Piezoelectric quartz wafer grinding control method and system based on grinding frequency monitoring

Abstract

The invention provides a piezoelectric quartz wafer grinding control method and system based on grinding frequency monitoring, and relates to the technical field of grinding control, the method comprises the following steps: executing multi-source data perception of piezoelectric quartz wafer grinding, and establishing an original behavior observation data stream; behavior modeling of each round of grinding is carried out; the time sequence grinding behavior data set is used for conducting grinding working condition sensing; activating a non-contact thickness sensing unit to execute real-time piezoelectric quartz wafer thickness sensing; with the target grinding requirement as a control target, grinding control reconstruction based on the real-time sensing result and the thickness sensing result is executed, and a reconstruction result is established; and carrying out grinding control management on the piezoelectric quartz wafer. The technical problem that in the prior art, due to the fact that grinding control depends on fixed parameters and lacks real-time sensing and feedback, the quality of the piezoelectric quartz wafer is unstable is solved, the grinding process is optimized and adjusted through the multiple sensing sources, and the machining precision and quality of piezoelectric quartz wafer grinding are improved.

Description

Technical Field

[0001] The present application relates to the field of grinding control technology, and in particular to a piezoelectric quartz wafer grinding control method and system based on grinding frequency monitoring. Background Art

[0002] As an important electronic component, the performance of piezoelectric quartz wafers is closely related to processing accuracy, among which the grinding process is a core link that determines the final quality of the wafers. At present, the existing piezoelectric quartz wafer grinding control methods often rely on preset fixed process parameters, which can ensure a certain processing quality under static conditions. However, in actual applications, since the grinding process itself is a dynamic process, it is easily affected by multiple factors such as grinding wheel wear, abrasive particle size changes, temperature and humidity fluctuations, and the equipment's own status. It is difficult for fixed parameters to always adapt to the actual working conditions throughout the entire grinding process, resulting in the inability to perceive and compensate for small fluctuations in the grinding process in real time, thereby affecting the accuracy of the grinding process and the final quality of the wafer.

[0003] In summary, the prior art has a technical problem in that the quality of piezoelectric quartz wafers is unstable due to the reliance of grinding control on fixed parameters and the lack of real-time perception and feedback. Summary of the Invention

[0004] The purpose of this application is to provide a piezoelectric quartz wafer grinding control method and system based on grinding frequency monitoring, so as to solve the technical problem in the prior art that the grinding control relies on fixed parameters and lacks real-time perception and feedback, resulting in unstable quality of piezoelectric quartz wafers.

[0005] In view of the above problems, the present application provides a piezoelectric quartz wafer grinding control method and system based on grinding frequency monitoring.

[0006] In the first aspect, the present application provides a piezoelectric quartz wafer grinding control method based on grinding frequency monitoring, which is implemented by a piezoelectric quartz wafer grinding control system based on grinding frequency monitoring, wherein the piezoelectric quartz wafer grinding control method based on grinding frequency monitoring includes: performing multi-source data perception of piezoelectric quartz wafer grinding, establishing an original behavior observation data stream, and multi-source data perception includes grinding frequency perception, step perception, and contact perception; performing behavior modeling of each round of grinding based on the original behavior observation data stream, and establishing a time-series grinding behavior data set; using the time-series grinding behavior data set to perform grinding condition perception, and establish real-time perception results; activating a non-contact thickness perception unit to perform real-time piezoelectric quartz wafer thickness perception, and establish a thickness perception result; taking the target grinding requirement as the control target, performing grinding control reconstruction based on the real-time perception result and the thickness perception result, and establishing a reconstruction result; and performing piezoelectric quartz wafer grinding control management according to the reconstruction result.

[0007] Optionally, multi-dimensional working condition feature extraction is performed on the time-series grinding behavior data set to establish a working condition feature vector; the labeled historical data is used to train a working condition recognition model, the working condition feature vector is sent to the working condition recognition model, and a working condition label sequence is output; the working condition label sequence is used to perform stable trend perception, and the working condition label of the current node is enhanced using the stable trend perception result to establish a real-time perception result.

[0008] Optionally, a non-contact thickness sensing unit is enabled to obtain the original thickness data of the piezoelectric quartz wafer; input variables are extracted based on the original behavior observation data stream, and the input variables include grinding frequency data, unit time step displacement, and contact pressure response; thickness fitting analysis is performed based on the input variables and the original thickness data to establish a thickness prediction result; and the thickness prediction result is used to calibrate and compensate the thickness sensing result.

[0009] Optionally, the real-time perception result is used to evaluate the working condition stability of the current node and establish a first evaluation anomaly; the grinding stability anomaly analysis of the piezoelectric quartz wafer is performed to establish a static evaluation threshold; based on the static evaluation threshold, anomaly trigger identification of the first evaluation anomaly is performed, and if the anomaly trigger identification result is a trigger result, a reconstruction instruction is generated, and according to the reconstruction instruction, the target grinding requirement is used as the control target, and the grinding control reconstruction based on the real-time perception result and the thickness perception result is performed.

[0010] Optionally, if the abnormal trigger identification result is a non-triggering result, the timing grinding behavior data set is called to perform timing fitting prediction and establish a timing fitting prediction result; the static evaluation threshold is used to perform trigger analysis of the timing fitting prediction result to identify the abnormal trigger node; based on the abnormal trigger node, a prediction reconstruction instruction is configured, and grinding control reconstruction is performed according to the prediction reconstruction instruction.

[0011] Optionally, if the abnormal triggering node is a node outside a preset time period, a holding instruction is generated; and the original polishing parameters are maintained according to the holding instruction to perform piezoelectric quartz wafer polishing control management.

[0012] Optionally, the equipment state evolution, grinding pressure fluctuation, and vibration signal response data recorded in the real-time perception results are used to perform a traceability evolution analysis of R&D anomalies, extract the key parameter evolution pattern that triggers the anomaly, and establish a reconstruction taboo constraint based on the key parameter evolution pattern; obtain the current grinding roughness based on the real-time perception results; and execute control optimization under the reconstruction taboo constraint according to the thickness perception results, the current grinding roughness, and the target grinding requirements to establish a reconstruction result.

[0013] Optionally, an objective function is established, and the evaluation items of the objective function include a thickness evaluation item, a roughness evaluation item, and an operating condition stability evaluation item; after reconstructing the control space using the reconstruction taboo constraint, control optimization based on the objective function is performed in the reconstructed control space to establish a reconstruction result.

[0014] Optionally, the grinding quality detection data of the piezoelectric quartz wafer is recorded, and the real parameter control data is obtained; the grinding quality detection data and the real parameter control data are mapped into a batch parameter group and then saved to a control center.

[0015] In a second aspect, the present application also provides a piezoelectric quartz wafer grinding control system based on grinding frequency monitoring, which is used to execute the piezoelectric quartz wafer grinding control method based on grinding frequency monitoring as described in the first aspect, wherein the piezoelectric quartz wafer grinding control system based on grinding frequency monitoring includes: a multi-source data perception module, which is used to perform multi-source data perception of piezoelectric quartz wafer grinding and establish an original behavior observation data stream, and the multi-source data perception includes grinding frequency perception, step perception, and contact perception; a behavior modeling module, which is used to perform behavior modeling of each round of grinding based on the original behavior observation data stream and establish a time-series grinding behavior data set; a grinding condition perception module, which is used to use the time-series grinding behavior data set to perform grinding condition perception and establish real-time perception results; a thickness perception module, which is used to activate a non-contact thickness perception unit to perform real-time piezoelectric quartz wafer thickness perception and establish a thickness perception result; a grinding control reconstruction module, which is used to perform grinding control reconstruction based on the real-time perception result and thickness perception result with the target grinding requirement as the control target and establish a reconstruction result; and a grinding control management module, which is used to perform piezoelectric quartz wafer grinding control management according to the reconstruction result.

