A tool setting instrument control system for a CNC machining center

Abstract

The present invention discloses a tool setter control system for a numerical control machining center. The system includes: a data acquisition module for collecting a multi-dimensional measurement data set of the tool setter of the numerical control machining center through sensors; a data preprocessing module for generating a standardized multi-dimensional measurement data set; a feature extraction and noise reduction module for constructing an improved variational autoencoder model to obtain a low-dimensional latent feature representation data set and a noise-reduced multi-dimensional reconstructed measurement data set; a calibration optimization module for constructing a tool setter calibration optimization objective function based on the low-dimensional latent feature representation data set and the noise-reduced multi-dimensional reconstructed measurement data set; a global optimization module for determining a globally optimal calibration parameter set; and a calibration control module for making the tool state best match the calibration parameters. The present invention shows high adaptability under various tool types and machining conditions, ensuring the global optimality of the tool setter calibration control parameters.

Description

Technical Field

[0001] The present invention relates to the field of numerical control technology, and particularly to a tool setting instrument control system for a numerical control machining center. Background Art

[0002] With the development of numerical control machining technology, numerical control machining centers are increasingly widely used in the fields of precision manufacturing, aerospace, and automotive manufacturing. As an important measuring tool for numerical control machining centers, the main functions of the tool setting instrument are to measure the position, shape, and wear condition of the tool to ensure machining accuracy and tool service life.

[0003] Currently, the calibration of tool setting instruments for numerical control machining centers mainly relies on static measurement and simple deviation compensation techniques. Traditional calibration methods usually set parameters based on experience and perform calibration control by manually adjusting tool compensation values. Traditional methods can achieve a certain accuracy in an ideal environment, but under conditions such as high-speed cutting, complex machining paths, and dynamic environmental changes, it is often difficult to ensure stable calibration effects. The main problems include:

[0004] During the numerical control machining process, the tool will be affected by factors such as vibration, thermal deformation, and cutting force changes during high-speed rotation and feeding, resulting in a large amount of noise and irrelevant information in the data collected by the measurement sensor. Traditional data processing methods mainly use simple algorithms such as filtering or mean processing, but these methods are difficult to effectively remove non-linear noise and may also cause loss of effective tool feature information, thus affecting the calibration accuracy.

[0005] Existing tool setting instrument calibration systems usually rely on preset compensation parameters and use local optimization strategies to adjust calibration parameters, making it difficult to perform refined adjustments under complex machining conditions. Due to changes in different types of tools, machining materials, and machining environments, a single parameter setting often cannot be applied to all machining tasks, resulting in the accumulation of calibration errors and affecting machining quality.

[0006] Traditional calibration methods usually perform static adjustments before machining. However, during the actual machining process, factors such as tool wear and temperature changes will affect the tool state, causing the original calibration parameters to gradually become invalid. Due to the lack of real-time feedback and dynamic adjustment mechanisms, existing methods cannot perform online calibration during machining, resulting in the continuous accumulation of errors and reduced machining accuracy.

[0007] Currently, some optimization methods attempt to use genetic algorithms or particle swarm optimization algorithms to optimize calibration parameters, but these algorithms are prone to falling into local optima in a high-dimensional optimization space, resulting in calibration results that cannot reach the optimal accuracy. In addition, some methods have a large amount of calculation and slow optimization speed, making it difficult to meet the real-time requirements of numerical control machining centers.

[0008] In view of the bottlenecks in the existing technology, there is an urgent need for a new tool setter control method that can efficiently reduce noise and extract tool state features, automatically optimize and calibrate parameters by combining global optimization methods, and have the ability of dynamic adaptive adjustment, so as to improve the machining accuracy and production efficiency of CNC machining centers. Summary of the Invention

[0009] An object of the present invention is to propose a tool setter control system for a CNC machining center. The present invention shows high adaptability under various tool types and machining conditions, ensuring the global optimality of the calibration control parameters of the tool setter.

[0010] A tool setter control system for a CNC machining center according to an embodiment of the present invention includes the following modules:

[0011] A data acquisition module for collecting a multi-dimensional measurement data set of a tool setter in a CNC machining center through a sensor;

[0012] A data preprocessing module for performing data cleaning, outlier removal, time series alignment and standardization processing on the multi-dimensional measurement data set to generate a standardized multi-dimensional measurement data set;

[0013] A feature extraction and noise reduction module for constructing an improved variational autoencoder model and performing noise reduction processing and feature extraction on the standardized multi-dimensional measurement data set based on the improved variational autoencoder model to obtain a low-dimensional latent feature representation data set and a noise-reduced multi-dimensional reconstructed measurement data set;

[0014] A calibration optimization module for constructing an optimization objective function for tool setter calibration based on the low-dimensional latent feature representation data set and the noise-reduced multi-dimensional reconstructed measurement data set;

[0015] A global optimization module for globally optimizing and searching the optimization objective function for tool setter calibration by using the cuckoo search optimization algorithm to determine a global optimal calibration parameter set;

[0016] A calibration control module for generating a calibration control strategy for a tool setter of a CNC machining center based on the global optimal calibration parameter set and dynamically calibrating and adjusting according to the machining task and the real-time tool state of the tool setter to keep the tool state in the best match with the calibration parameters.

[0017] A tool setter control method for a CNC machining center, applied to a tool setter control system for a CNC machining center, includes the following steps:

[0018] S1. Collect a multi-dimensional measurement data set of a tool setter in a CNC machining center by using a sensor;

[0019] S2. Perform data cleaning, outlier removal, time series alignment and standardization processing on the multi-dimensional measurement data set to form a standardized multi-dimensional measurement data set;

[0020] S3. Construct an improved variational autoencoder model, and perform noise reduction processing and feature extraction on the standardized multi-dimensional measurement data set through the improved variational autoencoder model to obtain a low-dimensional latent feature representation data set for the tool setter calibration of the CNC machining center and a multi-dimensional reconstructed measurement data set after noise reduction;

[0021] S4. Construct an optimization objective function for tool setter calibration based on the low-dimensional latent feature representation data set and the multi-dimensional reconstructed measurement data set after noise reduction;

[0022] S5. Use the cuckoo search optimization algorithm to perform a global search on the optimization objective function for tool setter calibration to determine the global optimal calibration parameter set;

[0023] S6. Generate a calibration control strategy for the tool setter of the CNC machining center according to the global optimal calibration parameter set, and perform dynamic calibration adjustment on the tool setter to achieve the matching control of the tool state of the tool setter and the calibration parameters.

