A high-efficiency low-damage grinding processing method for quartz glass
By using precision CNC grinding equipment and a meta-learning framework to optimize grinding parameters in quartz glass grinding, the problem of balancing high efficiency and low damage in quartz glass grinding was solved, achieving multi-objective optimization of high efficiency and low damage, and improving processing efficiency and quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NORTHWESTERN POLYTECHNICAL UNIV
- Filing Date
- 2026-06-02
- Publication Date
- 2026-07-21
AI Technical Summary
Existing quartz glass grinding processes suffer from the difficulty of balancing high material removal efficiency with low damage quality. Furthermore, existing optimization methods lack accurate prediction and direct control over the material removal process, resulting in low efficiency, poor generalization ability, and high experimental costs in multi-objective optimization.
Grinding experiments were conducted using precision CNC grinding equipment and diamond grinding wheels of different grit sizes. Combined with observations using scanning electron microscopy and optical microscopy, an experimental database was constructed, a regression model was established and embedded into a meta-learning framework, and grinding parameters were optimized using a non-dominated sorting genetic algorithm to achieve plastic flow removal and low-damage processing.
It enables efficient and low-damage processing of quartz glass components, significantly improves surface roughness and subsurface damage depth, enhances mechanical strength and optical performance, reduces processing costs and experimental cycles, and is suitable for different grinding wheels and working conditions.
Smart Images
Figure CN122425582A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of quartz processing technology, specifically to a method for efficient and low-damage grinding of quartz glass. Background Technology
[0002] Quartz glass, as a high-purity, amorphous, brittle optical material, possesses excellent optical transmittance, a low coefficient of thermal expansion, and high chemical stability. It is widely used in high-end laser systems, semiconductor masks, and aerospace optical components, placing extremely high demands on surface roughness, surface accuracy, and subsurface damage depth. However, quartz glass has high hardness and low fracture toughness. In traditional grinding processes, material removal is primarily achieved through brittle fracture, easily generating deep surface and subsurface damage layers, severely reducing the mechanical strength, optical performance, and laser damage resistance of components. To achieve low-damage processing, existing technologies often employ reduced grinding depth, finer-grained grinding wheels, or auxiliary methods. While these can promote plastic flow removal to some extent, a significant contradiction exists between achieving high material removal efficiency and low-damage quality. Meanwhile, existing optimization methods mainly rely on empirical trial and error, response surface methodology, or traditional machine learning, optimizing only a single or few indicators such as surface roughness or material removal rate. They lack accurate prediction and direct control over the material removal process, resulting in low efficiency, poor generalization ability, and high experimental costs for multi-objective optimization, failing to achieve efficient and low-damage integrated processing dominated by plastic flow. To address this, we propose a high-efficiency, low-damage grinding method for quartz glass. Summary of the Invention
[0003] To address the aforementioned technical problems, this paper provides a method for efficient and low-damage grinding of quartz glass. This technical solution solves the problems of low efficiency, poor generalization ability, and high experimental cost associated with multi-objective optimization.
[0004] To achieve the above objectives, the technical solution adopted by this invention is: a method for efficient and low-damage grinding of quartz glass, the specific processing steps of which are as follows: S1. Grinding Experiment Data Acquisition: Precision CNC grinding equipment and diamond grinding wheels of different grit sizes were used to perform surface grinding on quartz glass samples under multiple grinding parameters. The surface morphology was observed by scanning electron microscopy and optical microscopy to distinguish brittle-plastic removal behavior. The surface roughness Ra was measured by white light interferometer. The subsurface damage depth was detected by chemical etching combined with cross-sectional polishing. The removal form was manually calibrated and the removal rate was recorded to construct an experimental database.
