Piezoelectric crystal grinding control system based on frequency test technology

Abstract

The application discloses a piezoelectric crystal grinding control system based on frequency test technology. The system comprises a hardware acquisition module, a frequency test compensation module, a dynamic mapping control module, an execution driving module and a closed-loop self-calibration module. The hardware acquisition module acquires original resonance frequency signals, vibration intensity and grinding head working parameters. The frequency test compensation module separates and compensates vibration interference, and outputs accurate resonance frequency and its change rate. The dynamic mapping control module internally has a dynamic mapping model, takes the accurate resonance frequency, the change rate and the vibration intensity as inputs, and outputs the grinding head rotating speed, pressure and feeding amount. The outputs are dynamically corrected and pre-judged through a vibration working condition correction unit and a frequency drift early prediction unit, and grinding parameter instructions are generated. The execution driving module drives the grinding head to adjust the working parameters according to the instructions. The closed-loop self-calibration module corrects compensation coefficients and model weights according to actual frequency convergence effects. The application realizes full-link closed-loop accurate control of the grinding process, effectively suppresses vibration interference, and improves frequency test precision and machining consistency.

Description

Technical Field

[0001] This invention relates to the fields of piezoelectric crystal processing and industrial automatic control technology, and in particular to a piezoelectric crystal grinding control system based on frequency testing technology. Background Technology

[0002] Piezoelectric crystals, due to their excellent piezoelectric effect and stable frequency characteristics, are widely used in various frequency control and signal processing electronic devices. Their resonant frequency parameter is one of the core indicators for evaluating product performance, and the thickness control accuracy of the crystal during the grinding process directly affects the consistency and stability of the final frequency. Therefore, achieving accurate testing and dynamic control of the resonant frequency during the grinding process is a crucial step in improving product yield and processing consistency.

[0003] In actual production processes, the grinding and processing of piezoelectric crystals is often accompanied by complex conditions such as high-speed rotation, mechanical contact, and fluid disturbance, resulting in multi-source vibration coupling in the system's operating environment. The resonant frequency test signal is prone to superposition or coupling with environmental vibration signals during acquisition, leading to unstable frequency measurement results. If the interference components in the test signal are not effectively distinguished and processed, the accuracy of subsequent control decisions will be affected.

[0004] Meanwhile, piezoelectric crystal materials exhibit anisotropic characteristics, and their frequency changes nonlinearly with thickness variations during processing. A complex coupling relationship exists between grinding parameters (such as rotational speed, pressure, and feed rate) and frequency. In practical control, if parameter adjustments are based solely on a single frequency value without comprehensively considering the frequency variation trend and current vibration conditions, it is difficult to achieve precise matching control of the frequency change process, easily leading to problems such as adjustment lag or overcorrection.

[0005] Furthermore, in existing processing systems, frequency testing and grinding control often belong to different functional units. Test results are directly used as control inputs, lacking a dynamic correction mechanism for the validity of test data and failing to form an adaptive feedback structure based on processing results. As equipment operating time increases or operating conditions change, system parameters may deviate from the initial settings. Without self-calibration and self-optimization capabilities, the accuracy of frequency control will gradually decrease.

[0006] Therefore, there is an urgent need to construct a piezoelectric crystal grinding control system that can effectively separate and compensate frequency signals under complex vibration conditions, deeply couple frequency test results with grinding parameter control, and have dynamic mapping and closed-loop self-calibration capabilities, so as to improve frequency control accuracy and processing stability. Summary of the Invention

[0007] To address the shortcomings of existing technologies, the present invention aims to provide a piezoelectric crystal grinding control system based on frequency testing technology. This system is used to separate and compensate frequency test signals under complex vibration conditions to eliminate vibration interference. It also deeply couples precise frequency data with grinding parameters through a dynamic mapping model and an advanced prediction unit, and combines a closed-loop self-calibration module to achieve self-optimization of system parameters, thereby improving frequency testing accuracy and grinding control stability.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a piezoelectric crystal polishing control system based on frequency testing technology, comprising:

[0009] The hardware acquisition module acquires the original resonant frequency signal, vibration intensity, and grinding head operating parameters.

[0010] The frequency test compensation module, connected to the hardware acquisition module, includes:

[0011] The signal separation unit is used to receive the original resonance frequency signal and vibration intensity, and output the crystal resonance signal and vibration interference signal.

[0012] The precise frequency calculation unit is connected to the signal separation unit and the preset compensation coefficient library, and is used to calculate the precise resonance frequency and its rate of change based on the crystal resonance signal, the vibration interference signal and the compensation coefficient corresponding to the vibration intensity.

[0013] The dynamic mapping control module, connecting the frequency test compensation module and the hardware acquisition module, includes:

[0014] The dynamic mapping model is used to receive precise resonant frequency, rate of change and vibration intensity, and output grinding head speed, pressure and feed rate.

[0015] A vibration condition correction unit, connected to the dynamic mapping model, is used to correct the model output when the vibration intensity exceeds the threshold.

[0016] The frequency drift advance prediction unit is connected to the vibration condition correction unit. It is used to predict the frequency drift amount based on the vibration intensity and correct its output in advance to generate grinding parameter instructions.

[0017] The execution drive module is connected to the dynamic mapping control module and is used to drive the grinding head to adjust the working parameters according to the grinding parameter instructions.

[0018] A closed-loop self-calibration module, connecting the frequency test compensation module and the dynamic mapping control module, includes:

[0019] The effect acquisition unit is used to acquire the actual frequency convergence effect of the crystal after grinding.

[0020] A compensation coefficient correction unit, connected to the effect acquisition unit and the compensation coefficient library, is used to correct the compensation coefficient according to the actual frequency convergence effect.

[0021] The model weight optimization unit, connected to the effect acquisition unit and the dynamic mapping model, is used to optimize the model weights based on the actual frequency convergence effect.

[0022] Furthermore, the hardware acquisition module includes:

[0023] The frequency test probe unit is located next to the grinding head of the grinding equipment and facing the processing surface of the piezoelectric crystal to be ground, and is used to collect the original resonant frequency signal of the piezoelectric crystal.

[0024] A miniature vibration acquisition unit is integrated on the frequency test probe unit, works synchronously with the frequency test probe unit and is located at the same spatial point, and is used to acquire environmental vibration data during the grinding process. The environmental vibration data includes at least vibration frequency, amplitude and vibration intensity.

[0025] The grinding head status sensing unit includes a speed sensor, a pressure sensor, and a feed sensor. The speed sensor is mounted on the driving mechanism of the grinding head to collect the grinding head speed. The pressure sensor is mounted on the pressure regulating mechanism of the grinding head to collect the grinding pressure. The feed sensor is mounted on the feeding mechanism of the grinding head to collect the grinding head feed amount.

[0026] The industrial bus unit, using the Profinet industrial Ethernet protocol, is connected to the frequency test probe unit, the micro vibration acquisition unit, and the grinding head status sensing unit, respectively. It is used to synchronously transmit the original resonant frequency signal, the environmental vibration data, and the grinding head operating parameters to the frequency test compensation module and the dynamic mapping control module, and the transmission delay is less than or equal to a preset delay threshold.

[0027] Furthermore, the signal separation unit includes:

[0028] The phase alignment subunit, connected to the frequency test probe unit and the micro vibration acquisition unit, is used to perform phase delay compensation on the environmental vibration data to synchronize it with the original resonant frequency signal in time.

[0029] An amplitude normalization subunit, connected to the phase alignment subunit, is used to normalize the amplitude of the time-aligned environmental vibration signal and output a normalized vibration signal.

[0030] The frequency domain interference identification subunit, connected to the amplitude normalization subunit, is used to perform fast Fourier transform on the time-aligned original resonance frequency signal and the normalized vibration signal to obtain their respective spectra, and to identify the frequency components in the original frequency spectrum that coincide with the main frequency of the vibration spectrum through frequency matching, and mark them as vibration coupling interference frequency components.

[0031] The adaptive cancellation reconstruction subunit is connected to the frequency domain interference identification subunit. It has a built-in adaptive filtering algorithm to generate a cancellation signal based on the frequency components of the vibration coupling interference and the normalized vibration signal. The cancellation signal is superimposed with the original resonance frequency signal to eliminate the vibration interference. The residual signal is then reconstructed by inverse Fourier transform to obtain the resonance frequency signal of the piezoelectric crystal itself in the time domain. The vibration coupling interference signal is then output to the precise frequency calculation unit.

[0032] Furthermore, the precise frequency calculation unit includes:

[0033] The waveform preprocessing subunit, connected to the signal separation unit, is used to perform time-domain analysis on the resonant frequency signal of the piezoelectric crystal itself, capture transient change moments with amplitude exceeding a preset threshold and mark them as suspected abnormal periods, remove abnormal data points, and then stitch the remaining effective segments together. Cubic spline interpolation is used to smooth the discontinuities and reconstruct a continuous and pure resonant frequency waveform.

[0034] The instantaneous frequency calculation subunit is connected to the waveform preprocessing subunit and is used to perform zero-crossing detection on the reconstructed waveform, calculate the instantaneous resonant frequency value, and output it continuously at a preset period.

[0035] The frequency change rate calculation subunit, connected to the instantaneous frequency calculation subunit, is used to calculate the difference between the current instantaneous resonant frequency value and the previous period value and divide it by the sampling period to obtain the real-time frequency change rate.

[0036] The filtering and compensation subunit is connected to the instantaneous frequency calculation subunit, the frequency change rate calculation subunit, and the compensation coefficient library, respectively. It is used to perform moving average filtering on the instantaneous resonant frequency value and real-time frequency change rate within a preset time window, and to match the corresponding frequency compensation coefficient from the compensation coefficient library as the filtering weight based on the real-time vibration intensity transmitted by the hardware acquisition module, and participate in the weighted filtering to finally output the accurate resonant frequency and accurate frequency change rate.

[0037] Furthermore, the dynamic mapping model includes:

[0038] The crystal characteristic matching subunit is connected to the precise frequency calculation unit and the hardware acquisition module. It is used to identify the crystal orientation and thickness of the crystal based on the initial precise resonance frequency, match the pre-stored crystal characteristic database, and output the model parameter set.

[0039] The feature fusion subunit, connected to the crystal characteristic matching subunit, is used to perform wavelet packet decomposition, difference and short-time Fourier transform on the precise resonance frequency, precise frequency change rate and real-time vibration intensity respectively, extract time-frequency features, trend features and energy distribution features, and fuse them into a fusion feature vector;

[0040] A deep neural network inference subunit, connected to the feature fusion subunit, has a built-in deep neural network configured according to the model parameter set, used to map the fused feature vector to the base rotation speed, base pressure and base feed rate;

[0041] The nonlinear correction and verification subunit, connected to the deep neural network inference subunit, is used to compensate for boundary errors through a nonlinear correction function, and then verify them through the process constraint rule base to clamp the out-of-limit parameters to a safe threshold, thereby generating grinding head speed, grinding pressure and grinding head feed that meet safety requirements.

