must complete large-angle switching and suppress oscillations in an extremely short time, and are significantly affected by mechanical inertia, electromagnetic drives, and external interference.

[0003] Traditional technologies typically employ fixed-parameter control strategies and simple signal processing methods. The mirror angle is acquired via an optical encoder or angle sensor, and control signals are sent to the electromagnetic driver using pulse width modulation (PWM). The controller often relies on pre-calibrated parameters or lookup tables to determine the drive current, or uses classic linear controllers such as PID for closed-loop regulation. To reduce noise and abrupt changes, simple filtering methods such as low-pass filtering and mean filtering are commonly used to smooth the control signals and measurement data.

[0004] Traditional techniques rely primarily on fixed mappings and empirical parameter adjustments, lacking in-depth modeling and prediction of real-time changes in the galvanometer's motion state and mechanical characteristics. This leads to overshoot and oscillations during large-angle or high-frequency switching, neglects the timing of driving force and mechanical coupling, and lacks online adaptive and predictive capabilities. Therefore, existing technologies are prone to path deviations during large-angle laser galvanometer switching, resulting in low accuracy. Summary of the Invention

[0005] This invention provides a control method and system for low response time of laser galvanometers, to solve the problem of lack of precision in existing technologies when switching laser galvanometers at large angles.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a method for controlling the low response time of a laser galvanometer, comprising: The system acquires real-time data on the angle switching of the laser galvanometer and mechanical inertial parameters, aligns the timestamps of each data point, and then fuses the data to obtain a precise description of the current motion state. Based on the precise description of the current motion state, noise during the angle switching process is filtered out to obtain a smoothed estimate of the mechanical inertia variation. The overshoot oscillation risk in a fast response scenario is predicted by the mechanical inertia variation estimate. If the predicted risk exceeds the preset risk threshold, the control parameters are adjusted to generate an optimized electromagnetic driving force sequence. Based on the electromagnetic driving force sequence and combined with the real-time acquired time series data, the dynamic response of the laser galvanometer is simulated and verified to obtain the positioning accuracy index in the simulation results and determine the correction coefficient. Based on the time series data, the driving force matching logic is updated in real time. If the angle switching data indicates an increase in mechanical inertia, the strength of the electromagnetic drive is enhanced, the correction coefficient is updated, and a precise stopping control signal is obtained. The actual angle adjustment of the laser galvanometer is performed through the control signal, and the occurrence of overshoot oscillation is monitored. If the oscillation amplitude is lower than the preset oscillation amplitude threshold, the improvement of dynamic response is confirmed, and the processing quality feedback data is recorded. Based on the feedback data of the processing quality, the parameters used for the simulation verification are iteratively optimized to obtain a more accurate driving force matching strategy for the next rapid response loop processing.

[0007] Secondly, the present invention provides a control system for low response time of a laser galvanometer, comprising: Motion state acquisition module: used to acquire real-time collected angle switching data and mechanical inertial parameters of the laser galvanometer, align the timestamps of each data point, and then perform data fusion to obtain an accurate description of the current motion state; Inertia estimation module: used to filter noise during the angle switching process based on the precise description of the current motion state, and obtain a smoothed estimate of the mechanical inertia variation; Oscillation prediction module: used to predict the risk of overshoot oscillation in a fast response scenario using the estimated value of mechanical inertia variation; if the predicted risk exceeds the preset risk threshold, the control parameters are adjusted to generate an optimized electromagnetic driving force sequence. Simulation verification module: used to simulate and verify the dynamic response of the laser galvanometer based on the electromagnetic driving force sequence and the real-time acquired time series data, to obtain the positioning accuracy index in the simulation results and determine the correction coefficient; Matching update module: Based on the time series data, the driving force matching logic is updated in real time. If the angle switching data indicates an increase in mechanical inertia, the strength of the electromagnetic drive is enhanced, the correction coefficient is updated, and a precise stopping control signal is obtained. The monitoring module is used to perform actual angle adjustment of the laser galvanometer through the control signal, monitor the occurrence of overshoot oscillation, and if the oscillation amplitude is lower than the preset oscillation amplitude threshold, confirm the improvement of dynamic response and record the processing quality feedback data. Strategy optimization module: Iteratively optimizes the parameters used for the simulation verification based on the feedback data of the processing quality, and obtains a more accurate driving force matching strategy for the next rapid response loop processing.

[0008] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention constructs an accurate motion state description based on multi-sensor fusion, collects galvanometer angle, acceleration and mechanical inertia parameters in real time and aligns the timestamps, and uses Kalman filtering and time-series feature extraction to obtain a smooth inertial variation estimate; combined with neural network to predict oscillation risk online and compare it with the abnormal threshold, forming a closed-loop process from data acquisition to risk identification, which significantly improves the accuracy and real-time performance of initial risk identification. (2) The present invention dynamically models the identified risks and constructs a risk propagation and priority assessment mechanism based on oscillation characteristics and time series analysis; verifies the effectiveness of the electromagnetic driving force sequence through Monte Carlo simulation and frequency domain characteristics, calculates the intensity and range of chain oscillations, extracts key hubs and quantifies system stability, thereby accurately predicting the overshoot direction and amplitude and prioritizing the suppression of key node oscillations, improving the anti-disturbance capability and positioning accuracy; (3) The present invention constructs a closed-loop adaptive adjustment mechanism: when the system stability is lower than the stability threshold or the predicted risk exceeds the risk threshold, the optimized electromagnetic driving force sequence is automatically generated and distributed, the driving force matching logic is updated in real time, the back-end monitoring of processing quality feedback is executed, and the simulation parameters and neural network model are iteratively adjusted accordingly, so as to shorten the recovery and stabilization time, improve resource utilization and enhance overall robustness. Attached Figure Description

[0009] Figure 1 This is a schematic flowchart of a control method for low response time of a laser galvanometer provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of a control system structure for low response time of a laser galvanometer provided in the second embodiment of the present invention. Detailed Implementation

[0010] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] Reference Figure 1 The first embodiment of the present invention provides a method for controlling the low response time of a laser galvanometer, comprising the following steps: S101: Acquire the real-time collected angle switching data and mechanical inertial parameters of the laser galvanometer, align the timestamps of each data point, and then perform data fusion to obtain a precise description of the current motion state. S102, Based on the precise description of the current motion state, the noise during the angle switching process is filtered to obtain a smoothed estimate of the mechanical inertia variation. S103, using the estimated mechanical inertia variation, predict the risk of overshoot oscillation in a fast response scenario. If the predicted risk exceeds a preset risk threshold, adjust the control parameters to generate an optimized electromagnetic driving force sequence. S104. Based on the electromagnetic driving force sequence and combined with the real-time acquired time series data, the dynamic response of the laser galvanometer is simulated and verified to obtain the positioning accuracy index in the simulation results and determine the correction coefficient. S105, based on the time series data, update the driving force matching logic in real time. If the angle switching data indicates an increase in mechanical inertia, enhance the strength of the electromagnetic drive, update the correction coefficient, and obtain a precise stop control signal. S106, through the control signal, the actual angle adjustment of the laser galvanometer is executed, the occurrence of overshoot oscillation is monitored, and if the oscillation amplitude is lower than the preset oscillation amplitude threshold, the improvement of dynamic response is confirmed, and the feedback data of processing quality is recorded. S107, based on the feedback data of the processing quality and the positioning accuracy index of the simulation output, a loss function is constructed. Then, based on the model differentiability and computational budget, the gradient method is used to iteratively optimize the parameters used for the simulation verification to obtain a more accurate driving force matching strategy for the next fast response loop processing.