[0005] S2. Model Construction: Using grinding parameters and wheel grit size as input features, a regression model is established to predict surface roughness Ra and subsurface damage depth SSD; the regression model is embedded into the meta-learning framework as a physical constraint; manually calibrated material removal forms are used as training data for the meta-task, and a physical information-enhanced meta-learning algorithm is used for training; the material removal forms are predicted under arbitrary grinding parameters and wheel grit size. S3. Multi-objective optimization: With the goals of plastic flow removal, high efficiency and low damage, a non-dominated sorting genetic algorithm is used to globally optimize the parameters under the prediction model to obtain the Pareto optimal solution set, and the optimal combination of process parameters is selected by combining engineering constraints. S4. Processing Verification: Grind the quartz glass element using the optimal combination of process parameters, repeat the S1 testing method, and verify whether the surface roughness Ra, subsurface damage depth SSD, and removal method meet the preset targets.
[0006] Preferably, the grinding experiment in step S1 specifically includes: Prepare and pre-treat standard planar quartz glass samples, adjust precision CNC grinding equipment and diamond grinding wheels of different grit sizes, and calibrate testing equipment; The pretreated sample is clamped and positioned, and surface grinding is performed sequentially according to multiple preset grinding parameters, while the equipment operation status is recorded in real time. Weigh the sample before and after grinding, calculate and record the material removal rate; observe the sample surface morphology using an optical microscope and a scanning electron microscope, manually calibrate and record the brittle-plastic removal behavior and distribution; measure and record the sample surface roughness Ra using a white light interferometer; perform cross-sectional polishing and chemical etching on the sample, observe and measure the subsurface damage depth and record the results. Organize all raw experimental data and construct a traceable and queryable experimental database.
[0007] Preferably, the multiple grinding parameters mentioned in step S1 include wheel grit size, spindle speed, table feed rate, grinding depth, and grinding method; the calibration standard for the brittle-plastic removal behavior is: By observing the microstructure using a scanning electron microscope, areas with cracks, chipping, and fracture pits were identified as brittle removal, while areas with cutting grooves, plastic flow, and no obvious cracks were identified as plastic removal. The ratio and distribution of the two were manually determined. The surface roughness Ra is measured by using a white light interferometer to select multiple representative areas on the sample surface for measurement, and the average value of multiple measurements is taken as the final Ra data. The experimental database includes experimental condition parameters, grinding parameters, grinding wheel grit size, and surface cross-sectional morphology image data.
[0008] Preferably, in step S2, the regression model is one of a BP neural network, a random forest regression model, or a support vector regression model; the parameters in the regression model include the number of hidden layers and neurons in the neural network and the number of decision trees in the random forest, and the model parameters are iteratively optimized using gradient descent; the meta-learning framework includes a task generation module, a meta-training module, and a meta-testing module, and the regression model is embedded as a physical constraint in the meta-training module within the meta-learning framework, wherein the physical constraint is that the prediction results in the form of material removal match the Ra and SSD prediction values output by the regression model.
[0009] Preferably, in step S2, the meta-task in the training of the meta-learning algorithm is constructed from labeled material removal data, and a physical information enhancement strategy is introduced, incorporating Ra and SSD prediction values and the brittle-plastic removal mechanism of quartz glass grinding as auxiliary constraints into the loss function. The grinding parameters to be predicted and the grinding wheel grit size are normalized and then input into the regression model. The regression model output includes the material removal form and the prediction confidence level. The material removal form includes brittle removal, plastic removal and mixed removal.
[0010] Preferably, the non-dominated sorting genetic algorithm described in step S3 has its parameters pre-set. The NSGA parameters are set in combination with the experimental data scale and optimization accuracy requirements. The parameters include: population size, number of iterations, crossover probability, and mutation probability. The objectives of plastic flow removal, high efficiency, and low damage optimization are integrated into a fitness function. A weighted summation method is used to transform the multi-objective into a calculable single-objective fitness value, and the weight allocation is adjusted according to engineering requirements.
