Self-adaptive clamping method and device for tool holder of machine tool and computer program product

Abstract

The invention relates to the technical field of machine tool cutter clamping, and discloses a machine tool cutter holder self-adaptive clamping method and device and a computer program product, and the method comprises the steps: solving a corrected kinetic equation through a fourth-order Runge-Kutta method, generating simulation data in different clamping states, and constructing a source domain data set; clamping data of the cutter handle in different clamping states are collected, and an actual measurement data set is constructed; a CS-GWO optimization VMD algorithm is adopted to carry out noise reduction on the measured data set, and a target domain data set is obtained; training an improved alternate transfer learning model based on a training set composed of the source domain data set and the target domain data set; and utilizing the trained improved alternate transfer learning model to adaptively adjust the clamping force of the tool holder according to the material, specification and clamping state of the tool holder. Closed-loop self-adaptive regulation and control of the clamping force are achieved, parameters can be dynamically adjusted according to the material, specification and machining working conditions of the cutter handle, the problem of over-tightness or over-looseness is avoided, the machining precision is improved, and the service life of the cutter is prolonged.

Description

Technical Field

[0001] This invention relates to the field of machine tool clamping technology, and in particular to a method, apparatus and computer program product for adaptive clamping of machine tool clamps. Background Technology

[0002] As a core component connecting the spindle and tool holder, the clamping reliability of machine tool clips directly affects machining accuracy, tool life, and production safety. Current mainstream clipping methods include hydraulic clamping, pneumatic clamping, and mechanical locking. These clips suffer from the following technical drawbacks: (1) The clamping force control relies on preset parameters, which is difficult to adapt to the different needs of tool holders of different materials and specifications. It is easy to cause the tool holder to deform due to excessive clamping or to cause cutting vibration due to excessive clamping. (2) The lack of accurate modeling of the dynamic characteristics during the clamping process makes it impossible to perceive the dynamic changes of key parameters such as contact stiffness and radial displacement in real time, resulting in insufficient clamping stability. (3) Fault diagnosis relies on human experience or single sensor data, making it difficult to quickly identify potential problems such as clamping failure and contact wear, and the adaptability of tool holders across working conditions and models is poor.

[0003] Therefore, there is an urgent need in this field for a tool clip adaptive clamping method to solve the problems of poor adaptability, low diagnostic accuracy and weak anti-interference ability of traditional methods. Summary of the Invention

[0004] The purpose of this invention is to provide a method, apparatus, and computer program product for adaptive clamping of machine tool clips, so as to solve or at least partially solve the technical problems mentioned in the background art.

[0005] To achieve this objective, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for adaptive clamping of machine tool clips based on improved alternating transfer learning, comprising: Based on Hertz contact theory, a nonlinear dynamic model of tool holder and tool clip is constructed according to the elastic contact characteristics of tool holder and tool clip. The radial displacement excitation function is modified according to the real-time position of the tool holder, and the contact stiffness and impact force parameters are modified according to the tool holder specifications and clamping state to obtain the modified dynamic equation. The modified dynamic equations are solved using the fourth-order Runge-Kutta method, generating simulation data including clamping force, radial displacement, and vibration signals under different clamping states, and constructing a source domain dataset. Clamping data of tool holders of different specifications and materials under different clamping conditions were collected to construct a measured dataset; wherein, the clamping data includes clamping force, radial displacement and vibration signal; The CS-GWO optimized VMD algorithm was used to denoise the measured dataset to obtain the target domain dataset. The improved alternating transfer learning model is trained using a training set consisting of source domain datasets and target domain datasets. Using a trained improved alternating transfer learning model, the clamping force of the tool holder is adaptively adjusted according to the material, specifications and clamping status of the tool holder.