[0024] Optionally, the S1 includes the following steps:

[0025] S11. Measure the tool state of the tool setter in the CNC machining center by using a sensor array, where the sensor array includes a displacement sensor, a vibration sensor, a temperature sensor and an environmental monitoring sensor, respectively collect the displacement signal, acceleration signal, temperature signal and environmental impact signal of the tool, and construct an original multi-dimensional measurement data set ;

[0026] S12. Divide the original multi-dimensional measurement data set into data acquisition windows, set a fixed time window, and group the measurement data in the original multi-dimensional measurement data set in chronological order, so that each time window contains groups of measurement data to form a windowed multi-dimensional measurement data set .

[0027] Optionally, the S2 includes the following steps:

[0028] S21. Perform data cleaning on the windowed multi-dimensional measurement data set to remove missing values, duplicate values and abnormal measurement data, and construct a cleaned multi-dimensional measurement data set;

[0029] S22. Calculate the mean and standard deviation of each measurement signal according to the tool motion state and the distribution characteristics of the environmental impact signal, and remove them according to the set abnormal determination threshold to obtain a multi-dimensional measurement data set after removing abnormal values;

[0030] S23. Align the multi-dimensional measurement data set after removing outliers according to the time stamp, and use the interpolation method to complete the missing data in the multi-dimensional measurement data set to construct a time series-aligned multi-dimensional measurement data set;

[0031] S24. Normalize the time series-aligned multi-dimensional measurement data set according to the standardization rules, and map each measurement signal to the interval , to form a standardized multi-dimensional measurement data set .

[0032] Optionally, the S3 includes the following steps:

[0033] S31. Using the standardized multi-dimensional measurement data set as the input, construct an improved variational autoencoder model for tool depth noise reduction and feature extraction of a CNC machining center :

[0034] ;

[0035] wherein, is the data point in the standardized multi-dimensional measurement data set, is the encoder network of the improved variational autoencoder model, is the encoder network parameter, which is used to adaptively capture the dynamic correlation characteristics between the displacement signal, acceleration signal, temperature signal and environmental influence signal of the tool, is the decoder network of the improved variational autoencoder model, is the decoder network parameter, which is used to reconstruct the noise-reduced measurement data, is the latent feature representation of the standardized multi-dimensional measurement data set;

[0036] S32. Construct a latent feature mapping function that includes a spatial attention mechanism and a temporal attention mechanism to highlight the key signal dimensions and critical moments of the tool state of the CNC machining center, and define the latent feature representation in the latent feature mapping function as:

[0037] ;

[0038] wherein, , respectively represent the mean and variance vectors of the tool state features output by the encoder network, describing the latent feature distribution of the tool state, is a random variable of the standard normal distribution, is the spatial attention factor utilization function, which is used to extract the spatial correlation features between the displacement signal, acceleration signal, temperature signal and environmental influence signal of the tool, is the spatial attention parameter, is a time attention factor utilization function, which is used to capture the sensitivity difference of the tool measurement signal at a preset moment during the machining process. is the time attention parameter, l represents the index of the tool state feature in the spatial dimension, and m represents the index of the tool state feature in the time dimension. and respectively represent the data points in the standardized multi-dimensional measurement data set in the spatial dimension and the time dimension. represents the Hadamard product;

[0039] S33. Construct a decoder reconstruction function that integrates a dynamic error feedback mechanism, using the previous-step reconstruction error real-time feedback during the tool aligner calibration process of the CNC machining center as a dynamic adjustment factor to define the reconstructed data

[0040] ;

[0041] where represents the initial reconstruction output of the decoder network based on the latent feature representation. is the error correction weight matrix. is a learnable parameter. is the current-step reconstruction error. is the dynamic adjustment intensity factor. is the previous-step reconstruction error;

[0042] S34. Construct a depth denoising feature extraction loss function with joint spatial-temporal attention constraints :

[0043] ;

[0044] where is the reconstruction error term, KL is the divergence loss function. is the KL divergence term, which is used to constrain the distribution of the tool state latent feature representation. is the total number of data in the standardized multi-dimensional measurement data set. is the standard normal distribution prior. and are the spatial attention sparsity constraint term and the time attention sparsity constraint term respectively. and are the weight coefficients;

[0045] S35. Perform backpropagation according to the loss function with joint spatial-temporal attention constraints to optimize the encoder network parameters , the decoder network parameters and the spatial attention parameters , Time attention parameter , generate an improved variational autoencoder model for the measurement data of the tool setter of the CNC machining center:

[0046] S36. Use the improved variational autoencoder model to perform deep noise reduction and feature extraction on the standardized multi-dimensional measurement data set, and obtain a low-dimensional latent feature representation data set for calibrating the tool setter of the CNC machining center and the denoised multi-dimensional reconstructed measurement data set .

[0047] Optionally, the S4 includes the following steps:

[0048] S41. Define a set of calibration parameters , and each calibration parameter in the set of calibration parameters is used to describe the tool position compensation, dynamic deviation adjustment, and signal compensation factor involved in the calibration of the tool setter of the CNC machining center;

[0049] S42. Construct a tool setter calibration mapping function , the input of the tool setter calibration mapping function is the low-dimensional latent feature representation data set and the set of calibration parameters , and the output is the predicted measurement data after the tool setter is calibrated :

[0050] ;

[0051] Among them, is the low-dimensional latent feature representation obtained after the th data point is dimensionally reduced by the improved variational autoencoder model, is the mapping weight matrix, is the bias vector, is a function used to characterize the influence of the th calibration parameter on the low-dimensional latent feature representation , is the total number of calibration parameters, represents the jth calibration parameter in the set of calibration parameters;

[0052] S43. Construct a tool setter calibration optimization objective function , and the tool setter calibration optimization objective function aims to minimize the mean square error between the predicted measurement data and the denoised reconstructed multi-dimensional measurement data set :

[0053] ;

[0054] Among them, represents the calibration prediction error of the th data point, is the multi-dimensional measurement data reconstructed after noise reduction, is the regularization coefficient, is the regularization term.