[0011] Preferably, the specific steps for obtaining the Pareto optimal solution set through global optimization in step S3 are as follows: Using grinding parameters and grinding wheel grit size as optimization variables, consistent with the input features of the regression model, including grinding wheel grit size, spindle speed, table feed rate, and grinding depth, the value range of each variable is defined; optimization variables are obtained; and two types of constraints are set based on engineering practice and experimental feasibility. The optimization variables, constraints, and fitness function are input into the NSGA algorithm, the regression model is called, and the plasticity removal ratio, Ra, SSD, and removal rate output by the model are used as the optimization criteria to start global optimization. Each generation of the population generates a new population through crossover and mutation operations. The optimization target value corresponding to each individual is calculated using a regression model. Non-dominated solutions are selected through non-dominated sorting. The process is iterated step by step until the convergence condition is met. After the iteration is completed, all non-dominated solutions are collected to form the Pareto optimal solution set, which contains multiple sets of parameter combinations, each of which can achieve a cooperative balance of the objective optimization.
[0012] Preferably, in step S3, the optimal process parameter combination is selected by combining engineering constraints. From the Pareto optimal solution set, solutions that do not meet the requirements of equipment capacity and production efficiency are eliminated by combining hard and soft constraints. Priority is given to selecting parameter combinations with high plasticity removal ratio, low Ra and SSD and moderate removal rate to form 3-5 sets of candidate optimal parameters. The candidate optimal parameter combination was applied to an actual quartz glass grinding experiment. Experimental data was collected and compared with the predicted values of the regression model to verify the feasibility and optimization effect of the parameter combination. The parameter combination with the best overall performance in the experimental verification was selected as the final optimal process parameters for quartz glass grinding and archived in the experimental database.
[0013] Preferably, in step S4, before processing verification, a quartz glass element with the same specifications and initial state as the previous grinding experiment is selected; the optimal combination of process parameters obtained from multi-objective optimization is extracted and organized into a parameter comparison table, so that the equipment operation and detection accuracy are consistent with the previous S1 detection standard.
[0014] Preferably, in step S4, the detected plastic removal ratio, surface roughness Ra, subsurface damage depth SSD and removal rate are compared with the preset verification indicators one by one to determine whether they all meet the standards; if they all meet the standards, the optimal process parameter combination is confirmed to have passed the verification; if they do not all meet the standards, the reasons for non-compliance are analyzed and targeted adjustments are made, and steps S2 to S4 are repeated until all indicators meet the standards.
[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention effectively suppresses the generation and propagation of brittle cracks during grinding by controlling the material removal process to be dominated by plastic flow. The subsurface damage depth is reduced, and the surface roughness Ra is significantly improved, greatly enhancing the mechanical strength, optical transmittance, and laser damage resistance threshold of quartz glass components. It overcomes the contradiction of "difficulty in achieving both high efficiency and low damage" in traditional methods. Under the same or higher material removal rate conditions, it can still obtain surface and subsurface quality far superior to existing processes, and the processing efficiency and quality are improved simultaneously.
[0016] This invention employs a physical information-enhanced meta-learning model, which requires only a small amount of experimental data to accurately predict the material removal form under different grinding parameters and grinding wheel grit sizes. This avoids the high cost and long cycle caused by traditional trial-and-error or a large number of repeated experiments. The optimization process is highly intelligent and has strong generalization ability, making it applicable to different grinding wheels and working conditions. With plastic flow removal, high efficiency, and low damage as joint optimization objectives, the optimal combination of process parameters can achieve Pareto optimality in multiple dimensions such as plastic flow probability, surface roughness, subsurface damage depth, and material removal rate, solving the problem that existing single-objective or local optimization is prone to getting trapped in suboptimal solutions.
[0017] This invention requires no major modifications to existing precision CNC grinding machines; it can be implemented simply by optimizing grinding parameters and selecting grinding wheel grit size. It is low-cost, easy to operate, and has good repeatability, making it suitable for large-scale application in the manufacturing of high-end quartz glass optical components. Attached Figure Description
[0018] Figure 1 This is a flowchart of the steps of the present invention. Detailed Implementation
[0019] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.
[0020] This embodiment discloses a high-efficiency, low-damage grinding method for quartz glass, which solves the technical problems of high surface roughness, severe subsurface damage, low processing efficiency, and difficulty in accurately controlling the removal of brittle plasticity in the existing quartz glass grinding process. The specific implementation process is as follows: all operations are carried out in a constant temperature and humidity laboratory (temperature 23±1℃, humidity 50±5%) to avoid the influence of environmental factors on the processing and testing results.