[0016] One or more technical solutions provided in this application have at least the following beneficial effects: By performing multi-source data perception of piezoelectric quartz wafer grinding, an original behavior observation data stream is established. The multi-source data perception includes grinding frequency perception, step perception, and contact perception. Based on the original behavior observation data stream, the behavior modeling of each round of grinding is performed to establish a time-series grinding behavior data set. The time-series grinding behavior data set is used to perform grinding working condition perception and establish real-time perception results. The non-contact thickness perception unit is activated to perform real-time piezoelectric quartz wafer thickness perception and establish thickness perception results. With the target grinding requirement as the control target, grinding control reconstruction based on the real-time perception results and thickness perception results is performed to establish a reconstruction result. The piezoelectric quartz wafer grinding control management is performed according to the reconstruction result. In other words, by introducing multiple perception sources, the behavior of the entire grinding process is modeled, the working condition changes during the grinding process are monitored in real time, and the grinding control parameters are dynamically reconstructed in combination with the non-contact thickness perception unit, thereby improving the processing accuracy and quality of piezoelectric quartz wafer grinding.

[0017] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, which can be implemented in accordance with the contents of the description, and to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically listed below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in this application or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely exemplary, and a person of ordinary skill in the art can obtain other drawings based on the provided drawings without creative work.

[0019] Figure 1 This is a flow chart of the piezoelectric quartz wafer polishing control method based on polishing frequency monitoring in this application.

[0020] Figure 2 This is a schematic diagram of the structure of the piezoelectric quartz wafer grinding control system based on grinding frequency monitoring in this application.

[0021] Explanation of the reference numerals: multi-source data perception module 11 , behavior modeling module 12 , grinding condition perception module 13 , thickness perception module 14 , grinding control reconstruction module 15 , grinding control management module 16 . DETAILED DESCRIPTION

[0022] This application provides a piezoelectric quartz wafer grinding control method and system based on grinding frequency monitoring, addressing the prior art technical issues of unstable piezoelectric quartz wafer quality, which arise from the reliance on fixed parameters and the lack of real-time sensing and feedback. By introducing multiple sensing sources, modeling the entire grinding process, and monitoring operating conditions in real time, combined with a non-contact thickness sensing unit, the grinding control parameters are dynamically reconstructed, improving the processing accuracy and quality of piezoelectric quartz wafer grinding.

[0023] Below, the technical solutions in this application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of this application, rather than all the embodiments of this application. It should be understood that this application is not limited to the example embodiments described herein. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. It should also be noted that, for the convenience of description, only the parts related to this application, rather than all of them, are shown in the accompanying drawings.

[0024] For example, see the attached Figure 1 The present application provides a piezoelectric quartz wafer grinding control method based on grinding frequency monitoring, wherein the piezoelectric quartz wafer grinding control method based on grinding frequency monitoring is executed by a piezoelectric quartz wafer grinding control system based on grinding frequency monitoring, and the piezoelectric quartz wafer grinding control method based on grinding frequency monitoring specifically includes the following steps: Perform multi-source data perception of piezoelectric quartz wafer grinding and establish the original behavior observation data stream. Multi-source data perception includes grinding frequency perception, step perception, and contact perception.

[0025] Specifically, the grinding process is comprehensively monitored using a variety of sensors, ensuring the capture of key dynamic information during the grinding process, including grinding frequency sensing, step sensing, and contact sensing. Piezoelectric sensors mounted on the grinding tool or wafer surface monitor the contact frequency between the tool and wafer in real time. Frequency fluctuations can reflect subtle changes in the grinding process, such as tool wear, changes in surface roughness, or changes in contact pressure. For example, when the wafer surface is uneven, frequency fluctuations increase, enabling adjustments to grinding parameters accordingly. Displacement sensors (such as laser displacement sensors or encoders) mounted on the wafer measure the displacement of the wafer during each feed during the grinding process. This is used to determine the relative position and feed rate between the tool and wafer, ensuring that the tool is operating at the set feed rate. Pressure sensors monitor the contact between the tool and wafer, including whether the grinding tool (grinding disc) and the piezoelectric quartz wafer are in contact, as well as the tightness of the contact and the amount of pressure. Changes in contact force and contact area are captured in real time to determine the uniformity of tool-wafer contact during grinding, allowing adjustments to grinding parameters such as pressure or speed.

[0026] Through multi-source data perception, a raw behavioral observation data stream is established. In other words, raw data collected from multiple sensing sources represents the real-time status of the grinding process. For example, during the grinding process between a diamond grinding wheel and a 100mm diameter piezoelectric quartz wafer, the grinding pressure is 2N and the grinding speed is 50rpm. During the grinding process, a vibration sensor monitors the frequency generated by the contact between the tool and the wafer, a displacement sensor records wafer displacement data, and a pressure sensor monitors contact pressure. Multi-source data perception reveals that during the initial grinding stage, the vibration frequency is 80Hz and the displacement is 0.1mm. During the grinding process, the frequency fluctuates significantly (90-110Hz), and the contact pressure gradually increases to 2.2N, indicating that the grinding wheel is beginning to wear and the grinding pressure needs to be reduced. When the contact pressure reaches 2.5N, the displacement sensor detects a wafer feed of 0.15mm, indicating unstable feed speed during grinding. The feed rate is automatically adjusted to maintain a stable feed rate. The surface roughness of the ground surface is reduced from Ra = 0.5µm (without feedback control) to Ra = 0.2µm (with feedback control).

[0027] Multi-source data perception can comprehensively reflect the grinding status, perceive and monitor tiny fluctuations in grinding in real time, and dynamically adjust various parameters in the grinding process through comprehensive analysis of frequency, displacement, and contact pressure to ensure consistent processing quality throughout the grinding process and avoid the impact of grinding wheel wear or temperature changes.

[0028] Behavioral modeling of each round of grinding is performed based on the original behavioral observation data stream to establish a time-series grinding behavior dataset.

[0029] Specifically, the raw behavioral observation data stream received is preprocessed, including noise removal (for example, using a low-pass filter to remove high-frequency interference in the sensor signal), filling in possible missing values, and data alignment (ensuring that data from different sensors are synchronized in time). The grinding process is typically not completed in one go, but rather in multiple stages or rounds, with behavioral modeling performed for each round. By recording and modeling data at multiple moments in each grinding round, a time-continuous behavioral dataset is formed, known as a time-series grinding behavior dataset.

[0030] Extract features such as fluctuation amplitude and frequency change rate from frequency data; extract features such as feed rate and change rate from step data; extract features such as pressure fluctuation amplitude and maximum pressure from contact data. The observation data per second is matched one by one with the timestamp to form a time series data set. Each record includes features such as timestamp, frequency, step, contact pressure, etc. For example, at the 10th second, the frequency is 90Hz, the feed amount is 0.05mm, and the contact pressure is 1.55N. The time series grinding behavior data set is a data set generated based on the raw data and behavior modeling of each round of grinding process. It includes grinding behavior information at different time points and can reflect the time evolution of the grinding process and the process characteristics of different stages.