[0006] Optionally, the nonlinear dynamic model of the tool holder-tool holder based on Hertz contact theory and the elastic contact characteristics of the tool holder and the tool clip is constructed, specifically including: Based on Hertz contact theory, the contact force and contact deformation between the tool holder and the tool clip exhibit a nonlinear relationship: , , ; in, This refers to the normal clamping force generated when the tool holder contacts the tool shank. For the equivalent elastic modulus, E1 and E2 are the equivalent radius of curvature, respectively, and the elastic moduli of the clip and shank materials. and R1 and R2 are the Poisson's ratios of the tool holder and tool shank materials, respectively; R1 is the radius of curvature of the tool shank cone surface, and R2 is the radius of curvature of the tool holder cone surface. This represents the normal deformation at the contact point. The original dynamic equations were: , ; Among them, M, C K I , , and The following are, in order: mass distribution matrix of the tool holder-tool holder system, acceleration at each point, damping distribution matrix, velocity at each point, linear stiffness matrix, nonlinear contact stiffness, displacement, and external excitation; The process involves correcting the radial displacement excitation function based on the real-time position of the tool holder, and correcting the contact stiffness and impact force parameters based on the tool holder specifications and clamping state to obtain the corrected dynamic equations, specifically including: Incorporate the real-time position coordinates of the tool holder into the radial displacement excitation function: u 修正 (t)=ecos(ωt)+u0(t); where, ecos(ωt) is the periodic radial displacement caused by rotational eccentricity, ω is the rotational angular velocity, e is the eccentricity, and u0(t) is the additional displacement caused by cutting force fluctuation. Introducing a deformation correction factor for the contact area, the nonlinear contact stiffness is corrected as follows: In the formula, η≤1, and the more severe the wear, the smaller η becomes; Based on the strong correlation between impact force and tool holder specifications and clamping state, the momentum theorem is used to... Revised to: In the formula, For correction factor, This represents the change in velocity before and after the impact. The duration of the impact.

[0007] Optionally, the step of solving the modified dynamic equations using the fourth-order Runge-Kutta method to generate simulation data including clamping force, radial displacement, and vibration signals under different clamping states, and constructing a source domain dataset, specifically involves: The modified dynamic equations are solved using the fourth-order Runge-Kutta method to generate simulation data of the tool holder under five clamping states: normal clamping, over-tightening, over-loosening, contact wear, and driver failure. Source domain datasets labeled with clamping states are then constructed.

[0008] Optionally, the step of using the CS-GWO optimized VMD algorithm to denoise the measured dataset to obtain the target domain dataset specifically includes: The amplitude spectrum of the clamped data in the measured dataset is selected as the fitness function of the CS-GWO algorithm, and the minimum value of the amplitude spectrum entropy is used as the fitness function to search for the optimal parameter combination in the VMD algorithm. Where K is the modality number K, It is a secondary penalty factor; The VMD algorithm with optimal parameter combination is used to decompose the clamping data in the measured dataset to obtain K modal components; Calculate the correlation between each modal component and the measured sample, retain the highly correlated modal components and sum them to reconstruct the target domain data used to construct the target domain dataset.

[0009] Optionally, the amplitude spectrum of the clamped data in the selected measured dataset is used as the fitness function of the CS-GWO algorithm, and the minimum value of the amplitude spectrum entropy is used as the fitness function to search for the optimal parameter combination in the VMD algorithm. Specifically, it includes: Initialize the parameters of the CS-GWO algorithm and VMD; Using the minimum amplitude spectral entropy as the fitness function, the optimal parameter combination in the VMD algorithm is searched using the CS-GWO algorithm. The formula for calculating the amplitude spectral entropy is: ; In the above formula, Li For modal components u i The amplitude spectrum, H i For modal components u i The amplitude spectral entropy, where N is the length of the modal component; Optionally, the method for constructing the improved alternating transfer learning model is as follows: Construct a CNN model containing five convolutional layers, two pooling layers, and two fully connected layers. Add a batch normalization layer after each convolutional layer. Use ReLU as the activation function for the hidden layers and Softmax as the activation function for the output layer. The first pooling layer is located between the second and third convolutional layers, and the second pooling layer is located after the fifth convolutional layer and before the first fully connected layer. The CORAL loss function is calculated after the first convolutional layer of the CNN model to reduce the difference in second-order statistics between the source and target domain datasets. The sum of the MMD loss function and the classification loss is calculated in the fully connected layer, and the network weights and bias parameters are updated by alternating backpropagation.

