A smart control method and system for a silicon carbide micro powder grinding mill

Abstract

This invention relates to the field of grinding mill control technology, and discloses an intelligent control method and system for a silicon carbide micro powder grinding mill. The method includes collecting data such as load distribution, fluctuation frequency, equipment speed, and grinding wheel wear; generating a load feature vector through feature extraction; inputting a load trend prediction model and outputting a load trend vector; calculating the cooling water flow rate and lubricant viscosity consumption rate based on this vector; weighting and combining pipeline pressure and liquid temperature parameters to generate real-time flow and viscosity values; generating an initial water replenishment command based on load distribution, nozzle angle, and spray coverage; iteratively optimizing the nozzle angle and pipeline pressure to form the final water replenishment command; driving the supply system to dynamically adjust liquid distribution and outputting status data; and optimizing the command based on load distribution and fluctuation frequency. This method improves processing accuracy and resource utilization efficiency.

Description

Technical Field

[0001] This invention relates to the field of grinding mill control technology, and in particular to an intelligent control method and system for a silicon carbide micro powder grinding mill. Background Technology

[0002] Currently, silicon carbide micron powder grinding technology is a core process in high-end manufacturing industries (such as semiconductors, photovoltaics, and precision ceramics), and the quality of its products directly affects the performance of key materials and the efficiency of end-production. Achieving efficient and stable operation of the grinding process is crucial for ensuring product precision and industrial competitiveness. With the development of intelligent manufacturing technology, intelligent and precise control of the grinding process has become a key direction for improving equipment performance. However, under the complex and ever-changing actual grinding conditions, existing control systems still face challenges in dynamically adapting to real-time load fluctuations and are unable to accurately match the real-time demands of key resources such as cooling water and lubricating fluid. Therefore, developing advanced control methods that can intelligently respond to load changes and dynamically optimize resource allocation is of great value in promoting the intelligent upgrading of silicon carbide micron powder grinding technology.

[0003] In one existing technology, based on a preset material hardness grade, the system initializes a fixed cooling water flow rate (e.g., 12 L / min) and a fixed lubricant replenishment interval (e.g., every 45 minutes). During operation, the motor load is monitored by a base current sensor. When the load value exceeds a preset threshold (e.g., 480 A), the cooling water flow rate is automatically increased in a stepwise manner (e.g., +2 L / min). Simultaneously, the lubricant replenishment pump automatically starts and runs for a fixed duration (e.g., 90 seconds) after the fixed interval timer expires. During operation, peak load and average temperature data are periodically recorded. If the average temperature exceeds 75°C for three consecutive cycles, the base cooling water flow rate setting is increased (e.g., +1 L / min) according to preset rules.

[0004] Existing technologies employ fixed parameter rules (such as preset cooling water flow rate and fixed lubricant replenishment cycle) and rely on simple threshold-triggered adjustments (such as step-wise current increase after current exceeds the limit), failing to dynamically adapt to real-time fluctuations in grinding load. Due to the lack of collaborative analysis of multi-dimensional variables such as material characteristics and equipment status, the system can only respond locally to load changes, resulting in a lag between cooling water and lubricant supply and actual demand. This static control mode exacerbates temperature fluctuations in the processing area, failing to stabilize grinding accuracy and causing waste due to excessive or insufficient resource allocation. Therefore, existing technologies limit the improvement of processing accuracy. Summary of the Invention

[0005] This invention provides an intelligent control method and system for a silicon carbide micro powder grinding mill to improve processing accuracy.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides an intelligent control method for a silicon carbide micro powder grinding mill, comprising:

[0007] Collect load distribution data, fluctuation frequency, equipment speed, abrasive wear data, material hardness parameters, processing accuracy requirements, pipeline pressure, liquid temperature, nozzle angle, and spray coverage.

[0008] The load distribution data, the fluctuation frequency, the equipment rotation speed, and the grinding wheel wear data are filtered and feature extracted to obtain a load feature vector.

[0009] The load characteristic vector, the material hardness parameter, and the processing accuracy requirement are input into a pre-built load trend prediction model, and the load trend vector is output.

[0010] The load trend vector is input into a pre-built consumption rate distribution model to calculate the cooling water flow rate consumption rate and the lubricant viscosity consumption rate.

[0011] The cooling water flow rate and lubricant viscosity are obtained by weighting the pipeline pressure and the liquid temperature, and then combining them with the cooling water flow rate and the lubricant viscosity consumption rate.

[0012] The initial water replenishment control command is generated by weighting the cooling water flow rate, the lubricant viscosity, and the load distribution data, and combining the nozzle angle and the spray coverage area.

[0013] Based on the initial water replenishment control command, the equipment rotation speed, and the wear data of the grinding wheel, the final water replenishment control command is generated by iteratively optimizing the nozzle angle and the pipeline pressure.

[0014] According to the final water replenishment control command, the grinding mill liquid supply system is driven to adjust the cooling water flow rate and lubricant viscosity distribution in real time, and output liquid supply status data including actual flow rate, viscosity and distribution status.

[0015] Based on the liquid supply status data, combined with the load distribution data and fluctuation frequency, the final water replenishment control command is optimized to obtain the optimized water replenishment control command.

[0016] In one optional implementation, the step of filtering and extracting features from the load distribution data, the fluctuation frequency, the equipment rotation speed, and the grinding wheel wear data to obtain a load feature vector includes:

[0017] Based on the load distribution data, the fluctuation frequency, the equipment rotation speed, and the grinding wheel wear data, filtered data is obtained by wavelet transform filtering.

[0018] Based on the filtered data, principal component analysis is used to extract features to obtain principal component features.

[0019] The principal component features are normalized to obtain the load feature vector.

[0020] In one optional implementation, the process of constructing the load trend prediction model includes:

[0021] Acquire historical load characteristics, historical material hardness parameters, historical processing accuracy requirements, and corresponding historical load data;

[0022] An initial load trend prediction model was constructed using the support vector machine algorithm. The classification boundary was initialized using the RBF kernel function, and a penalty coefficient was set to determine the tolerance error threshold.

[0023] Based on the historical load characteristics, the historical material hardness parameters, and the historical processing accuracy requirements, the curvature adjustment parameters of the classification boundary are calculated using the kernel density estimation method, and the curvature of the classification boundary is dynamically adjusted using the RBF kernel function.

[0024] Based on the historical load data, the penalty coefficient is optimized using a grid search method to determine the error tolerance threshold;

[0025] Cross-validate the adjusted classification boundaries and calculate the set of classification accuracy and error distribution after boundary adjustment;

[0026] When the error distribution set exceeds the error tolerance threshold, the gradient descent method is used to re-optimize the curvature parameter of the classification boundary, and the classification boundary is iteratively updated until the error distribution set does not exceed the error tolerance threshold.

[0027] The optimized classification boundary is matched with the load data in a multidimensional space by using RBF kernel function mapping to determine the hyperplane equation corresponding to each load data.

[0028] Based on the time series characteristics of the historical load data, the sliding window method is used to extract the dynamic trend parameters of load data changes. The load data is then correlated with the hyperplane equation to construct the final load trend prediction model.

[0029] In one optional implementation, the process of constructing the consumption rate distribution model includes:

[0030] The historical load trend vector, the corresponding historical cooling water flow rate and historical lubricant viscosity rate are obtained, and the historical load trend vector is cleaned and its features are extracted to obtain trend features.