[0012] In step S101, the angle switching data and mechanical inertial parameters of the laser galvanometer are acquired in real time, the timestamps of each data point are aligned, and then the data is fused to obtain a precise description of the current motion state, including: S1011: Acquire laser galvanometer angle data and switching frequency data, collect mechanical inertial parameters, align data timestamps, and obtain the original time series dataset; S1012, If the timestamp deviation of the original dataset is greater than the preset timestamp deviation threshold, then the angle data and inertial parameters of the original dataset are fused to generate corrected motion data. S1013: Extract angular position and velocity information from the corrected motion data to obtain a precise description of the current motion state.

[0013] In step S1011, laser galvanometer angle data and switching frequency data are acquired, mechanical inertial parameters are collected, and the timestamps of the data are aligned to obtain the original time series dataset.

[0014] In one implementation, within the laser galvanometer system, an optical encoder acquires angle values ​​at 1000Hz and calculates the switching frequency (Hz). A high-performance IMU (1000Hz) acquires triaxial acceleration (g) and angular velocity (rad / s) as mechanical inertial parameters. All sensors are time-stamped by a unified master clock (preferably PPS hardware trigger or PTP / IEEE-1588 network clock), with each timestamp deviation required to be no more than 1ms. If the deviation is greater than 1ms, low-sampled or missing data points are aligned using linear or spline interpolation. After alignment, multi-sensor fusion is performed using a discrete Kalman filter with angle and inertial measurements as observations and a state vector of [angle, angular velocity] (the measurement noise and process noise covariance are set according to sensor calibration). Outliers (such as angle mutations greater than 0.017rad or acceleration mutations greater than 0.5g) are removed, and finally, the data is sorted by field. Export the original time series dataset before correction as a CSV or binary table.

[0015] It should be noted that the high-performance IMU (Inertial Measurement Unit) refers to a multi-axis inertial sensor module that can provide high frequency, low noise, and precise time synchronization in high dynamic scenarios such as laser galvanometers. It typically includes a three-axis accelerometer and a three-axis gyroscope, as well as optional magnetometers and temperature sensors. It supports continuous sampling of no less than 1000Hz and bandwidth of several hundred Hz, has low random noise and small zero-bias drift, supports hardware PPS or network time synchronization (PTP / IEEE-1588 or NTP), and can control the timestamp accuracy to no more than 1ms or even better.

[0016] It should be noted that the meaning of each parameter in the original time series dataset before correction is explained, among which... This is the sampling timestamp, in milliseconds, used for time alignment of multiple sensors; It is the instantaneous angle of the galvanometer measured by an optical encoder or angle sensor, in degrees; The current angle scanning frequency represents the galvanometer switching rate, in Hertz; ax, ay, and az are the three-axis linear accelerations output by the IMU, in g (gravitational acceleration); gx, gy, and gz are the three-axis angular velocities output by the IMU (which can be measured by a gyroscope), in radians per second.

[0017] In step S1012, if the timestamp deviation of the original dataset is greater than the preset timestamp deviation threshold, the angle data and inertial parameters of the original dataset are fused to generate corrected motion data.

[0018] In one implementation, after the optical encoder (1000Hz) and IMU (1000Hz) acquire angle and triaxial acceleration or angular velocity in parallel and synchronize with the master clock via PPS hardware triggering or PTP, if the timestamp of either device deviates from the master clock by more than a preset timestamp deviation threshold of 1ms, then on a time axis with a window of ±5ms, missing points are filled in by nearest neighbor, linear, or cubic spline interpolation, and outliers (e.g., angle mutations greater than 0.017rad or acceleration mutations greater than 0.5g) are removed, based on the measurement noise covariance determined by sensor calibration. , With process noise Discrete Kalman filtering is performed, with the state vector set as [angle, angular velocity]. In the prediction-update loop, the timestamps generated by the high-precision, real-time measurement of the galvanometer angle by the optical encoder are fused with the IMU measurements to generate a smooth and corrected motion timing sequence (example field). Export the corrected motion data in CSV or binary format.

[0019] It should be noted that the preset timestamp deviation threshold is based on the 1kHz sampling period (approximately 1ms) of the optical encoder and IMU, and the maximum time offset allowed to ensure that the fusion of nearest neighbor interpolation and Kalman filtering does not introduce significant phase or positioning errors.

[0020] It should be noted that the two noise covariance variables mentioned above represent the variance used in Kalman filtering to characterize the uncertainty of gyroscope angular velocity measurement (e.g., ) and the variance used to characterize the uncertainty of linear acceleration measurements by accelerometers (e.g. ), used to weigh the confidence of each sensor measurement in the prediction-update step.

[0021] It should be noted that process noise models the incompleteness of the system model and the uncertainty of external disturbances, specifically including unmodeled angular acceleration, random walks of gyroscope bias, driving errors, temperature drift, and quantization errors; when the state vector is [angle, angular velocity], the commonly used discrete process noise covariance is... (q is the power spectral density of angular acceleration white noise), and additional terms representing bias walks can be added to the matrix.

[0022] It should be noted that the optical encoder is used to measure the galvanometer angle with high precision and in real time and generate a timestamp (in order to align with the IMU), providing a reliable angle reference for Kalman filtering to fuse inertial data, eliminate anomalies, and for precise stopping and drive force closed-loop adjustment.

[0023] In step S1013, angular position and velocity information are extracted from the corrected motion data to obtain a precise description of the current motion state.

[0024] In one implementation, the Kalman-filtered time series is first subjected to outlier removal (e.g., removing points with angle abrupt changes greater than 0.017 rad, i.e., 1°, and missing markers), and then denoised using Savitzky-Golay filtering. Subsequently, the angular velocity is calculated using the central difference formula, and the calculation results are smoothed by a 5-point moving average. The standard deviation of the velocity is calculated as the uncertainty within a sliding window (e.g., ±5 ms). Finally, the angle (rad), angular velocity (rad / s), uncertainty, and status flag (normal or abnormal) after unit conversion are output in CSV or binary format as an accurate description of the current motion state for subsequent control and decision-making.

[0025] It should be noted that the specific formula for the central difference is as follows: ; in, Indicates the first Angle values ​​at each sampling time; For discrete-time index; Δt is the sampling interval; numerator The denominator represents the sampled values ​​to the left and right of that point; This represents the total distance between the two points on the left and right.

[0026] In step S102, based on the precise description of the current motion state, noise during the angle switching process is filtered to obtain a smoothed estimate of the mechanical inertia variation, including: S1021, If ​​the noise level in the precise description of the motion state is greater than a preset noise threshold, then Kalman filtering is performed to obtain smoothed motion data. S1022, Based on the smoothed motion data, extract the mechanical inertia variation features to obtain the smoothed mechanical inertia variation estimate.

[0027] In step S1021, if the noise level in the precise description of the motion state is greater than a preset noise threshold, Kalman filtering is performed to obtain smoothed motion data.