[0042] The multi-objective optimization balance subunit, connected to the nonlinear correction and verification subunit, is used to perform multi-objective optimization on the three parameters according to the weight requirements of removal rate and surface quality in the current grinding stage, generate optimized grinding parameters, and output them to the vibration condition correction unit.

[0043] The online model update subunit is connected to the multi-objective optimization balancing subunit and the model weight optimization unit, respectively. It is used to receive the optimized parameter weights output by the model weight optimization unit and perform online incremental updates on the deep neural network.

[0044] Furthermore, the vibration condition correction unit includes:

[0045] The vibration analysis and grading subunit is connected to the hardware acquisition module and is used to perform short-time Fourier transform and wavelet packet decomposition on the real-time vibration intensity, extract the time-frequency distribution, main frequency, and energy accumulation characteristics, and classify the vibration intensity into normal, medium intensity, and strong impact levels according to the preset threshold range.

[0046] The decoupling and weight adjustment subunit is connected to the vibration analysis and grading subunit and the dynamic mapping model, respectively. It is used to analyze the differential influence coefficients of vibration on each output parameter based on the dominant frequency and energy characteristics, and generate model weight correction coefficients in combination with vibration level. These coefficients are applied to the hidden layer weights of the dynamic mapping model to realize the dynamic adjustment of parameter weights.

[0047] The parameter local correction subunit, connected to the decoupling and weight adjustment subunit, is used to recalculate the grinding head speed, grinding pressure and grinding head feed according to the adjusted model, and to perform local corrections such as speed reduction, pressure fine adjustment and deceleration.

[0048] The working condition judgment and switching subunit is connected to the parameter local correction subunit. It has a built-in working condition mode library and is used to judge the current working condition mode according to the vibration level and vibration attenuation trend, and execute the corresponding parameter correction strategy, including small adjustment in stable grinding mode, active speed reduction combined with pressure fine adjustment in vibration suppression mode, and rapid tool retraction combined with pressure unloading in impact protection mode.

[0049] The correction output and pre-evaluation subunit is connected to the parameter local correction subunit. It is used to limit the correction amplitude of the corrected parameters to ensure that the single correction amount does not exceed the safety threshold. The limited correction result is output to the frequency drift advance prediction unit. At the same time, the correction effect is pre-evaluated according to the vibration change trend, and the pre-evaluation result is sent to the model weight optimization unit as an optimization reference.

[0050] Furthermore, the frequency drift advance prediction unit includes:

[0051] The drift prediction model storage subunit has a built-in vibration intensity-frequency drift prediction model. The vibration intensity-frequency drift prediction model is constructed by experimentally collecting the actual frequency drift of the piezoelectric crystal within a preset lead time window under different vibration intensities, using a combination of linear fitting and nonlinear correction, and the model fitting degree is not lower than the preset fitting degree threshold.

[0052] The drift prediction subunit is connected to the vibration condition correction unit and the drift prediction model storage subunit. It is used to receive the real-time vibration intensity after being processed by the vibration condition correction unit, input it into the vibration intensity-frequency drift prediction model, and calculate the predicted frequency drift within the preset lead time window.

[0053] The advanced correction execution subunit is connected to the drift prediction subunit and the vibration condition correction unit. It is used to generate speed compensation value, pressure compensation value and feed compensation value according to the predicted frequency drift amount, and superimpose them on the corrected grinding parameters output by the vibration condition correction unit to generate the final grinding parameter command and output it to the execution drive module.

[0054] The time window adaptive subunit, connected to the drift prediction subunit, is used to dynamically adjust the length of the preset lead time window according to the rate of change of vibration intensity.

[0055] The prediction optimization subunit is connected to the advance correction execution subunit and the closed-loop self-calibration module, respectively. It is used to evaluate the prediction error based on the actual frequency convergence effect collected by the effect acquisition unit in the closed-loop self-calibration module, and to perform online parameter correction on the prediction model in the drift prediction model storage subunit based on the prediction error.

[0056] Furthermore, the execution driving module includes:

[0057] The instruction parsing and allocation unit, connected to the frequency drift advance prediction unit, is used to receive and parse the target rotation speed, pressure and feed value, and allocate them to the corresponding drive channel.

[0058] The feedforward compensation unit, connected to the instruction parsing and allocation unit, is used to perform feedforward compensation on the target value based on the inertia and response delay characteristics of the actuator, and generate a drive instruction with advance compensation amount.

[0059] The speed drive unit, connected to the feedforward compensation unit, is used to receive speed drive commands, drive the grinding head spindle motor to reach the target speed, and feed back the actual speed.

[0060] The pressure drive unit, connected to the feedforward compensation unit, is used to receive pressure drive commands, control the pressure execution cylinder to output the target pressure and feed back the actual pressure.

[0061] The feed drive unit is connected to the feedforward compensation unit and is used to receive feed drive commands, drive the feed servo motor to move according to the target feed amount and feed back the actual position.

[0062] The multi-axis coordination unit is connected to each drive unit and is used to monitor the response status of each drive unit. When the response of any unit lags behind by more than a preset synchronization deviation threshold, a waiting or deceleration command is issued to ensure timing coordination.

[0063] The status self-test unit is connected to the multi-axis collaborative unit and is used to collect the temperature and current status of each drive unit in real time. When the parameters exceed the operating safety threshold, a protective command is issued.

[0064] The execution effect feedback unit is connected to each drive unit and the effect acquisition unit respectively. It is used to package the actual achieved speed, pressure and feed rate into execution effect data and transmit it to the effect acquisition unit.

[0065] Furthermore, the effect acquisition unit includes:

[0066] The execution data capture subunit, connected to the execution drive module, is used to capture the actual rotational speed, pressure, and feed rate values ​​reached by the grinding head after the grinding parameters are adjusted, and to record the adjustment time and duration.

[0067] The frequency final value acquisition subunit is connected to the frequency test compensation module and is used to acquire the actual resonant frequency value and spatial position coordinates of multiple sampling points on the piezoelectric crystal processing surface after grinding.

[0068] The frequency convergence analysis subunit is connected to the frequency final value acquisition subunit and is used to calculate the average value, range and standard deviation of the crystal plane frequency, compare it with the target frequency threshold, and output the crystal plane frequency consistency index and frequency compliance rate.

[0069] The convergence rate calculation subunit is connected to the execution data capture subunit and the frequency final value acquisition subunit, respectively, and is used to calculate the actual frequency convergence rate based on the time difference between the adjustment time and the grinding completion time and the frequency change.

[0070] The spectral purity analysis subunit is connected to the frequency final value acquisition subunit and is used to perform Fourier transform on the actual frequency value, analyze the proportion of spurious frequency energy, and output the spectral purity coefficient.

[0071] The comprehensive evaluation subunit for grinding effect is connected to the frequency convergence analysis subunit, the convergence rate calculation subunit, and the spectral purity analysis subunit, respectively, and is used to weight and fuse the various indicators to generate a comprehensive score for grinding effect.

[0072] An abnormal grinding identification subunit is connected to the grinding effect comprehensive evaluation subunit and is used to output an abnormal mark when the comprehensive score is lower than a preset qualified threshold.

[0073] The data packaging and transmission subunit is connected to the comprehensive evaluation subunit for grinding effect, the abnormal grinding identification subunit, the compensation coefficient correction unit, and the model weight optimization unit, respectively. It is used to package the effect score, abnormal marker, and data of each sampling point into effect feedback data and transmit them synchronously to the compensation coefficient correction unit and the model weight optimization unit.

[0074] Furthermore, the compensation coefficient correction unit includes:

[0075] The deviation calculation subunit is connected to the effect acquisition unit and is used to receive the actual frequency convergence effect, compare the average crystal plane frequency with the target frequency threshold to obtain the steady-state frequency deviation, and compare the actual frequency convergence rate with the target convergence rate to obtain the dynamic rate deviation.

[0076] The vibration condition backtracking subunit is connected to the hardware acquisition module and is used to retrieve the real-time vibration intensity time series data of this grinding and extract the average vibration intensity and peak vibration intensity of key stages.

[0077] The deviation-vibration correlation analysis subunit is connected to the deviation calculation subunit and the vibration condition backtracking subunit, respectively. It is used to perform correlation analysis between frequency deviation, velocity deviation and vibration intensity, identify the under-compensation and over-compensation regions of the compensation coefficient under different vibration intensities, and generate a deviation distribution map.

[0078] The compensation coefficient correction calculation subunit is connected to the deviation-vibration correlation analysis subunit and is used to calculate the compensation coefficient correction corresponding to each vibration intensity level based on the deviation distribution map and the preset correction step size factor.

[0079] The coefficient library update subunit is connected to the compensation coefficient correction calculation subunit and the compensation coefficient library, respectively, and is used to incrementally update the frequency compensation coefficients in the compensation coefficient library corresponding to the vibration intensity level according to the correction amount.

[0080] The correction effect verification subunit is connected to the coefficient library update subunit. It is used to collect the actual frequency convergence effect after the next grinding is completed, compare the degree of deviation improvement, and trigger a second correction when the degree of improvement is lower than the preset improvement threshold.

[0081] The aging trend analysis subunit is connected to the compensation coefficient correction calculation subunit and is used to statistically analyze the changing trend of the compensation coefficient correction during multiple grinding processes and determine the drift law of the compensation coefficient as the equipment ages.

[0082] The synchronous output subunit is connected to the coefficient library update subunit and the model weight optimization unit, respectively, and is used to synchronously transmit the key parameters of this compensation coefficient correction to the model weight optimization unit as an optimization reference.

[0083] The beneficial effects of this invention are:

[0084] Based on the above technical solution, this invention constructs a piezoelectric crystal polishing control system that combines frequency testing compensation, dynamic mapping control, and closed-loop self-calibration, which has the following beneficial effects:

[0085] 1. Improve the stability and accuracy of frequency testing: By separating and processing the resonance signal and vibration interference signal, and combining them with compensation coefficients for dynamic correction, the influence of vibration conditions on frequency measurement results is effectively reduced, the reliability of frequency test data is improved, and a stable data foundation is provided for subsequent control.

[0086] 2. Achieve multivariable dynamic matching control of grinding parameters: Construct a dynamic mapping model by using precise resonant frequency, frequency change rate and vibration intensity as joint inputs, so that the grinding parameter output can adapt to the nonlinear change characteristics in the crystal processing process and improve the grinding process's response to frequency changes.