[0031] The trend features are input into the input layer of the initially constructed consumption rate distribution model for training, and the predicted cooling water flow consumption rate and the predicted lubricant viscosity consumption rate are obtained from the output layer of the consumption rate distribution model.

[0032] The predicted cooling water flow rate consumption rate, the predicted lubricant viscosity consumption rate, the historical cooling water flow rate consumption rate, and the historical lubricant viscosity consumption rate are substituted into the loss function to calculate the loss value;

[0033] The gradient of the output layer of the consumption rate distribution model is calculated based on the loss value, and the gradient is passed forward layer by layer through the chain rule to calculate the gradient of the parameters of each layer and obtain the gradient data.

[0034] Based on the gradient data and the preset learning rate, the parameters of each layer of the consumption rate distribution model are updated using the gradient descent method.

[0035] The parameters of each layer are iteratively updated until the number of training iterations of the consumption rate distribution model is greater than a preset number of iterations, or the loss value of the consumption rate distribution model is less than a preset loss threshold. At this point, the training is considered complete, and a trained consumption rate distribution model is obtained.

[0036] In one optional implementation, the step of weighting the cooling water flow rate, the lubricant viscosity, and the load distribution data, and combining this with the nozzle angle and the spray coverage area to generate an initial water replenishment control command, includes:

[0037] Based on the cooling water flow rate, the lubricant viscosity, and the load distribution data, weights are allocated using the entropy weighting method to obtain weight coefficients.

[0038] The weighting coefficients, cooling water flow rate, lubricant viscosity, and load distribution data are weighted and fused to obtain the water replenishment frequency;

[0039] Based on the cooling water flow rate and the load distribution data, the flow rate adjustment range is calculated using a PID algorithm.

[0040] Based on the water replenishment frequency and the flow rate adjustment range, combined with the nozzle angle and the spray coverage range, the control command is mapped into spatial coverage parameters through a projection transformation matrix algorithm, and the initial water replenishment control command is output.

[0041] In one optional implementation, the step of generating a final water replenishment control command by iteratively optimizing the nozzle angle and the pipeline pressure based on the initial water replenishment control command, the equipment rotation speed, and the abrasive wear data includes:

[0042] Based on the equipment rotation speed and the wear data of the grinding wheel, rotation speed weight and wear weight are assigned by entropy weight method, and weighted fusion is performed to obtain the operating status parameters;

[0043] Based on the initial water replenishment control command and the operating status parameters, a preliminary command set is output through iterative optimization using the Q-learning algorithm.

[0044] Based on the preliminary instruction set, combined with the nozzle angle and the pipeline pressure, the nozzle angle is calibrated through spatial projection transformation, and the pipeline pressure is compensated using a linear interpolation algorithm to generate the final water replenishment control instruction set.

[0045] In one optional implementation, the step of optimizing the final water replenishment control command based on the liquid supply status data, combined with the load distribution data and fluctuation frequency, to obtain an optimized water replenishment control command includes:

[0046] Calculate the execution deviation value based on the liquid supply status data and the preset target supply status;

[0047] No adjustment is required when the execution deviation value is less than or equal to the preset deviation threshold;

[0048] When the execution deviation value is greater than the preset deviation threshold, the parameters of the Q-learning algorithm are adjusted by combining the load distribution data and the fluctuation frequency, and the final water replenishment control command is optimized to obtain the optimized water replenishment control command.

[0049] Secondly, the present invention provides an intelligent control system for a silicon carbide micro powder grinding mill, comprising:

[0050] The data acquisition module is used to collect load distribution data, fluctuation frequency, equipment speed, abrasive wear data, material hardness parameters, processing accuracy requirements, pipeline pressure, liquid temperature, nozzle angle and spray coverage.

[0051] The load feature analysis module is used to filter and extract features from the load distribution data, the fluctuation frequency, the equipment speed, and the wear data of the grinding wheel to obtain a load feature vector.

[0052] The load trend analysis module is used to input the load feature vector, the material hardness parameter, and the processing accuracy requirement into a pre-built load trend prediction model and output the load trend vector.

[0053] The consumption rate analysis module is used to input the load trend vector into a pre-built consumption rate distribution model to calculate the cooling water flow consumption rate and the lubricant viscosity consumption rate.

[0054] The flow rate and viscosity calculation module is used to assign weights based on the pipeline pressure and the liquid temperature, and to perform weighted fusion based on the cooling water flow rate consumption rate and the lubricant viscosity consumption rate to obtain the cooling water flow rate and lubricant viscosity.

[0055] The initial instruction generation module is used to perform weight allocation based on the cooling water flow rate, the lubricant viscosity and the load distribution data, and generate an initial water replenishment control instruction in combination with the nozzle angle and the spray coverage range;

[0056] The final instruction generation module is used to generate the final water replenishment control instruction by iteratively optimizing the nozzle angle and the pipeline pressure based on the initial water replenishment control instruction, the equipment rotation speed and the mold wear data.

[0057] The output control module is used to drive the grinding mill liquid supply system according to the final water replenishment control command, adjust the cooling water flow rate and lubricant viscosity distribution in real time, and output liquid supply status data including actual flow rate, viscosity and distribution status.

[0058] A control optimization module is used to optimize the final water replenishment control command based on the liquid supply status data, combined with the load distribution data and fluctuation frequency, to obtain an optimized water replenishment control command. Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the intelligent control method for the silicon carbide micron powder grinding mill described in any one of the above embodiments.

[0059] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform the intelligent control method for the silicon carbide micro powder grinding mill described in any one of the above-mentioned methods.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] (1) By filtering and denoising load data and equipment status and extracting features, and using a prediction model that integrates material characteristics and accuracy requirements to predict dynamic load trends, the existing technology relies on preset fixed thresholds and rules to improve the accuracy of load fluctuation prediction under complex grinding conditions.

[0062] (2) The demand rate of cooling water and lubricant is calculated by using the load prediction results to drive the consumption rate model. The real-time parameters such as pipeline pressure and liquid temperature are integrated for weighted calculation to dynamically generate flow rate and viscosity value. This solves the problem of supply lag or excess in the existing technology and improves the utilization efficiency of cooling water and lubricant.

[0063] (3) When generating water replenishment instructions, multi-dimensional parameters such as flow rate, viscosity, load distribution, nozzle angle and spray range are comprehensively considered. The nozzle angle and pipeline pressure are dynamically adjusted through iterative optimization algorithms to achieve precise matching between the spray coverage and process requirements, thereby improving the stability and processing accuracy of the processing area.

[0064] (4) Based on the actual output status data of the liquid supply system, combined with load distribution and fluctuation information, the water replenishment control command is continuously optimized through the algorithm to form a closed-loop control mechanism of "execution-feedback-adjustment", so that the system can adapt to changes in operating conditions in real time and improve the dynamic response capability of the system. Attached Figure Description

[0065] Figure 1 This is a schematic diagram of the intelligent control method for a silicon carbide micro powder grinding mill provided in the first embodiment of the present invention;

[0066] Figure 2 This is a schematic diagram of the intelligent control system structure of the silicon carbide micro powder grinding machine provided in the second embodiment of the present invention. Detailed Implementation

[0067] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] Reference Figure 1 The first embodiment of the present invention provides an intelligent control method for a silicon carbide micro powder grinding mill, comprising the following steps:

[0069] S11 collects load distribution data, fluctuation frequency, equipment speed, abrasive wear data, material hardness parameters, processing accuracy requirements, pipeline pressure, liquid temperature, nozzle angle, and spray coverage.

