A method for dynamic control of pickling solution for high-purity quartz products

Abstract

This invention discloses a dynamic control method for pickling solutions of high-purity quartz products, belonging to the field of process control technology. The method includes: real-time acquisition of pickling solution parameters; assimilation analysis of these parameters to generate a concentration-conductivity-temperature mapping table; conversion of a compensation addition rate setpoint into a control signal for an acid replenishment pump to execute the acid replenishment operation; simultaneous prediction of the hydrofluoric acid concentration after the acid replenishment operation using a digital twin model based on the compensation addition rate setpoint; comparison of the new hydrofluoric acid concentration with the predicted concentration to generate a deviation value; and calibration and updating of the digital twin model parameters based on the deviation value. This invention achieves accurate and optimized control of pickling solution activity compensation by constructing a digital twin model and predicting the compensation addition rate value based on the effective active concentration.

Description

Technical Field

[0001] This invention relates to the field of process control technology, and in particular to a method for dynamic control of pickling solution for high-purity quartz products. Background Technology

[0002] In the field of industrial process automation control, precise control of pickling processes has always been a key technical challenge. Traditional methods mainly rely on real-time monitoring of key parameters of the pickling solution (such as concentration, temperature, and conductivity) and feedback control based on preset static models or empirical thresholds. With the development of sensing technology and computer control theory, especially the rise of model predictive control (MPC) and digital twin technology, new technical paths have been provided for building high-precision and predictable process control. By establishing a virtual model of the controlled object, it is possible to predict and optimize the future state to a certain extent.

[0003] However, the core limitation of existing technologies in handling complex dynamic processes such as pickling of high-purity quartz products lies in the fact that the parameter mapping relationships they rely on are mostly static or linear models, which are difficult to accurately characterize the nonlinear and time-varying characteristics of the concentration-conductivity-temperature relationship caused by factors such as component fluctuations, temperature changes, and impurity accumulation in actual operation. The model mismatch problem leads to deviations in the effective active concentration calculated based on a single parameter, which in turn makes the compensation control action lack precision and affects the stability and consistency of pickling quality. Summary of the Invention

[0004] In view of the problems existing in the prior art, the present invention is proposed.

[0005] This invention provides a method for dynamic control of pickling solution for high-purity quartz products, which solves the problems of inaccurate estimation of effective active concentration and lag in compensation control caused by the inability of static parameter mapping relationship to adapt to dynamic changes in pickling solution.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides a method for dynamic control of pickling solution for high-purity quartz products, comprising: acquiring pickling solution parameters in real time; performing assimilation analysis on the pickling solution parameters to generate a concentration-conductivity-temperature relationship mapping table; converting the measured conductivity and temperature of the pickling solution parameters into an effective active concentration of hydrofluoric acid based on the concentration-conductivity-temperature relationship mapping table; comparing the effective active concentration with a preset target active concentration threshold to generate a comparison result, and triggering an activity compensation mechanism based on the comparison result; constructing a digital twin model, and after the activity compensation mechanism is triggered, predicting a set value for the compensation addition rate of hydrofluoric acid based on the effective active concentration; converting the set value for the compensation addition rate into a control signal for an acid replenishment pump to execute the acid replenishment operation; simultaneously, predicting a predicted value for the hydrofluoric acid concentration after the acid replenishment operation based on the set value for the compensation addition rate; comparing the new hydrofluoric acid concentration with the predicted value to generate a deviation value, and calibrating and updating the parameters of the digital twin model based on the deviation value.

[0008] As a preferred embodiment of the dynamic control method for pickling solution of high-purity quartz products according to the present invention, the parameters of the pickling solution of high-purity quartz products include hydrofluoric acid concentration, pickling solution temperature and measured values ​​of pickling solution conductivity.

[0009] As a preferred embodiment of the dynamic control method for pickling solution of high-purity quartz products according to the present invention, the steps of performing assimilation analysis on the parameters of the pickling solution of high-purity quartz products to generate a concentration-conductivity-temperature relationship mapping table are as follows:

[0010] A multi-channel parallel processing architecture was constructed to independently filter the measured values ​​of hydrofluoric acid concentration, pickling solution temperature, and pickling solution conductivity to obtain smooth sequences of each parameter.

[0011] The smoothed sequence of each parameter is input into a dynamically coupled extended Kalman filter assimilation algorithm for assimilation, and the assimilated pickling solution state data point set is output.

[0012] Based on the pickling solution state data point set, a three-dimensional surface fitting was performed using a hybrid radial basis function interpolation algorithm to generate a concentration-conductivity-temperature relationship mapping table.

[0013] As a preferred embodiment of the dynamic control method for pickling solution of high-purity quartz products according to the present invention, the step of converting the measured value of the pickling solution conductivity and the pickling solution temperature in the parameters of the pickling solution of high-purity quartz products into the effective active concentration of hydrofluoric acid based on the concentration-conductivity-temperature relationship mapping table is as follows:

[0014] The filtered conductivity and temperature of the pickling solution are extracted from the smoothed sequence of each parameter. The concentration-conductivity-temperature relationship mapping table is queried by fuzzy inference algorithm to select the reference data point set.

[0015] Based on the reference data point set, the initial value of the effective active concentration of hydrofluoric acid was obtained by using the active concentration calculation model.

[0016] Real-time pH monitoring values ​​were collected as auxiliary verification to correct the initial value of the effective active concentration and generate the effective active concentration.

[0017] As a preferred embodiment of the dynamic control method for pickling solution of high-purity quartz products according to the present invention, the step of comparing the effective active concentration with a preset target active concentration threshold to trigger an activity compensation mechanism is as follows:

[0018] The effective activity concentration is compared with a preset target activity concentration threshold to generate a comparison result.

[0019] When the comparison result shows that the effective activity concentration exceeds the preset target activity concentration threshold, the activity compensation mechanism is not triggered.

[0020] When the comparison result shows that the effective activity concentration does not exceed the preset target activity concentration threshold, the activity compensation mechanism is triggered.

[0021] As a preferred embodiment of the dynamic control method for pickling solution of high-purity quartz products according to the present invention, the steps for constructing the digital twin model are as follows:

[0022] Collect historical high-purity quartz product pickling solution parameter datasets, construct a deep neural network architecture, learn the dynamic relationship between historical high-purity quartz product pickling solution parameters, and obtain the deep neural network;

[0023] A digital twin model is constructed by combining deep neural networks with the physical equations governing mass conservation in the pickling process.