[0028] In one implementation, the system first calculates a noise metric (such as the standard deviation of the angle) within a sliding window for the formatted and timestamped angle and inertial measurement sequences. If the metric exceeds a preset noise threshold, Kalman filtering is performed as follows: a standard Kalman or extended Kalman filter is selected based on the linearity assumption, and a state vector is constructed. The system motion model is used to determine the state transition and observation matrix. Based on the sensor calibration, the measurement noise covariance R, process noise covariance Q, and initial covariance P0 are set. Missing points are first filled with nearest neighbor interpolation and outliers (such as angle mutations greater than 0.017 rad or acceleration mutations greater than 0.5 g) are removed. Then, at each sampling time, a prediction-update cyclic fusion of optical encoder and IMU measurements is performed and the smoothed angle, angular velocity, and uncertainty are output as smoothed motion data.

[0029] It should be noted that the specific value of the noise threshold should be determined based on the system encoder resolution, installation error, and task accuracy requirements. Specifically, this involves first acquiring raw angle data through static or slow-motion calibration experiments and then calculating its standard deviation. Take the preset threshold as (For example, k=3 indicates a 3σ decision).

[0030] It should be noted that the system motion model adopts continuous second-order rotational dynamics, and the following continuous-time model is given under a small-angle approximation: ; In the formula, θ is the angular displacement; I is the moment of inertia, measured in kg·m², which refers to the equivalent mass distribution around the rotation axis. In common commercial galvanometer systems, it is approximately... b is the viscous damping coefficient; k is the elastic coefficient (k=0 when there is no elasticity); u(t) is the applied torque input.

[0031] Construct the state vector: Then the continuous-time state equation is: ; In this formula, if radians are considered dimensionless, then and The units have the same dimensions; This represents process noise. Zero-order hold (ZOH) discretization (sampling interval Δt) is used to obtain the discrete-time state equation: ; In the formula, , The observation equation is: ; in For observation vectors (e.g., angle measurements) obtained from encoders and IMUs With angular velocity measurement ), can be taken or It depends on the observed item; , These are discrete process noise and measurement noise, respectively, satisfying... , Q and R are the covariance matrices of process noise and measurement noise, respectively.

[0032] It should be noted that for a certain type of MEMS, the combination of IMU and encoder, the measurement noise covariance R can be taken as... Example , The process noise Q can be taken as... (The numerical values ​​need to be adjusted according to the specific system calibration). The initial covariance P0 can be a diagonal matrix. .

[0033] It should be noted that the sampling interval Δt, in the example, is taken as... (Sampling rate 1kHz) is suitable for systems under test where the main frequency components are below 500Hz (meeting the Nyquist condition). To avoid aliasing, a suitable analog front-end low-pass filter should be used before sampling, with its cutoff frequency slightly higher than the highest frequency of the signal under test but lower than half the sampling rate.

[0034] In step S1022, mechanical inertia variation features are extracted based on the smoothed motion data to obtain the smoothed mechanical inertia variation estimate.

[0035] In one implementation, based on the smoothed angle, angular velocity, and acceleration time series obtained by Kalman filtering, the data is first formatted and cleaned as necessary (e.g., outlier removal and timestamp interpolation alignment) to obtain the Kalman-filtered smoothed angle series. (Sampling interval Δt, for example, Δt = 1ms); Central difference is used for points within the sequence. Estimate the angular velocity and use the second-order difference. To estimate angular acceleration, backward difference is used at the sequence endpoints. And corresponding angular acceleration difference processing to obtain mechanical inertia variation characteristics; then further calculate statistics (e.g., fluctuation range, standard deviation). Spectral analysis should clearly define the following steps: apply a window function (e.g., Hanning window) to the smoothed time series to reduce spectral leakage; if higher frequency resolution is needed, zero-filling can be used; the spectral resolution is... ( Sampling frequency, (Number of valid points). The dominant frequency can be determined using peak detection, requiring the signal-to-noise ratio of the peak amplitude relative to the noise baseline. (For example This is used as a criterion for peak validity. Specifically, it can be expressed as: using the sampling rate... Window function Perform an FFT on the signal segment length N and determine the dominant frequency from the amplitude spectrum. The clock frequency is determined to be valid at that time. It can be estimated from the non-peak sections of the spectrum to form mechanical inertial variation characteristics including angular velocity fluctuation range and main frequency information; finally, based on the mechanical inertial variation feature set, the smoothed mechanical inertial variation estimate is output.

[0036] In step S103, the overshoot oscillation risk in a fast-response scenario is predicted using the estimated mechanical inertia variation. If the predicted risk exceeds a preset risk threshold, the control parameters are adjusted to generate an optimized electromagnetic driving force sequence, including: S1031, Based on the smoothed mechanical inertia variation estimate, extract the dynamic features related to oscillation risk to obtain the oscillation risk feature set; S1032, For the oscillation risk feature set, predict the overshoot oscillation risk value under the fast response scenario to obtain the predicted risk value; S1033, if the predicted risk value exceeds the preset risk threshold, the electromagnetic drive parameters are optimized by adjusting the control parameters to generate an optimized electromagnetic drive force sequence.

[0037] In step S1031, dynamic features related to oscillation risk are extracted based on the smoothed mechanical inertia variation estimate to obtain an oscillation risk feature set.

[0038] In one implementation, starting from the smoothed mechanical inertia variation estimate obtained by Kalman filtering, time-domain statistics and differencing are first performed on the equally spaced time series to extract time-domain features such as the standard deviation of angle / angular velocity and the fluctuation range; the frequency domain analysis is implemented as follows. The sampling rate is denoted as... (Example values ​​for example:) ,correspond To avoid aliasing, all frequency thresholds used for judgment must be strictly lower than the Nyquist frequency. (It is recommended to take no more than 0.9 times) (Example value 450Hz). Using an FFT of length N (Example N=2048) in the frequency domain, combined with a window function (Example Hanning window), the frequency resolution is... (Example) The window overlap rate and zero-fill strategy should be given in the embodiments. Peak detection should be based on a joint criterion of peak amplitude, peak prominence, and in-band signal-to-noise ratio (SNR). For example, the identification conditions are: peak amplitude greater than 3 times the mean background noise, peak prominence greater than 0.2 times the peak amplitude, and frequency less than 0.9 times fs / 2. The above parameters are implementation examples; actual applications should be adjusted based on the sensor noise spectrum and calibration experiments. Then, the above time-domain and frequency-domain features are summarized into vectorized feature entries according to predetermined normalization and weighting rules or logical criteria, and necessary labeling or threshold judgment is performed. Finally, an oscillation risk feature set containing angular velocity fluctuations, primary and secondary frequencies, and amplitudes is output for subsequent oscillation risk assessment.

[0039] It should be noted that the temporal feature extraction and terminology conventions are as follows. Angles are expressed in radians. The "estimated value of mechanical inertia variation" refers to the equivalent moment of inertia. or its relative change (Dimensionless), and is part of the Kalman filter state. Example of time-domain features mathematically defined as follows: Angular standard deviation. angular velocity standard deviation peak First-order difference and its standard deviation Example threshold can be taken. (Based on specific machining accuracy requirements) recalibration is required under different equipment or operating conditions.

[0040] In step S1032, for the oscillation risk feature set, the overshoot oscillation risk value under the fast response scenario is predicted to obtain the predicted risk value.