[0055] Optionally, the S5 includes the following steps:

[0056] S51. Initialize the cuckoo population, and define the initial position of each individual in the cuckoo population as the calibration parameter set , and set the population size to . The initialization of the cuckoo population is defined as:

[0057] ;

[0058] where, represents the initial calibration parameter set of the th cuckoo individual, represents the initial value of the th calibration parameter in the th cuckoo individual;

[0059] S52. Evaluate the fitness of each individual in the cuckoo population according to the tool setter calibration optimization objective function to obtain the initial fitness value:

[0060] ;

[0061] where, is the initial fitness value of the th cuckoo individual, reflecting the prediction error and stability of the corresponding calibration parameter set in the tool setter calibration;

[0062] S53. Update the position of each individual in the cuckoo population using the Lévy flight mechanism to obtain the updated calibration parameter set :

[0063] ;

[0064] where, is the calibration parameter set of the th cuckoo individual updated in the th generation, is the calibration parameter set of the th cuckoo individual updated in the th generation, representing the updated value of the tool setter calibration parameter of the CNC machining center, is the step size factor for position update, represents the random flight step vector obeying the Lévy distribution, is the calibration parameter set of the individual with the optimal fitness in the th generation population;

[0065] Recalculate the fitness value with the updated set of calibration parameters ; ;

[0066] S55. According to the principle of selecting the best in the cuckoo search optimization algorithm, perform fitness comparison and replacement within the population, and update the optimal positions of the population individuals :

[0067] ;

[0068] S56. Execute iterative search until the set termination condition is met, and obtain the globally optimal set of calibration parameters :

[0069] ;

[0070] wherein is the iteration termination time, is the calibration optimization objective function value corresponding to the q-th cuckoo individual in the T-th iteration

[0071] Optionally, the S6 includes the following steps:

[0072] S61. Extract the calibration control strategy from the globally optimal set of calibration parameters, and set the calibration strategy category of the tool setter of the CNC machining center according to the tool displacement compensation, tool wear correction, machining environment interference correction, and dynamic error adjustment parameters:

[0073] Static calibration strategy, applicable to standard working conditions, the tool displacement compensation parameter is within the normal range, the tool wear correction parameter ≤ the preset threshold, and the dynamic error adjustment parameter is in a stable state;

[0074] Adaptive calibration strategy, applicable to working conditions where the machining environment changes greater than the threshold or the tool wear rate is greater than the threshold. When the tool displacement compensation parameter exceeds the normal range or the tool wear correction parameter is higher than the preset threshold, the calibration control strategy is adjusted to the adaptive mode;

[0075] Emergency calibration strategy, applicable to working conditions such as abnormal tool vibration, decreased machining accuracy, or error exceeding the limit. When the dynamic error adjustment parameter is higher than the set warning value, the calibration control strategy is adjusted to the emergency correction mode and the real-time compensation mechanism is triggered;

[0076] S62. Construct a calibration control matrix based on the tool status of the tool setter of the CNC machining center and the type of machining task, classify the tool calibration accuracy requirements in combination with the tool calibration historical data, and dynamically match the optimal calibration parameters during task switching;

[0077] S63. Real-time monitor the tool displacement signal, acceleration signal, temperature signal and environmental impact signal. When the measured data deviates from the historical calibration mean by more than the preset error range, adaptively adjust the calibration control strategy to make the calibration parameters adapt to the dynamic changes of the machining environment;

[0078] S64. Adopt a feedback optimization mechanism. After performing the calibration adjustment, calculate the calibration error based on the real-time measured data and update the calibration control strategy. When the calibration error is lower than the set threshold, keep the current calibration parameters. If the error exceeds the limit, dynamically correct the calibration parameters and optimize the calibration control matrix to achieve the calibration control of the tool setter of the CNC machining center.

[0079] The beneficial effects of the present invention are as follows:

[0080] (1) In the preprocessing of the measurement data of the tool setter of the CNC machining center, the present invention constructs an improved variational autoencoder model combined with a spatio-temporal attention mechanism to effectively remove the noise interference in the machining environment and extract the key features of the tool status. The improved variational autoencoder uses a spatial attention mechanism to allocate feature weights to the displacement, vibration, temperature and environmental impact signals of the tool, enhancing the ability to extract key tool status information. At the same time, combined with the temporal attention mechanism, by dynamically adjusting the importance weights of data at different time points, the time series modeling ability of the model in a complex machining environment is improved.

[0081] (2) The present invention introduces the cuckoo search optimization algorithm, enhances the global search ability through the Lévy flight mechanism, and introduces an adaptive step size control strategy during the calibration parameter update process, enabling the algorithm to maintain a larger search range in the early exploration stage to avoid falling into local optimal solutions, and using a high-precision local search mechanism in the later stage to improve the optimization convergence speed. The calibration parameter convergence speed has been improved, and it shows high adaptability under various tool types and machining conditions, ensuring the global optimality of the tool setter calibration control parameters. Description of the Drawings

[0082] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0083] Figure 1 is a flowchart of a control method for a tool setter of a CNC machining center proposed by the present invention. Detailed Embodiments

[0084] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0085] Reference Figure 1 , a tool setter control system for a CNC machining center, includes the following modules:

[0086] A data acquisition module, configured to collect a multi-dimensional measurement data set of a tool setter in a CNC machining center through sensors;

[0087] A data preprocessing module, configured to perform data cleaning, outlier removal, time series alignment, and standardization processing on the multi-dimensional measurement data set to generate a standardized multi-dimensional measurement data set;

[0088] A feature extraction and noise reduction module, configured to construct an improved variational autoencoder model, and perform noise reduction processing and feature extraction on the standardized multi-dimensional measurement data set based on the improved variational autoencoder model to obtain a low-dimensional latent feature representation data set and a noise-reduced multi-dimensional reconstructed measurement data set;

[0089] A calibration optimization module, configured to construct an optimization objective function for tool setter calibration based on the low-dimensional latent feature representation data set and the noise-reduced multi-dimensional reconstructed measurement data set;

[0090] A global optimization module, configured to perform global optimization search on the optimization objective function for tool setter calibration using the cuckoo search optimization algorithm to determine a global optimal calibration parameter set;

[0091] A calibration control module, configured to generate a calibration control strategy for the tool setter of the CNC machining center based on the global optimal calibration parameter set, and perform dynamic calibration adjustment according to the machining task and real-time tool state of the tool setter to keep the tool state in the best match with the calibration parameters.