[0024] As a preferred embodiment of the dynamic control method for pickling solution of high-purity quartz products according to the present invention, the step of predicting the set value of the compensation addition rate of hydrofluoric acid based on the effective active concentration after the activity compensation mechanism is triggered is as follows:

[0025] After the activity compensation mechanism is triggered, the effective activity concentration is input into the digital twin model for multi-step state deduction to generate the future state trajectory.

[0026] Based on the future state trajectory, calculate the compensation addition rate value required for each prediction step and combine them to form a sequence of compensation addition rate values;

[0027] The compensation addition rate value at the first time position in the compensation addition rate value sequence is selected as the hydrofluoric acid compensation addition rate setpoint.

[0028] As a preferred embodiment of the dynamic control method for pickling solution of high-purity quartz products according to the present invention, the steps of converting the compensation addition rate setpoint into a control signal for the acid replenishment pump and performing the acid replenishment operation are as follows:

[0029] The compensation addition rate setting value is verified and tamper-proofed, and a security control command is generated.

[0030] The model predictive control rolling optimization algorithm optimizes the duty cycle and frequency of safety control commands in real time, and converts them into pulse width modulation signals to drive the acid replenishment pump.

[0031] The acid replenishment pump is driven by a pulse width modulation signal, and the pump's operation status is monitored in real time.

[0032] As a preferred embodiment of the dynamic control method for pickling solution of high-purity quartz products according to the present invention, the step of simultaneously predicting the hydrofluoric acid concentration after the acid replenishment operation using a digital twin model based on a compensation addition rate setpoint is as follows:

[0033] The effective active concentration of hydrofluoric acid after the acid replenishment operation was collected, and a fusion feature set was constructed by combining the compensation addition rate set value and historical high-purity quartz product pickling solution parameters.

[0034] The fused feature set is input into the digital twin model to perform multi-step concentration prediction and generate an initial concentration prediction sequence.

[0035] Uncertainty quantification and calibration are performed on the initial concentration prediction sequence, and the predicted value of hydrofluoric acid concentration after acid replenishment is output.

[0036] As a preferred embodiment of the dynamic control method for pickling solution of high-purity quartz products according to the present invention, the steps of obtaining the new hydrofluoric acid concentration and comparing it with the predicted hydrofluoric acid concentration to generate a deviation value, and calibrating and updating the parameters of the digital twin model based on the deviation value are as follows:

[0037] The actual measured value of hydrofluoric acid concentration after the acid replenishment operation is obtained as the new hydrofluoric acid concentration. The new hydrofluoric acid concentration is compared with the predicted value of hydrofluoric acid concentration to generate a composite deviation feature set.

[0038] Based on the composite bias feature set, a hierarchical calibration strategy with attention mechanism weighting is adopted to perform collaborative calibration of the digital twin model and generate calibration results;

[0039] The calibration effect is evaluated using a meta-learning framework, and the hierarchical calibration strategy is optimized based on the evaluation results to complete the parameter update of the digital twin model.

[0040] The beneficial effects of this invention are as follows: By constructing a digital twin model and predicting the compensation addition rate value based on the effective active concentration, the accuracy and optimized control of pickling solution activity compensation are achieved; the digital twin model is not a static mapping, but deeply integrates real-time effective active concentration, historical dynamics, and physical laws, enabling high-precision multi-step prediction of the response after acid replenishment; based on the prediction results, the compensation addition rate setpoint can be solved inversely by the model, which not only considers the instantaneous concentration deviation, but also anticipates the dynamic impact of the addition behavior on the future state; the control strategy based on model prediction transforms the traditional passive lag compensation into active adjustment, effectively avoiding overshoot or undercompensation, improving the control accuracy and stability of pickling solution activity concentration, thereby ensuring the high quality and consistency of the pickling process for high-purity quartz products. Attached Figure Description

[0041] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 A flowchart for the dynamic control method of pickling solution for high-purity quartz products.

[0043] Figure 2 A flowchart for generating a concentration-conductivity-temperature mapping table.

[0044] Figure 3 This is a flowchart for converting to an effective active concentration.

[0045] Figure 4 A flowchart for adding rate setting values ​​to predictive compensation for digital twin models. Detailed Implementation

[0046] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0047] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0048] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0049] Reference Figures 1-4 This is one embodiment of the present invention, which provides a method for dynamically controlling the pickling solution of high-purity quartz products, comprising the following steps:

[0050] S1. Real-time acquisition of pickling solution parameters for high-purity quartz products, assimilation analysis of these parameters, and generation of a concentration-conductivity-temperature relationship mapping table.

[0051] S1.1: The parameters of the pickling solution for high-purity quartz products include the concentration of hydrofluoric acid, the temperature of the pickling solution, and the measured value of the conductivity of the pickling solution.

[0052] It should be noted that hydrofluoric acid concentration refers to the content of hydrofluoric acid (HF) in the pickling solution of high-purity quartz products, usually expressed as a mass percentage or molar concentration, which directly affects the pickling ability.

[0053] Pickling solution temperature refers to the actual temperature of the pickling solution for high-purity quartz products during the pickling process, which affects the chemical reaction rate and the activity of hydrofluoric acid.

[0054] The conductivity measurement value of pickling solution refers to the electrical conductivity value of high-purity quartz pickling solution measured by a conductivity sensor. It can indirectly reflect the ion concentration in the solution and the effective activity state of hydrofluoric acid.

[0055] S1.2: Construct a multi-channel parallel processing architecture to independently filter the measured values ​​of hydrofluoric acid concentration, pickling solution temperature, and pickling solution conductivity to obtain smooth sequences of each parameter;

[0056] Specifically, separate data paths are established for hydrofluoric acid concentration, pickling solution temperature, and pickling solution conductivity measurements, respectively. Each data path has independent signal reception and filtering capabilities. The three data paths operate synchronously in time but do not interfere with each other structurally, forming a multi-channel parallel processing architecture. The hydrofluoric acid concentration data path, pickling solution temperature data path, and pickling solution conductivity measurement data path receive the hydrofluoric acid concentration measurement value, pickling solution temperature measurement value, and pickling solution conductivity measurement value, and smooth them through filtering methods to form smoothed sequences for hydrofluoric acid concentration, pickling solution temperature, and pickling solution conductivity measurements, respectively.