[0087] 3. Enhance the foresight and continuity of the control process: Through the frequency drift prediction mechanism, possible frequency deviations are corrected in advance, reducing control lag and making the grinding adjustment process more stable and controllable.

[0088] 4. Achieve synergistic coupling between testing and control: By constructing a linkage mechanism between the frequency testing compensation module and the dynamic mapping control module, the test results and control output form a data closed loop, avoiding the direct transmission of test errors and improving the overall control consistency.

[0089] 5. Possesses self-calibration and self-optimization capabilities: Based on the actual frequency convergence effect after grinding, the compensation coefficient and model weights are dynamically corrected, enabling the system to adapt to changes in equipment status and operating conditions, and maintain control accuracy during long-term operation.

[0090] 6. Improve processing consistency and product stability: By forming a closed-loop adjustment mechanism of frequency testing, parameter control and effect feedback, the frequency convergence consistency during the piezoelectric crystal grinding process is improved, and the product quality stability is enhanced. Attached Figure Description

[0091] Figure 1 This is a schematic diagram of the piezoelectric crystal polishing control system based on frequency testing technology in this invention;

[0092] Figure 2 This is a schematic diagram of the signal separation unit in this invention;

[0093] Figure 3 This is a schematic diagram of the precise frequency calculation unit in this invention;

[0094] Figure 4 This is a schematic diagram of the dynamic mapping model in this invention;

[0095] Figure 5 This is a schematic diagram of the vibration condition correction unit in this invention;

[0096] Figure 6 This is a schematic diagram of the frequency drift advance prediction unit in this invention;

[0097] Figure 7 This is a schematic diagram of the effect acquisition unit in this invention;

[0098] Figure 8 This is a schematic diagram of the compensation coefficient correction unit in this invention.

[0099] Figure reference numerals: 1. Hardware acquisition module; 11. Frequency test probe unit; 12. Miniature vibration acquisition unit; 13. Grinding head status sensing unit; 14. Industrial bus unit; 2. Frequency test compensation module; 21. Signal separation unit; 211. Phase alignment subunit; 212. Amplitude normalization subunit; 213. Frequency domain interference identification subunit; 214. Adaptive phase cancellation reconstruction subunit; 22. Precise frequency calculation unit; 221. Waveform preprocessing subunit; 222. Instantaneous frequency calculation subunit; 223. Frequency change rate calculation subunit; 224. Filtering and compensation subunit; 3. Dynamic mapping control module Block; 31. Dynamic Mapping Model; 311. Crystal Characteristic Matching Subunit; 312. Feature Fusion Subunit; 313. Deep Neural Network Inference Subunit; 314. Nonlinear Correction and Verification Subunit; 315. Multi-Objective Optimization Balancing Subunit; 316. Online Model Update Subunit; 32. Vibration Condition Correction Unit; 321. Vibration Analysis Grading Subunit; 322. Decoupling and Weight Adjustment Subunit; 323. Local Parameter Correction Subunit; 324. Condition Discrimination and Switching Subunit; 325. Correction Output and Pre-evaluation Subunit; 33. Frequency Drift Lead Prediction Unit; 331. Drift Prediction Model Storage Sub-units; 332. Drift prediction sub-unit; 333. Advance correction execution sub-unit; 334. Time window adaptive sub-unit; 335. Prediction optimization sub-unit; 4. Execution drive module; 41. Instruction parsing and allocation unit; 42. Feedforward compensation unit; 43. Speed ​​drive unit; 44. Pressure drive unit; 45. Feed drive unit; 46. Multi-axis coordination unit; 47. Status self-checking unit; 48. Execution effect feedback unit; 5. Closed-loop self-calibration module; 51. Effect acquisition unit; 511. Execution data capture sub-unit; 512. Frequency final value acquisition sub-unit; 513. Frequency convergence analysis sub-unit 514. Convergence Rate Calculation Subunit; 515. Spectral Purity Analysis Subunit; 516. Grinding Effect Comprehensive Evaluation Subunit; 517. Abnormal Grinding Identification Subunit; 518. Data Packaging and Transmission Subunit; 52. Compensation Coefficient Correction Subunit; 521. Deviation Calculation Subunit; 522. Vibration Condition Retrospective Subunit; 523. Deviation-Vibration Correlation Analysis Subunit; 524. Compensation Coefficient Correction Amount Calculation Subunit; 525. Coefficient Library Update Subunit; 526. Correction Effect Verification Subunit; 527. Aging Trend Analysis Subunit; 528. Synchronous Output Subunit; 53. Model Weight Optimization Subunit. Detailed Implementation

[0100] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Identical components are denoted by the same reference numerals. It should be noted that the terms "front," "rear," "left," "right," "upper," and "lower" used in the following description refer to directions in the accompanying drawings, and the terms "bottom surface," "top surface," "inner," and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0101] like Figure 1 As shown in the figure, Embodiment 1 is the first embodiment of the present invention. This embodiment provides a piezoelectric crystal grinding control system based on frequency testing technology, which is applied to the grinding process of AT-cut quartz piezoelectric crystals (thickness 0.5mm).

[0102] The piezoelectric crystal grinding control system based on frequency testing technology includes: hardware acquisition module 1, frequency testing compensation module 2, dynamic mapping control module 3, execution drive module 4, and closed-loop self-calibration module 5. The modules are connected to each other via Profinet industrial control bus, and data processing and control command output are uniformly handled by PLC controller (model: S7-1200).

[0103] I. System Structure and Hardware Configuration;

[0104] 1. Hardware acquisition module 1:

[0105] Hardware acquisition module 1 is used to acquire the original resonant frequency signal, vibration intensity, and grinding head operating parameters. In this embodiment:

[0106] (1) Integrate a miniature acceleration vibration acquisition unit (model: ADXL345) on the frequency test probe of the existing quartz piezoelectric crystal grinding equipment to acquire vibration intensity signals in the range of 0-50Hz in real time, and set the vibration sampling frequency to 1kHz;

[0107] (2) Install a speed sensor (model: E6B2-CWZ6C) on the grinding head drive mechanism to collect the grinding head speed n, with a range of 0-3000 r / min and a resolution of 1 r / min;

[0108] (3) Install a pressure sensor (model: HM90) on the pressure regulating mechanism to collect the grinding pressure P. The range is 0-1MPa and the accuracy class is 0.5.

[0109] (4) Install a feed sensor (model: KTC-100) on the feed mechanism to collect the feed amount s of the grinding head. The range is 0-10mm and the resolution is 0.001mm.

[0110] The above-mentioned acquisition units are connected to the PLC controller through the Profinet industrial control bus to form a complete data acquisition link, ensuring that the original resonant frequency signal can be output stably, with a sampling period of 10ms.

[0111] 2. Frequency Test Compensation Module 2:

[0112] The frequency test compensation module 2 is connected to the hardware acquisition module 1 and includes a signal separation unit 21 and a precise frequency calculation unit 22.

[0113] (1) Signal separation unit 21;

[0114] The signal separation unit 21 receives the original resonant frequency signal and vibration intensity, and decomposes the original resonant frequency signal into a crystal resonant signal and a vibration interference signal. In this embodiment, a combination of phase cancellation algorithm and bandpass filtering algorithm is used to achieve vibration interference separation.

[0115] (2) Precision frequency calculation unit 22;

[0116] The precise frequency calculation unit 22 is connected to the signal separation unit 21 and a preset compensation coefficient library. The frequency compensation coefficients within the 0-50Hz vibration intensity range are calibrated experimentally to construct the compensation coefficient library. During the calibration process, under constant temperature conditions of 23±1℃ and humidity of 45%-55%RH, a standard AT-cut quartz piezoelectric crystal sample is selected for vibration loading experiments to obtain the correspondence between vibration intensity and frequency drift.

[0117] The precise frequency calculation unit 22 calculates the precise resonant frequency f and its rate of change v based on the crystal resonance signal, the vibration interference signal, and the compensation coefficient corresponding to the vibration intensity. In this embodiment, when the vibration intensity I = 15 Hz, the calculated precise resonant frequency f = 8000 Hz and the precise frequency rate of change v = 40 Hz / min.

[0118] II. Dynamic Mapping Control Module 3;

[0119] The dynamic mapping control module 3 connects the frequency test compensation module 2 and the hardware acquisition module 1, and includes a dynamic mapping model 31, a vibration condition correction unit 32, and a frequency drift advance prediction unit 33.

[0120] 1. Dynamic mapping model 31:

[0121] The dynamic mapping model 31 is used to receive the precise resonant frequency f, the rate of change v, and the vibration intensity I, and output the grinding head rotation speed n, pressure P, and feed rate s. In this embodiment, 100 AT-cut quartz piezoelectric crystals (0.5 mm thick) with different crystal orientation deviations are selected as experimental samples. Data is collected under vibration conditions of 0-50 Hz to construct a three-input-three-output dynamic mapping model 31, and the trained model is then fixed to the PLC controller.

[0122] When the inputs are f=8000Hz, v=40Hz / min, and I=15Hz, the model outputs: grinding head speed n=1200r / min, grinding pressure P=0.3MPa, and grinding head feed s=0.002mm / s.

[0123] 2. Vibration Condition Correction Unit 32:

[0124] In this embodiment, the threshold for vibration intensity exceeding the threshold is set to 20Hz. When the vibration intensity I > 20Hz, the vibration condition correction unit 32 corrects the output of the dynamic mapping model 31, reduces the feed rate and limits the speed increase rate to suppress frequency fluctuations.

[0125] 3. Frequency drift advance prediction unit 33:

[0126] The frequency drift advance prediction unit 33 is used to predict the frequency drift amount based on the vibration intensity and correct the output in advance. In this embodiment, a vibration intensity-frequency drift prediction model is constructed to achieve advance compensation of 0.3s.

[0127] When the vibration intensity I=15Hz, the predicted frequency drift Δf=8Hz within 0.3s is used to correct the feed rate from 0.002mm / s to 0.0018mm / s and generate grinding parameter instructions.

[0128] III. Execution Driver Module 4;

[0129] The execution drive module 4 is connected to the dynamic mapping control module 3 and is used to drive the grinding head to adjust the working parameters according to the grinding parameter instructions.

[0130] IV. Closed-loop self-calibration module 5;

[0131] The closed-loop self-calibration module 5 connects the frequency test compensation module 2 and the dynamic mapping control module 3, and includes an effect acquisition unit 51, a compensation coefficient correction unit 52 and a model weight optimization unit 53.

[0132] 1. Effect Acquisition Unit 51:

[0133] Used to collect the actual frequency convergence effect of the crystal after grinding.

[0134] 2. Compensation coefficient correction unit 52:

[0135] Used to correct the compensation coefficients in the compensation coefficient library based on the actual frequency convergence effect.