[0021] Example 1 S1. Grinding Experiment Data Acquisition A standard planar quartz glass sample with dimensions of 50mm × 50mm × 5mm was prepared. It was ultrasonically cleaned with anhydrous ethanol for 15 minutes to remove surface oil and impurities, and then dried for later use. A precision CNC surface grinding machine (model: MGA8425) was selected, equipped with diamond grinding wheels (grit sizes #120, #240, #400, and #800, with resin binder; wheel diameter Φ150mm, width 10mm). The machine tool spindle runout was adjusted to ≤0.002mm, and the table flatness to ≤0.001mm / 100mm. Testing equipment was calibrated, including a scanning electron microscope (model: SU8010) with a magnification range of 50-10000x, an optical microscope (model: Olympus BX53) with a magnification range of 50-1000x, and a white light interferometer (model: Zygo NewView). The 9000 has a measurement range of 0.1nm-100μm and a measurement accuracy of ±0.01nm, ensuring accurate and reliable detection data.
[0022] The pretreated quartz glass sample was clamped and positioned using a vacuum chuck. After clamping, the surface runout of the sample was ≤0.003mm. Multiple sets of grinding parameters were set, with the following specific parameter ranges: grinding wheel grit size: #120, #240, #400, #800; spindle speed: 10000r / min, 12000r / min, 14000r / min, 16000r / min; table feed speed: 50mm / min, 100mm / min, 150mm / min, 200mm / min; grinding depth: 0.01mm, 0.02mm, 0.03mm, 0.04mm; grinding method: dry grinding. Each set of parameters was repeated 3 times. The machine tool operating current, grinding sound, and other status parameters were recorded in real time to avoid abnormal working conditions affecting the experimental results.
[0023] The mass of the samples before and after grinding was measured using an electronic balance (accuracy 0.1 mg). The material removal rate was calculated according to the formula MRR=Δm / (ρ×A×t) (where Δm is the mass difference before and after grinding, ρ is the density of quartz glass 2.2 g / cm³, A is the grinding area, and t is the grinding time). The removal rate corresponding to each set of parameters was recorded. The surface morphology of the samples was observed using an optical microscope and a scanning electron microscope. The brittle-plastic removal behavior and distribution pattern were manually calibrated. The calibration criteria were: areas with cracks, chipping, and fracture pits observed under a scanning electron microscope were judged as brittle removal; areas with cutting grooves, plastic flow, and no obvious defects were judged as brittle removal. The areas showing cracks were identified as areas of plastic removal, and the ratio and distribution of the two were recorded. Five representative areas (each area 50μm×50μm) on the sample surface were selected using a white light interferometer for measurement, and the average value of the five measurements was taken as the final surface roughness Ra data. The samples were then subjected to cross-sectional polishing (polishing accuracy ≤0.001mm) and chemical etching (etching solution was a mixture of hydrofluoric acid and nitric acid in a volume ratio of 3:1, etching time 10s). The subsurface damage depth (SSD) was observed and measured using a scanning electron microscope. Three cross-sections were measured for each sample, and the average value was taken as the final SSD data.
[0024] All raw experimental data were collected, including experimental condition parameters (constant temperature and humidity environment parameters, equipment model), grinding parameters (grind wheel grit size, spindle speed, table feed rate, grinding depth, grinding method), grinding wheel grit size parameters, surface and cross-sectional morphology image data, removal rate, Ra, SSD and brittle-plastic removal form calibration results, and a traceable and searchable experimental database was constructed.
[0025] S2, Model Building 80% of the experimental data in the experimental database was selected as the training set, and 20% as the test set. Grinding parameters (spindle speed, table feed rate, grinding depth) and wheel grit size were used as input features, while surface roughness Ra and subsurface damage depth SSD were used as output features. A backpropagation (BP) neural network regression model was established. This BP neural network consists of one input layer (4 neurons, corresponding to the 4 input features), two hidden layers (12 neurons in the first hidden layer and 8 neurons in the second hidden layer), and one output layer (2 neurons, corresponding to Ra and SSD). Gradient descent was used to iteratively optimize the model parameters. The learning rate was set to 0.01, the number of iterations was 1000, and the iteration termination condition was a loss function value ≤ 0.001. After optimization, the Ra prediction error on the test set was ≤ 3%, and the SSD prediction error was ≤ 4%, meeting the prediction accuracy requirements.