[0070] S12, the load distribution data, the fluctuation frequency, the equipment rotation speed and the grinding wheel wear data are filtered and feature extracted to obtain a load feature vector;

[0071] S13, input the load characteristic vector, the material hardness parameter and the processing accuracy requirement into the pre-built load trend prediction model, and output the load trend vector;

[0072] S14, Input the load trend vector into the pre-built consumption rate distribution model to calculate the cooling water flow consumption rate and lubricant viscosity consumption rate;

[0073] S15, weight allocation is performed based on the pipeline pressure and the liquid temperature, and weighted fusion is performed based on the cooling water flow rate consumption rate and the lubricant viscosity consumption rate to obtain the cooling water flow rate and lubricant viscosity.

[0074] S16, weight allocation is performed based on the cooling water flow rate, the lubricant viscosity and the load distribution data, and an initial water replenishment control command is generated in combination with the nozzle angle and the spray coverage range;

[0075] S17, Based on the initial water replenishment control command, the equipment rotation speed and the mold wear data, the final water replenishment control command is generated by iteratively optimizing the nozzle angle and the pipeline pressure;

[0076] S18, according to the final water replenishment control command, drive the grinding mill liquid supply system, adjust the cooling water flow rate and lubricant viscosity distribution in real time, and output liquid supply status data including actual flow rate, viscosity and distribution status;

[0077] S19, Based on the liquid supply status data, combined with the load distribution data and fluctuation frequency, optimize the final water replenishment control command to obtain the optimized water replenishment control command.

[0078] In step S11, load distribution data, fluctuation frequency, equipment speed, abrasive wear data, material hardness parameters, processing accuracy requirements, pipeline pressure, liquid temperature, nozzle angle, and spray coverage are collected.

[0079] Specifically, the grinding machine employs a wet grinding process, using a nozzle system to spray a water-based polishing slurry onto the workpiece. This slurry contains silicon carbide micropowder as an abrasive carrier, a water-based component for cooling (called the cooling water component), and an additive component for lubrication (called the lubricant component). During this process, load distribution data characterizing the spatial distribution of grinding force is collected by a pressure sensor array in the grinding area. The grinding load fluctuation frequency (e.g., the dominant frequency value) is extracted from the Fourier transform results of a current sensor. The machine speed (e.g., the speed pulse signal) is obtained by the encoder of the equipment control unit. Finally, the wear data (e.g., the wear rate percentage) is generated using laser ranging data from the grinding wheel surface detection device. The grinding wheel surface detection device employs a non-contact measurement method. Specifically, a laser rangefinder (such as a laser displacement sensor) is fixedly installed inside or near the grinding wheel body at a safe location. Its laser beam penetrates a transparent observation window on the grinding wheel substrate or is projected onto the working surface of the grinding wheel through a pre-reserved safety gap. The distance change from the working surface of the grinding wheel to the sensor is measured by receiving the reflected laser signal. Based on the difference between the initial calibration distance and the real-time measurement distance, and combined with a grinding wheel material wear characteristic model constructed from historical operating data (a dataset consisting of the actual wear amount of the grinding wheel recorded in the equipment operation log and corresponding operating parameters), grinding wheel wear data (such as wear rate percentage) is calculated and generated. Simultaneously, material hardness parameters (such as Mohs hardness scale values) and processing accuracy requirements (such as surface roughness Ra values) are retrieved from the material database. The liquid pressure value of the polishing fluid (i.e., the grinding working medium containing silicon carbide micropowder) is collected through a pipeline pressure sensor, the liquid temperature of the polishing fluid is obtained through a temperature sensor, the nozzle angle is recorded using an angle encoder, and the spray coverage area is measured using a laser rangefinder. This step fully acquires ten types of real-time parameters and preset parameters required for the grinding process, providing a multi-dimensional data input basis for subsequent dynamic control. It solves the supply lag problem caused by incomplete data acquisition in the background technology, and achieves the goal of improving processing accuracy and resource utilization efficiency.

[0080] In step S12, the load distribution data, the fluctuation frequency, the equipment rotation speed, and the grinding wheel wear data are filtered and feature extracted to obtain a load feature vector.

[0081] In one specific implementation, the step of filtering and extracting features from the load distribution data, the fluctuation frequency, the equipment rotation speed, and the grinding wheel wear data to obtain a load feature vector includes:

[0082] Based on the load distribution data, the fluctuation frequency, the equipment rotation speed, and the grinding wheel wear data, filtered data is obtained by wavelet transform filtering.

[0083] Based on the filtered data, principal component analysis is used to extract features to obtain principal component features.

[0084] The principal component features are normalized to obtain the load feature vector.

[0085] Specifically, firstly, wavelet transform is used to filter the collected raw load distribution data (such as the spatial pressure matrix), fluctuation frequencies (such as the dominant frequency of the current spectrum), equipment speed (such as encoder pulse signals), and wear data (such as laser ranging wear rate). Specifically, wavelet decomposition and reconstruction algorithms are used to separate high-frequency noise components. High-frequency interference is filtered out based on an energy threshold (determined by statistically analyzing the noise energy distribution of the optimal operating conditions from a historical database), and an effective feature value range (e.g., load intensity of 450-550 Newtons) is extracted. The purified load and fluctuation data are then output.

[0086] Secondly, the purified load data, fluctuation data, equipment speed data, and wear rate data are input into Principal Component Analysis (PCA) for feature fusion. By calculating the covariance matrix of the multidimensional data, principal component feature vectors representing the direction of data variation are extracted. Principal components with a cumulative contribution rate exceeding 85% (determined based on historical operating condition feature contribution rates) are retained. Dimensionality reduction is then used to generate four-dimensional principal component features including the mean load intensity, the dominant frequency of fluctuation, the speed value, and the wear rate. These principal components are linear combinations of the original features (load intensity, dominant frequency of fluctuation, speed, and wear rate), reflecting the main variation patterns of these key physical quantities. Finally, the principal component features are normalized: the mean load intensity is linearly mapped to the [0,1] interval (determined based on the equipment's rated load range), the dominant frequency of fluctuation is standardized using Z-score (based on the historical mean and standard deviation of the dominant frequency), and the speed value and wear rate are scaled using maximum and minimum values ​​(based on the equipment's speed design range and wear rate threshold range, respectively). Finally, a normalized load feature vector is output.

[0087] This step eliminates sensor noise interference and extracts core operating condition features through denoising filtering and feature fusion processing, solving the load prediction deviation problem caused by noise in the original data in the background technology, providing high-precision input for subsequent trend prediction models, and improving the reliability of machining accuracy control.

[0088] In step S13, the load feature vector, the material hardness parameter, and the processing accuracy requirement are input into a pre-built load trend prediction model, and the load trend vector is output.

[0089] In one specific implementation, the process of constructing the load trend prediction model includes:

[0090] Acquire historical load characteristics, historical material hardness parameters, historical processing accuracy requirements, and corresponding historical load data;

[0091] An initial load trend prediction model was constructed using the support vector machine algorithm. The classification boundary was initialized using the RBF kernel function, and a penalty coefficient was set to determine the tolerance error threshold.