[0057] S1.3: Input the smoothed sequence of each parameter into the dynamically coupled extended Kalman filter assimilation algorithm for assimilation, and output the assimilated pickling solution state data point set;

[0058] Specifically, the smoothed sequences of hydrofluoric acid concentration, pickling solution temperature, and pickling solution conductivity measurements are fed into a dynamically coupled extended Kalman filter assimilation algorithm (EKF). Based on the state evolution law and observation relationship of the pickling process, the algorithm predicts and corrects the current state reflected by the smoothed sequences of hydrofluoric acid concentration, pickling solution temperature, and pickling solution conductivity measurements. After each correction, a data point containing consistent information of hydrofluoric acid concentration, pickling solution temperature, and pickling solution conductivity measurements is formed. The data points formed sequentially at multiple time points constitute the assimilated pickling solution state data point set.

[0059] It should be noted that the dynamically coupled extended Kalman filter assimilation algorithm is an existing assimilation method that combines the time-varying coupling relationship between the hydrofluoric acid concentration, pickling solution temperature and the measured conductivity of the pickling solution during the pickling process, and uses extended Kalman filtering to perform state estimation and data fusion on multi-source smooth sequences to generate a consistent pickling solution state data point set.

[0060] It should be noted that the state evolution law of the pickling process refers to the physical dynamic relationship between the measured values ​​of hydrofluoric acid concentration, pickling solution temperature, and pickling solution conductivity and time; the observation relationship refers to the correspondence between the smoothed sequence of hydrofluoric acid concentration, the smoothed sequence of pickling solution temperature, and the smoothed sequence of pickling solution conductivity and the actual state of the pickling solution.

[0061] S1.4: Based on the pickling solution state data point set, a three-dimensional surface fitting is performed using a hybrid radial basis function interpolation algorithm to generate a concentration-conductivity-temperature relationship mapping table.

[0062] Specifically, the hybrid radial basis function (RBF) interpolation algorithm uses the hydrofluoric acid concentration, measured conductivity, and temperature of each data point in the pickling solution state data set as known sample points in three-dimensional space. In three-dimensional space, for any location to be determined, the RBF algorithm generates corresponding weights using radial basis functions based on the distance from the location to be determined to each known sample point. The corresponding weights are then weighted and combined with the hydrofluoric acid concentration of the known sample points, and an auxiliary interpolation strategy is used to compensate for the overall trend, so that the interpolated hydrofluoric acid concentration maintains detailed response locally and smooth continuity globally. For all combinations of pickling solution conductivity and temperature covering the effective operating conditions of the pickling process, the corresponding hydrofluoric acid concentration is determined one by one, forming a concentration-conductivity-temperature relationship mapping table.

[0063] It should be noted that the radial basis function adopts a multiquadric kernel function form, the expression of which depends on the shape parameter. The shape parameter is adaptively selected on the pickling solution state data point set through leave-one-out cross-validation to minimize the mean absolute error of the prediction. The interpolation process is a global fit, and a Tikhonov regularization term is introduced to suppress overfitting and improve numerical stability. The regularization coefficient is also determined through cross-validation.

[0064] Auxiliary interpolation strategies refer to existing interpolation methods used in hybrid radial basis function interpolation algorithms to supplement the insufficient local approximation ability of radial basis functions and enhance the overall surface smoothness and physical rationality, such as low-order polynomial terms.

[0065] The effective operating range refers to the combination of values ​​that are actually possible and have process significance, including the measured values ​​of hydrofluoric acid concentration, pickling solution conductivity, and pickling solution temperature during the pickling process.

[0066] S2. Based on the concentration-conductivity-temperature relationship mapping table, the measured values ​​of the pickling solution conductivity and the pickling solution temperature in the pickling solution parameters of high-purity quartz products are converted into the effective active concentration of hydrofluoric acid.

[0067] S2.1: Extract the filtered conductivity measurement value and pickling temperature of the pickling solution from the smoothed sequence of each parameter, and use the fuzzy inference algorithm to query the concentration-conductivity-temperature relationship mapping table to select the reference data point set;

[0068] Specifically, the current moment's pickling solution conductivity measurement and temperature are extracted from the smoothed sequence of each parameter and used as query conditions to locate the neighboring region in the concentration-conductivity-temperature relationship mapping table. A fuzzy inference algorithm is used to evaluate the matching degree of data points within the neighboring region, establishing fuzzy membership relationships for the pickling solution conductivity measurement and temperature, and mapping them to the fuzzy linguistic variables corresponding to the pickling solution conductivity measurement and temperature for each data point in the concentration-conductivity-temperature relationship mapping table. Based on a pre-set fuzzy rule base, the membership degree of each data point in the pickling solution conductivity measurement dimension and pickling solution temperature dimension is combined to form the comprehensive matching degree between the data point and the current operating condition. All data points within the neighboring region are sorted according to the comprehensive matching degree, and the data points with the highest comprehensive matching degree are selected to form a reference data point set.

[0069] It should be noted that fuzzy membership relationships were established for the measured conductivity and temperature of the pickling solution, respectively. The fuzzy linguistic variable for the measured conductivity was divided into three levels: "low," "medium," and "high," while the fuzzy linguistic variable for the temperature was divided into three levels: "low temperature," "medium temperature," and "high temperature." Triangular membership functions were used for both, with the vertices and base endpoints determined based on the effective operating range of the pickling process. For example, the vertex of the "low" level triangle for the measured conductivity was located at 800 µS / cm (base...). The pickling solution temperature is as follows: 600–1000 µS / cm at the edge; 1400 µS / cm at the top edge (1200–1600 µS / cm at the bottom edge); 2000 µS / cm at the top edge (1800–2200 µS / cm at the bottom edge); 25°C at the top edge (20–30°C at the bottom edge); 40°C at the top edge (35–45°C at the bottom edge); and 55°C at the top edge (50–60°C at the bottom edge).

[0070] The pre-defined fuzzy rule base refers to a set of fuzzy conditional statements pre-established in the fuzzy inference algorithm to describe the empirical correlation between the measured conductivity of the pickling solution and the temperature and concentration of the pickling solution. It includes empirical rules such as "If the measured conductivity of the pickling solution is high and the temperature is medium, then the concentration of hydrofluoric acid tends to be low," which support the logic for determining the matching degree of data points. Examples: If the measured conductivity of the pickling solution is high and the temperature is high, then the concentration of hydrofluoric acid tends to be low; if the measured conductivity of the pickling solution is medium and the temperature is medium, then the concentration of hydrofluoric acid tends to be medium; if the measured conductivity of the pickling solution is low and the temperature is low, then the concentration of hydrofluoric acid tends to be high.