[0136] 3. Model weight optimization unit 53:

[0137] This is used to optimize the model weights of the dynamic mapping model 31 based on the actual frequency convergence effect.

[0138] In this embodiment, the closed-loop self-calibration cycle is set to calibrate once every 50 crystals are ground.

[0139] V. Working Principle;

[0140] The piezoelectric crystal grinding control system based on frequency testing technology in this embodiment achieves precise grinding through a multi-level linkage control mechanism of "real-time frequency testing - vibration compensation - dynamic mapping control - advanced prediction correction - closed-loop self-calibration".

[0141] During the grinding process, the hardware acquisition module 1 synchronously acquires the original resonant frequency signal, vibration intensity, and grinding head operating parameters; the frequency test compensation module 2 uses the signal separation unit 21 to remove vibration interference and combines the compensation coefficient library to perform accurate frequency calculation, obtaining the accurate resonant frequency f and its rate of change v; the dynamic mapping control module 3 takes f, the rate of change v, and the vibration intensity I as input, generates basic grinding parameters through the dynamic mapping model 31, and performs safety correction by the vibration condition correction unit 32 when the vibration intensity exceeds 20Hz; the frequency drift advance prediction unit 33 predicts the frequency drift trend 0.3s in advance and performs advance correction on the grinding parameters; the execution drive module 4 completes the parameter response within ≤0.2s; and the closed-loop self-calibration module 5 continuously optimizes the compensation coefficient and model weights based on the actual frequency convergence effect.

[0142] VI. Technical Effects;

[0143] After grinding and processing in this embodiment, the actual resonant frequency of the AT-cut quartz piezoelectric crystal is 8002Hz, the frequency convergence accuracy is ±2Hz, and the frequency deviation between the center and edge of the crystal surface is 12Hz.

[0144] Compared to traditional grinding methods that do not employ vibration compensation and dynamic mapping control, this embodiment has the following technical advantages:

[0145] 1. Vibration interference is removed through the frequency test compensation module 2, significantly improving the accuracy of frequency testing;

[0146] 2. Through a three-input-three-output dynamic mapping model 31, precise linkage control between frequency change and grinding parameters is achieved;

[0147] 3. By using a 0.3s advance prediction mechanism, the frequency drift trend can be corrected in advance, thereby improving the frequency convergence speed;

[0148] 4. Through a closed-loop self-calibration mechanism, adaptive optimization is achieved during long-term system operation, thereby improving the yield rate;

[0149] 5. Based on mature industrial-grade hardware, the modification cost is low and the industrial adaptability is strong.

[0150] Any modifications or equivalent substitutions made by those skilled in the art to the sensor model, model algorithm, self-calibration cycle, etc., without departing from the core technical concept of this invention—namely, achieving precise control of piezoelectric crystal grinding based on frequency testing technology—should fall within the protection scope of this invention. This invention is not only applicable to quartz piezoelectric crystals, but also to the grinding and processing control of piezoelectric crystals such as lithium niobate and lithium tantalate, as well as other precision brittle materials.

[0151] Example 2 is the second embodiment of the present invention. Based on Example 1, this embodiment further elaborates on the structure and signal processing flow of the hardware acquisition module 1, the signal separation unit 21, and the precise frequency calculation unit 22, so as to improve the anti-vibration interference capability and frequency calculation accuracy of the frequency testing technology in the piezoelectric crystal grinding process.

[0152] I. Structure and working principle of hardware acquisition module 1;

[0153] In the piezoelectric crystal grinding control system based on frequency testing technology in this embodiment, the hardware acquisition module 1 includes: a frequency testing probe unit 11, a micro vibration acquisition unit 12, a grinding head status sensing unit 13, and an industrial bus unit 14.

[0154] 1. Frequency test probe unit 11:

[0155] The frequency test probe unit 11 is located beside the grinding head of the grinding equipment, facing the processing surface of the piezoelectric crystal to be ground, and is used to acquire the original resonant frequency signal of the piezoelectric crystal in real time. As the core execution component of the frequency testing technology, the frequency test probe unit 11 continuously outputs a frequency response signal reflecting the change in crystal thickness during the grinding process.

[0156] 2. Miniature vibration acquisition unit 12:

[0157] The miniature vibration acquisition unit 12 is integrated on the frequency test probe unit 11, operates synchronously with the frequency test probe unit 11 and is located at the same spatial point, and is used to acquire environmental vibration data during the grinding process. The environmental vibration data includes at least the vibration frequency, amplitude and vibration intensity.

[0158] Since the miniature vibration acquisition unit 12 and the frequency test probe unit 11 are arranged at the same point, it can be ensured that the signals acquired by the two are consistent in the spatial dimension, thereby avoiding phase errors introduced by spatial differences and providing a basis for subsequent phase compensation and vibration coupling separation.

[0159] 3. Grinding head status sensing unit 13:

[0160] The grinding head status sensing unit 13 includes a speed sensor, a pressure sensor, and a feed sensor, which are respectively installed on the driving mechanism, pressure regulating mechanism, and feed mechanism of the grinding head. They are used to collect the grinding head speed, grinding pressure, and grinding head feed to form a data stream of grinding head working parameters.

[0161] The working parameters of the grinding head are correlated with the precise resonance frequency and the precise frequency change rate on the same time axis to provide synchronous input data for the dynamic mapping control module 3.

[0162] 4. Industrial Bus Unit 14:

[0163] The industrial bus unit 14 adopts the Profinet industrial Ethernet protocol and is connected to the frequency test probe unit 11, the micro vibration acquisition unit 12 and the grinding head status sensing unit 13 respectively. It is used to synchronously transmit the original resonant frequency signal, environmental vibration data and grinding head working parameters to the frequency test compensation module 2 and the dynamic mapping control module 3.

[0164] In this embodiment, the preset delay threshold is set to 0.1s, and the transmission delay of the industrial bus unit 14 is less than or equal to 0.1s. By setting this preset delay threshold, the frequency test data and the grinding head operating parameters are strictly synchronized in the time dimension, thereby avoiding control errors caused by data asynchrony.

[0165] The technical benefits are: by using common-point acquisition and low-latency synchronous transmission, strict time alignment between frequency signals and vibration signals is achieved, laying the foundation for high-precision vibration coupling separation.

[0166] II. Structure and working principle of signal separation unit 21;

[0167] Reference Figure 2 The signal separation unit 21 includes: a phase alignment subunit 211, an amplitude normalization subunit 212, a frequency domain interference identification subunit 213, and an adaptive destructive reconstruction subunit 214.

[0168] 1. Phase alignment subunit 211:

[0169] The phase alignment subunit 211 connects the frequency test probe unit 11 and the micro vibration acquisition unit 12, and is used to perform phase delay compensation on the environmental vibration data so that it is precisely synchronized with the original resonant frequency signal on the time axis.

[0170] Specifically, the phase offset is obtained by calculating the cross-correlation function of the two signals, and the phase shift compensation is performed on the environmental vibration signal to eliminate the slight delay difference caused by the acquisition link and signal propagation.

[0171] 2. Amplitude normalization subunit 212:

[0172] The amplitude normalization subunit 212 is connected to the phase alignment subunit 211 and is used to perform amplitude normalization processing on the time-aligned environmental vibration signal, output the normalized vibration signal, and unify its amplitude range to the standard range to facilitate subsequent frequency domain matching.

[0173] 3. Frequency domain interference identification subunit 213:

[0174] The frequency domain interference identification subunit 213 performs fast Fourier transform on the time-aligned original resonance frequency signal and the normalized vibration signal to obtain their respective spectra; and identifies the frequency components in the original frequency spectrum that coincide with the main frequency of the vibration spectrum through a frequency matching algorithm, and marks them as vibration coupling interference frequency components.

[0175] This step enables accurate identification of vibrational coupling interference, avoiding the mistaken elimination of the crystal's own resonance components as interference.

[0176] 4. Adaptive cancellation reconstruction subunit 214:

[0177] The adaptive cancellation reconstruction subunit 214 incorporates an adaptive filtering algorithm to generate a cancellation signal based on the frequency components of the vibration coupling interference and the normalized vibration signal, and then superimposes it with the original resonance frequency signal to achieve interference cancellation.

[0178] The residual signal is then subjected to inverse Fourier transform to reconstruct the resonant frequency signal of the piezoelectric crystal in the time domain, and the vibration coupling interference signal is output to the precise frequency calculation unit 22.

[0179] The technical benefits are as follows: by using frequency domain identification and adaptive cancellation mechanism, vibration coupling interference is eliminated from the source, improving the anti-interference capability of frequency testing and significantly enhancing the stability of subsequent frequency calculation.

[0180] III. Structure and working principle of the precision frequency calculation unit 22;

[0181] Reference Figure 3 The precise frequency calculation unit 22 includes: waveform preprocessing subunit 221, instantaneous frequency calculation subunit 222, frequency change rate calculation subunit 223, and filtering and compensation subunit 224.

[0182] 1. Waveform preprocessing subunit 221:

[0183] The waveform preprocessing subunit 221 performs time-domain analysis on the resonant frequency signal of the piezoelectric crystal itself, captures transient change moments with amplitude exceeding a preset threshold and marks them as suspected abnormal periods.

[0184] In this embodiment, the preset threshold is set to 20% of the peak amplitude of the normal resonant waveform. When the instantaneous amplitude exceeds this threshold, it is determined to be an abnormal data point and is removed.

[0185] After removing outlier data, the remaining valid segments are stitched together, and cubic spline interpolation is used to smooth the discontinuities, thus reconstructing a continuous and pure resonant frequency waveform.

[0186] 2. Instantaneous frequency calculation subunit 222:

[0187] The instantaneous frequency calculation subunit 222 performs zero-crossing detection on the reconstructed waveform, calculates the instantaneous resonant frequency value, and outputs it continuously at a preset period.

[0188] 3. Frequency change rate calculation subunit 223:

[0189] The frequency change rate calculation subunit 223 calculates the real-time frequency change rate by dividing the difference between the current instantaneous resonant frequency value and the previous period value by the sampling period.

[0190] 4. Filtering and Compensation Subunit 224:

[0191] The filtering and compensation subunit 224 performs a moving average filter on the instantaneous resonant frequency value and real-time frequency change rate within a preset time window; the preset time window is set to 0.5s.

[0192] Meanwhile, based on the real-time vibration intensity transmitted by hardware acquisition module 1, the corresponding frequency compensation coefficient is matched from the compensation coefficient library as the filtering weight to participate in the weighted filtering, and finally the accurate resonance frequency and accurate frequency change rate are output.

[0193] The technical benefits are as follows: by eliminating anomalies, weighted filtering, and adaptive compensation mechanisms for vibration intensity, the accuracy and consistency of the calculation of precise resonant frequency and precise frequency change rate are improved, ensuring stable and reliable frequency test results under different vibration conditions.