[0092] A tool setter control method for a CNC machining center, applied to a tool setter control system for a CNC machining center, includes the following steps:

[0093] S1. Collect a multi-dimensional measurement data set of a tool setter in a CNC machining center using sensors;

[0094] S2. Perform data cleaning, outlier removal, time series alignment, and standardization processing on the multi-dimensional measurement data set to form a standardized multi-dimensional measurement data set;

[0095] S3. Construct an improved variational autoencoder model, and perform noise reduction processing and feature extraction on the standardized multi-dimensional measurement data set through the improved variational autoencoder model to obtain a low-dimensional latent feature representation data set for tool setter calibration of the CNC machining center and a noise-reduced multi-dimensional reconstructed measurement data set;

[0096] S4. Construct an optimized objective function for tool setter calibration based on the low-dimensional latent feature representation dataset and the denoised multi-dimensional reconstruction measurement dataset;

[0097] S5. Use the cuckoo search optimization algorithm to globally search the optimized objective function for tool setter calibration and determine the global optimal calibration parameter set;

[0098] S6. Generate a calibration control strategy for the tool setter of the CNC machining center according to the global optimal calibration parameter set, and perform dynamic calibration adjustment on the tool setter to achieve matching control of the tool state and calibration parameters of the tool setter.

[0099] In this embodiment, S1 includes the following steps:

[0100] S11. Measure the tool state of the tool setter in the CNC machining center using a sensor array. The sensor array includes a displacement sensor, a vibration sensor, a temperature sensor, and an environmental monitoring sensor, and respectively collect the displacement signal, acceleration signal, temperature signal, and environmental impact signal of the tool to construct an original multi-dimensional measurement dataset ;

[0101] S12. Divide the original multi-dimensional measurement dataset into data acquisition windows, set a fixed time window, and group the measurement data in the original multi-dimensional measurement dataset according to the time sequence, so that each time window contains groups of measurement data to form a windowed multi-dimensional measurement dataset .

[0102] In this embodiment, S2 includes the following steps:

[0103] S21. Clean the windowed multi-dimensional measurement dataset to remove missing values, duplicate values, and abnormal measurement data, and construct a cleaned multi-dimensional measurement dataset;

[0104] S22. Calculate the mean and standard deviation of each measurement signal according to the tool motion state and the distribution characteristics of the environmental impact signal, and remove them according to the set abnormal determination threshold to obtain a multi-dimensional measurement dataset after removing abnormal values;

[0105] S23. Align the time series of the multi-dimensional measurement dataset after removing abnormal values according to the timestamp, and use the interpolation method to fill in the missing data in the multi-dimensional measurement dataset to construct a time series-aligned multi-dimensional measurement dataset;

[0106] S24. Normalize the time series-aligned multi-dimensional measurement dataset according to the standardization rule, and map each measurement signal to the interval to form a standardized multi-dimensional measurement dataset 。

[0107] In this embodiment, S3 includes the following steps:

[0108] S31. Using the standardized multi-dimensional measurement data set as the input, construct an improved variational autoencoder model for depth noise reduction and feature extraction of the tool setter of the CNC machining center :

[0109] ;

[0110] Among them, is the data point in the standardized multi-dimensional measurement data set, is the encoder network of the improved variational autoencoder model, is the encoder network parameter, which is used to adaptively capture the dynamic correlation characteristics between the displacement signal, acceleration signal, temperature signal and environmental influence signal of the tool, is the decoder network of the improved variational autoencoder model, is the decoder network parameter, which is used to reconstruct the noise-reduced measurement data, is the latent feature representation of the standardized multi-dimensional measurement data set;

[0111] S32. Construct a latent feature mapping function that includes a spatial attention mechanism and a temporal attention mechanism to highlight the key signal dimensions and critical moments of the tool state of the CNC machining center, and define the latent feature representation in the latent feature mapping function as:

[0112] ;

[0113] Among them, , respectively represent the mean and variance vectors of the tool state features output by the encoder network, describing the latent feature distribution of the tool state, is a random variable of the standard normal distribution, is the spatial attention factor utilization function, which is used to extract the spatial correlation features between the displacement signal, acceleration signal, temperature signal and environmental influence signal of the tool, is the spatial attention parameter, is the temporal attention factor utilization function, which is used to capture the sensitivity difference of the tool measurement signal at the preset moment during the machining process, is the temporal attention parameter, l represents the index of the tool state feature in the spatial dimension, m represents the index of the tool state feature in the temporal dimension, and respectively represent the data points in the standardized multi-dimensional measurement data set in the spatial dimension and the temporal dimension, denotes the Hadamard product;

[0114] The formula is used to calculate the potential feature representation of the tool state. By introducing a spatial attention mechanism and a temporal attention mechanism, the feature extraction ability of the variational autoencoder in a complex machining environment is improved. The spatial attention is used to highlight the influence of the key measurement signal dimensions, and the temporal attention is used to enhance the feature expression ability at specific moments (the moment when the tool first contacts the workpiece and the moment when cutting ends).

[0115] S33. Construct a decoder reconstruction function incorporating a dynamic error feedback mechanism, using the previous-step reconstruction error real-time feedback during the tool setter calibration process in a CNC machining center as a dynamic adjustment factor to define the reconstructed data

[0116] ;

[0117] where represents the initial reconstruction output of the decoder network based on the latent feature representation, is the error correction weight matrix, is a learnable parameter, is the current-step reconstruction error, is the dynamic adjustment intensity factor, is the previous-step reconstruction error;

[0118] The formula introduces a dynamic error feedback mechanism in the decoder stage, dynamically adjusting the current reconstructed data through the previous-step reconstruction error, optimizing the reconstruction error as the calibration parameters change, and improving the reconstruction accuracy of the tool measurement data.