[0031] Different modeling methods are used to model various types of data and analyze their changing trends and patterns. For example, time series analysis (such as ARIMA models) is used to analyze the trends of polishing frequency over time and identify its periodicity or decay patterns. This allows for predicting frequency changes during the polishing process and revealing its temporal evolution. State-space models are used to describe the relationship between step sensing data (such as wafer descent distance) and polishing time and force, revealing the interplay between different polishing stages. Clustering algorithms (such as K-means) are used to classify the polishing process into different stages (such as initial contact, stable polishing, and nearing completion) based on contact pressure and contact area change data, identifying different polishing conditions. Data at each moment (including raw data, extracted features, and modeling results) are stored by timestamp to form a time-series polishing behavior dataset. This dataset includes: raw data for each polishing cycle (such as frequency, step length, and contact pressure); features extracted at each moment (such as frequency fluctuation, step rate, and contact pressure peak); and process stages and states (such as initial contact, stable polishing, and nearing completion). For example, data per second is stored by timestamp, and some sample data are obtained as follows: at time 0s, the frequency is 75Hz, the feed amount is 0.05mm, the contact pressure is 1.50N, the frequency change is NaN (no predecessor data), the frequency change std (i.e., the standard deviation of the frequency change) is NaN (insufficient data), the step rate (feed amount / time interval) is 0.05mm / s, the step change rate (derivative of the step rate) is NaN (no predecessor data), the maximum contact pressure is 1.5N, the average contact pressure is 1.5N, and the contact state is initial contact; at time 1s, the frequency is 78Hz, the feed amount is 0.05mm, the contact pressure is 1.52N, the frequency change is 3.0Hz / s, the frequency change std is NaN (std cannot be calculated for a single point), the step rate is 0.05mm / s, and the step change rate is 0.0mm / s. ², the maximum contact pressure is 1.52N, the average contact pressure is 1.51N, and the contact state is initial contact; when the time is 2s, the frequency is 80Hz, the feed amount is 0.05mm, the contact pressure is 1.53N, the frequency change is 2.0Hz / s, the frequency change std is 0.71Hz, the step rate is 0.05mm / s, the step change rate is 0.0mm / s², the maximum contact pressure is 1.53N, the average contact pressure is 1.52N, and the contact state is stable grinding; when the time is 3s, the frequency is 85Hz, the feed amount is 0.05mm, the contact pressure is 1.55N, the frequency change is 5.0Hz / s, the frequency change std is 2.12Hz, the step rate is 0.05mm / s, the step change rate is 0.0mm / s², the maximum contact pressure is 1.55N, the average contact pressure is 1.53N, and the contact state is stable grinding. The std in frequency change std is the industry-wide abbreviation for standard deviation, which is used to measure the degree of dispersion between each data point in a set of data and the mean.Frequency change std refers to the standard deviation of the frequency change. The data at time 0 seconds has no predecessor data, so the frequency change and standard deviation cannot be calculated. Therefore, it is marked as NaN, indicating that insufficient data is available for calculation.

[0032] By extracting and modeling features from frequency, stepping, contact pressure, and other data, we accurately identify subtle changes in grinding, enabling precise grinding control and reducing quality fluctuations. As time-series data sets accumulate, we continuously adapt to varying grinding conditions (such as disc wear, temperature and humidity fluctuations, etc.) to ensure consistent machining quality.

[0033] The time series grinding behavior data set is used to perceive the grinding condition and establish a real-time perception result.

[0034] Furthermore, the present application also includes the following steps: performing multi-dimensional working condition feature extraction on the time-series grinding behavior data set to establish a working condition feature vector; using the labeled historical data to train the working condition recognition model, sending the working condition feature vector to the working condition recognition model, and outputting a working condition label sequence; using the working condition label sequence to perform stable trend perception, using the stable trend perception result to enhance the working condition label of the current node, and then establishing a real-time perception result.

[0035] Specifically, multidimensional working condition feature extraction is performed on the time-series grinding behavior dataset. Features with different physical or statistical significance that reflect the current state or conditions of the grinding process are extracted, including multiple dimensions such as frequency, stepping, contact pressure, and temperature. These data are then combined into a working condition feature vector, which comprehensively represents the different states of the grinding process. The working condition feature vector is a vector formed by extracting multidimensional feature information from the time-series grinding behavior dataset. It represents the working condition (i.e., operating state) of the grinding process, including changes in frequency, fluctuations in contact pressure, and changes in stepping rate, and can comprehensively reflect important parameters of the grinding process. For example, if the vibration frequency fluctuates between 80 and 100 Hz, the standard deviation and mean of the fluctuation amplitude can be extracted; if the stepping rate changes from 0.05 mm / s to 0.08 mm / s, the trend and amplitude of the rate change are extracted; and if the contact pressure changes from 1.5 N to 2.0 N during the grinding process, the range of change and its fluctuation pattern are extracted. The working condition feature vector is a vector composed by extracting multidimensional feature information from the time series grinding behavior dataset, which represents the working condition (i.e., operating status) of the grinding process.

[0036] A working condition recognition model is trained using annotated historical data. Each moment in the historical data is labeled with a corresponding working condition label, such as initial contact, stable grinding, nearing completion, abnormal vibration, and excessive pressure. Labels (also called working condition labels) are qualitative descriptions of the state of the grinding process at that moment. The annotated historical data includes raw data and working condition labels at each time point during the grinding process. Raw data typically includes sensor data (such as vibration frequency, displacement, and contact pressure), while working condition labels indicate the grinding state at each time point (such as initial contact, stable grinding, and nearing completion). Raw data may be affected by noise or contain missing data due to sensor issues. De-noising methods (such as filtering) and interpolation methods (such as linear interpolation) are required to repair the data. Because data scales may vary from sensor to sensor (for example, the contact pressure and vibration frequency ranges vary significantly), the data must be standardized or normalized to make it suitable for model input. Features that characterize the working condition are extracted from the raw data, such as the amplitude of vibration frequency fluctuations and the maximum contact pressure.

[0037] Based on the labeled historical data, an appropriate operating condition recognition model, such as a support vector machine, is selected for training. Using supervised learning methods, the features of the historical data are paired with the corresponding operating condition labels and input into the model. The model learns the relationship between the input features and the operating condition labels based on this data. The dataset is divided into a training set and a validation set. Typically, the training set is used to train the model, while the validation set is used to evaluate the model's performance and prevent overfitting. Using gradient descent or other optimization algorithms, the model continuously adjusts its parameters to minimize prediction error (such as classification error). In each round of training, the model adjusts itself based on the relationship between input features and labels until the desired accuracy is achieved. Model performance is evaluated using methods such as cross-validation, and hyperparameters (such as the learning rate and regularization parameter) are adjusted to further optimize the model. After model training, the test set is used to evaluate model performance. Metrics such as classification accuracy, precision, and recall are used to verify the model's effectiveness. If the model performs well on the test set, it is considered successfully trained and can be used for real-time operating condition recognition.

[0038] The working condition feature vector is input into the trained working condition recognition model. The model outputs a working condition label for each moment, forming a complete sequence of working condition labels representing the state of the grinding process at each point in time. This sequence of working condition labels is generated by the working condition recognition model and represents the working condition state of the grinding process at different points in time. By analyzing the working condition label sequence and utilizing stability trend perception, it is determined whether the current grinding state remains stable. By monitoring the trend of label changes, it is possible to identify any abnormal changes (such as premature grinding termination or sudden grinding anomalies) and make adjustments based on this. If the working condition in the label sequence experiences an abnormal fluctuation (such as a sudden change from stable grinding to nearing completion), the label is enhanced. For example, if the model predicts that the current moment is nearing completion, but the historical label sequence indicates that the current grinding state is still stable, trend perception will be used to enhance the current label to stable grinding to avoid misjudgment.

[0039] Specifically, by analyzing the changing patterns in the condition tag sequence, stability trend detection is performed to determine whether the current grinding process is stable. The duration and changing trends of each condition in the tag sequence are analyzed. For example, if the condition tag "stable grinding" remains constant at the most recent time points, the current state is determined to be stable. Based on historical data analysis, the transition patterns between condition tags are identified. For example, if the current tag suddenly jumps to an unstable state from initial contact to stable grinding to nearing completion, this may indicate an abnormal change. The current condition tag is determined to conform to a stable pattern, identifying potential anomalies or fluctuations. For example, if the current tag is "stable grinding" while the previous tag was "nearing completion," an unstable trend is considered. If stability trend detection identifies an abnormal change in the current condition tag, the current condition tag is enhanced. This is done by adjusting the current condition tag to a state more consistent with a stable trend. For example, if the system detects that the current tag is "nearing completion," but the tag sequence and trend analysis from the previous few seconds indicate that the grinding process is still in the stable grinding phase, the current tag can be enhanced to "stable grinding" to ensure the continuity and stability of the grinding process. After stable trend perception and label enhancement, a real-time perception result is generated, indicating the working condition of the current node. The real-time perception result refers to the working condition of the current node output based on the analysis of the working condition label sequence and trend perception, reflecting the current state of the grinding process.