[0071] S2.2: Based on the reference data point set, the initial value of the effective active concentration of hydrofluoric acid is calculated using the active concentration calculation model. The expression is as follows:

[0072] ;

[0073] In the formula, This indicates the initial effective active concentration of hydrofluoric acid. Indicates the total number of reference data points. Indicates the reference data point index. Indicates reference data point The weight, Indicates reference data point The concentration of hydrofluoric acid, Indicates the temperature adjustment factor. Indicates reference data point The temperature of the pickling solution, Indicates the reference pickling solution temperature. Represents the square of the temperature difference. This indicates the adjustment factor for the measured conductivity of the pickling solution. Indicates reference data point The measured conductivity value of the pickling solution. This indicates the measured conductivity value of the reference pickling solution. It represents the square of the difference in conductivity measurements.

[0074] It should be noted that the concentration of hydrofluoric acid and The dimensions are mass percentage and temperature difference. and conductivity difference The corresponding units are Celsius (degrees Celsius). ) and micro Siemens per centimeter ( Temperature adjustment factor The unit is 1 / Adjustment factor for the conductivity measurement value of pickling solution The unit is 1 / Therefore, the exponential decay term has no unit, remains dimensionless, and the weights... It is dimensionless, thus ensuring the consistency of all dimensions in the formula.

[0075] It should be noted that the reference data points The weighting takes into account the quality of the data points (such as measurement error and equipment accuracy), matching degree (the degree of consistency between the current operating conditions and the data points) and the reliability of historical data. The setting formula is ,in Indicates the quality factor. Represents the matching degree factor. This represents the adjustment parameter used to balance the importance of the quality factor and the matching degree factor. Specifically, data points with higher quality have a larger weight, while data points that match the current working conditions well also receive a larger weight. (Reference data points) The weights are obtained by weighted summation of quality and matching degree. For example, when the quality of a data point is 0.8 and the matching degree is 0.9, then the reference data point... The weight is 0.72. The weight setting can flexibly reflect the importance of each reference data point in the calculation, while ensuring that the entire weight allocation process is reproducible.

[0076] Temperature adjustment factor This method is used to control the effect of pickling solution temperature on the calculation of hydrofluoric acid concentration. Temperature has a significant impact on the rate of chemical reactions and the activity of hydrofluoric acid during pickling; therefore, it needs to be set based on historical experimental data. The value of the temperature adjustment factor is determined by measuring the variation of hydrofluoric acid concentration at different temperatures using historical experimental data and determining the relationship between temperature and concentration through fitting methods. For example, if historical experimental data shows that the hydrofluoric acid concentration increases by 2% for every 1°C increase in temperature, then the temperature adjustment factor can be set to... =0.02;

[0077] Adjustment factor for conductivity measurement of pickling solution To control the impact of conductivity changes on hydrofluoric acid concentration, the conductivity measurement value of the pickling solution is closely related to the ion concentration in the solution. Changes in ion concentration directly affect the concentration of hydrofluoric acid. Therefore, the adjustment factor for the conductivity measurement value of the pickling solution is set based on historical experimental data. The change in hydrofluoric acid concentration is measured under different conductivity measurements of the pickling solution, and the mathematical relationship between the conductivity measurement value of the pickling solution and the hydrofluoric acid concentration is obtained through a fitting method. Then, the deviation in the measured conductivity of the pickling solution can be converted into a correction for the hydrofluoric acid concentration; for example, if historical experimental data shows that for every 10 µS / cm increase in the measured conductivity of the pickling solution, the hydrofluoric acid concentration increases by 1%, then a correction can be set. =0.1.

[0078] S2.3: Collect real-time pH monitoring values ​​as auxiliary verification, correct the initial value of effective active concentration, and generate effective active concentration.

[0079] Specifically, real-time pH (acidity / alkalinity) monitoring values ​​are collected, and the real-time pH monitoring values ​​are compared with the initial effective active concentration of hydrofluoric acid. If the acidity trend reflected by the real-time pH monitoring values ​​deviates from the expected pH range corresponding to the initial effective active concentration of hydrofluoric acid, the initial effective active concentration of hydrofluoric acid is adjusted according to the preset pH-active concentration correspondence to form the corrected effective active concentration of hydrofluoric acid.

[0080] It should be noted that the expected pH range refers to the pH monitoring range determined based on the historical or theoretical acid-base performance of the pickling solution corresponding to the initial effective active concentration of hydrofluoric acid.

[0081] The pH-active concentration correspondence refers to the mapping relationship between the real-time pH monitoring value and the effective active concentration of hydrofluoric acid, which is established in advance in the pickling process. It is formed by statistical induction based on the real-time pH monitoring values ​​collected synchronously during historical pickling processes and the assimilated and verified effective active concentration data of hydrofluoric acid.

[0082] S3. Compare the effective active concentration with the preset target active concentration threshold, generate the comparison result, and trigger the activity compensation mechanism based on the comparison result.

[0083] S3.1: Compare the effective active concentration with the preset target active concentration threshold and generate the comparison result;

[0084] Specifically, the effective active concentration is directly compared with the preset target active concentration threshold. If the effective active concentration is greater than or equal to the preset target active concentration threshold, a comparison result of "effective active concentration exceeds preset target active concentration threshold" is generated; if the effective active concentration is less than the preset target active concentration threshold, a comparison result of "effective active concentration does not exceed preset target active concentration threshold" is generated.

[0085] It should be noted that the target active concentration threshold is set based on the balance requirements of surface impurity removal efficiency and quartz body corrosion control in the pickling process of high-purity quartz products. The specific setting steps include: determining the range of hydrofluoric acid active concentration that can effectively dissolve metal impurities without causing excessive etching of quartz during production; combining the corresponding effective active concentration data of qualified products in historical pickling batches; and selecting the median value with stable performance within this range as the target active concentration threshold. An exemplary value range is 0.8 mol / L to 1.5 mol / L. Within the range of 0.8 mol / L to 1.5 mol / L, the pickling solution can maintain sufficient reactivity to remove trace metal impurities such as sodium, potassium, and iron, while avoiding damage to the quartz network structure or excessive surface roughness due to excessive activity. When the effective active concentration exceeds 1.5 mol / L, the pickling solution is too corrosive, which may cause excessive etching, microcracks, or a decrease in purity on the surface of high-purity quartz products. When the effective active concentration is below 0.8 mol / L, the pickling reaction kinetics are insufficient, and impurities cannot be effectively removed, resulting in substandard product cleanliness.