[0026] A meta-learning framework comprising a task generation module, a meta-training module, and a meta-testing module was constructed. The aforementioned BP neural network regression model was embedded as a physical constraint into the meta-training module of the meta-learning framework. The physical constraint was that the predicted results of the material removal form matched the Ra and SSD predicted values output by the regression model (i.e., when the proportion of plastic removal is high, Ra and SSD should be at a low level, and when the proportion of brittle removal is high, Ra and SSD should be at a high level). The manually calibrated material removal forms (brittle removal, plastic removal, and mixed removal) in the experimental database were used as training data for the meta-tasks. Each meta-task contained 10 sets of input and output samples. A physical information enhancement strategy was introduced, and the Ra and SSD predicted values and the brittle-plastic removal mechanism of quartz glass grinding (the greater the grinding force and the coarser the grinding wheel grit, the easier it is to produce brittle removal) were incorporated into the loss function as auxiliary constraints. The physical information enhancement meta-learning algorithm was used for training, with 500 training iterations. After training, the prediction accuracy of the material removal form of the model was ≥92%.
[0027] The grinding parameters and wheel grit size to be predicted are normalized (normalized to the [0,1] interval) and then input into the regression model. The model output includes the material removal form and the prediction confidence. When the prediction confidence is ≥85%, the prediction result is considered valid and can be used for subsequent multi-objective optimization.
[0028] S3, Multi-objective optimization The optimization objectives were plastic flow removal, high efficiency, and low damage. Specifically, the optimization objectives were defined as follows: plastic removal rate ≥ 90%, surface roughness Ra ≤ 20 nm, subsurface damage depth SSD ≤ 2 μm, and material removal rate ≥ 2.0 mm³ / min. A non-dominated sorting genetic algorithm (NSGA-Ⅲ) was used. Algorithm parameters were pre-set, and based on the experimental data scale and optimization accuracy requirements, the parameters were set as follows: population size 100, iterations 200, crossover probability 0.8, and mutation probability 0.05. Grinding parameters (spindle speed, table feed rate, and grinding depth) and grinding wheel grit size were used as optimization variables, with the value range of each variable consistent with the experimental parameters in S1. Considering engineering practice and experimental feasibility, two types of constraints were set: hard constraints (spindle speed ≤ 16000 r / min, table feed rate ≥ 50 mm / min, grinding depth ≤ 0.04 mm) and soft constraints (equipment operating current ≤ 10 A, and no significant vibration during grinding).
[0029] The objectives of plastic flow removal, high efficiency, and low damage optimization are integrated into a fitness function. A weighted summation method is used to transform the multi-objective into a calculable single-objective fitness value. The weight allocation is adjusted according to engineering requirements. In this embodiment, the weight allocation is: plastic flow removal percentage (0.4), Ra (0.25), SSD (0.25), and removal rate (0.1). The fitness function expression is: F = 0.4 × (plastic removal percentage / 100) + 0.25 × (20-Ra) / 20 + 0.25 × (2-SSD) / 2 + 0.1 × (removal rate / 2.78). The higher the fitness value, the better the parameter combination.