[0194] IV. Overall Working Principle and Comprehensive Technical Effects;

[0195] During the grinding process, the hardware acquisition module 1 achieves synchronous acquisition at common points; the signal separation unit 21 removes vibration coupling interference through phase alignment, frequency domain recognition and adaptive cancellation mechanism; the precise frequency calculation unit 22 outputs precise resonance frequency and precise frequency change rate through anomaly elimination and vibration intensity weighted compensation; the above results are transmitted to the dynamic mapping control module 3 in real time to realize the dynamic adjustment of grinding parameters.

[0196] Through the above structural design and signal processing flow, this embodiment achieves:

[0197] 1. Accurate identification and active cancellation of vibration coupling interference;

[0198] 2. Frequency testing stability under high-frequency vibration conditions is significantly improved;

[0199] 3. Frequency test data and control parameters are highly synchronized;

[0200] 4. Provides high-confidence input data for dynamic mapping control.

[0201] This embodiment improves the overall accuracy and stability of the piezoelectric crystal polishing control system based on frequency testing technology by using modular signal processing and thresholding control strategies without adding complex hardware structures.

[0202] Example 3 is the third embodiment of the present invention. Based on the previous embodiments, this embodiment further elaborates on the structure, construction method, reasoning logic rules, vibration condition correction and frequency drift advance prediction mechanism of the dynamic mapping model 31 in the dynamic mapping control module 3, and gives an integrated mathematical expression to achieve precise control of piezoelectric crystal grinding based on frequency testing technology.

[0203] I. Structure and working principle of dynamic mapping model 31;

[0204] Reference Figure 4 , Figure 5 and Figure 6 After the frequency test compensation module 2 transmits the calculated precise resonant frequency f and precise frequency change rate v to the dynamic mapping control module 3, the dynamic mapping model 31 first activates the crystal characteristic matching subunit 311. This subunit connects the precise frequency calculation unit 22 and the hardware acquisition module 1. Based on the initial precise resonant frequency value acquired at the start of grinding, it identifies the crystal orientation type (e.g., AT-cut, BT-cut, or SC-cut) and thickness range of the piezoelectric crystal to be ground. Since crystals with different crystal orientations and thicknesses have different inherent frequencies and processing characteristics, the crystal characteristic matching subunit 311 queries the pre-stored crystal characteristic database based on the identification results, matches the exclusive model parameter set corresponding to the crystal, and outputs it to the subsequent deep neural network inference subunit 313. This ensures that the initial configuration of the deep neural network matches the current processing object, achieving adaptive adaptation to crystals of different specifications.

[0205] Subsequently, the feature fusion subunit 312 connects to the crystal characteristic matching subunit 311, receiving the precise resonant frequency f, the precise frequency change rate v, and the real-time vibration intensity I transmitted by the hardware acquisition module 1 as raw inputs. To fully exploit the rich information contained in these parameters, the feature fusion subunit 312 performs feature extraction on the three input quantities at different scales: wavelet packet decomposition is performed on the precise resonant frequency f to extract time-frequency features in different frequency bands, reflecting the detailed fluctuations of the frequency signal; first-order difference operation is performed on the precise frequency change rate v to extract its trend features, reflecting the acceleration and continuity of the frequency change direction; and short-time Fourier transform is performed on the real-time vibration intensity I to extract its energy distribution features and dominant frequency components, reflecting the intensity and frequency composition of the current vibration. After feature extraction, the feature fusion subunit 312 fuses the extracted time-frequency features, trend features, and energy distribution features at the feature layer to construct a fused feature vector that comprehensively characterizes the current crystal properties, frequency state, and operating environment.

[0206] The deep neural network inference subunit 313 is connected to the feature fusion subunit 312. It contains a pre-built deep neural network model, whose network structure, number of layers, and initial weights are dynamically configured based on the model parameter set output by the crystal characteristic matching subunit 311. The deep neural network inference subunit 313 inputs the fused feature vector into the configured deep neural network. After nonlinear transformations and mapping calculations through multiple hidden layers, it outputs a set of intermediate control parameters: basic rotational speed, basic pressure, and basic feed rate. These basic parameters preliminarily reflect the theoretically appropriate grinding parameters under the current frequency and vibration conditions.

[0207] Because deep neural networks may exhibit mapping errors under boundary conditions outside the training sample coverage, the nonlinear correction and verification subunit 314 is connected to the deep neural network inference subunit 313 to further process the basic parameters. First, this subunit incorporates a nonlinear correction function calibrated using extensive historical grinding data to compensate for the deep neural network's output error under boundary conditions, outputting the corrected basic control parameters. Subsequently, a process constraint rule base performs safety verification on the corrected parameters. This rule base stores the safety threshold ranges for the grinding process, such as a rotational speed safety threshold range of 0 r / min to 5000 r / min, a pressure safety threshold range of 0 N to 200 N, and a feed rate safety threshold range of 0 mm / min to 10 mm / min. When the corrected parameters exceed the corresponding safety threshold range, the nonlinear correction and verification subunit 314 automatically clamps the out-of-limit parameter values ​​to the nearest threshold boundary, generating grinding head rotational speed, grinding pressure, and grinding head feed rate that meet process safety requirements.

[0208] In actual grinding processes, material removal rate and surface quality are often two mutually restrictive indicators. Rapid material removal is prioritized in the initial stages of grinding, while surface precision is sought in the later stages. To address this, a multi-objective optimization balancing subunit 315 is connected to a nonlinear correction and verification subunit 314. Based on the weighting requirements of removal rate and surface quality at the current grinding stage (the weights can be set by the host computer or automatically calculated based on the remaining thickness), the subunit performs multi-objective optimization balancing of the three parameters. For example, in the rough grinding stage, a higher weight is assigned to the removal rate, resulting in a parameter combination that favors efficient removal; in the fine grinding stage, a higher weight is assigned to surface quality, resulting in a parameter combination that prioritizes precision. After optimization balancing, the optimized three grinding parameters are generated and output to the vibration condition correction unit 32.

[0209] When the optimized grinding parameters enter the vibration condition correction unit 32, the vibration analysis and grading subunit 321 first connects to the hardware acquisition module 1 to perform in-depth time-frequency analysis on the real-time vibration intensity I. This subunit uses short-time Fourier transform and wavelet packet decomposition techniques to extract the time-frequency distribution characteristics of the current vibration signal (i.e., the distribution of vibration energy over time at various frequencies), the dominant frequency (the frequency component with the highest energy), and the energy concentration characteristics (the degree of concentration of vibration energy in a specific frequency band). Simultaneously, based on the preset vibration intensity grading threshold range, the real-time vibration intensity is divided into different levels: when the vibration intensity is less than 2 m / s², the vibration intensity is classified into different levels. 2 When the vibration intensity is 2 m / s, it is judged as normal vibration level; when the vibration intensity is 2 m / s 2 Up to 5m / s 2 When the vibration intensity is between 5 m / s², it is classified as a moderate intensity vibration; when the vibration intensity is greater than 5 m / s², it is classified as a moderate intensity vibration. 2 At that time, it was determined to be a strong impact vibration level.

[0210] The decoupling and weight adjustment subunit 322 is connected to the vibration analysis and grading subunit 321 and the dynamic mapping model 31, respectively. Based on the dominant vibration frequency and energy concentration characteristics obtained from the analysis, it analyzes the degree of differential influence of the current vibration on each output parameter (rotation speed, pressure, feed rate) and generates corresponding differential influence coefficients. For example, if the dominant vibration frequency is close to the grinding head rotation frequency, it indicates that the vibration has a greater impact on the rotation speed; if the vibration energy is concentrated in the low frequency range, it may have a greater impact on the stability of pressure control. Combining the vibration level and differential influence coefficients, the decoupling and weight adjustment subunit 322 generates model weight correction coefficients that match the current vibration condition and applies these correction coefficients to the hidden layer node weights of the dynamic mapping model 31, realizing the dynamic adjustment of the model parameter weights, which is equivalent to making the model "aware" of the current vibration environment and adjusting its internal sensitivity.

[0211] The parameter local correction subunit 323 is connected to the decoupling and weight adjustment subunit 322. Based on the weighted dynamic mapping model 31, it recalculates the grinding head speed, grinding pressure, and grinding head feed rate, and performs targeted local corrections. Specifically, when vibration interference is identified, the parameter local correction subunit 323 reduces the grinding head speed to decrease the excitation source, fine-tunes the grinding pressure to maintain a stable removal rate, and reduces the feed rate to prevent overcutting.

[0212] To address vibration conditions of varying severity, the condition discrimination and switching subunit 324 is connected to a parameter local correction subunit 323, which has a pre-built condition mode library including a smooth grinding mode, vibration suppression mode, and impact protection mode. The condition discrimination and switching subunit 324 automatically determines the appropriate condition mode to switch to based on the vibration level and vibration attenuation trend (i.e., whether the vibration tends to increase or decrease), and executes the corresponding parameter correction strategy: in smooth grinding mode, a small-scale weight adjustment strategy is used to maintain processing efficiency; in vibration suppression mode, an active speed reduction combined with pressure fine-tuning correction strategy is used to actively suppress vibration generation; in impact protection mode, a rapid tool retraction combined with pressure unloading correction strategy is used to prioritize the protection of the crystal and equipment safety.

[0213] The correction output is connected to the pre-evaluation subunit 325 via a parameter local correction subunit 323. First, the correction amplitude of the parameters is limited to ensure that the amount of a single correction does not exceed a safe threshold (e.g., a single speed correction does not exceed 200 r / min, a single pressure correction does not exceed 20 N, and a single feed rate correction does not exceed 0.5 mm / min), preventing drastic parameter fluctuations that could lead to system instability. After completing the limitation, this subunit outputs the limited correction result to the frequency drift lead prediction unit 33. Simultaneously, this subunit pre-evaluates the degree of improvement that this correction will have on the subsequent vibration suppression effect based on the current trend of vibration intensity changes, and sends the pre-evaluation result to the model weight optimization unit 53 as a reference for subsequent model weight optimization.

[0214] The corrected grinding parameters, processed by the vibration condition correction unit 32, are transmitted to the frequency drift advance prediction unit 33. The core of this unit lies in its built-in drift prediction model storage subunit 331, which stores a pre-constructed vibration intensity-frequency drift prediction model. This model is constructed using extensive experimental data: under different vibration intensities, the actual frequency drift of the piezoelectric crystal is collected within a preset advance time window (e.g., 0.3 s). A method combining linear fitting and nonlinear correction is used to establish a model relating vibration intensity I to frequency drift Δf, with a model fit of no less than 0.98, ensuring the accuracy of the prediction.