[0119] S34. Construct a depth denoising feature extraction loss function with joint spatial-temporal attention constraints :

[0120] ;

[0121] where is the reconstruction error term, KL is the divergence loss function, is the KL divergence term, used to constrain the distribution of the latent feature representation of the tool state, is the total number of data in the standardized multi-dimensional measurement dataset, is the prior of the standard normal distribution, and are the spatial attention sparsity constraint term and the temporal attention sparsity constraint term respectively, and are the weight coefficients;

[0122] The formula is used to optimize the loss function of the VAE model. By combining the reconstruction error term, KL divergence loss, and attention sparsity constraint term, it ensures that the model can efficiently denoise and extract the key features of the tool state.

[0123] S35. Perform backpropagation according to the loss function with joint space-time attention constraints to optimize the encoder network parameters , the decoder network parameters and the spatial attention parameters , the temporal attention parameters , to generate an improved variational autoencoder model for the measurement data of the tool setter in the CNC machining center:

[0124] S36. Use the improved variational autoencoder model to perform deep denoising and feature extraction on the standardized multi-dimensional measurement data set, and obtain a low-dimensional latent feature representation data set for calibrating the tool setter in the CNC machining center and the denoised multi-dimensional reconstructed measurement data set .

[0125] In this embodiment, during the preprocessing of the measurement data of the tool setter in the CNC machining center, an improved variational autoencoder model combined with a space-time attention mechanism is constructed, which is used to effectively remove the noise interference in the machining environment and extract the key features of the tool state. The improved variational autoencoder uses the spatial attention mechanism to allocate feature weights to the displacement, vibration, temperature, and environmental impact signals of the tool, enhancing the ability to extract key tool state information. At the same time, combined with the temporal attention mechanism, by dynamically adjusting the importance weights of data at different time points, the time series modeling ability of the model in a complex machining environment is improved.

[0126] In this embodiment, S4 includes the following steps:

[0127] S41. Define the calibration parameter set , and each calibration parameter in the calibration parameter set is used to describe the tool position compensation, dynamic deviation adjustment, and signal compensation factor involved in the calibration of the tool setter in the CNC machining center;

[0128] S42. Construct the tool setter calibration mapping function , the input of the tool setter calibration mapping function is the low-dimensional latent feature representation data set and the calibration parameter set , and the output is the predicted measurement data after calibrating the tool setter :

[0129] ;

[0130] Among them, is the The low-dimensional latent feature representation obtained after dimensionality reduction of is the mapping weight matrix, is the bias vector, is used to characterize the influence of the th calibration parameter on the low-dimensional latent feature representation function, is the total number of calibration parameters, represents the jth calibration parameter in the calibration parameter set;

[0131] The formula is used to construct the mapping relationship of tool calibration, mapping the low-dimensional feature representation data set and the calibration parameter set into the calibrated tool measurement data to ensure the accuracy of tool calibration.

[0132] S43. Construct the calibration optimization objective function of the tool setter , and the calibration optimization objective function of the tool setter aims to minimize the mean square error between the predicted measurement data and the denoised reconstructed multi-dimensional measurement data set :

[0133] ;

[0134] Among them, represents the calibration prediction error of the th data point, is the denoised reconstructed multi-dimensional measurement data, is the regularization coefficient, is the regularization term.

[0135] The formula is used to construct the calibration optimization objective function to minimize the error between the calibrated tool measurement data and the denoised measurement data to ensure the high precision of tool calibration.

[0136] In this embodiment, by constructing the calibration optimization objective function of the tool setter and combining the low-dimensional latent feature representation data set and the denoised multi-dimensional reconstructed measurement data set, accurate modeling of the tool setter calibration parameters is achieved, which can effectively reduce the calibration error, improve the stability of the calibration parameters, avoid the error accumulation caused by manually setting parameters in the traditional method, the optimized model can adapt to different tools and machining conditions, improve the calibration accuracy while reducing the calibration time, and enhance the intelligence and adaptability of the tool setter of the CNC machining center.

[0137] In this embodiment, S5 includes the following steps:

[0138] S51. Initialize the cuckoo population, and define the initial position of the cuckoo population individuals as the calibration parameter set , set the population size to , and the initialization of the cuckoo population is defined as:

[0139] ;

[0140] Among them, represents the initial calibration parameter set of the -th cuckoo individual, represents the initial value of the -th calibration parameter in the -th cuckoo individual;

[0141] S52. Evaluate the fitness of each individual in the cuckoo population according to the tool setter calibration optimization objective function to obtain the initial fitness value:

[0142] ;

[0143] Among them, is the initial fitness value of the -th cuckoo individual, reflecting the prediction error and stability of the corresponding calibration parameter set in the tool setter calibration;

[0144] S53. Update the position of each individual in the cuckoo population using the Lévy flight mechanism to obtain the updated calibration parameter set :

[0145] ;

[0146] Among them, is the calibration parameter set of the -th generation updated -th cuckoo individual, is the calibration parameter set of the -th generation updated -th cuckoo individual, representing the updated value of the tool setter calibration parameter of the CNC machining center, is the step size factor for position update, represents the random flight step vector obeying the Lévy distribution, is the calibration parameter set of the individual with the optimal fitness in the -th generation population;

[0147] The formula is used to update the position of the cuckoo individual, guiding the search through the Lévy flight mechanism, ensuring the search ability for the global optimal solution, and avoiding falling into the local optimum.

[0148] S54. Recalculate the fitness value with the updated calibration parameter set ;

[0149] S55. According to the principle of selecting the best in the cuckoo search optimization algorithm, perform fitness comparison and replacement within the population, and update the optimal position of the population individuals. :

[0150] ;

[0151] S56. Perform iterative search until the set termination condition is met, and obtain the global optimal calibration parameter set. :

[0152] ;

[0153] Among them, is the iteration termination time, is the calibration optimization objective function value corresponding to the q-th cuckoo individual in the T-th iteration.

[0154] In this embodiment, the cuckoo search optimization algorithm is introduced, the global search ability is enhanced through the Levy flight mechanism, and an adaptive step size control strategy is introduced in the process of updating the calibration parameters, so that the algorithm maintains a large search range in the early exploration stage, avoiding falling into local optimal solutions, and using a high-precision local search mechanism in the later stage to improve the optimization convergence speed. The convergence speed of the calibration parameters is improved, and it shows high adaptability under various tool types and machining conditions, ensuring the global optimality of the tool setter calibration control parameters.