[0040] By extracting multi-dimensional working condition features and training working condition recognition models, we can accurately identify various states in the grinding process, eliminating manual intervention and improving the degree of recognition automation. Stable trend perception and label enhancement effectively avoid grinding state instability caused by short-term fluctuations or abnormal misjudgments, maintaining a smooth grinding process.

[0041] Activate the non-contact thickness sensing unit to perform real-time piezoelectric quartz wafer thickness sensing and establish thickness sensing results.

[0042] Furthermore, the present application also includes the following steps: enabling a non-contact thickness sensing unit to obtain the original thickness data of the piezoelectric quartz wafer; extracting input variables based on the original behavior observation data stream, the input variables including grinding frequency data, unit time step displacement, and contact pressure response; performing thickness fitting analysis based on the input variables and the original thickness data to establish a thickness prediction result; and using the thickness prediction result to calibrate and compensate the thickness sensing result.

[0043] Specifically, a non-contact thickness sensing unit is a sensor used to measure the thickness of materials (such as piezoelectric quartz wafers) without direct contact with the target surface. These sensors, including laser rangefinders, optical sensors (such as white light interferometers), and capacitive sensors, measure the thickness of an object by reflecting light or using electric fields, eliminating the need for direct contact with the surface and potentially causing surface damage or measurement errors. The non-contact thickness sensor monitors the thickness of the piezoelectric quartz wafer in real time, calculating the thickness based on reflected light or other physical quantities (such as electric field changes or laser reflections) from the wafer surface without requiring direct contact with the wafer surface. Data from the non-contact thickness sensing unit is received in real time to determine the wafer thickness at each moment. The non-contact thickness sensor samples the wafer once or multiple times per second, calculating the actual wafer thickness based on laser reflections and other methods. For example, suppose a wafer gradually decreases in thickness from 2.000 mm to 1.990 mm during the grinding process.

[0044] Record the thickness value at each time point (such as 2.000mm, 1.990mm, etc.) and store it as real-time thickness sensing results. Analyze the trend of the thickness value data, such as whether there are abnormal fluctuations and whether it is within the expected range. For example, a laser displacement sensor records thickness data once per second, and the initial wafer thickness is 2.000mm. During the grinding process, the thickness data is acquired in real time. Suppose that in the next 5 seconds, the thickness changes as follows: 2.000mm, 1.998mm, 1.995mm, 1.993mm, 1.990mm. The thickness change trend is from 2.000mm to 1.990mm, indicating that the thickness of the wafer gradually decreases during the grinding process.

[0045] Raw thickness data of the piezoelectric quartz wafer is acquired through a non-contact thickness sensing unit. Key input variables influencing thickness variation are extracted from the raw behavioral observation data stream, including grinding frequency data, displacement per unit time step, and contact pressure response. Grinding frequency data represents the vibration characteristics of the contact between the grinding tool and the wafer, typically measured by a piezoelectric sensor. Displacement per unit time step represents the wafer's advance distance per time unit, typically measured by a displacement sensor and reflecting the wafer's feed rate. Contact pressure response represents the contact force between the tool and the wafer, typically measured by a pressure sensor. Contact pressure directly affects grinding efficiency and wafer thickness variation.

[0046] A thickness fitting model is established by analyzing the relationship between input variables (grinding frequency, step displacement, contact pressure, etc.) and actual measured thickness raw data. The collected input variables (such as frequency, step displacement, and pressure) are paired with the raw thickness data to form a training dataset. This data is used to train the fitting model, identifying the relationship between the input variables and the thickness data. For example, it may be found that an increase in contact pressure leads to an accelerated rate of thickness change, or that an increase in step displacement is associated with a certain correlation between thickness reduction. Once the model is trained, new real-time data (such as the current frequency, step displacement, and pressure) is input, and the model can predict the current thickness value. For example, based on historical data training, the model may derive the following prediction: when the grinding frequency is in the range of 1000 to 1200 Hz, the step displacement is 0.05 mm / min, and the contact pressure is 5 N, approximately 0.002 mm of material can be removed per minute. The current grinding frequency collected in real time is 1100Hz, the step displacement is 0.05mm / min, the contact pressure is 5N, and the grinding has been going on for 3 minutes. The model can predict that approximately 0.006mm of material has been removed. If the initial thickness is 0.7mm, then the predicted current thickness is 0.694mm. Compare this predicted result with the original data just measured by the non-contact thickness sensing unit (assuming it is 0.695mm). If it is known that this model of sensor has a systematic error of +0.0005mm near 0.7mm, then subtract 0.0005mm from the original measurement of 0.695mm to obtain the calibrated thickness sensing result of 0.6945mm.

[0047] The predicted thickness is compared with the thickness sensed by the non-contact thickness sensing unit. If there is a discrepancy between the predicted and actual thickness, a calibration compensation is applied to the predicted result to ensure the final thickness sensed result is as accurate as possible. The calibrated and compensated result is more reliable than either the raw measured value or the predicted value alone. Fitting analysis captures the complex relationship between grinding behavior and thickness variation, allowing the predicted result to reflect the dynamic characteristics of the grinding process. Using the predicted result to calibrate and compensate the directly measured value effectively corrects for deviations caused by sensor errors, environmental interference, and other factors.

[0048] Taking the target grinding requirement as the control goal, the grinding control reconstruction based on the real-time perception results and thickness perception results is performed to establish the reconstruction results.

[0049] Furthermore, the present application also includes the following steps: using the real-time perception result to evaluate the working condition stability of the current node and establish a first evaluation abnormality; performing grinding stability abnormality analysis of the piezoelectric quartz wafer and establishing a static evaluation threshold; performing abnormal trigger identification of the first evaluation abnormality based on the static evaluation threshold, and if the abnormal trigger identification result is a trigger result, generating a reconstruction instruction, and according to the reconstruction instruction, performing grinding control reconstruction based on the real-time perception result and the thickness perception result with the target grinding requirement as the control target.

[0050] Furthermore, the present application also includes the following steps: if the abnormal trigger identification result is a non-triggering result, calling the timing grinding behavior data set to perform timing fitting prediction and establish a timing fitting prediction result; using the static evaluation threshold to perform trigger analysis of the timing fitting prediction result and identify the abnormal trigger node; configuring a prediction reconstruction instruction based on the abnormal trigger node, and performing grinding control reconstruction according to the prediction reconstruction instruction.

[0051] Furthermore, the present application also includes the following steps: if the abnormal trigger node is a node outside the preset time period, a hold instruction is generated; and the original grinding parameters are maintained according to the hold instruction to perform piezoelectric quartz wafer grinding control management.

[0052] Specifically, during the grinding process, real-time sensing results (such as frequency data, step data, contact pressure data, thickness sensing results, etc.) are continuously obtained to reflect the current grinding status. By analyzing the real-time sensing results, check whether there are abnormal fluctuations or trends in the grinding process. For example, whether the frequency fluctuation amplitude is too large, whether the contact pressure exceeds the safety range, whether the step displacement is abnormal, etc. Based on historical data, stability indicators are set, such as the frequency fluctuation amplitude should be within 5%, the contact pressure should be maintained between 1.5N and 2.0N, etc. By comparing with the stability indicators, it is judged whether the current grinding process is stable. The first evaluation anomaly refers to certain abnormal behaviors or states detected during the working condition stability evaluation process. When an anomaly is detected in the working condition stability evaluation process, the first evaluation anomaly will be recorded.