[0030] The optimization variables, constraints, and fitness function are input into the NSGA-III algorithm. A regression model is invoked, and the plasticity removal ratio, Ra, SSD, and removal rate output by the model are used as the optimization criteria to initiate global optimization. Each generation of the population generates a new population through crossover and mutation operations. The optimization objective value for each individual is calculated using the regression model. Non-dominated solutions are selected through non-dominated sorting, and the process iterates until the convergence condition is met (the change in the optimal fitness value ≤ 0.001 over 10 consecutive generations). After iteration, all non-dominated solutions are collected to form a Pareto optimal solution set containing 8 parameter combinations. Combining hard and soft constraints, solutions that do not meet the equipment capacity and production efficiency requirements are eliminated. Parameter combinations with a high plasticity removal ratio, low Ra and SSD, and a moderate removal rate are prioritized, forming 3 candidate optimal parameter sets. Three sets of candidate optimal parameter combinations were applied to actual quartz glass grinding experiments. Experimental data were collected and compared with the predicted values of the regression model to verify the feasibility and optimization effect of the parameter combinations. The experimental results showed that the actual detection value of candidate parameter 1 had the smallest deviation from the predicted value and the best overall performance. It was determined to be the final optimal process parameter for quartz glass grinding, specifically: grinding wheel grit size #800, spindle speed 15000 r / min, table feed speed 180 mm / min, and grinding depth 0.03 mm. This parameter combination was archived in the experimental database.
[0031] S4, Processing Verification Quartz glass elements (50mm×50mm×5mm in size, ultrasonically cleaned with anhydrous ethanol) consistent with the specifications and initial state of the previous grinding experiments were selected. The optimal combination of process parameters obtained from multi-objective optimization was extracted and compiled into a parameter comparison table. The precision CNC surface grinding machine, diamond grinding wheel and testing equipment were debugged to ensure that the operation and testing accuracy of the equipment are consistent with the previous S1 testing standard, that is, the spindle runout is ≤0.002mm and the white light interferometer measurement accuracy is ±0.01nm.
[0032] The quartz glass element was ground using the optimal combination of process parameters. The detection method in S1 was repeated, and the processed element was tested. The test results are as follows: plastic removal rate 92.3%, surface roughness Ra 17.6 nm, subsurface damage depth SSD 1.38 μm, and material removal rate 2.58 mm³ / min. The preset verification indicators are: plastic removal rate ≥ 90%, Ra ≤ 20 nm, SSD ≤ 2 μm, and removal rate ≥ 2.0 mm³ / min. The test results were compared with the preset verification indicators one by one. All indicators met the preset requirements, and the optimal combination of process parameters was verified as successful.
[0033] Example 2 The difference between this embodiment and embodiment 1 is that: in S2, the regression model adopts the random forest regression model, and in S3, the weight allocation of the optimization objective is adjusted to: plasticity removal ratio (0.35), Ra (0.3), SSD (0.25), and removal rate (0.1). The remaining steps and parameter settings are the same as in embodiment 1.
[0034] Experimental verification revealed the optimal combination of process parameters to be: grinding wheel grit #800, spindle speed 15500 r / min, table feed rate 175 mm / min, and grinding depth 0.03 mm. The processing verification results were as follows: plasticity removal rate 91.8%, surface roughness Ra 17.9 nm, subsurface damage depth SSD 1.42 μm, and material removal rate 2.52 mm³ / min. All indicators met the preset verification requirements, and the verification was successful.
[0035] Example 3 The difference between this embodiment and embodiment 1 is that: wet grinding is used in S1, and the NSGA-Ⅲ algorithm parameters in S3 are adjusted to: population size 120, number of iterations 250, crossover probability 0.75, mutation probability 0.06. The remaining steps and parameter settings are the same as in embodiment 1.
[0036] Experimental verification revealed the optimal combination of process parameters to be: grinding wheel grit #800, spindle speed 14500 r / min, table feed rate 185 mm / min, and grinding depth 0.03 mm. The processing verification results showed: plasticity removal rate 93.1%, surface roughness Ra 16.8 nm, subsurface damage depth SSD 1.32 μm, and material removal rate 2.61 mm³ / min. All indicators met the preset verification requirements, and the surface quality was further improved compared to Example 1, indicating that wet grinding can effectively improve the grinding effect.
[0037] Comparative experiment A conventional grinding method (without model prediction and parameter optimization, using a grinding wheel grit size of #400, a spindle speed of 12000 r / min, a table feed rate of 100 mm / min, and a grinding depth of 0.02 mm) was used to grind quartz glass samples of the same specification. The test results were as follows: plastic removal rate of 68%, surface roughness Ra of 52.3 nm, subsurface damage depth SSD of 4.9 μm, and material removal rate of 1.76 mm³ / min.