[0041] In one implementation, the Oscillation Risk Prediction (MLP) process is as follows: within the sampling window, if the peak angle or angular velocity exceeds a reference threshold (example...), ... ,Right now If the corresponding processing positioning error exceeds the error threshold X set based on industry production conditions (example X=0.1%), it is labeled as a "overshoot oscillation" positive sample; otherwise, it is a negative sample. The example training set size is N=5000 (approximately 1000 positive samples and 4000 negative samples), using hierarchical 5-fold cross-validation for evaluation. The training data should come from real-world records collected simultaneously by multiple sensors. The training set size can be adjusted according to actual data availability and can be expanded using data augmentation techniques. The example MLP structure is as follows: input dimension d, hidden layer 128 / 64 (ReLU), output unit Sigmoid; training uses binary cross-entropy loss, optimizer Adam, ... , , Stop early Evaluation was conducted using ROC-AUC, recall, precision, and F1 score; results were available on the example validation set. (Example). Threshold The choice should be based on cost-sensitive analysis on the validation set or the Youden index (example). To ensure a recall rate of at least 0.9, online updates employ a controlled fine-tuning strategy: labeled online samples are cached in a local data pool to reach the retraining scale. (Example 2000) After retraining and validating in an offline environment, the online model should be replaced only when the performance is improved or does not decrease. At the same time, a rollback mechanism must be provided to deal with erroneous updates.

[0042] It should be noted that the labeling rules for positive / negative samples are as follows: if the peak angle or angular velocity within the window exceeds the baseline threshold, and the corresponding processing positioning error exceeds the industry-set error threshold X, then the window is labeled as a "overshoot oscillation" positive sample; otherwise, it is labeled as a negative sample. To reduce labeling noise, the peak-error correlation can be required to be within the same event time window and satisfy frequency domain significance (to avoid mislabeling solely due to high-frequency sensor noise). To address class imbalance, hierarchical sampling, class weights, or minority class oversampling / augmentation can be used when constructing the dataset. Before training, features are standardized using unified normalized parameters (based on the training set). Training and validation are evaluated using hierarchical 5-fold cross-validation (metrics include ROC-AUC, recall, precision, and F1). The threshold selection is based on the cost Youden index on the validation set. Online samples are cached with labeled samples according to a controlled strategy and retrained and validated offline after reaching the retraining scale (example 2000). The online model is replaced when performance is improved or maintained without decline and a rollback / grayscale release mechanism is implemented.

[0043] It should be noted that the benchmark determination (such as time-domain threshold) The time and frequency domain features of the filtered angle / angular velocity are extracted, and the sensor noise and resolution, galvanometer material and structural stiffness, and processing conditions are combined with an appropriate safety margin to determine the result.

[0044] In step S1033, if the predicted risk value exceeds a preset risk threshold, the electromagnetic drive parameters are optimized by adjusting the control parameters to generate an optimized electromagnetic drive force sequence.

[0045] In one implementation, multiple sensors first collect angle switching and inertial / oscillation data of the laser galvanometer in parallel, and perform time synchronization and cleaning on the data (e.g., median filtering, outlier replacement). Then, frequency features related to the dynamic response are extracted based on the formatted time series and used as input to output a predicted risk value by the constructed neural network or corresponding prediction model. When the predicted risk value exceeds a preset risk threshold, the information processing unit optimizes the electromagnetic driving force parameters by adjusting the control parameters based on the model output and historical features (e.g., adjusting the current amplitude and driving frequency according to the example to form a new electromagnetic driving force parameter set). Monte Carlo simulation can be used to simulate and verify the candidate driving force sequence, and correction coefficients are calculated using methods such as linear interpolation based on the positioning accuracy index to generate the final optimized electromagnetic driving force sequence. Subsequently, the sequence is sent to the driver in real time and the driving force matching logic is updated for verification and iteration in the closed loop.

[0046] It should be noted that the optimization process is as follows: Explanation of control parameter optimization and Monte Carlo verification implementation. The optimization problem is formalized as follows: Let the control sequence... Solve ; in For risk estimation based on candidate sequences, For positioning error indicators, For energy consumption estimation, These are the weighting coefficients. A fast approximate optimization method can be used to solve this problem; to ensure real-time performance, the prediction step size can be limited (example). For each candidate sequence, perform M Monte Carlo simulations (Example: M=1000). Sample system parameters according to a preset distribution (e.g., J variation ±5%, damping b variation ±10%), and calculate the positioning error statistic (e.g., 95th percentile). If the simulation statistic exceeds the target, calculate the correction coefficient γ proportionally (Example: ...). The candidate sequences are adjusted, and spline or linear interpolation is performed on the adjusted sequences to ensure smooth time axis and meet driver bandwidth limits. Safety checks (limit violation detection and step limits) are performed before deployment, and the response is monitored in real time after deployment, with rollback or switching to safe mode as necessary.

[0047] In step S104, based on the electromagnetic driving force sequence and combined with the real-time acquired time series data, the dynamic response of the laser galvanometer is simulated and verified to obtain the positioning accuracy index in the simulation results, and the correction coefficient is determined, including: S1041: Acquire real-time angle and acceleration data of the laser galvanometer, and perform Kalman filtering to remove noise and outliers, resulting in a formatted time series dataset. S1042, Based on the formatted time series dataset, extract frequency features related to the dynamic response to obtain a dynamic response feature set; S1043, For the dynamic response feature set, the electromagnetic driving force sequence is simulated and verified to obtain a positioning accuracy index set; S1044, if the maximum deviation value in the positioning accuracy index set exceeds the preset deviation threshold, then linear interpolation is performed to adjust the correction coefficients to obtain an optimized correction coefficient set.

[0048] In step S1041, real-time angle and acceleration data of the laser galvanometer are acquired and Kalman filtering is performed to remove noise and outliers, resulting in a formatted time series dataset.

[0049] In one implementation, real-time angle data is first acquired in parallel from the optical encoder and acceleration data from the inertial measurement unit via a multi-sensor interface. The two data streams are synchronized in time (e.g., hardware triggering is used to control the time deviation within 0.5ms) to obtain the original time series. If the noise level or anomalies in the original series exceed a preset anomaly threshold (e.g., the standard deviation of angle switching data is greater than 8.7e-4rad, i.e., 0.05°, or a sudden change point is greater than 0.017rad, or a sudden change in acceleration is greater than 0.5g), the anomalies are first marked and replaced with the mean or weighted average of the nearest points. Then, in the case of high noise, the Kalman filter algorithm is applied to fuse and smooth the angle and inertial data (the prediction-update step is used to suppress measurement noise and correct deviations). Finally, the cleaned and smoothed data is aligned to a unified time reference (e.g., aligned to a 0.05ms or 0.1ms timestamp) using spline interpolation to form a continuous and formatted time series dataset.

[0050] It should be noted that the aforementioned abnormal threshold is based on the system design indicators and sensor specifications as an initial reference. Baseline data is collected under normal operating conditions and statistics (such as mean ± N times standard deviation or extreme value determination) are calculated. This is supplemented by experimental calibration and Monte Carlo simulation verification. The threshold can also be dynamically adjusted during closed-loop operation.

[0051] In step S1042, frequency features related to the dynamic response are extracted based on the formatted time series dataset to obtain a dynamic response feature set.