[0010] Optionally, the loss function of the improved alternating transfer learning model is set as follows: The distance between the second-order statistics of the data features in the source and target datasets is defined as the CORAL loss function L. CORAL : , is the Frobenius norm of the mean square matrix; Where d is the feature dimension of the output of the first convolutional layer, and the source domain dataset D s , and the target domain dataset D t The covariance matrices of the features are C s With C t ; MMD is used to measure the feature set D. s and D t Distributional differences in the regenerating nucleus Hilbert space: , A feature mapping function to map the output of a fully connected layer to a high-dimensional space; Where, n s n t D respectively s and D t The number of samples, , D respectively s The i-th sample and D t The features of the j-th sample; For the i-th target domain data sample, predict the probability that it belongs to category c. : ; in, This is the raw output of the fully connected layer for the i-th target domain data sample in category c. It is a natural constant; The total loss of the fully connected layer is L total =L MMD +λL CE Where λ is a preset weighting coefficient; ; Where N is the total number of samples, and c is the category of clamping state. Let be the true label of the i-th target domain data sample; The weights and biases of the fully connected layer are updated first through backpropagation, and then the parameters of the convolutional layer are updated by combining the CORAL loss of the first convolutional layer. This process is repeated iteratively.

[0011] Secondly, the present invention provides a machine tool clip adaptive clamping device based on improved alternating transfer learning, comprising a tool holder, an elastic clip for holding the tool holder, a piezoelectric ceramic actuator, and a wedge block transmission mechanism; the piezoelectric ceramic actuator is drivenly connected to the elastic clip via the wedge block transmission mechanism; the piezoelectric ceramic actuator realizes the identification of the clamping state and the adaptive adjustment of the clamping force of the clip through an improved alternating transfer learning machine tool clip adaptive clamping method as described above.

[0012] Optionally, when the clamping state is identified as "too tight", the piezoelectric ceramic actuator reduces the output voltage by 5% to 8%; When the clamping state is identified as "too loose", the piezoelectric ceramic actuator is controlled to increase the output voltage by 8% to 12%. When the clamping condition is identified as "too tight", the piezoelectric ceramic actuator is controlled to reduce the output voltage by 5% to 8%. When the clamping condition is identified as "contact wear", the compensation coefficient of the contact stiffness is dynamically adjusted; the adjustment range of the compensation coefficient is 1.0~1.15.

[0013] Thirdly, the present invention also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements a machine tool clip adaptive clamping method based on improved alternating transfer learning as described above.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: This application constructs a dynamic model of the machine tool clip and tool holder based on Hertz contact theory, and combines it with the CS-GWO optimized VMD algorithm to extract features, effectively improving the anti-interference ability of feature extraction and solving the problem of cutting noise masking key signals. By improving the alternating transfer learning algorithm, it solves the feature transfer adaptation problem between source domain data and target domain data, improving the diagnostic accuracy under small sample conditions and demonstrating strong generalization ability across working conditions and tool holder models. It achieves closed-loop adaptive control of clamping force, which can dynamically adjust parameters according to the tool holder material, specifications and machining conditions to avoid excessive tightness or looseness, thereby improving machining accuracy and tool life. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 The flowchart illustrates a method for adaptive clamping of machine tool clips based on improved alternating transfer learning, as provided in an embodiment of the present invention.

[0017] Figure 2 This is a schematic diagram of a nonlinear dynamic model of a tool holder-tool shank provided in an embodiment of the present invention.

[0018] Figure 3 This is a comparison chart of the clamping state diagnosis accuracy of different algorithms provided in an embodiment of the present invention.

[0019] Figure 4 This is a partial structural diagram of an adaptive clamping mechanism for a machine tool holder provided in an embodiment of the present invention.