[0038] Compared with the processing results of Example 1 of the present invention, the quartz glass components processed by the method of the present invention have a 35.7% higher plasticity removal rate, a 66.3% lower surface roughness Ra, a 71.8% lower subsurface damage depth SSD, and a 46.6% higher material removal rate. This significantly achieves efficient and low-damage grinding of quartz glass, solves the technical defects of traditional grinding methods, and has good engineering application value.
[0039] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.
Claims
1. A method for efficient and low-damage grinding of quartz glass, characterized in that, The specific processing steps are as follows: S1. Grinding Experiment Data Acquisition: Precision CNC grinding equipment and diamond grinding wheels of different grit sizes were used to perform surface grinding on quartz glass samples under multiple sets of grinding parameters; the surface morphology was observed by scanning electron microscope and optical microscope to distinguish brittle-plastic removal behavior; the surface roughness Ra was measured by white light interferometer; the subsurface damage depth was detected by chemical etching combined with cross-sectional polishing; the removal form was manually calibrated and the removal rate was recorded; and an experimental database was constructed. S2. Model Construction: Using grinding parameters and wheel grit size as input features, a regression model is established to predict surface roughness Ra and subsurface damage depth SSD; the regression model is embedded into the meta-learning framework as a physical constraint; manually calibrated material removal forms are used as training data for the meta-task, and a physical information-enhanced meta-learning algorithm is used for training; the material removal forms are predicted under arbitrary grinding parameters and wheel grit size. S3. Multi-objective optimization: With the goals of plastic flow removal, high efficiency and low damage, a non-dominated sorting genetic algorithm is used to globally optimize the parameters under the prediction model to obtain the Pareto optimal solution set, and the optimal combination of process parameters is selected by combining engineering constraints. S4. Processing Verification: Grind the quartz glass element using the optimal combination of process parameters, repeat the S1 testing method, and verify whether the surface roughness Ra, subsurface damage depth SSD, and removal method meet the preset targets.
2. The method for efficient and low-damage grinding of quartz glass according to claim 1, characterized in that, The grinding experiment described in step S1 specifically includes: Prepare and pre-treat standard planar quartz glass samples, adjust precision CNC grinding equipment and diamond grinding wheels of different grit sizes, and calibrate testing equipment; The pretreated sample is clamped and positioned, and surface grinding is performed sequentially according to multiple preset grinding parameters, while the equipment operation status is recorded in real time. Weigh the sample before and after grinding, calculate and record the material removal rate; observe the sample surface morphology using an optical microscope and a scanning electron microscope, manually calibrate and record the brittle-plastic removal behavior and distribution; measure and record the sample surface roughness Ra using a white light interferometer; perform cross-sectional polishing and chemical etching on the sample, observe and measure the subsurface damage depth and record the results. Organize all raw experimental data and construct a traceable and queryable experimental database.
3. The method for efficient and low-damage grinding of quartz glass according to claim 1, characterized in that: The multiple grinding parameters mentioned in step S1 include wheel grit size, spindle speed, table feed rate, depth of grinding, and grinding method; the calibration standard for the brittle-plastic removal behavior is: By observing the microstructure using a scanning electron microscope, areas with cracks, chipping, and fracture pits were identified as brittle removal, while areas with cutting grooves, plastic flow, and no obvious cracks were identified as plastic removal. The ratio and distribution of the two were manually determined. The surface roughness Ra is measured by using a white light interferometer to select multiple representative areas on the sample surface for measurement, and the average value of multiple measurements is taken as the final Ra data. The experimental database includes experimental condition parameters, grinding parameters, grinding wheel grit size, and surface cross-sectional morphology image data.
4. The method for efficient and low-damage grinding of quartz glass according to claim 1, characterized in that: In step S2, the regression model is one of the following: a backpropagation neural network, a random forest regression model, or a support vector regression model. The parameters of the regression model include the number of hidden layers and neurons in the neural network and the number of decision trees in the random forest. The gradient descent method is used to iteratively optimize the model parameters. The meta-learning framework includes a task generation module, a meta-training module, and a meta-testing module. The regression model is embedded as a physical constraint in the meta-training module of the meta-learning framework. The physical constraint is that the prediction results in the form of material removal match the Ra and SSD prediction values output by the regression model.