[0052] In one implementation, based on the formatted time series, firstly, a window function is applied to each data window to be analyzed (a fixed-length window can be selected) and a Fast Fourier Transform (FFT) is performed to convert the time-domain signal into a spectrum. Secondly, feature terms related to the dynamic response are extracted from the spectrum, such as the dominant frequency, secondary frequency, amplitude of the corresponding frequency, harmonic distribution, and bandwidth or energy ratio. Significant components can be screened according to a preset amplitude threshold (in the example, an amplitude threshold of 0.4-0.5 can be used as the discrimination criterion). Finally, the dominant frequency, secondary frequency, amplitude, harmonic ratio, and frequency band energy ratio are organized into a dynamic response feature set in a unified format. Specifically, for each analysis window, the features extracted by FFT are normalized by min-max according to predefined fields (e.g., dominant frequency (Hz), secondary frequency (Hz), corresponding amplitude, harmonic energy ratio, energy ratio of each frequency band, timestamp, sampling rate) to form a unified feature item, thus obtaining the dynamic response feature set.

[0053] It should be noted that the amplitude threshold is initially determined by taking the galvanometer material properties and processing speed as references. It is determined by collecting baseline time series under normal operating conditions and experimentally calibrating the spectrum statistics (or setting an empirical lower limit such as 0.4-0.5), supplemented by Monte Carlo simulation verification, and finally iteratively adjusted in closed-loop operation.

[0054] In step S1043, the electromagnetic driving force sequence is simulated and verified for the dynamic response feature set to obtain a positioning accuracy index set.

[0055] In one implementation, several candidate electromagnetic driving force sequences are first generated based on the feature set (example parameters include current of 1.1A and frequency of 950Hz). Then, Monte Carlo simulation is used, with the key system parameter in the simulation set as the equivalent moment of inertia. Damping coefficient (During Monte Carlo sampling, the distribution is perturbed by ±3% and ±10% of the nominal value to reflect uncertainty.) These parameters are identified through offline systems (such as impulse response / frequency response testing, least squares, or frequency domain fitting) and experimentally calibrated, and the sampling distribution is determined based on the calibration uncertainty. During the simulation process, each candidate sequence is subjected to multiple random sampling simulations (e.g., 1000 simulations), and the positioning deviation of each simulation is recorded and a set of positioning accuracy indicators (such as maximum deviation, average deviation, and deviation distribution; in the example, the maximum deviation may be 5.2e-4 rad, i.e., 0.03°) are statistically obtained.

[0056] It should be noted that when using the Monte Carlo simulation for random sampling, the selection of the sample size N must be based on statistical evidence: for example, if the goal is to estimate the 95% confidence upper limit of the P99 positioning error, an initial N=2000 is recommended. Record the static or steady-state positioning deviation for each simulation. (Units are uniformly in degrees), and the statistics include the sample maximum value, mean, standard deviation, P90 / P95 / P99 and other higher quantiles and their confidence intervals.

[0057] In step S1044, if the maximum deviation value in the positioning accuracy index set exceeds the preset deviation threshold, then linear interpolation is performed to adjust the correction coefficients to obtain an optimized correction coefficient set.

[0058] In one implementation, the positioning accuracy index set obtained from Monte Carlo simulation is first evaluated. If the maximum deviation value obtained from the simulation (e.g., the simulation result is 5.2e-4 rad, or 0.03°) exceeds the preset deviation threshold (e.g., 3.5e-4 rad, or 0.02°), a correction is triggered. Then, the adjustment amount of the correction coefficient is calculated using the linear interpolation method to generate an optimized correction coefficient set. This optimized correction coefficient set is then applied to the driving force sequence and the simulation is repeated to verify that the maximum deviation is reduced to within the maximum deviation threshold. Finally, the verified optimized correction coefficient set is sent to the driver in real time or incorporated into the closed-loop iteration for continuous monitoring and adjustment.

[0059] It should be noted that the deviation threshold is determined by the characteristics of the galvanometer device (such as moment of inertia and damping), the sensor resolution and noise level, the accuracy requirements of the specific application, combined with the high quantile statistics obtained from Monte Carlo simulation (such as the 95% confidence limit of P99) and an appropriate safety margin.

[0060] In step S105, the driving force matching logic is updated in real time based on the time series data. If the angle switching data indicates an increase in mechanical inertia, the strength of the electromagnetic drive is increased, the correction coefficient is updated, and a precise stopping control signal is obtained, including: S1051, Based on the formatted time series dataset, extract frequency features related to mechanical inertia. If the frequency features exceed a preset frequency threshold, determine that mechanical inertia has increased, and obtain a mechanical inertia feature set. S1052, For the set of mechanical inertial features, information processing is performed to adjust the signal strength of the electromagnetic drive to obtain an enhanced set of electromagnetic drive parameters. S1053, based on the enhanced electromagnetic drive parameter set, update the correction coefficient in real time to generate a set of control signals for precise stopping.

[0061] In step S1051, frequency features related to mechanical inertia are extracted based on the formatted time series dataset. If the frequency features exceed a preset frequency threshold, it is determined that mechanical inertia has increased, and a mechanical inertia feature set is obtained.

[0062] In one implementation, starting from formatted and equally aligned time-series data, a Fourier transform is first performed on the angle and acceleration sequences to convert the time-domain signal into a frequency-domain spectrum. The spectral lines are mapped to frequencies, and significant spectral peaks are detected on the power / amplitude spectrum to extract quantitative indicators such as the primary frequency, secondary frequency, amplitude, energy ratio, or bandwidth. Then, the primary frequency is compared with a preset frequency threshold (example threshold 450Hz). If the primary frequency or other key frequencies exceed the threshold, it is determined that the mechanical inertia has increased. The primary frequency, secondary frequency, corresponding amplitude, and energy ratio are then grouped into a mechanical inertia feature set for use in subsequent drive matching and control strategies. The threshold setting should be determined in conjunction with the galvanometer material properties and processing speed to ensure specificity.

[0063] It should be noted that the preset frequency threshold must be strictly less than the Nyquist frequency (for a sampling rate of 1000Hz, the Nyquist frequency is 500Hz, and it is recommended to take no more than 0.9 times the Nyquist frequency, with an example value of 450Hz). The setting is based on the combination of the galvanometer material properties and inherent bandwidth, processing speed, and the main frequency characteristics extracted by Fourier spectrum extraction of the formatted time series. It is determined based on offline calibration experiments and Monte Carlo statistics (such as upper confidence limits of high quantiles), and a safety margin (such as 0.1 times the Nyquist frequency as a margin) is added.

[0064] In step S1052, for the mechanical inertial feature set, information processing is performed to adjust the signal strength of the electromagnetic drive to obtain an enhanced electromagnetic drive parameter set.

[0065] In one implementation, the information processing starts from the obtained mechanical inertia feature set. First, a Fourier transform is performed on the time series after cleaning and Kalman filtering to identify the main frequency and secondary frequency and their corresponding amplitudes. The energy ratio (energy percentage) of the frequency band where the main frequency is located to the full frequency band or target frequency band is calculated. Then, these quantized features are compared with pre-calibrated quantized feature thresholds to assess and determine the degree of increase in mechanical inertia. Subsequently, linear interpolation is performed to adjust the correction coefficients to compensate for the direction and amplitude of the deviation. Based on the updated correction coefficients, a control signal with a precise timestamp is generated and sent to the driver. Closed-loop monitoring stops the error and overshoot, and iterative updates are performed based on feedback until the error meets the threshold or the iteration termination condition is reached, resulting in an enhanced electromagnetic drive parameter set.