[0053] Before calling, a static evaluation threshold is obtained through a large amount of historical data analysis. The static evaluation threshold is a constant or interval set through historical data analysis, working condition stability evaluation and other methods to determine whether the current grinding process meets the expected stability requirements. Static thresholds are usually set based on the experience of long-term data and engineering requirements, such as the maximum value of contact pressure, frequency fluctuation amplitude, etc. For example, the contact pressure is set between 1.5N and 2.0N. Exceeding this range may indicate excessive or uneven grinding; the frequency fluctuation amplitude should be kept within 5%. Exceeding this threshold may indicate uneven contact between the tool and the wafer; the step rate change should be set between ±0.05mm / s. Excessive changes may lead to inconsistent processing.

[0054] Based on the static evaluation threshold, the system performs anomaly trigger identification for the first evaluation anomaly. If the current operating condition exceeds the static evaluation threshold, an anomaly alarm is triggered and the reconstruction instruction generation process is initiated. If the current anomaly exceeds the threshold, a trigger result is output, indicating that action is required. If it does not exceed the threshold, a non-trigger result may be output, indicating that the current anomaly is within the acceptable range and no immediate intervention is required.

[0055] Once an anomaly is identified and triggers abnormality recognition, a reconfiguration command is generated, and the grinding process control is reconfigured according to the target grinding requirements. A reconfiguration command is generated to adjust parameters in the grinding process to restore a stable state. This may include slowing the grinding speed, adjusting the feed rate, and adjusting the tool contact force. Based on the reconfiguration command, the control parameters or strategies of the grinding equipment are dynamically adjusted. The target grinding requirement is the control objective, meaning that the target grinding requirement is the ultimate guide for all adjustments, ensuring that any parameter changes do not deviate from this overall goal. When performing adjustments, the current real-time sensing results and thickness sensing results are comprehensively considered. The real-time sensing results provide immediate information on the current grinding process status, while the thickness sensing results provide progress information. Based on the reconfiguration command, the grinding equipment's operating parameters (such as pressure, speed, time, amplitude, etc.) are adjusted in real time to change the behavior of the grinding process and optimize it towards the target requirements.

[0056] Through real-time perception and anomaly recognition, unstable factors in the grinding process can be promptly detected and grinding parameters can be adjusted through reconstruction instructions to ensure smooth operation of the grinding process. The generated reconstruction instructions are used to adjust grinding parameters (such as contact pressure and feed rate), thereby optimizing the control strategy during the grinding process and improving processing quality.

[0057] If no abnormalities are detected during the current polishing process, a time-series polishing behavior dataset is used for time-series fitting prediction. Patterns and trends within the historical data are analyzed to predict possible changes in operating conditions at a specific point in the future. An appropriate time-series fitting model, such as the Autoregressive Integrated Moving Average (ARIMA) model, is selected and trained on the historical data to generate a mathematical model that can predict future conditions. The time-series polishing behavior dataset contains a series of time-series data from the polishing process, recording key polishing parameters such as contact pressure, frequency, and step displacement at different time points. By analyzing the historical data and behavioral patterns within the time-series dataset, a mathematical model is constructed to predict future conditions. A prediction model is invoked, fed with the latest time-series data. Based on the patterns inherent in this data, the model predicts the likely changing trends or specific values ​​of key parameters (such as wafer thickness, polishing pressure, and vibration amplitude) over a period of time (e.g., the next 10 seconds or the next polishing cycle). This time-series fitting prediction result represents the predicted future condition and includes the predicted values ​​of the relevant parameters at each future time point. For example, the contact pressure predicted in the next 5s is 1.55N, the frequency fluctuation is 4.5%, and the step displacement is 0.06mm / s.

[0058] Static evaluation thresholds are used to analyze the timing fitting prediction results to determine whether anomalies will occur at certain moments in the future. For example, whether the predicted contact pressure exceeds the set upper limit, whether the frequency fluctuation exceeds the predetermined range, or whether the step displacement changes significantly. The static threshold for contact pressure might be 1.8N, the maximum threshold for frequency fluctuation is 5%, and the maximum range of step displacement variation is ±0.1mm / s. If the predicted contact pressure exceeds 1.8N at some point in the future, it will be identified as an abnormal trigger node. For example, the predicted contact pressure at the 9th second is 1.9N, exceeding the threshold and triggering an anomaly.

[0059] Through trigger analysis, abnormal trigger nodes are identified. These nodes are the time points where abnormalities may occur in the grinding process. At the time of the trigger node, the grinding process may experience abnormal phenomena such as excessive contact pressure and excessive frequency fluctuations. According to the abnormal trigger node, predictive reconstruction instructions are configured to help adjust various parameters in the grinding process to avoid abnormalities and ensure the smooth progress of the grinding process. For example, if the contact pressure exceeds the predetermined range, an instruction is generated to reduce the grinding pressure; if the step displacement is too large, the feed rate needs to be slowed down or the grinding wheel contact force needs to be adjusted. According to the predictive reconstruction instruction, the grinding control reconstruction is started in advance or before the prediction node, that is, the grinding parameters are dynamically adjusted. Actual adjustments are made according to the generated instructions to ensure that the contact pressure, step rate, etc. are maintained within the ideal range. After reconstruction, the parameters in the grinding process continue to be monitored to ensure stability and quality.

[0060] When abnormal triggering nodes are detected and fall outside the expected time interval, a hold command is generated to maintain the current grinding parameters, avoiding unnecessary adjustments and ensuring the stability of the grinding process. Abnormal triggering nodes in the grinding process are identified by comparing real-time sensing data (such as contact pressure, step displacement, frequency, etc.) with static evaluation thresholds. A preset time interval is defined. For example, the grinding process is considered stable from the 5th to the 15th second, meaning that the grinding process is considered stable within this interval.

[0061] If the abnormal trigger node occurs within the preset time interval (such as an abnormality at the 7th second), the relevant control mechanism will be triggered to adjust the grinding parameters. If the abnormal trigger node occurs outside the preset time interval (such as the 2nd or 18th second), the current abnormality is considered temporary and may not affect the entire grinding process. If the system detects that the abnormal trigger node occurs outside the preset time interval, it will generate a hold instruction. The purpose of the hold instruction is to instruct the system not to make any adjustments when the abnormality occurs, and to continue to maintain the current grinding parameters unchanged, avoiding excessive adjustments to the grinding process due to temporary abnormal fluctuations, which may lead to reduced grinding stability.

[0062] Based on the generated hold command, the original grinding parameters are maintained during the grinding process. This means that the current contact pressure, feed rate, or other control parameters are not adjusted. This ensures a stable grinding process and avoids unnecessary intervention or the introduction of instability. Even after the hold command is executed, various data during the grinding process will continue to be monitored. If anomalies persist over the subsequent period, the current control strategy will be reassessed and adjusted as necessary.

[0063] Furthermore, the present application also includes the following steps: using the equipment state evolution, grinding pressure fluctuation, and vibration signal response data recorded in the real-time perception results to conduct traceability evolution analysis of R&D anomalies, extracting the key parameter evolution pattern that triggers the anomaly, and establishing reconstruction taboo constraints based on the key parameter evolution pattern; obtaining the current grinding roughness based on the real-time perception results; executing control optimization under the reconstruction taboo constraints according to the thickness perception results, the current grinding roughness, and the target grinding requirements to establish a reconstruction result.

[0064] Furthermore, the present application also includes the following steps: establishing an objective function, wherein the evaluation items of the objective function include a thickness evaluation item, a roughness evaluation item, and an operating condition stability evaluation item; after reconstructing the control space using the reconstruction taboo constraint, performing control optimization based on the objective function in the reconstructed control space to establish a reconstruction result.