5. The method for efficient and low-damage grinding of quartz glass according to claim 1, characterized in that: In step S2, the meta-learning algorithm training meta-task is constructed from labeled material removal data. A physical information enhancement strategy is introduced, and Ra and SSD prediction values and the brittle-plastic removal mechanism of quartz glass grinding are incorporated into the loss function as auxiliary constraints. The grinding parameters to be predicted and the grinding wheel grit size are normalized and then input into the regression model. The regression model output includes the material removal form and the prediction confidence level. The material removal form includes brittle removal, plastic removal and mixed removal.
6. The method for efficient and low-damage grinding of quartz glass according to claim 1, characterized in that: The non-dominated sorting genetic algorithm described in step S3 pre-sets the parameters, setting the NSGA parameters. The parameter settings are combined with the experimental data scale and optimization accuracy requirements. The parameters include: population size, number of iterations, crossover probability, and mutation probability. The objectives of plastic flow removal, high efficiency, and low damage optimization are integrated into a fitness function. A weighted summation method is used to transform the multi-objective into a calculable single-objective fitness value, and the weight allocation is adjusted according to engineering requirements.
7. The method for efficient and low-damage grinding of quartz glass according to claim 1, characterized in that, The specific steps for obtaining the Pareto optimal solution set through global optimization in step S3 are as follows: Using grinding parameters and grinding wheel grit size as optimization variables, consistent with the input features of the regression model, including grinding wheel grit size, spindle speed, table feed rate, and grinding depth, the value range of each variable is defined; optimization variables are obtained; and two types of constraints are set based on engineering practice and experimental feasibility. The optimization variables, constraints, and fitness function are input into the NSGA algorithm, the regression model is called, and the plasticity removal ratio, Ra, SSD, and removal rate output by the model are used as the optimization criteria to start global optimization. Each generation of the population generates a new population through crossover and mutation operations. The optimization target value corresponding to each individual is calculated using a regression model. Non-dominated solutions are selected through non-dominated sorting. The process is iterated step by step until the convergence condition is met. After the iteration is completed, all non-dominated solutions are collected to form the Pareto optimal solution set, which contains multiple sets of parameter combinations, each of which can achieve a cooperative balance of the objective optimization.
8. The method for efficient and low-damage grinding of quartz glass according to claim 1, characterized in that: In step S3, the optimal process parameter combination is selected by combining engineering constraints. From the Pareto optimal solution set, solutions that do not meet the requirements of equipment capacity and production efficiency are eliminated by combining hard and soft constraints. Priority is given to selecting parameter combinations with high plasticity removal ratio, low Ra and SSD and moderate removal rate to form 3-5 sets of candidate optimal parameters. The candidate optimal parameter combination was applied to an actual quartz glass grinding experiment. Experimental data was collected and compared with the predicted values of the regression model to verify the feasibility and optimization effect of the parameter combination. The parameter combination with the best overall performance in the experimental verification was selected as the final optimal process parameters for quartz glass grinding and archived in the experimental database.
9. The method for efficient and low-damage grinding of quartz glass according to claim 1, characterized in that: In step S4, before processing verification, select quartz glass elements that are consistent with the specifications and initial state of the previous grinding experiment; extract the optimal combination of process parameters obtained from multi-objective optimization and organize them into a parameter comparison table to ensure that the equipment operation and testing accuracy are consistent with the previous S1 testing standards.
10. The method for efficient and low-damage grinding of quartz glass according to claim 1, characterized in that: In step S4, the detected plastic removal ratio, surface roughness Ra, subsurface damage depth SSD, and removal rate are compared with the preset verification indicators one by one to determine whether they all meet the standards. If they all meet the standards, the optimal process parameter combination is confirmed to have passed the verification. If they do not all meet the standards, the reasons for non-compliance are analyzed and targeted adjustments are made. Steps S2 to S4 are repeated until all indicators meet the standards.