[0066] It should be noted that the pre-calibrated quantization feature thresholds are determined through offline calibration experiments and statistical analysis. Specifically, the Fourier spectral features (major and minor frequencies, amplitudes, and energy ratios) of the formatted time series are extracted by combining the material and structural parameters of the galvanometer, the sensor resolution, and the noise performance. Then, high quantile statistics (such as the 95% upper confidence limit of P99) are obtained through Monte Carlo simulation, and an appropriate safety margin is added to finally determine the thresholds.

[0067] In step S1053, the correction coefficients are updated in real time according to the enhanced electromagnetic drive parameter set to generate a set of control signals for precise stopping.

[0068] In one implementation, the enhanced electromagnetic drive parameter set is updated in real time, and the correction coefficient is dynamically calculated with the enhanced parameters as input. First, the current enhanced parameters (e.g., current 1.2A, frequency 550Hz) and the real-time measured stop deviation are read, and a preset deviation threshold (e.g., 3.5e-4rad, i.e., 0.02°) is used as the criterion. The correction coefficient is adjusted according to the real-time update algorithm to compensate for the deviation direction and amplitude. Then, a control signal set with precise timestamps is generated based on the updated correction coefficient. The control signal set is sent to the driver in real time, and the actual stop error and overshoot are monitored in the closed loop. If the target is still not met, the correction coefficient is iteratively updated based on the feedback data until the threshold is met or the iteration termination condition is reached.

[0069] It should be noted that the maximum number of iterations in the iterative parameter adjustment process is [number missing]. (This can be adjusted during deployment based on real-time latency, convergence speed, and statistical confidence requirements). If the target is reached... If the target is still not met, implement the following fault protection and upgrade measures in sequence: roll back to the last validated stable parameters and enter a safe / conservative mode (e.g., reduce the drive amplitude, extend the pulse width); issue an alarm and record a complete log (including random seed, sample number, statistics for each round, and adjustment history) for manual analysis; increase offline validation efforts (increase the Monte Carlo sample size, use a more refined model, or higher-order sensitivity analysis) and retrain / fit the inverse model or adjust the gain in an offline environment; if software optimization cannot resolve the issue, prompt for hardware repair or replacement (e.g., check the sensor, driver, or mechanical components) and restart the online adjustment cycle after repair or recalibration.

[0070] In step S106, the actual angle adjustment of the laser galvanometer is performed via the control signal, and the occurrence of overshoot oscillation is monitored. If the oscillation amplitude is lower than a preset oscillation amplitude threshold, the improvement in dynamic response is confirmed, and the processing quality feedback data is recorded, including: S1061: Acquire real-time angle adjustment data and oscillation data, filter and remove noise to obtain smooth angle data and oscillation data; S1062, Based on the smoothed angle data and oscillation data, extract the angle oscillation frequency features. If the amplitude of the angle oscillation frequency features is lower than the preset oscillation amplitude threshold, determine that the dynamic response is improved and obtain the angle oscillation feature set. S1063, Based on the angular oscillation feature set, the parameters of the control signal are adjusted through signal processing to obtain an optimized control signal set, and quality feedback data during the processing is collected.

[0071] In step S1061, real-time angle adjustment data and oscillation data are acquired, filtered to remove noise, and smooth angle data and oscillation data are obtained.

[0072] In one implementation, real-time angle adjustment data and oscillation data are first acquired in parallel by an optical encoder and a high-sensitivity inertial sensor, and the multi-sensor data are time-synchronized and aligned (exemplary sampling interval is 0.05ms). Then, the original time series is cleaned: median filtering is used to remove impulse noise, outliers (such as angle abrupt changes exceeding 0.021rad or acceleration abrupt changes exceeding 0.5g) are removed or replaced with the nearest mean, and the data is aligned to a unified timestamp through linear interpolation. For the remaining noise, low-pass filtering (example cutoff frequency is about 1kHz) is preferentially used to filter out high-frequency interference, or Kalman filtering is used for multi-sensor fusion when the noise level is high (e.g., the standard deviation of angle data is greater than 8.7e-4rad, i.e., 0.05°) to obtain a more robust estimate. After the above processing, a smooth angle sequence and oscillation sequence are output.

[0073] In step S1062, based on the smoothed angle data and oscillation data, the angle oscillation frequency feature is extracted. If the amplitude of the angle oscillation frequency feature is lower than the preset oscillation amplitude threshold, it is determined that the dynamic response is improved, and the angle oscillation feature set is obtained.

[0074] In one implementation, for the filtered and smoothed angle data and oscillation data, the information processing unit first converts the time-domain signal into a frequency-domain spectrum through Fourier transform and detects significant spectral peaks on the amplitude spectrum to extract the dominant frequency, secondary frequency, and their corresponding amplitude, energy ratio, or bandwidth of the angle oscillation. Then, the amplitude of the extracted dominant frequency or key frequency is compared with a preset oscillation amplitude threshold (e.g., dominant frequency 550Hz, amplitude 0.3, and oscillation amplitude threshold 0.45). If the amplitude of the frequency feature is lower than the oscillation amplitude threshold, it is determined that the dynamic response is improved, and the dominant frequency, amplitude, energy ratio, and other items are merged into the angle oscillation feature set for use by subsequent drive matching and control strategies.

[0075] It should be noted that the oscillation amplitude threshold is determined by extracting the main / secondary frequency and amplitude features from the Fourier spectrum of the filtered time series, combined with factors such as the material and structural stiffness of the galvanometer, sensor resolution and noise, and processing conditions, based on offline calibration experiments and Monte Carlo statistics (e.g., taking the upper confidence limit of the high quantile) and adding an appropriate safety margin.

[0076] In step S1063, based on the angular oscillation feature set, the parameters of the control signal are adjusted through signal processing to obtain an optimized control signal set, and quality feedback data during the processing is collected.

[0077] In one implementation, the signal processing adjustment refers to, for the obtained angular oscillation feature set (such as dominant and secondary frequencies, corresponding amplitudes and energy ratios), firstly, using adaptive filtering based on the features, calculating the control signal parameters to be adjusted, and generating an optimized control signal set accordingly. Specifically, this can be achieved by adjusting the pulse width modulation parameters and applying smoothing processing to the resulting signal sequence to reduce the impact of abrupt changes on the galvanometer. Subsequently, the smoothed optimized control signal drives the galvanometer to perform angle adjustment, and quality feedback data is collected in real time during the processing. The feedback data is used to verify the effectiveness of the control signal and serves as the basis for subsequent iterations (such as further fine-tuning the pulse width or updating the driving force parameters).

[0078] In step S107, a loss function is constructed by aligning the feedback data of the processing quality with the positioning accuracy index of the simulation output. Then, based on the model's differentiability and computational budget, the gradient method is used to iteratively optimize the parameters used for the simulation verification, resulting in a more accurate driving force matching strategy for the next rapid response loop processing, including: S1071, Based on the feedback data of the processing quality, obtain the processing error value and oscillation data, extract the error oscillation frequency features, and obtain the error oscillation feature set; S1072, if the frequency amplitude of the error oscillation feature set is lower than the preset error oscillation threshold, then adjust the driving force parameters to obtain an optimized driving force parameter set.