[0065] Specifically, from the real-time sensing results, we obtain the data on the evolution of equipment status, grinding pressure fluctuations, and vibration signal responses, conduct a traceable evolution analysis, analyze the time when the anomaly occurs during the grinding process, and identify the root cause of the anomaly. By comparing the data of normal grinding processes and abnormal processes, we can find the key parameters that cause the anomaly. For example, does the fluctuation of grinding pressure suddenly increase at a certain point in time, does the vibration signal show abnormal peaks, etc. If the fluctuation amplitude of grinding pressure or vibration signal at a specific moment is abnormal, it may mean that the equipment is wearing out, the operating conditions have changed, or other problems. Through traceable evolution analysis, these anomalies can be accurately identified and their root causes can be found.

[0066] Equipment state evolution refers to the changes in equipment status during the grinding process. This is typically achieved through the collection of sensor data (such as vibration, temperature, and pressure) to track equipment health and analyze whether there are issues such as wear or failure. Grinding pressure fluctuation refers to the changes in contact pressure between the tool and the wafer during the grinding process. Stable grinding pressure is crucial for ensuring wafer surface uniformity and processing quality. Vibration signal response data, which refers to the vibration signal generated by the grinding tool during contact with the wafer, is used to assess the dynamic behavior and stability of the grinding process. Abnormal vibration signals indicate equipment issues or an unstable grinding process.

[0067] By tracing the changing trends of abnormal data during the grinding process, the root cause of the anomaly can be identified, along with the key factors contributing to the anomaly, such as equipment wear and operating condition fluctuations. By analyzing real-time sensor data, the changing patterns of key parameters influencing the grinding process (such as grinding pressure, vibration signals, and temperature) before and after the anomaly occurs are identified. This key parameter evolution pattern is extracted, and reconstruction taboo constraints are set to avoid the region that caused the anomaly when adjusting control parameters. Reconstruction taboo constraints are a set of rules or restrictions extracted from the evolution pattern of the key parameters of the anomaly. These constraints are used to limit the control parameter range during the reconstruction process, preventing the selection of unreasonable or unstable grinding parameters during the control optimization process and thus preventing new anomalies. For example, if it is determined that contact pressures above 1.8N cause vibration anomalies, a taboo constraint can be set to prevent contact pressures exceeding 1.8N. Wafer roughness is monitored and evaluated in real time during the grinding process using surface roughness measurement equipment (such as a profilometer). For example, if real-time monitoring shows a wafer roughness value of 0.5µm, this value will serve as the basis for control optimization.

[0068] Based on the target grinding requirements, an objective function is established, including thickness evaluation items, roughness evaluation items, and working condition stability evaluation items. Evaluation items are components of the objective function, representing different aspects of the grinding process that need to be optimized. For example, the thickness evaluation item assesses whether the wafer thickness meets the requirements, such as ensuring that the wafer thickness is between 1.990mm and 2.000mm. The roughness evaluation item assesses the surface quality, such as ensuring that the wafer roughness is less than 0.3µm. The working condition stability evaluation item evaluates the stability of the grinding process.

[0069] Based on the reconstruction taboo constraints, the adjustable parameter range (reconstructed control space) is trimmed or restricted. Reconstructing the control space of grinding parameters means limiting the variable range of grinding parameters to avoid entering abnormal areas. Reconstructing the control space involves adjusting and restricting the possible grinding parameter space based on the current grinding conditions and target requirements, thereby generating a new control space within which optimization is performed. The optimal grinding control parameters are found within the reconstructed control space to achieve the predetermined target requirements (such as reducing roughness or ensuring stable working conditions). Within the reconstructed control space, control optimization is performed using the objective function, optimizing the grinding parameters based on the objective function's evaluation criteria (such as thickness, roughness, and stability). Assume that the objective function can be expressed as: objective function = α (thickness error) + β (roughness error) + γ (working condition stability error), where α, β, and γ are weighting coefficients adjusted according to the importance of different evaluation criteria.

[0070] Within the reconstructed control space, adjust grinding parameters (such as contact pressure, feed rate, and rotational speed). Use optimization algorithms (such as gradient descent and genetic algorithms) to search for the optimal solution within the control space. For example, adjust the contact pressure to 1.8 N, the feed rate to 0.05 mm / s, and the rotational speed to 55 rpm to minimize the error term of the objective function, ensuring a roughness less than 0.3 µm and a thickness that meets design requirements.

[0071] After the control optimization process, a reconstruction result is obtained: the optimized and adjusted grinding control parameter configuration. These parameter configurations meet the target grinding requirements and ensure the grinding process is carried out under optimal conditions. Once the reconstruction result is determined, these adjusted parameters are immediately implemented to reconstruct the grinding control. Based on the optimization results, various parameters in the grinding process are adjusted in real time to ensure efficient and stable grinding.

[0072] Through traceability analysis and extraction of key parameter evolution patterns, potential anomalies are identified and eliminated, optimizing the grinding process. By optimizing within the reconstructed control space, the optimal parameter configuration for the current grinding conditions is found, optimizing wafer thickness and roughness. Through real-time feedback and control optimization, the grinding process is precisely controlled to ensure that wafer thickness, roughness, and operating stability meet predetermined requirements.

[0073] The piezoelectric quartz wafer polishing control management is performed based on the reconstruction result.

[0074] Furthermore, the present application also includes the following steps: recording the grinding quality detection data of the piezoelectric quartz wafer and obtaining the real parameter control data; mapping the grinding quality detection data and the real parameter control data into a batch parameter group, and saving them to the control center.

[0075] Specifically, the polishing process is controlled in real time based on the reconstruction results—optimized polishing control parameters such as contact pressure, feed rate, and rotational speed. This means that control parameters are set and adjusted based on the reconstruction results to ensure that parameters such as contact pressure, speed, and disc speed remain within optimal ranges during polishing, thereby ensuring wafer quality and operating stability. During polishing control, sensor data is continuously monitored to ensure that all control parameters remain within their set ranges, preventing instabilities caused by changing operating conditions.

[0076] Through sensor fusion technology, a set of real-time sensors monitors various dynamic signals during the grinding process, enabling real-time roughness estimation rather than direct measurement. A suite of real-time sensors (vibration accelerometers / piezoelectric sensors, acoustic emission (AE) sensors, cutting / grinding force or contact pressure sensors, displacement / step sensors, spindle current or speed encoders, etc.) captures signals related to the grinding process, such as equipment vibration, cutting force, and contact pressure fluctuations. Feature extraction (RMS, spectral energy, spectral slope, envelope characteristics, peak value, kurtosis, and time series statistics) is performed on these signals. These features are then mapped to the actual roughness values ​​(Ra / Rq / Rz) obtained from offline experiments (measured using a profilometer under static or intermittent conditions). A regression model (SVR / RandomForest / GBDT, or LSTM / 1D-CNN, etc.) is then trained. This feature mapping to the actual roughness values ​​(Ra / Rq / Rz) obtained from offline experiments enables real-time roughness prediction.

[0077] To ensure the accuracy of the roughness prediction model, both offline calibration and on-site calibration are essential. Extensive offline experiments, encompassing various conditions such as grinding wheel wear, grinding pressure, speed, and initial workpiece roughness, are used to train and optimize the model, enabling it to accurately map real-time signal characteristics to actual roughness values. To prevent model drift and error accumulation over long-term operation, periodic on-site calibration is implemented. For example, short machine stops are performed periodically, or true roughness measurements are taken at specific measurement points to correct the model's predicted roughness, ensuring long-term stability and accuracy. In addition to providing roughness predictions, uncertainty or confidence levels are also output, quantifying the reliability of the predictions. This allows for adaptive determination of the reliability of the current predictions. If unreliable, actual roughness measurements are triggered for correction. To enhance the accuracy and interference immunity of roughness measurements, signal processing is also implemented, including the integration of vibration sensors to monitor the vibration status of the grinding equipment and workpiece in real time and dynamically correct the roughness measurements. A reference sensor (an accelerometer mounted on the machine bed) is used for common-mode cancellation and other functions to improve sensitivity. Despite interference from vibration and other factors during the grinding process, real-time roughness measurement does not rely on traditional contact measurement tools. Instead, it achieves real-time roughness estimation during the grinding process through sensor fusion and machine learning methods, combined with extensive offline calibration experiments and online calibration. Actual control parameters may vary from the preset target parameters due to equipment wear, environmental factors, or other process fluctuations.