[0155] In this embodiment, S6 includes the following steps:

[0156] S61. Extract the calibration control strategy from the global optimal calibration parameter set, and set the calibration strategy category of the tool setter of the CNC machining center according to the tool displacement compensation, tool wear correction, machining environment interference correction, and dynamic error adjustment parameters:

[0157] Static calibration strategy, applicable to standard working conditions, the tool displacement compensation parameter is within the normal range, the tool wear correction parameter ≤ the preset threshold, and the dynamic error adjustment parameter is in a stable state;

[0158] Adaptive calibration strategy, applicable to working conditions where the machining environment change is greater than the threshold or the tool wear speed is greater than the threshold. When the tool displacement compensation parameter exceeds the normal range or the tool wear correction parameter is higher than the preset threshold, the calibration control strategy is adjusted to the adaptive mode;

[0159] Emergency calibration strategy, applicable to working conditions such as abnormal tool vibration, decreased machining accuracy, or error exceeding the limit. When the dynamic error adjustment parameter When it is higher than the set warning value, the calibration control strategy is adjusted to the emergency correction mode, and the real-time compensation mechanism is triggered;

[0160] S62. Construct a calibration control matrix based on the tool state and machining task category of the tool setter of the CNC machining center, classify the tool calibration accuracy requirements in combination with the tool calibration historical data, and dynamically match the optimal calibration parameters during task switching;

[0161] S63. Real-time monitor the tool displacement signal, acceleration signal, temperature signal and environmental impact signal. When the measured data deviates from the historical calibration mean by more than the preset error range, adaptively adjust the calibration control strategy to make the calibration parameters adapt to the dynamic changes of the machining environment;

[0162] S64. Adopt a feedback optimization mechanism. After performing the calibration adjustment, calculate the calibration error based on the real-time measured data and update the calibration control strategy. When the calibration error is lower than the set threshold, maintain the current calibration parameters. If the error exceeds the limit, dynamically correct the calibration parameters and optimize the calibration control matrix to achieve the calibration control of the tool setter of the CNC machining center.

[0163] This embodiment realizes the dynamic calibration adjustment of the tool setter through the calibration control strategy based on the global optimal calibration parameter set, ensures the real-time matching of the tool state and calibration parameters, can adaptively adjust the tool offset compensation and error correction, effectively reduces the tool measurement error, improves the stability and accuracy of tool calibration, thereby improving the machining consistency and production efficiency of the CNC machining center, reducing human intervention, improving the automation level, and is applicable to high-precision manufacturing environments.

[0164] Example 1:

[0165] In the CNC machining workshop of an aviation parts manufacturing plant, when a Haas UMC-750 five-axis machining center was batch machining aviation engine blades, the operator found that the tool life was much lower than expected and the size deviation of the machined blades frequently exceeded the tolerance range. After preliminary inspection, it was found that the problem might be related to the insufficient calibration accuracy of the tool setter. The traditional calibration method could not meet the dynamically changing machining environment, resulting in the continuous accumulation of machining errors. In order to optimize the calibration process, the enterprise decided to apply the method of the present invention to improve the tool calibration accuracy, reduce errors and improve production efficiency.

[0166] On March 15, 2024, Technician Li in the workshop noticed during the production process that for a batch of blades processed with a φ12mm cemented carbide ball end mill, their dimensional deviations all exceeded the tolerance range (within ±5μm). Further analysis found that the traditional tool setter calibration method relied on fixed offset compensation parameters and could not be adjusted in real time during tool wear, resulting in the continuous accumulation of calibration errors. Technician Li decided to use the method of the present invention for experiments and enabled the improved variational autoencoder model and cuckoo search optimization algorithm in the control system for tool calibration optimization.

[0167] First, in the control system of the CNC machining center, enable the data acquisition module to record the displacement, vibration, temperature, and environmental impact signals of the tool during calibration in real time, sampling every 100ms to form a measurement data set. From 10:00 to 12:00 on March 15, 2024, a total of 30,000 measurement data were collected. The data is shown as follows in the embodiment:

[0168]

[0169] Input the collected data into the improved variational autoencoder model of the present invention for noise reduction processing to remove non-linear noise interference. After the model runs for 2.5 seconds, a low-dimensional latent feature data set is generated, and the denoised reconstructed measurement data is output. In the embodiment, for the data collected by tool No. 04 at 10:00:04, the displacement signal in the original data was greatly affected by vibration noise. After noise reduction, the signal-to-noise ratio increased by 18.6%, and the accuracy of tool state feature extraction increased by 14.3%, providing high-quality input data for subsequent calibration optimization.

[0170] Subsequently, the system starts the cuckoo search optimization algorithm to optimize the calibration parameters from 10:05 to 10:20, performing a total of 100 rounds of search iterations, and finally obtaining the optimal set of calibration parameters. During the optimization process, the algorithm automatically adjusts the tool offset compensation amount, temperature compensation factor, and dynamic error correction coefficient, reducing the final calibration error to within ±1.2μm. After optimization, the calibration error is reduced by 68.4%, and the parameter convergence time is reduced by 43.3%, significantly improving the calibration accuracy of the tool setter.

[0171] After completing the parameter optimization, the system activates the adaptive calibration adjustment function and enters the actual machining process at 10:30. According to the dynamic monitoring data, the system detects that the tool wear exceeds the preset threshold (the tool offset increases by 2.4μm) after 2 hours, and automatically adjusts the compensation parameters to avoid the further accumulation of calibration errors.

[0172] After continuous machining for 6 hours, the changes in calibration accuracy when using the method of the present invention and the traditional method are compared as follows:

[0173]

[0174] It can be seen that during the long-term processing, the error of the traditional method gradually increases, while the method of the present invention can adjust the calibration parameters in real time, keeping the error always within ±2.1 μm, thus improving the long-term stability of tool calibration.

[0175] Through a 3-day experiment, a total of 120 blade processing tasks were completed. By comparing the processing efficiency and qualification rate of using the traditional method and the method of the present invention, the experimental results show that after adopting the method of the present invention, the tool calibration time is shortened by 43.0%, and the processing qualification rate is increased by 6.2%, improving the production efficiency of the CNC machining center.