[0079] In step S1071, based on the feedback data of the processing quality, the processing error value and oscillation data are obtained, the error oscillation frequency characteristics are extracted, and the error oscillation feature set is obtained.

[0080] In one implementation, the optical inspection system first collects processing quality feedback data (such as deviation values, surface roughness, etc.) and the inertial sensor records oscillation data in parallel. The information processing unit performs time synchronization and cleaning on the multi-source data (e.g., median filtering to remove impulse noise, eliminating or replacing outliers with the nearest mean and aligning to a unified timestamp) to obtain a formatted processing error value sequence and a filtered oscillation time sequence. Then, the processing error values ​​(such as the deviation sequence between the actual position and the target position) are calculated and the oscillation time sequence is sent to the frequency domain analysis module. The amplitude spectrum is generated using Fourier transform or equivalent spectrum analysis methods. Significant spectral peaks are detected on the spectrum to extract quantitative indicators such as primary and secondary frequencies, corresponding amplitudes, energy proportions, and bandwidth. Finally, the error value entries and the extracted oscillation frequency features are merged to form an error oscillation feature set.

[0081] In step S1072, if the frequency amplitude of the error oscillation feature set is lower than the preset error oscillation threshold, the driving force parameters are adjusted to obtain an optimized driving force parameter set.

[0082] In one implementation, processing quality feedback data and oscillation time series are first acquired in parallel by optical detection and inertial sensors. After time synchronization and data cleaning (e.g., median filtering), a formatted error sequence and a filtered oscillation sequence are obtained. The information processing unit sends the oscillation sequence to a frequency domain analysis module (e.g., Fourier transform) to extract oscillation features such as primary and secondary frequencies and corresponding amplitudes, and compares them with a preset error oscillation amplitude threshold. When the frequency-corresponding amplitude of the error oscillation feature set is lower than the error oscillation amplitude threshold, the system combines historical features and error values ​​to predict the optimal adjustment amount, calculates and outputs an optimized driving force parameter set (for example, parameters such as current and frequency can be adjusted according to model suggestions to form a new driving force parameter set). This parameter set can be used to generate control signals and verify and iterate in a closed loop.

[0083] It should be noted that the preset error oscillation amplitude threshold is determined by extracting the main / secondary frequency and amplitude characteristics from the Fourier spectrum of the filtered oscillation time series, combined with engineering factors such as the material and structural stiffness of the galvanometer, sensor resolution and noise, and processing conditions. It is based on baseline data under normal operating conditions and offline calibration experiments, supplemented by Monte Carlo statistics (e.g., taking the high quantile or the upper confidence limit of P99), and an appropriate safety margin is added. It can be iteratively corrected in closed-loop operation.

[0084] In summary, this invention discloses a method for controlling the low response time of a laser galvanometer, comprising: The system acquires real-time data on the angle switching of the laser galvanometer and mechanical inertial parameters, aligns the timestamps of each data point, and then fuses the data to obtain a precise description of the current motion state. Based on the precise description of the current motion state, noise during the angle switching process is filtered out to obtain a smoothed estimate of the mechanical inertia variation. The overshoot oscillation risk in a fast response scenario is predicted by the mechanical inertia variation estimate. If the predicted risk exceeds the preset risk threshold, the control parameters are adjusted to generate an optimized electromagnetic driving force sequence. Based on the electromagnetic driving force sequence and combined with the real-time acquired time series data, the dynamic response of the laser galvanometer is simulated and verified to obtain the positioning accuracy index in the simulation results and determine the correction coefficient. Based on the time series data, the driving force matching logic is updated in real time. If the angle switching data indicates an increase in mechanical inertia, the strength of the electromagnetic drive is increased, the correction coefficient is updated, and a precise stopping control signal is obtained. The actual angle adjustment of the laser galvanometer is performed through the control signal, and the occurrence of overshoot oscillation is monitored. If the oscillation amplitude is lower than the preset oscillation amplitude threshold, the improvement of dynamic response is confirmed, and the processing quality feedback data is recorded. Based on the feedback data of the processing quality, the parameters used for the simulation verification are iteratively optimized to obtain a more accurate driving force matching strategy for the next rapid response loop processing.

[0085] Reference Figure 2 The second embodiment of the present invention provides a control system for low response time of laser galvanometers, comprising: Motion state acquisition module: used to acquire real-time collected angle switching data and mechanical inertial parameters of the laser galvanometer, align the timestamps of each data point, and then perform data fusion to obtain an accurate description of the current motion state; Inertia estimation module: used to filter noise during the angle switching process based on the precise description of the current motion state, and obtain a smoothed estimate of the mechanical inertia variation; Oscillation prediction module: used to predict the risk of overshoot oscillation in a fast response scenario using the estimated value of mechanical inertia variation; if the predicted risk exceeds the preset risk threshold, the control parameters are adjusted to generate an optimized electromagnetic driving force sequence. Simulation verification module: used to simulate and verify the dynamic response of the laser galvanometer based on the electromagnetic driving force sequence and the real-time acquired time series data, to obtain the positioning accuracy index in the simulation results and determine the correction coefficient; Matching update module: Based on the time series data, the driving force matching logic is updated in real time. If the angle switching data indicates an increase in mechanical inertia, the strength of the electromagnetic drive is enhanced, the correction coefficient is updated, and a precise stopping control signal is obtained. The monitoring module is used to perform actual angle adjustment of the laser galvanometer through the control signal, monitor the occurrence of overshoot oscillation, and if the oscillation amplitude is lower than the preset oscillation amplitude threshold, confirm the improvement of dynamic response and record the processing quality feedback data. Strategy optimization module: Based on the feedback data of the processing quality and the positioning accuracy index of the simulation output, a loss function is constructed. Then, based on the model differentiability and computational budget, the gradient method is used to iteratively optimize the parameters used for the simulation verification to obtain a more accurate driving force matching strategy for the next fast response loop processing.

[0086] It should be noted that the control system for low response time of laser galvanometer provided in the embodiments of the present invention is used to execute all the process steps of the control method for low response time of laser galvanometer in the above embodiments. The working principle and beneficial effect of the two are one-to-one, so they will not be described again.

[0087] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a control method for low response time of a laser galvanometer. When the processor executes the computer program, it implements the steps in the various embodiments of the control method for low response time of a laser galvanometer described above, for example... Figure 1 The step S101 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above system embodiments, such as the oscillation prediction module.

[0088] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0089] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0090] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0091] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0092] The above specific embodiments have further described in detail the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and do not limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for controlling the low response time of a laser galvanometer, characterized in that, include: The system acquires real-time data on the angle switching of the laser galvanometer and mechanical inertial parameters, aligns the timestamps of each data point, and then fuses the data to obtain a precise description of the current motion state. Based on the precise description of the current motion state, noise during the angle switching process is filtered out to obtain a smoothed estimate of the mechanical inertia variation. The overshoot oscillation risk in a fast response scenario is predicted by the mechanical inertia variation estimate. If the predicted risk exceeds the preset risk threshold, the control parameters are adjusted to generate an optimized electromagnetic driving force sequence. Based on the electromagnetic driving force sequence and combined with the real-time acquired time series data, the dynamic response of the laser galvanometer is simulated and verified to obtain the positioning accuracy index in the simulation results and determine the correction coefficient. Based on the time series data, the driving force matching logic is updated in real time. If the angle switching data indicates an increase in mechanical inertia, the strength of the electromagnetic drive is enhanced, the correction coefficient is updated, and a precise stopping control signal is obtained. The actual angle adjustment of the laser galvanometer is performed through the control signal, and the occurrence of overshoot oscillation is monitored. If the oscillation amplitude is lower than the preset oscillation amplitude threshold, the improvement of dynamic response is confirmed, and the processing quality feedback data is recorded. Based on the feedback data of the processing quality and the positioning accuracy index of the simulation output, a loss function is constructed. Then, based on the model differentiability and computational budget, the gradient method is used to iteratively optimize the parameters used for the simulation verification, resulting in a more accurate driving force matching strategy for the next fast response loop processing.