[0086] S3.2: When the comparison result shows that the effective activity concentration exceeds the preset target activity concentration threshold, the activity compensation mechanism is not triggered;

[0087] Specifically, when the comparison result shows that the effective active concentration exceeds the preset target active concentration threshold, it indicates that the reactivity of hydrofluoric acid in the current pickling solution has met the process requirements. At this time, no acid replenishment action is performed, the active compensation mechanism remains closed, no control signal is sent to the acid replenishment pump, and the pickling process continues according to the existing pickling solution state.

[0088] S3.3: When the comparison result shows that the effective activity concentration does not exceed the preset target activity concentration threshold, the activity compensation mechanism is triggered.

[0089] Specifically, when the comparison result shows that the effective active concentration does not exceed the preset target active concentration threshold, it indicates that the reactivity of hydrofluoric acid in the current pickling solution is insufficient. At this time, the active compensation mechanism is activated. The activation of the active compensation mechanism is manifested by entering the set value of the hydrofluoric acid compensation addition rate predicted by the digital twin model and preparing to generate a control signal for driving the acid replenishment pump. The active compensation mechanism acts as a condition trigger node, enabling the entire acid replenishment operation to be executed in order to restore the process activity of the pickling solution.

[0090] S4. Construct a digital twin model. After the activity compensation mechanism is triggered, the digital twin model predicts the compensation addition rate setting value of hydrofluoric acid based on the effective activity concentration.

[0091] S4.1: Collect historical high-purity quartz product pickling solution parameter dataset, construct a deep neural network architecture, learn the dynamic relationship between historical high-purity quartz product pickling solution parameters, and obtain the deep neural network;

[0092] Specifically, a historical dataset of pickling solution parameters for high-purity quartz products was collected, including measurements of hydrofluoric acid concentration, pickling solution temperature, and pickling solution conductivity at multiple time points. Based on this dataset, a deep neural network architecture was constructed, consisting of multiple layers of neurons, capable of modeling time-series data. Within this deep neural network architecture, the hydrofluoric acid concentration, pickling solution temperature, and pickling solution conductivity measurements at each time point in the historical dataset were used as training samples. By adjusting the internal connection weights of the network, the deep neural network was able to reflect the dynamic correlation between hydrofluoric acid concentration, pickling solution temperature, and pickling solution conductivity measurements. After training, a deep neural network with mapping capabilities was obtained.

[0093] It should be noted that time-series data refers to a sequence of measurements of hydrofluoric acid concentration, pickling solution temperature, and pickling solution conductivity recorded in chronological order. Each data point corresponds to a specific time point, reflecting the evolution of parameters over time during the pickling process.

[0094] S4.2: Combine deep neural networks with the physical equations for mass conservation in the pickling process to construct a digital twin model.

[0095] Specifically, based on the learning results of deep neural networks on the dynamic relationship between the measured values ​​of hydrofluoric acid concentration, pickling solution temperature, and pickling solution conductivity in historical high-purity quartz pickling solution parameter datasets, the material change constraints described by the mass conservation physical equation of the pickling process are embedded in the prediction logic. The mass conservation physical equation of the pickling process expresses the total amount conservation relationship of hydrofluoric acid caused by factors such as reaction consumption, acid replenishment, and volume changes during the pickling process. By combining the nonlinear mapping capability of deep neural networks with the mechanistic constraints of the mass conservation physical equation of the pickling process, a digital twin model with both data-driven characteristics and physical consistency is formed.

[0096] It should be noted that the deep neural network adopts a Long Short-Term Memory (LSTM) architecture, which includes three LSTM hidden layers, two fully connected layers, and one output layer, corresponding to the measured values ​​of hydrofluoric acid concentration, pickling solution temperature, and pickling solution conductivity at the next time step. The mean squared error is used as the optimization target during the pre-training stage, and training is terminated early when the loss does not decrease significantly after several consecutive rounds (e.g., 10 rounds).

[0097] It should be noted that the pre-training process of the digital twin model is based on a historical dataset of pickling solution parameters for high-purity quartz products. This dataset includes measurements of hydrofluoric acid concentration, pickling solution temperature, and pickling solution conductivity. A deep neural network architecture is constructed and trained using this dataset to understand the dynamic relationship between hydrofluoric acid concentration, pickling solution temperature, and pickling solution conductivity. After training, the deep neural network is obtained. This deep neural network is then combined with the mass conservation physical equations of the pickling process to form a digital twin model with physical consistency and data-driven capabilities. This process is completed before the activity compensation mechanism is triggered, providing a foundation for multi-step state extrapolation.

[0098] S4.3: After the activity compensation mechanism is triggered, the effective activity concentration is input into the digital twin model for multi-step state deduction to generate the future state trajectory;

[0099] Specifically, after the activity compensation mechanism is triggered, the effective activity concentration is fed into the digital twin model. Starting with the effective activity concentration, the digital twin model combines the physical equation of mass conservation in the pickling process and the learning results of the deep neural network on the dynamic relationship between the parameters of the pickling solution of historical high-purity quartz products to sequentially deduce the hydrofluoric acid concentration, pickling solution temperature and pickling solution conductivity measurements at multiple time points. Each step of the deduction is based on the state at the previous moment, gradually forming a sequence of hydrofluoric acid concentration, pickling solution temperature and pickling solution conductivity measurements covering a period of time in the future, which is the future state trajectory.

[0100] S4.4: Based on the future state trajectory, calculate the compensation addition rate value required for each prediction step and combine them to form a sequence of compensation addition rate values;

[0101] Specifically, based on the future state trajectory, for the hydrofluoric acid concentration corresponding to each prediction step, the required compensation addition rate value for each prediction step is calculated, resulting in multiple compensation addition rate values ​​that correspond one-to-one with the time points of the future state trajectory; the compensation addition rate values ​​are arranged in chronological order and combined to form a sequence of compensation addition rate values.

[0102] Based on the future state trajectory, the compensation addition rate value required for each prediction step is calculated, as expressed by:

[0103] ;

[0104] ;

[0105] In the formula, Indicates the first The compensation addition rate value required for each prediction step size. Indicates the first The target hydrofluoric acid concentration at each prediction step size. Indicates the first The actual hydrofluoric acid concentration at each predicted step size. Indicates the time step. Indicates liquid flow rate, Indicates the volume of the tank. This indicates the concentration of the mother liquor.