[0176] This experiment verified the effectiveness of the method of the present invention in a real production environment. Compared with the traditional method, the present invention has significant advantages in the following aspects:

[0177] It has strong data noise reduction ability, increasing the signal-to-noise ratio by 18.6% and ensuring the reliability of tool state characteristics;

[0178] It optimizes the calibration parameters, reducing the calibration error by 68.4% and the convergence time by 43.3%;

[0179] It supports dynamic calibration adjustment, and during the 6-hour long-term processing, the error is always controlled within ±2.1 μm;

[0180] It improves the production efficiency, shortening the tool calibration time by 43.0% and increasing the processing qualification rate by 6.2%.

[0181] In summary, the method of the present invention can significantly improve the calibration accuracy and stability of the tool aligner of the CNC machining center in practical applications, is applicable to high-precision manufacturing scenarios, and has broad industrial application value.

[0182] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes, should be covered by the protection scope of the present invention.

Claims

1. A tool setter control method for a CNC machining center, including a control system, characterized in that, The control system includes the following modules: A data acquisition module, which is used to collect a multi-dimensional measurement data set of the tool setter in the CNC machining center through sensors; A data preprocessing module, which is used to perform data cleaning, outlier removal, time series alignment and normalization processing on the multi-dimensional measurement data set to generate a normalized multi-dimensional measurement data set; A feature extraction and noise reduction module, which is used to construct an improved variational autoencoder model, and perform noise reduction processing and feature extraction on the normalized multi-dimensional measurement data set based on the improved variational autoencoder model to obtain a low-dimensional latent feature representation data set and a denoised multi-dimensional reconstructed measurement data set; A calibration optimization module, which is used to construct an optimization objective function for tool setter calibration based on the low-dimensional latent feature representation data set and the denoised multi-dimensional reconstructed measurement data set; A global optimization module, which is used to perform a global optimization search on the optimization objective function for tool setter calibration using the cuckoo search optimization algorithm to determine a global optimal calibration parameter set; A calibration control module, which is used to generate a calibration control strategy for the tool setter in the CNC machining center based on the global optimal calibration parameter set, and perform dynamic calibration adjustment according to the machining task and real-time tool state of the tool setter to make the tool state best match the calibration parameters; It also includes the following steps: S1. Use sensors to collect a multi-dimensional measurement data set of the tool setter in the CNC machining center; S2. Perform data cleaning, outlier removal, time series alignment and normalization processing on the multi-dimensional measurement data set to form a normalized multi-dimensional measurement data set; S3. Construct an improved variational autoencoder model, and perform noise reduction processing and feature extraction on the normalized multi-dimensional measurement data set through the improved variational autoencoder model to obtain a low-dimensional latent feature representation data set for tool setter calibration in the CNC machining center and a denoised multi-dimensional reconstructed measurement data set; S4. Construct an optimization objective function for tool setter calibration based on the low-dimensional latent feature representation data set and the denoised multi-dimensional reconstructed measurement data set; S5. Use the cuckoo search optimization algorithm to perform a global search on the optimization objective function for tool setter calibration to determine a global optimal calibration parameter set; S6. Generate a calibration control strategy for the tool setter in the CNC machining center according to the global optimal calibration parameter set, and perform dynamic calibration adjustment on the tool setter to realize the matching control of the tool state of the tool setter and the calibration parameters; The S1 includes the following steps: S11. Measure the tool state of the tool setter in the CNC machining center using a sensor array. The sensor array includes a displacement sensor, a vibration sensor, a temperature sensor, and an environmental monitoring sensor, which respectively collect the displacement signal, acceleration signal, temperature signal, and environmental impact signal of the tool, and construct an original multi-dimensional measurement data set ; S12. Divide the original multi-dimensional measurement data set to obtain data acquisition windows, set a fixed time window, and group the measurement data in the original multi-dimensional measurement data set in chronological order, so that each time window contains groups of measurement data, forming a windowed multi-dimensional measurement data set ; The S2 includes the following steps: S21. Perform data cleaning on the windowed multi-dimensional measurement data set to remove missing values, duplicate values, and abnormal measurement data, and construct a cleaned multi-dimensional measurement data set; S22. According to the tool motion state and the distribution characteristics of the environmental impact signal, calculate the mean and standard deviation of each measurement signal, and perform rejection according to the set outlier determination threshold to obtain a multi-dimensional measurement data set after outlier removal; S23. Align the time series of the multi-dimensional measurement data set after outlier removal according to the time stamp, use the interpolation method to fill in the missing data in the multi-dimensional measurement data set, and construct a multi-dimensional measurement data set with time series alignment; S24. Normalize the multi-dimensional measurement data set aligned with the time series according to the standardization rules, and map each measurement signal to the interval , to form a standardized multi-dimensional measurement data set ; The S3 includes the following steps: S31. Using a standardized multi-dimensional measurement data set as input, an improved variational autoencoder model for depth noise reduction and feature extraction of the tool setter of a CNC machining center is constructed : ； Among them, is a data point in the standardized multi-dimensional measurement data set, is the encoder network of the improved variational autoencoder model, are the encoder network parameters, used to adaptively capture the dynamic correlation characteristics among the displacement signal, acceleration signal, temperature signal and environmental influence signal of the tool, is the decoder network of the improved variational autoencoder model, are the decoder network parameters, used to reconstruct the denoised measurement data, is the latent feature representation of the standardized multi-dimensional measurement data set; S32. Construct a potential feature mapping function that includes a spatial attention mechanism and a temporal attention mechanism to highlight the key signal dimensions and critical moments of the tool state of the CNC machining center, and define the potential feature representation in the potential feature mapping function as follows: ； Among them, , respectively represent the mean and variance vectors of the tool state features output by the encoder network, describing the potential feature distribution of the tool state, is a random variable of the standard normal distribution, is the spatial attention factor utilization function, used to extract the spatial correlation features between the displacement signal, acceleration signal, temperature signal and environmental impact signal of the tool, is the spatial attention parameter, is the temporal attention factor utilization function, used to capture the sensitivity difference of the tool measurement signal at the preset moment during the machining process, is the temporal attention parameter, l represents the index of the tool state feature in the spatial dimension, m represents the index of the tool state feature in the temporal dimension, and respectively represent the data points in the standardized multi-dimensional measurement data set in the spatial dimension and the temporal dimension, represents the Hadamard product; S33. Construct a decoder reconstruction function that incorporates a dynamic error feedback mechanism, using the previous reconstruction error real-time feedback during the calibration process of the tool setter of the CNC machining center as the dynamic adjustment factor, and define the reconstruction data : ； Among them, represents the initial reconstruction output of the decoder network based on the latent feature representation, is the error correction weight matrix, is a learnable parameter, is the reconstruction error at the current step, is the dynamic adjustment intensity factor, is the reconstruction error at the previous step; S34. Construct a depth denoising feature extraction loss function with joint space-time attention constraints : ； Among them, is the reconstruction error term, KL is the divergence loss function, is the KL divergence term, which is used to constrain the distribution of the potential feature representation of the tool state, is the total number of data in the standardized multi-dimensional measurement dataset, is the prior of the standard normal distribution, and are the spatial attention sparse constraint term and the temporal attention sparse constraint term respectively, and are the weight coefficients; Backpropagation is performed according to the loss function constrained by joint spatial-temporal attention to optimize the encoder network parameters , the decoder network parameters and the spatial attention parameters , the temporal attention parameters , to generate an improved variational autoencoder model for the tool setting instrument measurement data of the CNC machining center: S36. Use the improved variational autoencoder model to perform deep noise reduction and feature extraction on the standardized multi-dimensional measurement data set, and obtain a low-dimensional latent feature representation data set for the tool setter calibration of the CNC machining center and the multi-dimensional reconstructed measurement data set after noise reduction .