2. The control method for low response time of laser galvanometers according to claim 1, characterized in that, The process involves acquiring real-time collected angle switching data and mechanical inertial parameters of the laser galvanometer, aligning the timestamps of each data point, and then fusing the data to obtain a precise description of the current motion state, including: Acquire laser galvanometer angle data and switching frequency data, collect mechanical inertial parameters, align the timestamps of the data, and obtain the original time series dataset; If the timestamp deviation of the original dataset is greater than the preset timestamp deviation threshold, the angle data and inertial parameters of the original dataset are fused to generate corrected motion data. The angle, position, and velocity information are extracted from the corrected motion data to obtain a precise description of the current motion state.

3. The control method for low response time of laser galvanometers according to claim 1, characterized in that, The step of filtering noise during the angle switching process based on a precise description of the current motion state to obtain a smoothed estimate of the mechanical inertia variation includes: If the noise level in the precise description of the motion state is greater than a preset noise threshold, then Kalman filtering is performed to obtain smoothed motion data. Based on the smoothed motion data, mechanical inertia variation features are extracted to obtain the smoothed mechanical inertia variation estimate.

4. The control method for low response time of laser galvanometers according to claim 1, characterized in that, The step of predicting overshoot oscillation risk in a fast-response scenario using the estimated mechanical inertia variation, and adjusting control parameters to generate an optimized electromagnetic driving force sequence if the predicted risk exceeds a preset risk threshold, includes: Based on the smoothed mechanical inertia variation estimate, dynamic features related to oscillation risk are extracted to obtain an oscillation risk feature set; For the oscillation risk feature set, the overshoot oscillation risk value under the fast response scenario is predicted to obtain the predicted risk value; If the predicted risk value exceeds the preset risk threshold, the electromagnetic drive parameters are optimized by adjusting the control parameters to generate an optimized electromagnetic drive force sequence.

5. The control method for low response time of laser galvanometers according to claim 1, characterized in that, The step involves simulating and verifying the dynamic response of the laser galvanometer based on the electromagnetic driving force sequence and real-time acquired time series data, obtaining the positioning accuracy index in the simulation results, and determining the correction coefficient, including: Real-time angle and acceleration data of the laser galvanometer are acquired and Kalman filtered to remove noise and outliers, resulting in a formatted time series dataset. Based on the formatted time series dataset, frequency features related to the dynamic response are extracted to obtain the dynamic response feature set; Based on the dynamic response feature set, the electromagnetic driving force sequence is simulated and verified to obtain a positioning accuracy index set; If the maximum deviation value in the positioning accuracy index set exceeds the preset deviation threshold, then linear interpolation is performed to adjust the correction coefficients to obtain an optimized set of correction coefficients.

6. The control method for low response time of laser galvanometers according to claim 5, characterized in that, The driving force matching logic is updated in real time based on the time series data. If the angle switching data indicates an increase in mechanical inertia, the strength of the electromagnetic drive is increased, the correction coefficient is updated, and a precise stopping control signal is obtained. Based on the formatted time series dataset, frequency features related to mechanical inertia are extracted. If the frequency features exceed a preset frequency threshold, it is determined that mechanical inertia has increased, and a mechanical inertia feature set is obtained. For the set of mechanical inertial features, information processing is performed to adjust the signal strength of the electromagnetic drive, thereby obtaining an enhanced set of electromagnetic drive parameters. Based on the enhanced electromagnetic drive parameter set, the correction coefficients are updated in real time to generate a set of control signals for precise stopping.

7. The control method for low response time of laser galvanometers according to claim 1, characterized in that, The control signal is used to adjust the actual angle of the laser galvanometer, monitor the occurrence of overshoot oscillation, and if the oscillation amplitude is lower than a preset oscillation amplitude threshold, the improvement in dynamic response is confirmed, and processing quality feedback data is recorded, including: Acquire real-time angle adjustment data and oscillation data, filter to remove noise, and obtain smooth angle data and oscillation data; Based on the smoothed angle data and oscillation data, the angle oscillation frequency feature is extracted. If the amplitude of the angle oscillation frequency feature is lower than the preset oscillation amplitude threshold, the dynamic response is improved, and the angle oscillation feature set is obtained. Based on the angular oscillation feature set, the parameters of the control signal are adjusted through signal processing to obtain an optimized control signal set, and quality feedback data during the processing is collected.

8. The control method for low response time of laser galvanometers according to claim 1, characterized in that, The process involves aligning the feedback data on processing quality with the positioning accuracy index output from the simulation and constructing a loss function. Then, based on model differentiability and computational cost, a gradient method is used to iteratively optimize the parameters used for the simulation verification, resulting in a more accurate driving force matching strategy for the next rapid response loop. This includes: Based on the feedback data of the processing quality, the processing error value and oscillation data are obtained, the error oscillation frequency features are extracted, and the error oscillation feature set is obtained; If the frequency amplitude of the error oscillation feature set is lower than the preset error oscillation threshold, the driving force parameters are adjusted to obtain an optimized driving force parameter set.

9. A control system for low response time of laser galvanometers, characterized in that, include: Motion state acquisition module: used to acquire real-time collected angle switching data and mechanical inertial parameters of the laser galvanometer, align the timestamps of each data point, and then perform data fusion to obtain an accurate description of the current motion state; Inertia estimation module: used to filter noise during the angle switching process based on the precise description of the current motion state, and obtain a smoothed estimate of the mechanical inertia variation; Oscillation prediction module: used to predict the risk of overshoot oscillation in a fast response scenario using the estimated value of mechanical inertia variation; if the predicted risk exceeds the preset risk threshold, the control parameters are adjusted to generate an optimized electromagnetic driving force sequence. Simulation verification module: used to simulate and verify the dynamic response of the laser galvanometer based on the electromagnetic driving force sequence and the real-time acquired time series data, to obtain the positioning accuracy index in the simulation results and determine the correction coefficient; Matching update module: Based on the time series data, the driving force matching logic is updated in real time. If the angle switching data indicates an increase in mechanical inertia, the strength of the electromagnetic drive is enhanced, the correction coefficient is updated, and a precise stopping control signal is obtained. The monitoring module is used to perform actual angle adjustment of the laser galvanometer through the control signal, monitor the occurrence of overshoot oscillation, and if the oscillation amplitude is lower than the preset oscillation amplitude threshold, confirm the improvement of dynamic response and record the feedback data of processing quality. Strategy optimization module: Based on the feedback data of the processing quality and the positioning accuracy index of the simulation output, a loss function is constructed. Then, based on the model differentiability and computational budget, the gradient method is used to iteratively optimize the parameters used for the simulation verification to obtain a more accurate driving force matching strategy for the next fast response loop processing.

Images