[0106] It should be noted that, and It is usually expressed as molar concentration (mol / L) or mass percentage (%), so both have the same units, ensuring that differences between them can be compared effectively. This indicates the time step, usually in seconds (s) or hours (h). Ensure consistency in time units, as concentration units are in mol / L or %, while time units are in seconds or hours. The unit is molar concentration difference (mol / L or %), while the time step is used to measure the difference. After normalization, The unit for the (compensation addition rate value) is mol / L / s or % / s. The unit is L or m³. The unit is mol / L or %. The units are L / s or L / h; therefore, the dimensions of all parameters in the formula are consistent and can be converted to each other.

[0107] S4.5: Select the compensation addition rate value at the first time position in the compensation addition rate value sequence as the hydrofluoric acid compensation addition rate set value.

[0108] Specifically, in the sequence of compensation addition rate values, the compensation addition rate value corresponding to the first time position is located in chronological order. The compensation addition rate value corresponds to the acid addition intensity required for the most recent prediction step after the current time. The compensation addition rate value of the first time position is directly determined as the hydrofluoric acid compensation addition rate set value.

[0109] S5. Convert the compensation addition rate setpoint into a control signal for the acid replenishment pump and execute the acid replenishment operation; at the same time, the digital twin model predicts the hydrofluoric acid concentration after the acid replenishment operation based on the compensation addition rate setpoint.

[0110] S5.1: Verify and prevent tampering of the compensation addition rate setting value, and generate a security control command;

[0111] Specifically, the process involves verifying whether the compensation addition rate setting falls within the preset process safety range, checking the data format and transmission integrity of the compensation addition rate setting, and confirming that it has not been lost or unauthorized modified during transmission. If the hydrofluoric acid compensation addition rate setting passes the process safety range verification and integrity check, the compensation addition rate setting is encapsulated into a safety control instruction conforming to the control protocol specification. If any verification step fails, no safety control instruction is generated, and the acid replenishment operation is suspended.

[0112] It should be noted that the process safety range is set based on the safety and effectiveness requirements of acid replenishment operations in the pickling process of high-purity quartz products. The specific setting steps include: determining the maximum acid replenishment rate that will not cause violent reactions of the pickling solution, aggravated equipment corrosion, or damage to the quartz surface during production, and determining the minimum acid replenishment rate that can effectively improve the active concentration; combining the hydrofluoric acid compensation addition rate values ​​actually used in multiple batches of stable operation data, selecting the range with good performance as the process safety range; the exemplary value range is 0.5 liters to 3.0 liters per minute. When it is below 0.5 liters / minute, the acid replenishment effect is not obvious, and it is difficult to restore the effective active concentration within a reasonable time; when it is above 3.0 liters / minute, it may cause local excessive concentration, sudden temperature rise, or acid mist overflow, affecting process stability and operational safety.

[0113] It should be noted that the correspondence between the compensation addition rate setpoint and the pulse width modulation signal of the acid replenishment pump is determined by a pre-established acid replenishment pump flow calibration curve. The acid replenishment pump flow calibration curve is obtained by applying pulse width modulation signals with different duty cycles and frequency combinations to the acid replenishment pump in offline mode and measuring the corresponding stable output flow. The calibration covers a duty cycle range of 20% to 100% and a frequency range of 0.5Hz to 10Hz. The flow measurement results are recorded in liters per minute (L / min). After generating the safety control command, the model predictive control rolling optimization algorithm maps the compensation addition rate setpoint to a duty cycle and frequency combination that meets the flow requirements based on the acid replenishment pump flow calibration curve. This is used for the subsequent generation of the pulse width modulation signal to drive the acid replenishment pump.

[0114] S5.2: The duty cycle and frequency of the safety control command are optimized in real time by the model predictive control rolling optimization algorithm and converted into a pulse width modulation signal to drive the acid replenishment pump;

[0115] Specifically, based on the hydrofluoric acid compensation addition rate setpoint contained in the safety control command, and combined with the prediction of the future pickling solution state by the digital twin model, the duty cycle and frequency of the pulse width modulation (PWM) signal are adjusted within the current control cycle. The model predictive control (MPC) rolling optimization algorithm replans the acid replenishment pump action in each control cycle, so that the pulse width modulation signal can accurately reflect the requirements of the hydrofluoric acid compensation addition rate setpoint and adapt to the dynamic changes in the pickling solution state. The optimized duty cycle and frequency are directly used to generate the pulse width modulation signal driving the acid replenishment pump.

[0116] It should be noted that during operation, the model predictive control rolling optimization algorithm needs to comprehensively consider the acid replenishment accuracy and the smoothness of the actuator. The optimization objectives include making the actual acid replenishment flow rate as close as possible to the set value of the hydrofluoric acid compensation addition rate, while suppressing drastic fluctuations in the acid replenishment pump control signal. The model predictive control rolling optimization algorithm sets multiple constraints, covering the allowable range of duty cycle and frequency, the maximum change amplitude of a single adjustment, and the physically feasible range of the acid replenishment rate, to ensure that the generated pulse width modulation signal meets both process requirements and equipment safety operation specifications. For example, in a certain control cycle, the target hydrofluoric acid compensation addition rate is 2.0 L / min. The model predictive control rolling optimization algorithm predicts the pickling solution state for the next 5 steps, and optimizes it to gradually increase the duty cycle from 60% to 75%, while keeping the frequency constant at 5Hz. Since the change amplitude is less than 20%, the constraint is met, and the corresponding PWM signal is generated.

[0117] S5.3: Based on pulse width modulation signals, drive the acid replenishment pump to work and monitor the execution status of the acid replenishment pump in real time.

[0118] Specifically, based on pulse width modulation (PWM) signals, the drive circuit of the acid replenishment pump controls the power supply switching according to the high-level duration and period of the PWM signal, enabling the acid replenishment pump to operate at the corresponding working intensity. The duty cycle of the PWM signal determines the average output flow rate of the acid replenishment pump per unit time, and the frequency determines the speed of the switching action. Together, they ensure that the actual rate of hydrofluoric acid addition by the acid replenishment pump matches the compensation addition rate setting value. During the operation of the acid replenishment pump, the current, pressure, or flow signals of the acid replenishment pump are collected in real time by sensors to form the acid replenishment pump execution status information. The consistency of the acid replenishment pump execution status information with the expected action is checked to confirm that the acid replenishment pump is working normally according to the instructions.