[0092] Based on the historical load characteristics, the historical material hardness parameters, and the historical processing accuracy requirements, the curvature adjustment parameters of the classification boundary are calculated using the kernel density estimation method, and the curvature of the classification boundary is dynamically adjusted using the RBF kernel function.

[0093] Based on the historical load data, the penalty coefficient is optimized using a grid search method to determine the error tolerance threshold;

[0094] Cross-validate the adjusted classification boundaries and calculate the set of classification accuracy and error distribution after boundary adjustment;

[0095] When the error distribution set exceeds the error tolerance threshold, the gradient descent method is used to re-optimize the curvature parameter of the classification boundary, and the classification boundary is iteratively updated until the error distribution set does not exceed the error tolerance threshold.

[0096] The optimized classification boundary is matched with the load data in a multidimensional space by using RBF kernel function mapping to determine the hyperplane equation corresponding to each load data.

[0097] Based on the time series characteristics of the historical load data, the sliding window method is used to extract the dynamic trend parameters of load data changes. The load data is then correlated with the hyperplane equation to construct the final load trend prediction model.

[0098] Specifically, historical load characteristics (normalized mean load intensity, dominant frequency of fluctuation, rotational speed, and wear rate), historical material hardness parameters (such as Mohs hardness scale values), historical processing accuracy requirements (such as surface roughness Ra values), and corresponding historical load data (time-series load values) are obtained from historical databases. The model is initialized using a support vector regression algorithm. Multidimensional features are mapped to a high-dimensional space using a radial basis function kernel to initialize the classification boundary. Simultaneously, an initial penalty coefficient is set based on the 3σ principle of historical load data error distribution (three times the standard deviation of historical errors), and a tolerance error threshold (such as ±5% relative error) is determined.

[0099] Based on historical load characteristics, material hardness parameters, and processing accuracy requirements, a kernel density estimation algorithm is used to calculate the curvature of dense regions of the feature distribution, and the curvature parameters of the radial basis function kernel are dynamically adjusted (the curvature correction coefficient is calculated based on the eigenvalues ​​of the covariance matrix of the historical data feature space distribution). According to the prediction error of historical load data, a grid search method is used to traverse and optimize within a preset parameter space (penalty coefficient range 0.1-10, which is set based on the effective range of equipment control parameters. This effective range is obtained through parameter sensitivity analysis of 200 sets of historical operating data. The analysis results show that when the penalty coefficient is below 0.1, the model's response to load fluctuations is insufficient, while exceeding 10 leads to overfitting). The penalty coefficient with the smallest cross-validation mean square error is selected, and the error tolerance threshold is updated.

[0100] K-fold cross-validation (K value 5, determined by the historical data sample size) was performed on the adjusted classification boundary to calculate the classification accuracy (the acceptable threshold of 85% is set according to the minimum requirement of the industry acceptance standard) and the error distribution set. When the error distribution exceeds the error tolerance threshold, the gradient descent algorithm was used to iteratively optimize the curvature parameters until the error converged within the threshold. The optimized classification boundary was matched with historical load data through radial basis function kernel mapping to determine the hyperplane equation (decision function) corresponding to each load data point in the feature space. Finally, combined with the timestamps of the historical load data, the sliding window method was used (the window length is calculated by the reciprocal of the operating condition dominant frequency, where the operating condition dominant frequency refers to the dominant frequency component identified after the equipment vibration signal is transformed by fast Fourier transform during the grinding process. This frequency reflects the inherent working rhythm of the grinding system. Its extraction method is: by collecting time-domain signals through vibration sensors installed on the grinding machine spindle, and taking the first dominant frequency with an energy share of more than 60% as the operating condition dominant frequency) to extract the dynamic trend parameters (such as the slope of change) of the load data within the window, thus completing the construction of the load trend prediction model.

[0101] During the model application phase, the real-time load feature vector (including the average load intensity, the dominant frequency of fluctuation, the rotational speed, and the wear rate), material hardness parameters, and processing accuracy requirements are input into the model. The load intensity change rate (such as the rising slope value) and the fluctuation frequency change rate (such as the falling slope value) are calculated through the decision function, and a two-dimensional load dynamic trend vector (including the load intensity change rate and the fluctuation frequency change rate) is output.

[0102] This step solves the problem of load trend prediction deviation caused by static rules in the background technology by integrating multi-dimensional historical data and dynamic optimization mechanism, and provides high-precision trend input for subsequent calculation of cooling water and lubricant consumption rate, directly improving the timeliness of resource supply and the stability of processing accuracy.

[0103] In step S14, the load trend vector is input into the pre-built consumption rate distribution model to calculate the cooling water flow rate consumption rate and the lubricant viscosity consumption rate.

[0104] In one specific implementation, the process of constructing the consumption rate distribution model includes:

[0105] The historical load trend vector, the corresponding historical cooling water flow rate and historical lubricant viscosity rate are obtained, and the historical load trend vector is cleaned and its features are extracted to obtain trend features.

[0106] The trend features are input into the input layer of the initially constructed consumption rate distribution model for training, and the predicted cooling water flow consumption rate and the predicted lubricant viscosity consumption rate are obtained from the output layer of the consumption rate distribution model.

[0107] The predicted cooling water flow rate consumption rate, the predicted lubricant viscosity consumption rate, the historical cooling water flow rate consumption rate, and the historical lubricant viscosity consumption rate are substituted into the loss function to calculate the loss value;

[0108] The gradient of the output layer of the consumption rate distribution model is calculated based on the loss value, and the gradient is passed forward layer by layer through the chain rule to calculate the gradient of the parameters of each layer and obtain the gradient data.

[0109] Based on the gradient data and the preset learning rate, the parameters of each layer of the consumption rate distribution model are updated using the gradient descent method.

[0110] The parameters of each layer are iteratively updated until the number of training iterations of the consumption rate distribution model is greater than a preset number of iterations, or the loss value of the consumption rate distribution model is less than a preset loss threshold. At this point, the training is considered complete, and a trained consumption rate distribution model is obtained.

[0111] Specifically, historical load trend vectors (containing two dimensions: load intensity change rate and fluctuation frequency change rate), corresponding historical cooling water flow rate (liters / minute), and historical lubricant viscosity consumption rate (centipoise / minute) are obtained from the historical database. Data cleaning is performed on the historical load trend vectors: based on the defined rated operating range (load intensity change rate threshold ±10 N / s, fluctuation frequency change rate threshold ±5 Hz / s, determined by the maximum mechanical load test calibration of the equipment), out-of-limit abnormal data points are removed. After cleaning, multi-dimensional trend features are extracted using sliding window mean filtering (window length of 5 sampling points, determined based on historical data spectrum analysis), including the 10-minute moving average of the load intensity change rate and the 10-minute moving average of the fluctuation frequency change rate. The cleaned two-dimensional trend features (mean of load intensity change rate and mean of fluctuation frequency change rate) are input into a three-layer fully connected neural network model (input layer node count is 2, matching the feature dimensions; the hidden layer activation function uses ReLU). After nonlinear transformation of the hidden layer, predicted cooling water flow rate and predicted lubricant viscosity consumption rate are generated at the output layer. The predicted and historical true values ​​are substituted into the mean squared error loss function to calculate the loss value (the loss function is the mean of the squared differences between the predicted and true values). Backpropagation is performed based on the loss value: starting from the output layer, the partial derivatives of the loss function with respect to the weights are calculated, and the gradient is propagated layer by layer using the chain rule, calculating the gradient tensor of each layer's parameters (such as the gradient value of the weight matrix). Gradient descent combined with a learning rate (0.01, verified as the optimal convergence parameter on the historical training dataset through grid search) is used to update all model weight parameters. Training terminates when the number of iterations reaches the upper limit (1000 times, set according to computational resource constraints) or the loss value falls below a threshold (0.05, determined based on the critical value at which the loss function converges to a stable state during historical training).