[0215] The drift prediction subunit 332 is connected to the vibration condition correction unit 32 and the drift prediction model storage subunit 331. It receives the real-time vibration intensity I after it has been processed by the vibration condition correction unit 32, and inputs it into the vibration intensity-frequency drift prediction model to quickly calculate the predicted frequency drift within a preset lead time window. This step realizes the "prediction" of future frequency changes, improving control from "post-event feedback" to "pre-event prediction".

[0216] The advanced correction execution subunit 333 connects to the drift prediction subunit 332 and the vibration condition correction unit 32. Based on the predicted frequency drift, it generates corresponding speed compensation values, pressure compensation values, and feed compensation values. The calculation principle for the compensation values ​​is: when the predicted frequency will drift upward (i.e., become thicker), the pressure or speed is appropriately increased to accelerate removal; when the predicted frequency will drift downward (i.e., become thinner), the pressure or speed is appropriately decreased to prevent over-cutting. Subsequently, these compensation values ​​are superimposed on the corrected grinding parameters output by the vibration condition correction unit 32 to generate the final grinding parameter command, which is then output to the execution drive module 4.

[0217] To adapt to varying vibration rates, the time window adaptive subunit 334 is connected to the drift prediction subunit 332 to monitor the rate of change of vibration intensity in real time. When the vibration rate of change is large, indicating a rapid deterioration or improvement in the operating conditions, the subunit automatically shortens the length of the preset lead time window (e.g., from 0.3s to 0.15s) to improve the timeliness of the prediction. When the vibration rate of change is small, indicating a relatively stable operating condition, the lead time window is appropriately extended (e.g., extended to 0.5s) to obtain more sufficient adjustment time.

[0218] Finally, the prediction optimization subunit 335 is connected to the advance correction execution subunit 333 and the closed-loop self-calibration module 5 to realize the self-learning and continuous optimization of the prediction model. After each grinding process, the prediction optimization subunit 335 obtains the actual frequency convergence effect of this grinding from the effect acquisition unit 51 in the closed-loop self-calibration module 5, compares the actual frequency drift with the previously predicted drift, and evaluates the prediction error. Based on this prediction error, an online learning algorithm is used to fine-tune and correct the parameters of the prediction model in the drift prediction model storage subunit 331, making the model increasingly accurate in subsequent processing and realizing the self-evolution of prediction capabilities.

[0219] II. Model building steps;

[0220] S1: Crystal property matching: based on the initial precise resonance frequency Given the crystal thickness h, match the crystal type and parameter set in the pre-stored crystal property database. This forms the model parameter set.

[0221] S2: Feature Fusion: Time-frequency analysis is performed on the precise resonant frequency f, precise frequency change rate v, and real-time vibration intensity I: trend features are extracted from f and v through wavelet packet decomposition. , The energy distribution is obtained by short-time Fourier transform. ; Fuse all features into a vector .

[0222] S3: Deep Neural Network Training: Employing a Multi-Layer Feedforward Network The activation function is a regularized error function. and normalized exponential function The combination of these parameters outputs the basic grinding parameters. The loss function combines frequency convergence error and grinding surface quality index, and is trained to the output error. .

[0223] S4: Nonlinear Correction and Verification: Through the nonlinear correction function Boundary error compensation is performed on the initial output parameters, and the output is clamped using the process constraint rule library to ensure output accuracy. Do not exceed the safety threshold.

[0224] S5: Multi-objective optimization balance: Define removal rate weights based on the current grinding stage. With surface quality weight By optimizing the function The parameters are weighted and adjusted to generate the final optimized parameters. .

[0225] S6: Online Model Update: The online model update subunit 316 receives the optimized weights output by the closed-loop self-calibration module 5 and performs incremental training updates. To adapt to new crystal specifications or vibration conditions.

[0226] III. Dynamic Mapping Model 31 Core Algorithm Formula;

[0227] Based on the above steps, this embodiment constructs the following innovative dynamic mapping formula:

[0228] ;

[0229] Where Y is the optimized grinding parameter vector, including the grinding head rotation speed. Grinding pressure and grinding head feed rate ;

[0230] The frequency integration space is represented by the integration variable, which is used for higher-order frequency domain weighted calculations.

[0231] This is the Gamma function, used to adjust the weights of the nonlinear response of high-frequency vibrations to the output parameters;

[0232] K is the number of segmented sampling points, representing the length of the fused feature vector;

[0233] For Bessel functions, used to describe the response coupling of vibrational energy at different frequencies;

[0234] The precise resonance frequency represents the currently acquired crystal resonance frequency;

[0235] This is the initial crystal resonance frequency, used as a reference for crystal property matching subunits;

[0236] This is the standard deviation of frequency fluctuation, used to normalize frequency error and ensure numerical stability.

[0237] The precise rate of change of frequency represents the instantaneous change of frequency over time;

[0238] The standard deviation of the rate of change of frequency is used to normalize the rate of change of frequency.

[0239] This is the error function, used for nonlinear mapping of vibration influence weights;

[0240] I represents the real-time vibration intensity, which serves as a normalization factor for the control parameters, adjusting the dynamic correction of the grinding parameters by the vibration.

[0241] The Riemann Zeta function value is used to normalize the weighting coefficients of the fused features, ensuring a balance between the weights of the integral and summation terms.

[0242] For frequency trend features, it represents the time-frequency trend of the i-th segment;

[0243] The frequency change rate trend feature represents the change rate trend of the i-th segment;

[0244] For vibration energy distribution, denoted as the normalized value of vibration energy at frequency.

[0245] Explanation of formula range:

[0246] Integral term This indicates the continuous influence of vibration coupling on grinding parameters;

[0247] Summation term This represents the normalized contribution of the fused feature vectors;

[0248] Overall Results It reflects the optimized adjustment of grinding parameters within the safe threshold.

[0249] This formula comprehensively maps frequency, rate of change of frequency, and vibration intensity through higher-order integrals, Bessel functions, error functions, and Riemann Zeta functions, taking into account individual crystal differences and vibration coupling interference, to achieve dynamic, nonlinear, and proactive control. At the same time, this formula realizes a nonlinear, higher-order mapping relationship between input data and grinding parameters, enabling the piezoelectric crystal grinding control system to dynamically, accurately, and proactively adjust grinding parameters, significantly improving frequency convergence accuracy, reducing the impact of vibration interference, and improving crystal processing yield and surface quality. It also supports adaptive control for different crystal orientations, thicknesses, and vibration conditions.

[0250] IV. Technical Effects;

[0251] Through the collaborative work of the dynamic mapping model 31, vibration condition correction unit 32, and frequency drift advance prediction unit 33, this embodiment realizes full-link intelligent control from "crystal characteristic adaptation" to "multi-dimensional feature extraction", and then to "vibration condition adaptive correction" and "frequency drift advance prediction", which significantly improves the accuracy, response speed and adaptive capability of frequency control in the piezoelectric crystal grinding process.

[0252] Example 4 is the fourth embodiment of the present invention. In this embodiment, the execution drive module 4 and the closed-loop self-calibration module 5 in the aforementioned piezoelectric crystal grinding control system based on frequency testing technology are further optimized to achieve closed-loop control of grinding parameter execution and effect feedback, thereby realizing high precision, real-time response and adaptive optimization of grinding processing.

[0253] I. Working principle of execution driver module 4;

[0254] 1. Instruction parsing and allocation unit 41:

[0255] The execution drive module 4 first receives grinding parameter commands from the frequency drift advance prediction unit 33, including the target grinding head speed. Grinding pressure and grinding head feed rate The instruction parsing and allocation unit parses these three target parameters and allocates them to the corresponding drive channels, providing input signals to each drive unit.

[0256] 2. Feedforward compensation processing:

[0257] The feedforward compensation unit 42 performs advance compensation calculations on the target parameters based on the inertia, response delay characteristics, and real-time vibration environment of each drive mechanism, and generates drive commands with compensation. , , This ensures that the grinding head can reach the target value in a timely manner during actual response, avoiding a decrease in frequency control accuracy due to drive lag. The feedforward compensation formula can be expressed as:

[0258] , , ;

[0259] in, , , These are the compensated grinding head speed, pressure, and feed rate, respectively.

[0260] , , These are the inertia compensation coefficients for each drive unit;

[0261] , , These are the drive response delays, respectively.

[0262] The feedforward compensation effect of the above formula is that the grinding head can reach the target parameters earlier in the actual response, reducing frequency lag error.

[0263] 3. Multi-axis collaborative control:

[0264] The speed drive unit 43, pressure drive unit 44, and feed drive unit 45 each receive the feedforward compensated commands, perform real-time adjustments, and simultaneously transmit feedback signals (actual speed, actual pressure, and actual feed rate) back to the multi-axis coordination unit 46. The multi-axis coordination unit 46 monitors the response status of each drive unit. When the response lag of any unit exceeds a preset synchronization deviation threshold (e.g., ...), ... When the time reaches 0.05s, a waiting or deceleration command is automatically issued to ensure the timing coordination of the grinding head speed, pressure and feed rate, and to avoid single-axis abnormalities affecting the overall grinding effect.

[0265] 4. Status self-check and safety protection:

[0266] The status self-test unit 47 collects the temperature, current and other safety parameters of each drive unit in real time. When any parameter exceeds the preset operating safety threshold (such as the upper temperature limit), it will detect the temperature. =80℃, upper limit of current When the voltage reaches 5A, a protective command is immediately issued to trigger deceleration, shutdown, or unloading operations to ensure the safety of the equipment and the crystal.

[0267] 5. Feedback on implementation results:

[0268] The execution effect feedback unit 48 packages and sends the actual achieved speed, pressure and feed data to the effect acquisition unit 51 to form grinding parameter execution effect data, which is also used for frequency convergence analysis of the closed-loop self-calibration module 5.

[0269] II. Closed-loop self-calibration principle;

[0270] 1. Effect Acquisition Unit 51:

[0271] Reference Figure 7 The effect acquisition unit 51 includes:

[0272] Execute data capture subunit 511: record the actual rotational speed, pressure, and feed rate of the grinding head and their time series;

[0273] Frequency final value acquisition subunit 512: Acquires the actual resonant frequency and spatial coordinates of multiple sampling points on the piezoelectric crystal processing surface;

[0274] Frequency convergence analysis subunit 513: Calculate the average crystal plane frequency Range and standard deviation, relative to the target frequency threshold Comparisons were made to generate crystal plane frequency consistency indicators and frequency compliance rates;

[0275] Convergence rate calculation subunit 514: Calculates the frequency convergence rate based on the time difference between the adjustment time and the grinding completion time, and the frequency change. To achieve the target convergence rate =40Hz / min comparison;

[0276] Spectral purity analysis subunit 515: Performs Fourier transform on frequency sampling points to analyze the proportion of spurious frequency energy. Output spectral purity coefficient ;

[0277] Grinding effect comprehensive evaluation subunit 516: The above indicators are weighted and integrated to generate a comprehensive score. The calculation formula for this process is configured as follows: ;

[0278] in, This is a comprehensive score.