[0020] In the diagram: 11, clamping arm; 12, roller bearing; 20, tool holder. Detailed Implementation

[0021] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0022] Example 1: Please refer to Figure 1 , Figure 1A flowchart illustrating a method for adaptive clamping of machine tool clips based on improved alternating transfer learning, provided in an embodiment of the present invention; the method includes: Step 110: Based on Hertz contact theory, construct a nonlinear dynamic model of tool holder and tool clip according to the elastic contact characteristics of the tool holder and tool clip, and modify the radial displacement excitation function according to the real-time position of the tool holder. Modify the contact stiffness and impact force parameters according to the tool holder specifications and clamping state to obtain the modified dynamic equation.

[0023] like Figure 4 As shown, the tool holder includes two clamping arms, each equipped with a roller bearing for elastic contact with the tool holder. Hertz contact theory is a classic theory describing the nonlinear relationship between contact force and contact deformation when two elastic bodies are in contact. It is applicable to point / surface contact scenarios between the tool holder's conical surface and the tool holder's conical surface. The nonlinear dynamic model of the tool holder-tool holder constructed in this embodiment is as follows: Figure 2 As shown; Based on Hertz contact theory, the contact force and contact deformation between the tool holder and the tool clip exhibit a nonlinear relationship: , , ; in, This refers to the normal clamping force generated when the tool holder contacts the tool shank. For the equivalent elastic modulus, E1 and E2 are the equivalent radius of curvature, respectively, and the elastic moduli of the clip and shank materials. and R1 and R2 are the Poisson's ratios of the tool holder and tool shank materials, respectively; R1 is the radius of curvature of the tool shank cone surface, and R2 is the radius of curvature of the tool holder cone surface. This represents the normal deformation at the contact point. The original dynamic equations were: , ; Among them, M, C K I , , and The following are, in order: mass distribution matrix of the tool holder-tool holder system, acceleration at each point, damping distribution matrix, velocity at each point, linear stiffness matrix, nonlinear contact stiffness, displacement, and external excitation; The process involves correcting the radial displacement excitation function based on the real-time position of the tool holder, and correcting the contact stiffness and impact force parameters based on the tool holder specifications and clamping state to obtain the corrected dynamic equations, specifically including: Incorporate the real-time position coordinates of the tool holder into the radial displacement excitation function: u 修正 (t)=ecos(ωt)+u0(t); where, ecos(ωt) is the periodic radial displacement caused by rotational eccentricity, ω is the rotational angular velocity, e is the eccentricity, and u0(t) is the additional displacement caused by cutting force fluctuation. Introducing a deformation correction factor for the contact area, the nonlinear contact stiffness is corrected as follows: In the formula, η≤1, and the more severe the wear, the smaller η becomes; Based on the strong correlation between impact force and tool holder specifications and clamping state, the momentum theorem is used to... Revised to: In the formula, For correction factor, This represents the change in velocity before and after the impact. The duration of the impact.

[0024] Substituting the modified parameters into the dynamic equation yields the modified dynamic equation.

[0025] Step 120: Solve the modified dynamic equations using the fourth-order Runge-Kutta method to generate simulation data including clamping force, radial displacement, and vibration signals under different clamping states, and construct the source domain dataset.

[0026] Specifically, the modified dynamic equations are solved using the fourth-order Runge-Kutta method to generate simulation data of the tool holder under five clamping states: normal clamping, over-tightening, over-loosening, contact wear, and driver failure. Source domain datasets labeled with clamping states are then constructed. The fourth-order Runge-Kutta method (RK4) is a classical time-domain numerical method for solving nonlinear ordinary differential equations. It is suitable for calculating the transient vibration response of tool holder-tool shank systems (without analytical solutions).

[0027] Taking a single degree of freedom as an example, the core steps to solve the problem are: The second-order dynamic equation Transformed into a system of first-order equations: ; The solution at each time step is obtained by iterating through a four-step slope-weighted average: ; Where h is the time step. For the nth step displacement, For the speed at step n, This is the nth step.