[0078] Grinding quality inspection data (such as roughness and thickness) and actual control data (such as contact pressure and feed rate) are combined to form a batch parameter group. Each batch parameter group contains a complete data set, recording the quality and control data during the grinding process of a specific batch. The generated batch parameter group is saved to the control center, a centralized data management platform that stores parameter data for all production batches and supports quality analysis, process improvement, and production traceability. The control center archives batch parameter groups based on information such as batch number and date to ensure data traceability and security. A batch parameter group refers to all relevant parameter data (such as grinding quality data and control data) recorded during the grinding process for a batch of piezoelectric quartz wafers. For example, if a 50mm diameter piezoelectric quartz wafer is ground at a grinding pressure of 1.5N and a grinding speed of 60rpm, control management is performed based on the reconstructed results to set the contact pressure to 1.8N, the feed rate to 0.05mm / s, and the rotation speed to 55rpm. During the grinding process, the recorded roughness was 0.2µm, the thickness was 1.995mm, the contact pressure was 1.8N, the feed rate was 0.05mm / s, and the grinding wheel speed was 55rpm. These data were mapped to a batch parameter group and saved to the control center.

[0079] Control instructions (reconstruction results) are converted into actual operations (control management), and the control effects are verified through quality inspection data, forming a complete closed loop from instructions to execution to results, ensuring the actual effectiveness of the control strategy. By recording actual parameter control data and quality inspection data and forming batch parameter groups, the processing process and final results of each batch of products can be documented, facilitating production traceability and quality control.

[0080] In summary, the piezoelectric quartz wafer grinding control method based on grinding frequency monitoring provided in this application has the following beneficial effects: By performing multi-source data perception of piezoelectric quartz wafer grinding, an original behavior observation data stream is established. The multi-source data perception includes grinding frequency perception, step perception, and contact perception. Based on the original behavior observation data stream, the behavior modeling of each round of grinding is performed to establish a time-series grinding behavior data set. The time-series grinding behavior data set is used to perform grinding working condition perception and establish real-time perception results. The non-contact thickness perception unit is activated to perform real-time piezoelectric quartz wafer thickness perception and establish thickness perception results. With the target grinding requirement as the control target, grinding control reconstruction based on the real-time perception results and thickness perception results is performed to establish a reconstruction result. The piezoelectric quartz wafer grinding control management is performed according to the reconstruction result. In other words, by introducing multiple perception sources, the behavior of the entire grinding process is modeled, the working condition changes during the grinding process are monitored in real time, and the grinding control parameters are dynamically reconstructed in combination with the non-contact thickness perception unit, thereby improving the processing accuracy and quality of piezoelectric quartz wafer grinding.

[0081] Example 2: Based on the same inventive concept as the piezoelectric quartz wafer grinding control method based on grinding frequency monitoring in the aforementioned Example 1, this application also provides a piezoelectric quartz wafer grinding control system based on grinding frequency monitoring. Figure 2 , the piezoelectric quartz wafer grinding control system based on grinding frequency monitoring includes: The multi-source data perception module 11 is used to perform multi-source data perception of piezoelectric quartz wafer grinding and establish an original behavior observation data stream. The multi-source data perception includes grinding frequency perception, step perception, and contact perception; the behavior modeling module 12 is used to perform behavior modeling of each round of grinding based on the original behavior observation data stream and establish a time-series grinding behavior data set; the grinding condition perception module 13 is used to use the time-series grinding behavior data set to perform grinding condition perception and establish real-time perception results; the thickness perception module 14 is used to activate the non-contact thickness perception unit to perform real-time piezoelectric quartz wafer thickness perception and establish thickness perception results; the grinding control reconstruction module 15 is used to perform grinding control reconstruction based on the real-time perception results and thickness perception results with the target grinding requirement as the control target, and establish a reconstruction result; the grinding control management module 16 is used to perform piezoelectric quartz wafer grinding control management according to the reconstruction result.

[0082] Furthermore, the grinding condition perception module 13 in the piezoelectric quartz wafer grinding control system based on grinding frequency monitoring is also used to: perform multi-dimensional condition feature extraction on the time-series grinding behavior data set to establish a condition feature vector; use the labeled historical data to train the condition recognition model, send the condition feature vector to the condition recognition model, and output a condition label sequence; use the condition label sequence to perform stable trend perception, use the stable trend perception result to enhance the condition label of the current node, and establish a real-time perception result.

[0083] Furthermore, the thickness sensing module 14 in the piezoelectric quartz wafer grinding control system based on grinding frequency monitoring is also used to: enable a non-contact thickness sensing unit to obtain the original thickness data of the piezoelectric quartz wafer; extract input variables based on the original behavior observation data stream, the input variables including grinding frequency data, unit time step displacement, and contact pressure response; perform thickness fitting analysis based on the input variables and the original thickness data to establish a thickness prediction result; and use the thickness prediction result to calibrate and compensate the thickness sensing result.

[0084] Furthermore, the grinding control reconstruction module 15 in the piezoelectric quartz wafer grinding control system based on grinding frequency monitoring is also used to: use the real-time perception result to evaluate the working condition stability of the current node and establish a first evaluation abnormality; perform grinding stability abnormality analysis of the piezoelectric quartz wafer and establish a static evaluation threshold; perform abnormal trigger identification of the first evaluation abnormality based on the static evaluation threshold, and if the abnormal trigger identification result is a trigger result, generate a reconstruction instruction, and according to the reconstruction instruction, perform grinding control reconstruction based on the real-time perception result and the thickness perception result with the target grinding requirement as the control target.

[0085] Furthermore, the grinding control reconstruction module 15 in the piezoelectric quartz wafer grinding control system based on grinding frequency monitoring is also used to: if the abnormal trigger identification result is a non-triggering result, call the timing grinding behavior data set to perform timing fitting prediction and establish a timing fitting prediction result; use the static evaluation threshold to perform trigger analysis of the timing fitting prediction result and identify the abnormal trigger node; configure a prediction reconstruction instruction based on the abnormal trigger node, and perform grinding control reconstruction according to the prediction reconstruction instruction.

[0086] Furthermore, the grinding control reconstruction module 15 in the piezoelectric quartz wafer grinding control system based on grinding frequency monitoring is also used to: generate a holding instruction if the abnormal trigger node is a node outside the preset time period; and maintain the original grinding parameters according to the holding instruction to perform piezoelectric quartz wafer grinding control management.

[0087] Furthermore, the grinding control reconstruction module 15 in the piezoelectric quartz wafer grinding control system based on grinding frequency monitoring is also used to: use the equipment state evolution, grinding pressure fluctuation, and vibration signal response data recorded in the real-time sensing results to perform traceability evolution analysis of R&D anomalies, extract the key parameter evolution pattern that triggers the anomaly, and establish reconstruction taboo constraints based on the key parameter evolution pattern; obtain the current grinding roughness based on the real-time sensing results; perform control optimization under the reconstruction taboo constraints according to the thickness sensing results, the current grinding roughness, and the target grinding requirements to establish a reconstruction result.

[0088] Furthermore, the grinding control reconstruction module 15 in the piezoelectric quartz wafer grinding control system based on grinding frequency monitoring is also used to: establish an objective function, the evaluation items of the objective function include a thickness evaluation item, a roughness evaluation item, and an operating condition stability evaluation item; after reconstructing the control space using the reconstruction taboo constraint, perform control optimization based on the objective function in the reconstructed control space to establish a reconstruction result.