[0279] An abnormal grinding identification subunit 517 is connected to a grinding effect comprehensive evaluation subunit 516, and is used to output an abnormal mark when the comprehensive score is lower than a preset qualified threshold (60 points);

[0280] The data packaging and transmission subunit 518 is connected to the comprehensive evaluation subunit 516 for grinding effect, the abnormal grinding identification subunit 517, the compensation coefficient correction unit 52, and the model weight optimization unit 53, respectively. It is used to package the effect score, abnormal marker, and data of each sampling point into effect feedback data and transmit them synchronously to the compensation coefficient correction unit 52 and the model weight optimization unit 53.

[0281] 2. Compensation coefficient correction unit 52:

[0282] The compensation coefficient correction unit 52 dynamically optimizes the frequency test compensation module 2 based on the effect acquisition results, referring to... Figure 8 ,include:

[0283] Deviation calculation subunit 521 is used to receive the actual frequency convergence effect and calculate the average crystal plane frequency. With target frequency threshold The frequency steady-state deviation was obtained by comparison. The process is configured as follows: ; and the actual frequency convergence rate convergence rate with target The dynamic deviation of the rate was obtained by comparison. The process is configured as follows: ;

[0284] Vibration condition backtracking subunit 522 is used to retrieve real-time vibration intensity time-series data for this grinding process and extract the average vibration intensity of key stages. and peak vibration intensity ;

[0285] The deviation-vibration correlation analysis subunit 523 is connected to the deviation calculation subunit 521 and the vibration condition backtracking subunit 522 respectively. It is used to perform correlation analysis between frequency deviation, velocity deviation and vibration intensity, identify the under-compensation and over-compensation regions of compensation coefficient under different vibration intensities, and generate deviation distribution map.

[0286] The compensation coefficient correction calculation subunit 524 is connected to the deviation-vibration correlation analysis subunit 523. It is used to calculate the compensation coefficient correction corresponding to each vibration intensity level based on the deviation distribution map and a preset correction step size factor. This process is configured as follows:

[0287] ;

[0288] in, This is the correction amount for the compensation coefficient. For steady-state frequency deviation, The target frequency threshold, =15Hz is the vibration threshold. This formula realizes nonlinear correction of frequency and velocity deviations under different vibration conditions, and improves the adaptability and accuracy of the compensation coefficient.

[0289] The coefficient library update subunit 525 is connected to the compensation coefficient correction calculation subunit 524 and the compensation coefficient library respectively, and is used to incrementally update the frequency compensation coefficients in the compensation coefficient library corresponding to the vibration intensity level according to the correction amount.

[0290] The correction effect verification subunit 526 and the connection coefficient library update subunit 525 are used to collect the actual frequency convergence effect after the next grinding is completed, compare the degree of deviation improvement, and trigger a second correction when the degree of improvement is lower than the preset improvement threshold (0.1).

[0291] The aging trend analysis subunit 527 is connected to the compensation coefficient correction calculation subunit 524, which is used to statistically analyze the changing trend of the compensation coefficient correction during multiple grinding processes and to determine the drift law of the compensation coefficient as the equipment ages.

[0292] The synchronous output subunit 528 is connected to the coefficient library update subunit 525 and the model weight optimization unit 53 respectively, and is used to synchronously transmit the key parameters of this compensation coefficient correction to the model weight optimization unit 53 as optimization reference.

[0293] III. Technical Effects;

[0294] 1. Rapid response and coordination: Feedforward compensation combined with multi-axis collaborative monitoring enables synchronous adjustment of grinding head speed, pressure and feed rate, with system response delay ≤0.2s, significantly reducing frequency lag error.

[0295] 2. Safety protection: The status self-test unit 47 realizes temperature, current and multi-axis synchronous safety monitoring to avoid crystal damage or equipment failure caused by operation exceeding the threshold.

[0296] 3. Closed-loop self-optimization: The effect acquisition unit 51, the compensation coefficient correction unit 52 and the model weight optimization unit 53 form a closed loop, enabling dynamic self-calibration of frequency test accuracy, grinding parameter control and crystal processing quality, adaptive adjustment of compensation coefficient, and long-term stable operation.

[0297] 4. Frequency convergence and consistency improvement: Through comprehensive evaluation of multiple indicators (average value, range, convergence rate, spectral purity), the compensation coefficient and model weight are corrected, so that the crystal surface frequency convergence accuracy reaches ±2Hz and the crystal surface consistency index is improved by more than 95%, which meets the industrial batch grinding requirements of high-precision piezoelectric crystals.

[0298] 5. Aging trend monitoring: Statistical analysis of the compensation coefficient correction amount can identify the aging pattern of the equipment, ensuring frequency accuracy and grinding quality under long-term operation.

[0299] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A piezoelectric crystal polishing control system based on frequency testing technology, characterized in that, include: The hardware acquisition module acquires the original resonant frequency signal, vibration intensity, and grinding head operating parameters. The frequency test compensation module, connected to the hardware acquisition module, includes: The signal separation unit is used to receive the original resonance frequency signal and vibration intensity, and output the crystal resonance signal and vibration interference signal. The precise frequency calculation unit is connected to the signal separation unit and the preset compensation coefficient library, and is used to calculate the precise resonance frequency and the precise frequency change rate based on the crystal resonance signal, the vibration interference signal and the compensation coefficient corresponding to the vibration intensity. The dynamic mapping control module, connecting the frequency test compensation module and the hardware acquisition module, includes: The dynamic mapping model is used to receive precise resonant frequency, precise frequency change rate and vibration intensity, and output grinding head speed, pressure and feed rate. Vibration condition correction unit, connected to the dynamic mapping model, is used to correct the model output when the vibration intensity exceeds the threshold. The frequency drift advance prediction unit is connected to the vibration condition correction unit. It is used to predict the frequency drift amount based on the vibration intensity and correct its output in advance to generate grinding parameter instructions. The execution drive module is connected to the dynamic mapping control module and is used to drive the grinding head to adjust the working parameters according to the grinding parameter instructions. A closed-loop self-calibration module, connecting the frequency test compensation module and the dynamic mapping control module, includes: The effect acquisition unit is used to acquire the actual frequency convergence effect of the crystal after grinding. A compensation coefficient correction unit, connected to the effect acquisition unit and the compensation coefficient library, is used to correct the compensation coefficient according to the actual frequency convergence effect. The model weight optimization unit, connected to the effect acquisition unit and the dynamic mapping model, is used to optimize the model weights based on the actual frequency convergence effect.

2. The piezoelectric crystal polishing control system based on frequency testing technology according to claim 1, characterized in that, The hardware acquisition module includes: The frequency test probe unit is located next to the grinding head of the grinding equipment and facing the processing surface of the piezoelectric crystal to be ground, and is used to collect the original resonant frequency signal of the piezoelectric crystal. A miniature vibration acquisition unit is integrated on the frequency test probe unit, works synchronously with the frequency test probe unit and is located at the same spatial point, and is used to acquire environmental vibration data during the grinding process. The environmental vibration data includes at least vibration frequency, amplitude and vibration intensity. The grinding head status sensing unit includes a speed sensor, a pressure sensor, and a feed sensor. The speed sensor is mounted on the driving mechanism of the grinding head to collect the grinding head speed. The pressure sensor is mounted on the pressure regulating mechanism of the grinding head to collect the grinding pressure. The feed sensor is mounted on the feeding mechanism of the grinding head to collect the grinding head feed amount. The industrial bus unit, using the Profinet industrial Ethernet protocol, is connected to the frequency test probe unit, the micro vibration acquisition unit, and the grinding head status sensing unit, respectively. It is used to synchronously transmit the original resonant frequency signal, the environmental vibration data, and the grinding head operating parameters to the frequency test compensation module and the dynamic mapping control module, and the transmission delay is less than or equal to a preset delay threshold.

3. The piezoelectric crystal polishing control system based on frequency testing technology according to claim 2, characterized in that, The signal separation unit includes: The phase alignment subunit, connected to the frequency test probe unit and the micro vibration acquisition unit, is used to perform phase delay compensation on the environmental vibration data to synchronize it with the original resonant frequency signal in time. An amplitude normalization subunit, connected to the phase alignment subunit, is used to normalize the amplitude of the time-aligned environmental vibration signal and output a normalized vibration signal. The frequency domain interference identification subunit, connected to the amplitude normalization subunit, is used to perform fast Fourier transform on the time-aligned original resonance frequency signal and the normalized vibration signal to obtain their respective spectra, and to identify the frequency components in the original frequency spectrum that coincide with the main frequency of the vibration spectrum through frequency matching, and mark them as vibration coupling interference frequency components. The adaptive cancellation reconstruction subunit is connected to the frequency domain interference identification subunit. It has a built-in adaptive filtering algorithm to generate a cancellation signal based on the frequency components of the vibration coupling interference and the normalized vibration signal. The cancellation signal is superimposed with the original resonance frequency signal to eliminate the vibration interference. The residual signal is then reconstructed by inverse Fourier transform to obtain the resonance frequency signal of the piezoelectric crystal itself in the time domain. The vibration coupling interference signal is then output to the precise frequency calculation unit.

4. The piezoelectric crystal polishing control system based on frequency testing technology according to claim 1, characterized in that, The precise frequency calculation unit includes: The waveform preprocessing subunit, connected to the signal separation unit, is used to perform time-domain analysis on the resonant frequency signal of the piezoelectric crystal itself, capture transient change moments with amplitude exceeding a preset threshold and mark them as suspected abnormal periods, remove abnormal data points, and then stitch the remaining effective segments together. Cubic spline interpolation is used to smooth the discontinuities and reconstruct a continuous and pure resonant frequency waveform. The instantaneous frequency calculation subunit is connected to the waveform preprocessing subunit and is used to perform zero-crossing detection on the reconstructed waveform, calculate the instantaneous resonant frequency value, and output it continuously at a preset period. The frequency change rate calculation subunit, connected to the instantaneous frequency calculation subunit, is used to calculate the difference between the current instantaneous resonant frequency value and the previous period value and divide it by the sampling period to obtain the real-time frequency change rate. The filtering and compensation subunit is connected to the instantaneous frequency calculation subunit, the frequency change rate calculation subunit, and the compensation coefficient library, respectively. It is used to perform moving average filtering on the instantaneous resonant frequency value and real-time frequency change rate within a preset time window, and to match the corresponding frequency compensation coefficient from the compensation coefficient library as the filtering weight based on the real-time vibration intensity transmitted by the hardware acquisition module, and participate in the weighted filtering to finally output the accurate resonant frequency and accurate frequency change rate.