[0028] For example, in this embodiment, the simulation data includes features such as clamping force, radial displacement, vibration signal (frequency), and temperature. The source domain dataset has 1200 samples and a sample length of 1024.

[0029] Step 130: Collect clamping data of tool holders of different specifications and materials under different clamping states to construct a test dataset.

[0030] The sensor module collects clamping data of tool holders of different materials and specifications under different clamping states in actual processing scenarios under varying working conditions (different speeds and loads), including clamping force, radial displacement, vibration signals, and temperature data, to construct a target domain dataset; wherein the clamping data includes clamping force, radial displacement, vibration signals, and temperature data.

[0031] Step 140: Use the CS-GWO optimized VMD algorithm to reduce noise in the measured dataset to obtain the target domain dataset.

[0032] This step uses the CS-GWO optimized VMD algorithm to extract features from the vibration signal, specifically including: The amplitude spectrum of the clamped data in the measured dataset is selected as the fitness function of the CS-GWO algorithm, and the minimum value of the amplitude spectrum entropy is used as the fitness function to search for the optimal parameter combination in the VMD algorithm. Where K is the modality number K, It is a secondary penalty factor; The VMD algorithm with optimal parameter combination is used to decompose the clamping data in the measured dataset to obtain K modal components; Calculate the correlation between each modal component and the measured sample, retain the highly correlated modal components and sum them to reconstruct the target domain data used to construct the target domain dataset.

[0033] Among them, the Grey Wolf Algorithm (GWO) is widely used in function optimization and other problems due to its strong search ability and high efficiency. However, the position update method of the GWO algorithm makes its global search ability relatively weak and it is prone to getting trapped in local minima. The Cuckoo Search Algorithm (CS) has fewer parameters and is easy to jump out of the current region, thus achieving a global search. The CS algorithm is inspired by Levi's flight and the parasitic feeding mechanism of the cuckoo. The location of the host's nest, parasitized by the cuckoo, is iteratively updated according to a formula based on the cuckoo's unique search method: ; In the above formula, i = 1, 2, ..., N, where N represents the number of host bird nests; This is an inner product operation; This represents the coordinates of the i-th host nest during the t-th iteration; Let these be the coordinates of the i-th host nest after iteration; =1 is the step size factor; It follows a Lévy distribution with a step size of .

[0034] Combining the advantages of the CS algorithm, such as its ability to easily jump from the computational region to the uncomputed region during global search, the GWO algorithm, which is improved upon, effectively overcomes the problem of the GWO algorithm easily getting trapped in local optima during the optimization process. This improvement allows the algorithm to better explore the search space, enhances its global search capability, and thus makes it more effective in solving various optimization problems.

[0035] The amplitude spectrum of the clamped data in the selected measured dataset is used as the fitness function of the CS-GWO algorithm, and the minimum value of the amplitude spectrum entropy is used as the fitness function to search for the optimal parameter combination in the VMD algorithm. Specifically, it includes: Initialize the parameters of the CS-GWO algorithm and VMD; Using the minimum amplitude spectral entropy as the fitness function, the optimal parameter combination in the VMD algorithm is searched using the CS-GWO algorithm. The formula for calculating the amplitude spectral entropy is: ; In the above formula, L i For modal components u i The amplitude spectrum, H i For modal components u i The amplitude spectral entropy, where N is the length of the modal component.

[0036] Among the many parameters of VMD, the number of decomposition modes K and the second-order penalty factor have the greatest impact on the results. Choose the VMD parameter combination. Using it as the fitness function optimization objective of the CS-GWO algorithm can effectively avoid obtaining suboptimal solutions due to improper parameter combinations.

[0037] In this step, the change in amplitude spectral entropy can reflect whether the signal contains a lot of fault characteristic information. When the modal components obtained by VMD decomposition contain a lot of noise components and fewer fault impact signal components, the amplitude spectral entropy of the modal component will be large, as can be seen from the formula calculation results. Conversely, when the modal component contains less noise signal, the regular periodic fault impact signal response is obvious, that is, the modal component is rich in fault characteristic information and has a small amplitude spectral entropy.