[0119] S5.4: Collect the effective active concentration of hydrofluoric acid after the acid replenishment operation, and construct a fusion feature set by combining the compensation addition rate set value and historical high-purity quartz product pickling solution parameters;

[0120] Specifically, after the acid replenishment operation is completed, the conductivity, temperature, and pH of the pickling solution are acquired in real time. Based on the concentration-conductivity-temperature mapping table and the active concentration calculation model, the effective active concentration of hydrofluoric acid after the acid replenishment operation is obtained. The effective active concentration of hydrofluoric acid after the acid replenishment operation, the compensation addition rate setting value used this time, and the corresponding hydrofluoric acid concentration, pickling solution temperature, and pickling solution conductivity measurement values ​​in the historical high-purity quartz product pickling solution parameter dataset are combined to form a multi-dimensional data set that includes the current state and historical background. This multi-dimensional data set is the fusion feature set.

[0121] S5.5: Input the fused feature set into the digital twin model to perform multi-step concentration prediction and generate an initial concentration prediction sequence;

[0122] Specifically, the fused feature set is fed into the digital twin model. The digital twin model is based on the effective active concentration of hydrofluoric acid after the acid replenishment operation, the compensation addition rate setting value, and the historical pickling solution parameters of high-purity quartz products contained in the fused feature set. It combines the physical equation of mass conservation in the pickling process with the learning results of the deep neural network on the dynamic relationship of parameters to predict the hydrofluoric acid concentration at multiple future time points in turn. Each step of the prediction is based on the state at the previous moment, gradually forming a hydrofluoric acid concentration sequence covering the prediction period. The hydrofluoric acid concentration sequence is the initial concentration prediction sequence.

[0123] S5.6: Perform uncertainty quantification and calibration on the initial concentration prediction sequence, and output the predicted value of hydrofluoric acid concentration after acid replenishment operation.

[0124] Specifically, based on the deviation distribution between the prediction results and actual measured values ​​in historical operation using a digital twin model, the reliability of the hydrofluoric acid concentration prediction values ​​at each time point in the initial concentration prediction sequence is assessed. The values ​​in the initial concentration prediction sequence are adjusted according to the reliability, reducing the prediction impact of periods with high uncertainty and increasing the weight of parts consistent with historical patterns, thus generating a calibrated concentration prediction sequence. The hydrofluoric acid concentration value corresponding to the current moment is extracted from the calibrated concentration prediction sequence as the predicted hydrofluoric acid concentration value after the acid replenishment operation.

[0125] S6. Obtain the new hydrofluoric acid concentration and compare it with the predicted hydrofluoric acid concentration to generate a deviation value. Based on the deviation value, calibrate and update the parameters of the digital twin model.

[0126] S6.1: Obtain the actual measured value of hydrofluoric acid concentration after the acid replenishment operation as the new hydrofluoric acid concentration, compare the new hydrofluoric acid concentration with the predicted value of hydrofluoric acid concentration, and generate a composite deviation feature set;

[0127] Specifically, the actual measured value of hydrofluoric acid concentration after the acid replenishment operation is obtained as the new hydrofluoric acid concentration. The new hydrofluoric acid concentration is compared with the predicted value of hydrofluoric acid concentration item by item to generate the numerical deviation value between the two. At the same time, the historical deviation sequence formed in previous acid replenishment operations is retrieved to observe the position and direction of change of the deviation value on the time axis, forming a trend of change in the time dimension. The measured values ​​of the conductivity of the pickling solution, the temperature of the pickling solution, and the real-time pH monitoring value at the time of the deviation value are associated as the corresponding operating conditions. The deviation value, the trend of change in the time dimension, and the corresponding operating conditions are combined to form a composite deviation feature set.

[0128] S6.2: Based on the composite bias feature set, a hierarchical calibration strategy with attention mechanism weighting is adopted to perform collaborative calibration on the digital twin model and generate calibration results;

[0129] Specifically, the deviation values, time-dimensional trends, and corresponding operating conditions in the composite deviation feature set are sent to different calibration levels. In each level, the attention mechanism assigns different weights to the deviation values, time-dimensional trends, and corresponding operating conditions based on the correlation strength between the deviation values, time-dimensional trends, and corresponding operating conditions and historical calibration records, highlighting factors with a greater impact on the deviation. Each level adjusts the connection parameters of the deep neural network in the digital twin model based on the weighted deviation values, time-dimensional trends, and corresponding operating conditions. The adjustment results from multiple levels are simultaneously applied to the digital twin model to form the calibration effect.

[0130] It should be noted that the hierarchical calibration strategy comprises three levels. The first level adjusts the weights of the last fully connected layer of the deep neural network based on the deviation value itself. The second level adjusts the recurrent connection parameters of the LSTM hidden layer based on the changing trend over time. The third level adjusts the mapping parameters between the input layer and the first LSTM layer based on the corresponding operating conditions. Each level uses Huber loss as the optimization basis during calibration.

[0131] S6.3: Evaluate the calibration effect through a meta-learning framework, and optimize the hierarchical calibration strategy based on the evaluation results to complete the parameter update of the digital twin model.

[0132] Specifically, the calibrated digital twin model is applied to multiple historical hydrofluoric acid replenishment operation scenarios to obtain the deviation between the predicted hydrofluoric acid concentration and the corresponding new hydrofluoric acid concentration in each scenario. The meta-learning framework judges whether the calibration effect improves the prediction consistency based on the deviation performance. If the calibration effect is good, the current hierarchical calibration strategy structure is maintained. If the calibration effect is poor, the main feature types that cause the deviation are identified, and the weight allocation method or calibration response intensity of the corresponding features in the attention mechanism-weighted hierarchical calibration strategy is adjusted. The adjusted hierarchical calibration strategy is applied to the digital twin model to update the connection parameters of the deep neural network, forming a new round of digital twin model state and completing the parameter update of the digital twin model.

[0133] It should be noted that the meta-learning framework employs a model-agnostic meta-learning (MAML) algorithm. The task is divided into two loops: an inner loop for performing hierarchical calibration on a single task (e.g., with a learning rate of 0.0005 and 3 iterations); and an outer loop for updating the initialization parameters of the hierarchical calibration strategy across tasks (e.g., with a learning rate of 0.001 and 8 historical tasks sampled per round). The pre-training process utilizes multiple batches of acid replenishment operation records from a historical high-purity quartz product pickling solution parameter dataset. It extracts the corresponding composite deviation feature set for each batch, the deviation changes between the predicted and new hydrofluoric acid concentrations before and after calibration, and uses these as training samples. This allows the meta-learning framework to learn to judge the calibration effect based on task characteristics and generate effective strategy adjustment instructions.