[0112] Model application phase: The real-time load trend vector (including the rate of change of load intensity and the rate of change of fluctuation frequency) is input into the trained model. The model then uses neural network weights to calculate and output the cooling water flow rate consumption rate (liters / minute) and the lubricant viscosity consumption rate (centipoises / minute). For example, if the rate of increase in load intensity change rate (+0.02 / minute) triggers an increase in armature cooling demand, the model outputs the increment in cooling water flow rate consumption rate (baseline value 12 liters / minute + 1.5 liters / minute). This step addresses the prediction bias caused by fixed rules through a data-driven modeling approach, providing an accurate benchmark for fluid supply control.

[0113] In step S15, the cooling water flow rate and the lubricant viscosity are weighted and fused together according to the pipeline pressure and the liquid temperature, and the weighted fusion is performed in combination with the cooling water flow rate and the lubricant viscosity.

[0114] Specifically, the pipeline pressure and liquid temperature data are first processed using a dynamic weighting algorithm: the percentage deviation between the pipeline pressure and the preset benchmark pressure (automatically calculated based on the average pressure under stable operating conditions during the first 100 hours of equipment trial operation) is calculated, and the deviation value is input into the Sigmoid function to convert it into a pressure weighting coefficient (e.g., a deviation of ±10% corresponds to a weight of 0.6); at the same time, it is detected whether the liquid temperature exceeds the threshold range (e.g., 25-30℃, determined based on the best operating conditions statistically from historical databases). If it exceeds the threshold, the temperature weighting coefficient is adjusted according to preset rules (e.g., an increase of 0.05 weight for every 1℃ increase in temperature) (e.g., a weight of 0.4 at 32℃).

[0115] In the parameter fusion calculation process, the cooling water flow rate consumption rate (12 liters / minute) is first multiplied by the pressure weighting coefficient (0.6) to obtain the pressure-corrected base flow rate value (7.2 liters / minute). Simultaneously, a temperature compensation amount (0.3 liters / minute × 2 = 0.6 liters / minute) is calculated based on the degree of liquid temperature exceeding the threshold (e.g., exceeding the threshold by 2°C). These are then combined to generate the final estimated cooling water flow rate value (7.2 + 0.6 = 7.8 liters / minute). Similarly, the lubricant viscosity calculation first multiplies the consumption rate (85 centipoise / minute) by the temperature weighting coefficient (0.4) to obtain the base viscosity value (34 centipoise). Then, based on the pressure deviation (e.g., below the baseline by 0.3 bar), a pressure compensation amount (0.5 centipoise × 3 = 1.5 centipoise) is deducted to finally obtain the corrected viscosity estimate value (34 - 1.5 = 32.5 centipoise).

[0116] The weighting coefficients play a role in three aspects: First, the pressure weighting coefficient (0.6) and temperature weighting coefficient (0.4) quantify the influence weights of pipeline pressure and liquid temperature on system demand, respectively, reflecting the differences in the importance of parameters under different operating conditions; second, the compensation value (e.g., 0.3 liters / minute·℃) establishes the nonlinear adjustment relationship when parameters exceed limits, ensuring the system's sensitive response to abnormal operating conditions; finally, through step-by-step calculation by multiplying by the weights and then superimposing the compensation, the dominant role of the main parameters is preserved, while the dynamic correction of auxiliary parameters is realized. This composite calculation method can more accurately reflect the synergistic influence mechanism of various parameters under complex operating conditions than the traditional weighted average.

[0117] This step addresses the supply deviation issue caused by environmental interference in the background technology through weight allocation, ensuring that the supply of cooling water and lubricant precisely matches the real-time operating conditions, thereby improving the temperature stability and grinding accuracy of the machining area. It should be noted that the cooling water and lubricant components are mixed in a preset ratio in the polishing slurry distribution system to form a complete grinding working medium. The independent control objective for both is to adjust the mixing ratio and flow rate in real time so that the polishing slurry simultaneously meets the dual requirements of grinding cooling and surface lubrication, thereby improving machining accuracy and equipment lifespan.

[0118] In step S16, a weighted distribution is performed based on the cooling water flow rate, the lubricant viscosity, and the load distribution data, and an initial water replenishment control command is generated by combining the nozzle angle and the spray coverage range.

[0119] In one specific implementation, the step of weighting the cooling water flow rate, the lubricant viscosity, and the load distribution data, and combining this with the nozzle angle and the spray coverage area to generate an initial water replenishment control command includes:

[0120] Based on the cooling water flow rate, the lubricant viscosity, and the load distribution data, weights are allocated using the entropy weighting method to obtain weight coefficients.

[0121] The weighting coefficients, cooling water flow rate, lubricant viscosity, and load distribution data are weighted and fused to obtain the water replenishment frequency;

[0122] Based on the cooling water flow rate and the load distribution data, the flow rate adjustment range is calculated using a PID algorithm.

[0123] Based on the water replenishment frequency and the flow rate adjustment range, combined with the nozzle angle and the spray coverage range, the control command is mapped into spatial coverage parameters through a projection transformation matrix algorithm, and the initial water replenishment control command is output.

[0124] Specifically, the weighting coefficients are first assigned using the entropy weighting method: based on historical operating data analysis, the dispersion of the cooling water flow sequence is calculated (coefficient of variation 0.25) to obtain the flow weighting coefficient (0.3), the viscosity weighting coefficient is determined by analyzing the viscosity fluctuation amplitude of the lubricant (coefficient of variation 0.18), and the load weighting coefficient (0.5) is obtained by evaluating the spatial non-uniformity of the pressure matrix of the grinding area (16-point spatial distribution data collected by the pressure sensor array) (coefficient of variation 0.32). The normalized parameter values ​​(flow rate 0-1 scale corresponds to 5-15 L / min, viscosity 0-1 scale corresponds to 50-100 cP, load 0-1 scale corresponds to 300-500 N) are weighted and fused: 0.3 × flow rate scale value + 0.2 × viscosity scale value + 0.5 × load scale value, outputting a fused index value in the 0-1 range. This value is then converted into a water replenishment frequency (e.g., a fused value of 0.73 corresponds to 2.2 times / hour) through a calibration curve (based on a quadratic polynomial regression model established from 200 sets of historical data). This frequency value has been experimentally verified to match the water replenishment requirements within the optimal processing accuracy range (error ±0.3 times / h).