[0089] Furthermore, the grinding control management module 16 in the piezoelectric quartz wafer grinding control system based on grinding frequency monitoring is also used to: record the grinding quality detection data of the piezoelectric quartz wafer and obtain real parameter control data; map the grinding quality detection data and the real parameter control data into a batch parameter group and save them to the control center.

[0090] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. Figure 1 The piezoelectric quartz wafer grinding control method based on grinding frequency monitoring and the specific examples in Example 1 are also applicable to the piezoelectric quartz wafer grinding control system based on grinding frequency monitoring in this embodiment. Through the above detailed description of the piezoelectric quartz wafer grinding control method based on grinding frequency monitoring, those skilled in the art can clearly understand the piezoelectric quartz wafer grinding control system based on grinding frequency monitoring in this embodiment, so for the sake of brevity of the specification, it will not be described in detail here.

[0091] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

[0092] Obviously, for those skilled in the art, several improvements and modifications can be made to the present application without departing from the principles of the present application, and these improvements and modifications also fall within the scope of protection of the present application.

Claims

1. A piezoelectric quartz wafer grinding control method based on grinding frequency monitoring, characterized in that: include: Perform multi-source data perception for piezoelectric quartz wafer grinding and establish a raw behavior observation data stream. Multi-source data perception includes grinding frequency perception, step perception, and contact perception. Performing behavioral modeling for each round of grinding based on the original behavioral observation data stream to establish a time-series grinding behavior dataset; Using the time series grinding behavior data set to perceive the grinding condition and establish real-time perception results; Activate the non-contact thickness sensing unit to perform real-time piezoelectric quartz wafer thickness sensing and establish thickness sensing results; Taking the target grinding requirement as the control target, perform grinding control reconstruction based on real-time perception results and thickness perception results, and establish the reconstruction results; The piezoelectric quartz wafer polishing control management is performed based on the reconstruction result.

2. The piezoelectric quartz wafer polishing control method based on polishing frequency monitoring according to claim 1, characterized in that: The performing of the grinding control reconstruction based on the real-time sensing result and the thickness sensing result, and establishing the reconstruction result, includes: Using the equipment state evolution, grinding pressure fluctuations, and vibration signal response data recorded in the real-time sensing results to conduct a traceability evolution analysis of R&D anomalies, extract the evolution pattern of key parameters that trigger the anomalies, and establish reconstruction taboo constraints based on the evolution pattern of key parameters; Acquire the current grinding roughness based on the real-time sensing result; According to the thickness sensing result, the current grinding roughness, and the target grinding requirement, a control optimization is performed under the reconstruction taboo constraint to establish a reconstruction result.

3. The piezoelectric quartz wafer polishing control method based on polishing frequency monitoring according to claim 2, characterized in that: The step of performing control optimization under reconstruction taboo constraints according to the thickness sensing result, the current grinding roughness, and the target grinding requirement to establish a reconstruction result includes: Establishing an objective function, wherein evaluation items of the objective function include a thickness evaluation item, a roughness evaluation item, and an operating condition stability evaluation item; After the control space is reconstructed using the reconstruction taboo constraint, control optimization based on the objective function is performed in the reconstructed control space to establish a reconstruction result.

4. The piezoelectric quartz wafer polishing control method based on polishing frequency monitoring according to claim 1, characterized in that: Before the target grinding requirement is used as the control target and the grinding control reconstruction based on the real-time sensing result and the thickness sensing result is performed, the method includes: Using the real-time sensing result, the working condition stability evaluation of the current node is performed to establish a first evaluation anomaly; Performing grinding stability anomaly analysis of the piezoelectric quartz wafer and establishing a static evaluation threshold; Based on the static evaluation threshold, abnormal trigger identification of the first evaluation abnormality is performed. If the abnormal trigger identification result is a trigger result, a reconstruction instruction is generated. According to the reconstruction instruction, the target grinding requirement is used as the control target, and the grinding control reconstruction based on the real-time perception result and the thickness perception result is executed.

5. The piezoelectric quartz wafer polishing control method based on polishing frequency monitoring according to claim 4, characterized in that: The abnormality trigger identification of the first evaluation abnormality based on the static evaluation threshold includes: If the abnormal trigger identification result is a non-trigger result, calling the time series grinding behavior data set to perform time series fitting prediction and establish a time series fitting prediction result; Using the static evaluation threshold to perform trigger analysis on the time series fitting prediction results to identify abnormal trigger nodes; A prediction reconstruction instruction is configured based on the abnormal trigger node, and grinding control reconstruction is performed according to the prediction reconstruction instruction.

6. The piezoelectric quartz wafer polishing control method based on polishing frequency monitoring according to claim 5, characterized in that: The trigger analysis of the time series fitting prediction result using the static evaluation threshold to identify abnormal trigger nodes includes: If the abnormal triggering node is a node outside the preset time period, a hold instruction is generated; The original polishing parameters are maintained according to the maintenance instruction to perform piezoelectric quartz wafer polishing control management.

7. The piezoelectric quartz wafer polishing control method based on polishing frequency monitoring according to claim 1, characterized in that: The method of using the time series grinding behavior data set to perceive the grinding condition and establish a real-time perception result includes: Performing multi-dimensional working condition feature extraction on the time series grinding behavior data set to establish a working condition feature vector; Using the annotated historical data to train a working condition recognition model, sending the working condition feature vector to the working condition recognition model, and outputting a working condition label sequence; The working condition label sequence is used to perform stable trend perception, and the working condition label of the current node is enhanced using the stable trend perception result to establish a real-time perception result.

8. The piezoelectric quartz wafer polishing control method based on polishing frequency monitoring according to claim 1, characterized in that: The piezoelectric quartz wafer grinding control management according to the reconstruction result includes: Record the grinding quality test data of piezoelectric quartz wafers and obtain real parameter control data; After the grinding quality detection data and the real parameter control data are mapped into a batch parameter group, the group is saved in a control center.

9. The piezoelectric quartz wafer polishing control method based on polishing frequency monitoring according to claim 1, wherein: The activating the non-contact thickness sensing unit to perform real-time piezoelectric quartz wafer thickness sensing and establish a thickness sensing result includes: Enable the non-contact thickness sensing unit to obtain the original thickness data of the piezoelectric quartz wafer; Extracting input variables according to the original behavior observation data stream, wherein the input variables include grinding frequency data, unit time step displacement, and contact pressure response; Perform thickness fitting analysis based on the input variables and the thickness original data to establish a thickness prediction result; The thickness prediction result is used to perform calibration compensation for the thickness sensing result.

10. A piezoelectric quartz wafer grinding control system based on grinding frequency monitoring, characterized in that: The steps for implementing the piezoelectric quartz wafer grinding control method based on grinding frequency monitoring according to any one of claims 1 to 9, wherein the piezoelectric quartz wafer grinding control system based on grinding frequency monitoring comprises: Multi-source data perception module, used to perform multi-source data perception of piezoelectric quartz wafer grinding and establish the original behavior observation data stream. Multi-source data perception includes grinding frequency perception, step perception, and contact perception; A behavior modeling module, configured to perform behavior modeling for each round of grinding based on the original behavior observation data stream, and establish a time series grinding behavior dataset; A grinding condition perception module, configured to perceive the grinding condition using the time series grinding behavior dataset and establish a real-time perception result; A thickness sensing module is used to activate a non-contact thickness sensing unit to perform real-time piezoelectric quartz wafer thickness sensing and establish thickness sensing results; A grinding control reconstruction module is used to perform grinding control reconstruction based on real-time sensing results and thickness sensing results with the target grinding requirement as the control target, and establish a reconstruction result; The grinding control management module is used to perform grinding control management of the piezoelectric quartz wafer according to the reconstruction result.

Images