2. The control method of the tool setter of a CNC machining center according to claim 1, wherein The S4 includes the following steps: S41. Define a set of calibration parameters , each calibration parameter in the set of calibration parameters is used to describe the tool position compensation, dynamic deviation adjustment, and signal compensation factor involved in the calibration of the tool setter of the CNC machining center; S42. Construct a tool setter calibration mapping function , where the input of the tool setter calibration mapping function is a low-dimensional latent feature representation dataset and a set of calibration parameters , and the output is the predicted measurement data after tool setter calibration : ； Among them, is the low-dimensional latent feature representation obtained after the dimensionality reduction of the th data point through the improved variational autoencoder model, is the mapping weight matrix, is the bias vector, is a function used to characterize the influence of the th calibration parameter on the low-dimensional latent feature representation , is the total number of calibration parameters, represents the jth calibration parameter in the set of calibration parameters; S43. Construct the calibration optimization objective function of the tool setter , and the calibration optimization objective function of the tool setter aims to minimize the mean square error between the predicted measurement data and the reconstructed multi-dimensional measurement data set after noise reduction : ； Among them, represents the calibration prediction error of the th data point, is the multi-dimensional measurement data reconstructed after noise reduction, is the regularization coefficient, is the regularization term.

3. A tool setter control method for a CNC machining center according to claim 2, characterized in that The S5 includes the following steps: S51. Initialize the cuckoo population and define the initial positions of the individuals in the cuckoo population as the calibration parameter set , set the population size to , and the initialization of the cuckoo population is defined as: ； Among them, represents the initial calibration parameter set of the th cuckoo individual, represents the initial value of the th calibration parameter among the th cuckoo individuals; S52. Optimize the objective function according to the tool setter calibration Evaluate the fitness of each individual in the cuckoo population to obtain the initial fitness value: ； Among them, is the initial fitness value of the th cuckoo individual, reflecting the prediction error and stability of the corresponding calibration parameter set in the tool setter calibration; S53. Update the position of each individual in the cuckoo population by using the Lévy flight mechanism to obtain an updated set of calibration parameters : ； Among them, is the calibration parameter set of the th updated cuckoo individual of the th generation, is the calibration parameter set of the th updated cuckoo individual of the th generation, representing the updated value of the calibration parameter of the tool setter of the CNC machining center, is the step factor of position update, represents the random flight step vector subject to the Levy distribution, is the calibration parameter set of the individual with the optimal fitness in the th generation population; Recalculate the fitness value with the updated set of calibration parameters again ; According to the principle of selecting the best in the cuckoo search optimization algorithm, perform fitness comparison and replacement within the population, and update the optimal positions of the individuals in the population : ； S56. Perform iterative search until a set termination condition is met to obtain a globally optimal set of calibration parameters : ； Among them, is the iteration termination time, is the calibrated optimization objective function value corresponding to the q-th cuckoo individual in the T-th iteration.

4. A tool setter control method for a CNC machining center according to claim 3, characterized in that, The S6 includes the following steps: S61. Extract the calibration control strategy from the set of globally optimal calibration parameters, and set the calibration strategy category of the tool setter of the CNC machining center according to the tool displacement compensation, tool wear correction, machining environment interference correction, and dynamic error adjustment parameters: Static calibration strategy, applicable to standard working conditions, tool displacement compensation parameters are within the normal range, tool wear correction parameters ≤ preset threshold, dynamic error adjustment parameters are in a stable state; Adaptive calibration strategy, applicable to working conditions where the change in the machining environment is greater than the threshold or the tool wear rate is greater than the threshold. When the tool displacement compensation parameter exceeds the normal range or the tool wear correction parameter is higher than the preset threshold, the calibration control strategy is adjusted to the adaptive mode; Emergency calibration strategy, applicable to working conditions such as abnormal tool vibration, decreased machining accuracy or exceeded error limits. When the dynamic error adjustment parameter is higher than the set warning value, the calibration control strategy is adjusted to the emergency correction mode, and the real-time compensation mechanism is triggered; S62. Construct a calibration control matrix based on the tool state of the tool setter of the CNC machining center and the machining task category, classify the tool calibration accuracy requirements in combination with the tool calibration historical data, and dynamically match the optimal calibration parameters during task switching; S63. Monitor the tool displacement signal, acceleration signal, temperature signal, and environmental impact signal in real time. When the measured data deviates from the historical calibration mean by more than the preset error range, adaptively adjust the calibration control strategy to make the calibration parameters adapt to the dynamic changes of the machining environment; S64. Adopt a feedback optimization mechanism. After performing the calibration adjustment, calculate the calibration error based on the real-time measured data and update the calibration control strategy. When the calibration error is lower than the set threshold, maintain the current calibration parameters. If the error exceeds the limit, dynamically correct the calibration parameters and optimize the calibration control matrix to achieve the calibration control of the tool setter of the CNC machining center.

Images