[0125] Next, the flow regulation amplitude is calculated using a proportional-integral-derivative (PID) algorithm: In the proportional stage, the difference between the real-time cooling water flow rate and the target value (determined by linear interpolation of the load distribution mean) is calculated and multiplied by the proportional coefficient (0.5); in the integral stage, the sum of flow deviations over the past 10 minutes (the time window is determined by statistical analysis of the operating condition change cycle) is accumulated and multiplied by the integral coefficient (0.2, determined based on multiple linear regression analysis of the past 6 months of operating condition data); in the derivative stage, the real-time gradient change rate of the load distribution data is extracted and multiplied by the derivative coefficient (0.3). The output values ​​of the three stages are then superimposed to generate the flow regulation amplitude (e.g., +4.8 liters / minute).

[0126] Finally, spatial mapping is performed by combining the nozzle angle (45 degrees) and the spray coverage area (2-meter diameter): the flow rate adjustment (+4.8 liters / minute) is converted into an increase in spray coverage area using a projection transformation matrix algorithm (each increase of 1 liter / minute corresponds to an area expansion of 0.1 square meters). Coverage uniformity is checked (current value 82%). If it is lower than the 85% threshold specified in Clause 4.5 of industry standard ISO 21940-2:2022, the nozzle angle is adjusted to 48 degrees according to the angle-coverage area relationship curve (generated from computational fluid dynamics simulation data). An initial water replenishment control vector is output, containing the flow rate adjustment value (+4.8 liters / minute), the water replenishment frequency (2.2 times / hour), and the angle command (48 degrees). This step solves the localized overheating problem caused by uneven spray coverage in the background technology by integrating fluid parameters, spatial constraints, and real-time load distribution.

[0127] In step S17, based on the initial water replenishment control command, the equipment rotation speed, and the wear data of the grinding wheel, the final water replenishment control command is generated by iteratively optimizing the nozzle angle and the pipeline pressure.

[0128] In one specific implementation, the step of generating a final water replenishment control command by iteratively optimizing the nozzle angle and the pipeline pressure based on the initial water replenishment control command, the equipment rotation speed, and the abrasive wear data includes:

[0129] Based on the equipment rotation speed and the wear data of the grinding wheel, rotation speed weight and wear weight are assigned by entropy weight method, and weighted fusion is performed to obtain the operating status parameters;

[0130] Based on the initial water replenishment control command and the operating status parameters, a preliminary command set is output through iterative optimization using the Q-learning algorithm.

[0131] Based on the preliminary instruction set, combined with the nozzle angle and the pipeline pressure, the nozzle angle is calibrated through spatial projection transformation, and the pipeline pressure is compensated using a linear interpolation algorithm to generate the final water replenishment control instruction set.

[0132] Specifically, the operating status parameters are first calculated using the entropy weight method: based on the fluctuation rate of the equipment speed sequence in historical operating conditions (e.g., ±2% speed fluctuation corresponds to an information entropy value of 0.8), the speed weight coefficient (e.g., 0.6) is calculated according to the entropy weight rule defined in Section 3.2 of the Equipment Condition Assessment Standard; simultaneously, the wear weight coefficient (e.g., 0.4) is calculated based on the growth slope of the wear rate (e.g., an increase of 0.7 per 5% wear rate). The real-time equipment speed value is multiplied by the speed weight coefficient, and the wear rate is multiplied by the wear weight coefficient to output the operating status parameters (e.g., 75 points).

[0133] Subsequently, the Q-learning algorithm was used for instruction optimization: the operating state parameters (75 points) and the initial water replenishment control instructions (flow rate +4.8 liters / minute, frequency 2.2 times / hour, angle 48 degrees) were input into the reinforcement learning framework. The state space was defined as 10 discrete intervals of the operating state parameters, and the action space was defined as nozzle angle adjustment step size ±1 degree and pipeline pressure adjustment step size ±0.1 bar (step size values ​​were determined by hydraulic system response characteristic experiments). The improvement in equipment heat dissipation efficiency (calculated based on temperature sensor feedback) was used as the reward function, and the Q-value table was iteratively updated using the Bellman equation: after executing an action, the new state was observed and the immediate reward (e.g., heat dissipation efficiency improvement of 5%) was calculated. Combined with the learning rate (0.01, determined by historical training convergence test reports) and the discount factor (0.9, set according to the equipment thermal response delay characteristics), the Q-value was updated until the change was less than the convergence threshold of 0.05 for 3 consecutive iterations (defined in the algorithm stability test report), and a preliminary instruction set (e.g., flow rate +5.5 liters / minute, frequency 2.3 times / hour) was output.

[0134] The final step involves spatial calibration and pressure compensation: Based on the initial command set's flow parameters (+5.5 L / min) and nozzle angle (49 degrees), a spatial projection transformation algorithm is used to detect the uniformity of spray coverage (currently 83%). If it falls below the 85% threshold specified in Clause 4.5 of ISO 21940-2:2022, the nozzle angle is calibrated to 48 degrees according to the angle-uniformity curve generated from computational fluid dynamics simulation data. Simultaneously, based on the pressure-flow relationship curve calibrated from hydraulic system characteristic experiments, a linear interpolation algorithm is used to calculate the pipeline pressure compensation value (e.g., +0.2 bar), generating a final water replenishment control command set that includes flow adjustment (+5.5 L / min), water replenishment frequency (2.3 times / hour), nozzle angle (48 degrees), and pipeline pressure (2.4 bar). This step addresses the uneven spray coverage problem through dynamic optimization and spatial coordination mechanisms.

[0135] In step S18, the grinding mill liquid supply system is driven according to the final water replenishment control command to adjust the cooling water flow rate and lubricant viscosity distribution in real time, and output liquid supply status data including actual flow rate, viscosity and distribution status.

[0136] Specifically, after receiving a flow rate adjustment command (e.g., +5.5 liters / minute), the liquid supply system actuator adjusts the pumping ratio of the polishing fluid lubricant component according to the viscosity command. The valve opening of the proportional valve for the polishing fluid cooling water component is adjusted (the opening increment is calculated based on the flow rate-opening calibration curve). The nozzle servo motor rotates to the target position according to the angle command (e.g., 48 degrees), and the pipeline pressure regulator controls the pneumatic booster device according to the pressure command (e.g., 2.4 bar).

[0137] During system operation, the flow sensor collects the actual flow rate of cooling water in real time (e.g., 44.5 liters / minute), and the viscosity sensor obtains the actual viscosity value of the lubricating fluid (e.g., 10.2 centipoise). The final output includes liquid supply status data containing the actual flow rate and its deviation from the target value, the actual viscosity value and its adjustment status, the nozzle angle execution status, and the pipeline pressure adjustment status. This step achieves stable control of the cooling conditions in the processing area through precise execution of control commands and real-time status feedback.

[0138] In step S19, based on the liquid supply status data, combined with the load distribution data and fluctuation frequency, the final water replenishment control command is optimized to obtain the optimized water replenishment control command.

[0139] In one specific implementation, the step of optimizing the final water replenishment control command based on the liquid supply status data, combined with the load distribution data and fluctuation frequency, to obtain an optimized water replenishment control command includes:

[0140] Calculate the execution deviation value based on the liquid supply status data and the preset target supply status;

[0141] No adjustment is required when the execution deviation value is less than or equal to the preset deviation threshold;

[0142] When the execution deviation value is greater than the preset deviation threshold, the parameters of the Q-learning algorithm are adjusted by combining the load distribution data and the fluctuation frequency, and the final water replenishment control command is optimized to obtain the optimized water replenishment control command.