5. The piezoelectric crystal polishing control system based on frequency testing technology according to claim 1, characterized in that, The dynamic mapping model includes: The crystal characteristic matching subunit is connected to the precise frequency calculation unit and the hardware acquisition module. It is used to identify the crystal orientation and thickness of the crystal based on the initial precise resonance frequency, match the pre-stored crystal characteristic database, and output the model parameter set. The feature fusion subunit, connected to the crystal characteristic matching subunit, is used to perform wavelet packet decomposition, difference and short-time Fourier transform on the precise resonance frequency, precise frequency change rate and real-time vibration intensity respectively, extract time-frequency features, trend features and energy distribution features, and fuse them into a fusion feature vector; A deep neural network inference subunit, connected to the feature fusion subunit, has a built-in deep neural network configured according to the model parameter set, used to map the fused feature vector to the base rotation speed, base pressure and base feed rate; The nonlinear correction and verification subunit, connected to the deep neural network inference subunit, is used to compensate for boundary errors through a nonlinear correction function, and then verify them through the process constraint rule base to clamp the out-of-limit parameters to a safe threshold, thereby generating grinding head speed, grinding pressure and grinding head feed that meet safety requirements. The multi-objective optimization balance subunit, connected to the nonlinear correction and verification subunit, is used to perform multi-objective optimization on the three parameters according to the weight requirements of removal rate and surface quality in the current grinding stage, generate optimized grinding parameters, and output them to the vibration condition correction unit. The online model update subunit is connected to the multi-objective optimization balancing subunit and the model weight optimization unit, respectively. It is used to receive the optimized parameter weights output by the model weight optimization unit and perform online incremental updates on the deep neural network.

6. The piezoelectric crystal polishing control system based on frequency testing technology according to claim 1, characterized in that, The vibration condition correction unit includes: The vibration analysis and grading subunit is connected to the hardware acquisition module and is used to perform short-time Fourier transform and wavelet packet decomposition on the real-time vibration intensity, extract the time-frequency distribution, main frequency, and energy accumulation characteristics, and classify the vibration intensity into normal, medium intensity, and strong impact levels according to the preset threshold range. The decoupling and weight adjustment subunit is connected to the vibration analysis and grading subunit and the dynamic mapping model, respectively. It is used to analyze the differential influence coefficients of vibration on each output parameter based on the dominant frequency and energy characteristics, and generate model weight correction coefficients in combination with vibration level. These coefficients are applied to the hidden layer weights of the dynamic mapping model to realize the dynamic adjustment of parameter weights. The parameter local correction subunit, connected to the decoupling and weight adjustment subunit, is used to recalculate the grinding head speed, grinding pressure and grinding head feed according to the adjusted model, and to perform local corrections such as speed reduction, pressure fine adjustment and deceleration. The working condition judgment and switching subunit is connected to the parameter local correction subunit. It has a built-in working condition mode library and is used to judge the current working condition mode according to the vibration level and vibration attenuation trend, and execute the corresponding parameter correction strategy, including small adjustment in stable grinding mode, active speed reduction combined with pressure fine adjustment in vibration suppression mode, and rapid tool retraction combined with pressure unloading in impact protection mode. The correction output and pre-evaluation subunit is connected to the parameter local correction subunit. It is used to limit the correction amplitude of the corrected parameters to ensure that the single correction amount does not exceed the safety threshold. The limited correction result is output to the frequency drift advance prediction unit. At the same time, the correction effect is pre-evaluated according to the vibration change trend, and the pre-evaluation result is sent to the model weight optimization unit as an optimization reference.

7. The piezoelectric crystal polishing control system based on frequency testing technology according to claim 1, characterized in that, The frequency drift advance prediction unit includes: The drift prediction model storage subunit has a built-in vibration intensity-frequency drift prediction model. The vibration intensity-frequency drift prediction model is constructed by experimentally collecting the actual frequency drift of the piezoelectric crystal within a preset lead time window under different vibration intensities, using a combination of linear fitting and nonlinear correction, and the model fitting degree is not lower than the preset fitting degree threshold. The drift prediction subunit is connected to the vibration condition correction unit and the drift prediction model storage subunit. It is used to receive the real-time vibration intensity after being processed by the vibration condition correction unit, input it into the vibration intensity-frequency drift prediction model, and calculate the predicted frequency drift within the preset lead time window. The advanced correction execution subunit is connected to the drift prediction subunit and the vibration condition correction unit. It is used to generate speed compensation value, pressure compensation value and feed compensation value according to the predicted frequency drift amount, and superimpose them on the corrected grinding parameters output by the vibration condition correction unit to generate the final grinding parameter command and output it to the execution drive module. The time window adaptive subunit, connected to the drift prediction subunit, is used to dynamically adjust the length of the preset lead time window according to the rate of change of vibration intensity. The prediction optimization subunit is connected to the advance correction execution subunit and the closed-loop self-calibration module, respectively. It is used to evaluate the prediction error based on the actual frequency convergence effect collected by the effect acquisition unit in the closed-loop self-calibration module, and to perform online parameter correction on the prediction model in the drift prediction model storage subunit based on the prediction error.

8. The piezoelectric crystal polishing control system based on frequency testing technology according to claim 1, characterized in that, The execution driver module includes: The instruction parsing and allocation unit, connected to the frequency drift advance prediction unit, is used to receive and parse the target rotation speed, pressure and feed value, and allocate them to the corresponding drive channel. The feedforward compensation unit, connected to the instruction parsing and allocation unit, is used to perform feedforward compensation on the target value based on the inertia and response delay characteristics of the actuator, and generate a drive instruction with advance compensation amount. The speed drive unit, connected to the feedforward compensation unit, is used to receive speed drive commands, drive the grinding head spindle motor to reach the target speed, and feed back the actual speed. The pressure drive unit, connected to the feedforward compensation unit, is used to receive pressure drive commands, control the pressure execution cylinder to output the target pressure and feed back the actual pressure. The feed drive unit is connected to the feedforward compensation unit and is used to receive feed drive commands, drive the feed servo motor to move according to the target feed amount and feed back the actual position. The multi-axis coordination unit is connected to each drive unit and is used to monitor the response status of each drive unit. When the response of any unit lags behind by more than a preset synchronization deviation threshold, a waiting or deceleration command is issued to ensure timing coordination. The status self-test unit is connected to the multi-axis collaborative unit and is used to collect the temperature and current status of each drive unit in real time. When the parameters exceed the operating safety threshold, a protective command is issued. The execution effect feedback unit is connected to each drive unit and the effect acquisition unit respectively. It is used to package the actual achieved speed, pressure and feed rate into execution effect data and transmit it to the effect acquisition unit.

9. The piezoelectric crystal polishing control system based on frequency testing technology according to claim 1, characterized in that, The effect acquisition unit includes: The execution data capture subunit, connected to the execution drive module, is used to capture the actual rotational speed, pressure, and feed rate values ​​reached by the grinding head after the grinding parameters are adjusted, and to record the adjustment time and duration. The frequency final value acquisition subunit is connected to the frequency test compensation module and is used to acquire the actual resonant frequency value and spatial position coordinates of multiple sampling points on the piezoelectric crystal processing surface after grinding. The frequency convergence analysis subunit is connected to the frequency final value acquisition subunit and is used to calculate the average value, range and standard deviation of the crystal plane frequency, compare it with the target frequency threshold, and output the crystal plane frequency consistency index and frequency compliance rate. The convergence rate calculation subunit is connected to the execution data capture subunit and the frequency final value acquisition subunit, respectively, and is used to calculate the actual frequency convergence rate based on the time difference between the adjustment time and the grinding completion time and the frequency change. The spectral purity analysis subunit is connected to the frequency final value acquisition subunit and is used to perform Fourier transform on the actual frequency value, analyze the proportion of spurious frequency energy, and output the spectral purity coefficient. The comprehensive evaluation subunit for grinding effect is connected to the frequency convergence analysis subunit, the convergence rate calculation subunit, and the spectral purity analysis subunit, respectively, and is used to weight and fuse the various indicators to generate a comprehensive score for grinding effect. An abnormal grinding identification subunit is connected to the grinding effect comprehensive evaluation subunit and is used to output an abnormal mark when the comprehensive score is lower than a preset qualified threshold. The data packaging and transmission subunit is connected to the comprehensive evaluation subunit for grinding effect, the abnormal grinding identification subunit, the compensation coefficient correction unit, and the model weight optimization unit, respectively. It is used to package the effect score, abnormal marker, and data of each sampling point into effect feedback data and transmit them synchronously to the compensation coefficient correction unit and the model weight optimization unit.

10. The piezoelectric crystal polishing control system based on frequency testing technology according to claim 1, characterized in that, The compensation coefficient correction unit includes: The deviation calculation subunit is connected to the effect acquisition unit and is used to receive the actual frequency convergence effect, compare the average crystal plane frequency with the target frequency threshold to obtain the steady-state frequency deviation, and compare the actual frequency convergence rate with the target convergence rate to obtain the dynamic rate deviation. The vibration condition backtracking subunit is connected to the hardware acquisition module and is used to retrieve the real-time vibration intensity time series data of this grinding and extract the average vibration intensity and peak vibration intensity of key stages. The deviation-vibration correlation analysis subunit is connected to the deviation calculation subunit and the vibration condition backtracking subunit, respectively. It is used to perform correlation analysis between frequency deviation, velocity deviation and vibration intensity, identify the under-compensation and over-compensation regions of the compensation coefficient under different vibration intensities, and generate a deviation distribution map. The compensation coefficient correction calculation subunit is connected to the deviation-vibration correlation analysis subunit and is used to calculate the compensation coefficient correction corresponding to each vibration intensity level based on the deviation distribution map and the preset correction step size factor. The coefficient library update subunit is connected to the compensation coefficient correction calculation subunit and the compensation coefficient library, respectively, and is used to incrementally update the frequency compensation coefficients in the compensation coefficient library corresponding to the vibration intensity level according to the correction amount. The correction effect verification subunit is connected to the coefficient library update subunit. It is used to collect the actual frequency convergence effect after the next grinding is completed, compare the degree of deviation improvement, and trigger a second correction when the degree of improvement is lower than the preset improvement threshold. The aging trend analysis subunit is connected to the compensation coefficient correction calculation subunit and is used to statistically analyze the changing trend of the compensation coefficient correction during multiple grinding processes and determine the drift law of the compensation coefficient as the equipment ages. The synchronous output subunit is connected to the coefficient library update subunit and the model weight optimization unit, respectively, and is used to synchronously transmit the key parameters of this compensation coefficient correction to the model weight optimization unit as an optimization reference.

Images