[0134] In summary, this invention achieves precise and optimized control of pickling solution activity compensation by constructing a digital twin model and predicting the compensation addition rate value based on the effective active concentration. The digital twin model is not a static mapping but deeply integrates real-time effective active concentration, historical dynamics, and physical laws, enabling high-precision multi-step prediction of the response after acid replenishment. Based on the prediction results, the model can inversely solve for the compensation addition rate setpoint, considering not only immediate concentration deviations but also anticipating the dynamic impact of the addition behavior on future states. The model-predictive control strategy transforms traditional passive lag compensation into active adjustment, effectively avoiding overshoot or undercompensation, improving the control precision and stability of the pickling solution's active concentration, and thus ensuring the high quality and consistency of the pickling process for high-purity quartz products.

[0135] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for dynamically controlling the pickling solution of high-purity quartz products, characterized in that: include, Real-time acquisition of pickling solution parameters for high-purity quartz products; assimilation analysis of these parameters to generate a concentration-conductivity-temperature relationship mapping table. Based on the concentration-conductivity-temperature relationship mapping table, the measured values ​​of the pickling solution conductivity and the pickling solution temperature in the pickling solution parameters of high-purity quartz products are converted into the effective active concentration of hydrofluoric acid. The effective activity concentration is compared with a preset target activity concentration threshold to generate a comparison result, and the activity compensation mechanism is triggered based on the comparison result. A digital twin model is constructed, and after the activity compensation mechanism is triggered, the digital twin model predicts the compensation addition rate set value of hydrofluoric acid based on the effective activity concentration. The compensation addition rate setpoint is converted into a control signal for the acid replenishment pump to execute the acid replenishment operation. Simultaneously, based on the compensation addition rate setpoint, a digital twin model predicts the hydrofluoric acid concentration after the acid replenishment operation. The steps are as follows. The effective active concentration of hydrofluoric acid after the acid replenishment operation was collected, and a fusion feature set was constructed by combining the compensation addition rate set value and historical high-purity quartz product pickling solution parameters. The fused feature set is input into the digital twin model to perform multi-step concentration prediction and generate an initial concentration prediction sequence. Uncertainty quantification and calibration are performed on the initial concentration prediction sequence, and the predicted value of hydrofluoric acid concentration after acid replenishment is output. The new hydrofluoric acid concentration is obtained and compared with the predicted hydrofluoric acid concentration to generate a deviation value. The parameters of the digital twin model are then calibrated and updated based on the deviation value. The steps are as follows. The actual measured value of hydrofluoric acid concentration after the acid replenishment operation is obtained as the new hydrofluoric acid concentration. The new hydrofluoric acid concentration is compared with the predicted value of hydrofluoric acid concentration to generate a composite deviation feature set. Based on the composite bias feature set, a hierarchical calibration strategy with attention mechanism weighting is adopted to perform collaborative calibration on the digital twin model and generate calibration results; The calibration effect is evaluated using a meta-learning framework, and the hierarchical calibration strategy is optimized based on the evaluation results to complete the parameter update of the digital twin model.

2. The method for dynamic control of pickling solution for high-purity quartz products as described in claim 1, characterized in that: The parameters of the pickling solution for high-purity quartz products include hydrofluoric acid concentration, pickling solution temperature, and measured values ​​of pickling solution conductivity.

3. The method for dynamic control of pickling solution for high-purity quartz products as described in claim 1, characterized in that: The steps for generating the concentration-conductivity-temperature relationship mapping table are as follows: A multi-channel parallel processing architecture was constructed to independently filter the measured values ​​of hydrofluoric acid concentration, pickling solution temperature, and pickling solution conductivity to obtain smooth sequences of each parameter. The smoothed sequence of each parameter is input into a dynamically coupled extended Kalman filter assimilation algorithm for assimilation, and the assimilated pickling solution state data point set is output. Based on the pickling solution state data point set, a three-dimensional surface fitting was performed using a hybrid radial basis function interpolation algorithm to generate a concentration-conductivity-temperature relationship mapping table.

4. The method for dynamic control of pickling solution for high-purity quartz products as described in claim 3, characterized in that: The steps for converting the measured conductivity and temperature of the pickling solution in the parameters of the high-purity quartz product pickling solution into the effective active concentration of hydrofluoric acid are as follows: The filtered conductivity and temperature of the pickling solution are extracted from the smoothed sequence of each parameter. The concentration-conductivity-temperature relationship mapping table is queried by fuzzy inference algorithm to select the reference data point set. Based on the reference data point set, the initial value of the effective active concentration of hydrofluoric acid was obtained by using the active concentration calculation model. Real-time pH monitoring values ​​were collected as auxiliary verification to correct the initial value of the effective active concentration and generate the effective active concentration.

5. The method for dynamic control of pickling solution for high-purity quartz products as described in claim 1, characterized in that: The steps for triggering the activity compensation mechanism are as follows: The effective activity concentration is compared with a preset target activity concentration threshold to generate a comparison result. When the comparison result shows that the effective activity concentration exceeds the preset target activity concentration threshold, the activity compensation mechanism is not triggered. When the comparison result shows that the effective activity concentration does not exceed the preset target activity concentration threshold, the activity compensation mechanism is triggered.

6. The method for dynamic control of pickling solution for high-purity quartz products as described in claim 1, characterized in that: The steps for constructing a digital twin model are as follows: Collect historical high-purity quartz product pickling solution parameter datasets, construct a deep neural network architecture, learn the dynamic relationship between historical high-purity quartz product pickling solution parameters, and obtain the deep neural network; A digital twin model is constructed by combining deep neural networks with the physical equations governing mass conservation in the pickling process.

7. The method for dynamic control of pickling solution for high-purity quartz products as described in claim 1, characterized in that: The steps for predicting the setpoint for the compensated addition rate of hydrofluoric acid are as follows: After the activity compensation mechanism is triggered, the effective activity concentration is input into the digital twin model for multi-step state deduction to generate the future state trajectory. Based on the future state trajectory, calculate the compensation addition rate value required for each prediction step and combine them to form a sequence of compensation addition rate values; The compensation addition rate value at the first time position in the compensation addition rate value sequence is selected as the hydrofluoric acid compensation addition rate setpoint.

8. The method for dynamic control of pickling solution for high-purity quartz products as described in claim 1, characterized in that: The steps for performing the acid replenishment operation are as follows: The compensation addition rate setting value is verified and tamper-proofed, and a security control command is generated. The model predictive control rolling optimization algorithm optimizes the duty cycle and frequency of safety control commands in real time, and converts them into pulse width modulation signals to drive the acid replenishment pump. The acid replenishment pump is driven by a pulse width modulation signal, and the pump's operation status is monitored in real time.

Images