[0143] Specifically, firstly, the liquid supply status data is derived from the output of step S18, including the actual cooling water flow rate, actual lubricant viscosity, and specific values ​​of the distribution status; the load distribution data is derived from the pressure distribution information of the grinding area collected in step S11; the fluctuation frequency is also derived from the load fluctuation main frequency value collected in step S11. The preset target supply status is determined by equipment specifications or historical best operating condition data, including the target cooling water flow rate and target lubricant viscosity. The deviation value is calculated by comparing the difference between the actual cooling water flow rate and the target cooling water flow rate, and the difference between the actual lubricant viscosity and the target lubricant viscosity, using the root mean square error method to obtain the overall deviation value. For example, the deviation value is equal to the square root of the sum of the square of the difference between the actual flow rate and the target flow rate plus the square of the difference between the actual viscosity and the target viscosity.

[0144] The deviation threshold is set based on historical operational data statistics: the deviation distribution under the past 100 normal operating conditions is analyzed, and the average value plus twice the standard deviation is taken as the upper limit of the threshold. If the execution deviation value is less than or equal to the threshold, it indicates that the current supply status meets the requirements, and no adjustment or optimization of the water replenishment control command is needed; if the execution deviation value is greater than the threshold, it is necessary to combine load distribution data (such as the spatial distribution characteristics of the pressure matrix) and fluctuation frequency (such as the real-time main frequency value) to adjust the parameters of the Q-learning algorithm through the gradient descent algorithm. The Q-learning algorithm is used to generate the final water replenishment control command in step S17, and its parameters include the learning rate and discount factor. The specific implementation process of the gradient descent algorithm is as follows: the execution deviation value is used as the input of the loss function, and the partial derivative of the loss function with respect to the Q-learning learning rate parameter is calculated; the gradient direction is adjusted by combining the non-uniformity coefficient of the load distribution data (calculated by the coefficient of variation of the pressure matrix) and the stability index of the fluctuation frequency (calculated by the rate of change of the main frequency value); the learning rate parameter value is updated by backpropagation through the chain rule; iterative updates are performed until the change in the loss function converges to the preset convergence threshold (such as 0.01, determined based on historical training data). The optimized Q-learning parameters are reapplied to the control command generation logic. Combined with the current operating status parameters, optimized water replenishment control commands are output, including adjusted flow rate, frequency, angle, and pressure values.

[0145] This step addresses the supply mismatch problem caused by sudden changes in operating conditions or sensor drift by dynamically correcting the reinforcement learning algorithm parameters through real-time feedback on the deviation between the liquid supply status and the target status, and by integrating load distribution and fluctuation characteristics. It forms a closed-loop control mechanism that enables the system to adaptively adjust the cooling water and lubricant supply strategies, reducing temperature fluctuations in the machining area and directly improving machining accuracy stability and resource utilization efficiency.

[0146] Reference Figure 2 The second embodiment of the present invention provides an intelligent control system for a silicon carbide micro powder grinding mill, comprising:

[0147] The data acquisition module is used to collect load distribution data, fluctuation frequency, equipment speed, abrasive wear data, material hardness parameters, processing accuracy requirements, pipeline pressure, liquid temperature, nozzle angle and spray coverage.

[0148] The load feature analysis module is used to filter and extract features from the load distribution data, the fluctuation frequency, the equipment speed, and the wear data of the grinding wheel to obtain a load feature vector.

[0149] The load trend analysis module is used to input the load feature vector, the material hardness parameter, and the processing accuracy requirement into a pre-built load trend prediction model and output the load trend vector.

[0150] The consumption rate analysis module is used to input the load trend vector into a pre-built consumption rate distribution model to calculate the cooling water flow consumption rate and the lubricant viscosity consumption rate.

[0151] The flow rate and viscosity calculation module is used to assign weights based on the pipeline pressure and the liquid temperature, and to perform weighted fusion based on the cooling water flow rate consumption rate and the lubricant viscosity consumption rate to obtain the cooling water flow rate and lubricant viscosity.

[0152] The initial instruction generation module is used to perform weight allocation based on the cooling water flow rate, the lubricant viscosity and the load distribution data, and generate an initial water replenishment control instruction in combination with the nozzle angle and the spray coverage range;

[0153] The final instruction generation module is used to generate the final water replenishment control instruction by iteratively optimizing the nozzle angle and the pipeline pressure based on the initial water replenishment control instruction, the equipment rotation speed and the mold wear data.

[0154] The output control module is used to drive the grinding mill liquid supply system according to the final water replenishment control command, adjust the cooling water flow rate and lubricant viscosity distribution in real time, and output liquid supply status data including actual flow rate, viscosity and distribution status.

[0155] The control optimization module is used to optimize the final water replenishment control command based on the liquid supply status data, combined with the load distribution data and fluctuation frequency, to obtain the optimized water replenishment control command.

[0156] It should be noted that the intelligent control device for a silicon carbide micro powder grinding mill provided in this embodiment of the invention is used to execute all the process steps of the intelligent control method for a silicon carbide micro powder grinding mill in the above embodiment. The working principle and beneficial effects of the two are one-to-one, so they will not be described again.

[0157] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as an intelligent control program for a silicon carbide micro-powder grinding mill. When the processor executes the computer program, it implements the steps in the aforementioned intelligent control method embodiments for silicon carbide micro-powder grinding mills, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as the intelligent control module of the silicon carbide micro powder grinding machine.

[0158] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0159] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0160] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0161] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0162] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0163] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0164] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. An intelligent control method for a silicon carbide micro powder grinding mill, characterized in that, include: Collect load distribution data, fluctuation frequency, equipment speed, abrasive wear data, material hardness parameters, processing accuracy requirements, pipeline pressure, liquid temperature, nozzle angle, and spray coverage. The load distribution data, the fluctuation frequency, the equipment rotation speed, and the grinding wheel wear data are filtered and feature extracted to obtain a load feature vector. The load characteristic vector, the material hardness parameter, and the processing accuracy requirement are input into a pre-built load trend prediction model, and the load trend vector is output. The load trend vector is input into a pre-built consumption rate distribution model to calculate the cooling water flow rate consumption rate and the lubricant viscosity consumption rate. The cooling water flow rate and lubricant viscosity are obtained by weighting the pipeline pressure and the liquid temperature, and then combining them with the cooling water flow rate and the lubricant viscosity consumption rate. The initial water replenishment control command is generated by weighting the cooling water flow rate, the lubricant viscosity, and the load distribution data, and combining the nozzle angle and the spray coverage area. Based on the initial water replenishment control command, the equipment rotation speed, and the wear data of the grinding wheel, the final water replenishment control command is generated by iteratively optimizing the nozzle angle and the pipeline pressure. According to the final water replenishment control command, the grinding mill liquid supply system is driven to adjust the cooling water flow rate and lubricant viscosity distribution in real time, and output liquid supply status data including actual flow rate, viscosity and distribution status. Based on the liquid supply status data, combined with the load distribution data and fluctuation frequency, the final water replenishment control command is optimized to obtain the optimized water replenishment control command.

2. The intelligent control method for the silicon carbide micro powder grinding mill according to claim 1, characterized in that, The process of filtering and extracting features from the load distribution data, the fluctuation frequency, the equipment rotation speed, and the grinding wheel wear data to obtain a load feature vector includes: Based on the load distribution data, the fluctuation frequency, the equipment rotation speed, and the grinding wheel wear data, filtered data is obtained by wavelet transform filtering. Based on the filtered data, principal component analysis is used to extract features to obtain principal component features. The principal component features are normalized to obtain the load feature vector.

3. The intelligent control method for the silicon carbide micro powder grinding mill according to claim 1, characterized in that, The process of constructing the load trend prediction model includes: Acquire historical load characteristics, historical material hardness parameters, historical processing accuracy requirements, and corresponding historical load data; An initial load trend prediction model was constructed using the support vector machine algorithm. The classification boundary was initialized using the RBF kernel function, and a penalty coefficient was set to determine the tolerance error threshold. Based on the historical load characteristics, the historical material hardness parameters, and the historical processing accuracy requirements, the curvature adjustment parameters of the classification boundary are calculated using the kernel density estimation method, and the curvature of the classification boundary is dynamically adjusted using the RBF kernel function. Based on the historical load data, the penalty coefficient is optimized using a grid search method to determine the error tolerance threshold; Cross-validate the adjusted classification boundaries and calculate the set of classification accuracy and error distribution after boundary adjustment; When the error distribution set exceeds the error tolerance threshold, the gradient descent method is used to re-optimize the curvature parameter of the classification boundary, and the classification boundary is iteratively updated until the error distribution set does not exceed the error tolerance threshold. The optimized classification boundary is matched with the load data in a multidimensional space by using RBF kernel function mapping to determine the hyperplane equation corresponding to each load data. Based on the time series characteristics of the historical load data, the sliding window method is used to extract the dynamic trend parameters of load data changes. The load data is then correlated with the hyperplane equation to construct the final load trend prediction model.

4. The intelligent control method for the silicon carbide micro powder grinding mill according to claim 1, characterized in that, The process of constructing the consumption rate distribution model includes: The historical load trend vector, the corresponding historical cooling water flow rate and historical lubricant viscosity rate are obtained, and the historical load trend vector is cleaned and its features are extracted to obtain trend features. The trend features are input into the input layer of the initially constructed consumption rate distribution model for training, and the predicted cooling water flow consumption rate and the predicted lubricant viscosity consumption rate are obtained from the output layer of the consumption rate distribution model. The predicted cooling water flow rate consumption rate, the predicted lubricant viscosity consumption rate, the historical cooling water flow rate consumption rate, and the historical lubricant viscosity consumption rate are substituted into the loss function to calculate the loss value; The gradient of the output layer of the consumption rate distribution model is calculated based on the loss value, and the gradient is passed forward layer by layer through the chain rule to calculate the gradient of the parameters of each layer and obtain the gradient data. Based on the gradient data and the preset learning rate, the parameters of each layer of the consumption rate distribution model are updated using the gradient descent method. The parameters of each layer are iteratively updated until the number of training iterations of the consumption rate distribution model is greater than a preset number of iterations, or the loss value of the consumption rate distribution model is less than a preset loss threshold. At this point, the training is considered complete, and a trained consumption rate distribution model is obtained.

5. The intelligent control method for the silicon carbide micro powder grinding mill according to claim 1, characterized in that, The step of weighting the cooling water flow rate, lubricant viscosity, and load distribution data, and combining this with the nozzle angle and spray coverage area, to generate an initial water replenishment control command includes: Based on the cooling water flow rate, the lubricant viscosity, and the load distribution data, weights are allocated using the entropy weighting method to obtain weight coefficients. The weighting coefficients, cooling water flow rate, lubricant viscosity, and load distribution data are weighted and fused to obtain the water replenishment frequency; Based on the cooling water flow rate and the load distribution data, the flow rate adjustment range is calculated using a PID algorithm. Based on the water replenishment frequency and the flow rate adjustment range, combined with the nozzle angle and the spray coverage range, the control command is mapped into spatial coverage parameters through a projection transformation matrix algorithm, and the initial water replenishment control command is output.

6. The intelligent control method for the silicon carbide micro powder grinding mill according to claim 1, characterized in that, The step of generating a final water replenishment control command by iteratively optimizing the nozzle angle and the pipeline pressure based on the initial water replenishment control command, the equipment rotation speed, and the abrasive wear data includes: Based on the equipment rotation speed and the wear data of the grinding wheel, rotation speed weight and wear weight are assigned by entropy weight method, and weighted fusion is performed to obtain the operating status parameters; Based on the initial water replenishment control command and the operating status parameters, a preliminary command set is output through iterative optimization using the Q-learning algorithm. Based on the preliminary instruction set, combined with the nozzle angle and the pipeline pressure, the nozzle angle is calibrated through spatial projection transformation, and the pipeline pressure is compensated using a linear interpolation algorithm to generate the final water replenishment control instruction set.

7. The intelligent control method for the silicon carbide micro powder grinding mill according to claim 1, characterized in that, The step of optimizing the final water replenishment control command based on the liquid supply status data, combined with the load distribution data and fluctuation frequency, yields the optimized water replenishment control command, including: Calculate the execution deviation value based on the liquid supply status data and the preset target supply status; No adjustment is required when the execution deviation value is less than or equal to the preset deviation threshold; When the execution deviation value is greater than the preset deviation threshold, the parameters of the Q-learning algorithm are adjusted by combining the load distribution data and the fluctuation frequency, and the final water replenishment control command is optimized to obtain the optimized water replenishment control command.

8. An intelligent control system for a silicon carbide micro powder grinding mill, characterized in that, include: The data acquisition module is used to collect load distribution data, fluctuation frequency, equipment speed, abrasive wear data, material hardness parameters, processing accuracy requirements, pipeline pressure, liquid temperature, nozzle angle and spray coverage. The load feature analysis module is used to filter and extract features from the load distribution data, the fluctuation frequency, the equipment speed, and the wear data of the grinding wheel to obtain a load feature vector. The load trend analysis module is used to input the load feature vector, the material hardness parameter, and the processing accuracy requirement into a pre-built load trend prediction model and output the load trend vector. The consumption rate analysis module is used to input the load trend vector into a pre-built consumption rate distribution model to calculate the cooling water flow consumption rate and the lubricant viscosity consumption rate. The flow rate and viscosity calculation module is used to assign weights based on the pipeline pressure and the liquid temperature, and to perform weighted fusion based on the cooling water flow rate consumption rate and the lubricant viscosity consumption rate to obtain the cooling water flow rate and lubricant viscosity. The initial instruction generation module is used to perform weight allocation based on the cooling water flow rate, the lubricant viscosity and the load distribution data, and generate an initial water replenishment control instruction in combination with the nozzle angle and the spray coverage range; The final instruction generation module is used to generate the final water replenishment control instruction by iteratively optimizing the nozzle angle and the pipeline pressure based on the initial water replenishment control instruction, the equipment rotation speed and the mold wear data. The output control module is used to drive the grinding mill liquid supply system according to the final water replenishment control command, adjust the cooling water flow rate and lubricant viscosity distribution in real time, and output liquid supply status data including actual flow rate, viscosity and distribution status. The control optimization module is used to optimize the final water replenishment control command based on the liquid supply status data, combined with the load distribution data and fluctuation frequency, to obtain the optimized water replenishment control command.

9. An electronic device, characterized in that, The system includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the intelligent control method for a silicon carbide micron powder grinding mill as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform the intelligent control method for the silicon carbide micro powder grinding mill as described in any one of claims 1 to 7.

Images