[0038] For example, the wolf pack size is set to 10, the maximum number of iterations is 30, and the fitness value converges to the optimal solution after 9 iterations. The modal components with high correlation are screened to reconstruct signals, and effective fault features such as amplitude spectral entropy and peak factor are extracted to filter out cutting noise interference.

[0039] Step 150: Train the pre-built improved alternating transfer learning model based on the training set composed of the source domain dataset and the target domain dataset.

[0040] In step 150, the improved method for constructing the alternating transfer learning model is as follows: Construct a CNN model containing five convolutional layers, two pooling layers, and two fully connected layers. Add a batch normalization layer after each convolutional layer. Use ReLU as the activation function for the hidden layers and Softmax as the activation function for the output layer. The first pooling layer is located between the second and third convolutional layers, and the second pooling layer is located after the fifth convolutional layer and before the first fully connected layer. The CORAL loss function is calculated after the first convolutional layer of the CNN model to reduce the difference in second-order statistics between the source and target domain datasets. The sum of the MMD loss function and the classification loss is calculated in the fully connected layer. The network weights and bias parameters are updated by alternating backpropagation to achieve feature transfer adaptation across working conditions and tool holder models.

[0041] The method for setting the loss function of the improved alternating transfer learning model is as follows: The distance between the second-order statistics of data features in the source and target datasets is defined as the CORAL (Correlation Alignment) loss function L. CORAL : , is the Frobenius norm of the mean square matrix; Where d is the feature dimension of the output of the first convolutional layer, and the source domain dataset D s , and the target domain dataset D t The covariance matrices of the features are C s With C t .

[0042] The core of CORAL is to reduce the difference in second-order statistics between the source and target domain features (second-order statistics mainly refer to the covariance matrix of features, which reflects the correlation between features, such as the correlation between "radial displacement and clamping force").

[0043] Convolutional layer 1 extracts the low-level features of the data (such as the time-domain waveform of radial displacement and the edge features of vibration signals). The covariance of the low-level features of simulated data (source domain) and real data (target domain) differs greatly (for example, the radial displacement fluctuation pattern in the simulation is different from that in the real machine tool). Calculating the CORAL loss can align this difference and make the low-level features more general.

[0044] MMD (Maximum Mean Discrepancy) is used to measure the feature set D. s and D t Distributional differences in the regenerating nucleus Hilbert space: , A feature mapping function to map the output of a fully connected layer to a high-dimensional space; Where, n s n t D respectively s and D t The number of samples, , D respectively s The i-th sample and D t The features of the j-th sample; The core of MMD is to measure the overall distribution difference of features between the source and target domains in a high-dimensional space (more comprehensive than CORAL, including not only second-order statistics but also higher-order statistical features).

[0045] The fully connected layer extracts high-level semantic features (such as the combination of "low clamping force + large radial displacement + abnormal vibration frequency" corresponding to "overly loose state"); MMD loss can further reduce the distribution difference of high-level features in the source domain (simulation) and the target domain (real), and solve the feature adaptation problem across working conditions / tool ​​holder models.

[0046] For the i-th target domain data sample, predict the probability that it belongs to category c. : ; in, This is the raw output of the fully connected layer for the i-th target domain data sample in category c. It is a natural constant; The total loss of the fully connected layer is L total =L MMD +λL CE Where λ is a preset weight coefficient; the weights and biases of the fully connected layer are updated first through backpropagation, and then the parameters of the convolutional layer are updated by combining the CORAL loss of the first convolutional layer. This process is repeated to balance "distribution alignment" and "classification accuracy". ; Where N is the total number of samples, and c is the category of clamping state. Let be the true label of the i-th target domain data sample; Classification loss (cross-entropy loss) is the core loss for classification tasks. It measures the difference between the class probability predicted by the model and the true class of the sample. The smaller the loss, the more accurate the classification. It ensures that the model can accurately distinguish between five states: "normal clamping, too tight, too loose, contact wear, and drive failure" while aligning the distribution of the source / target domain, thus avoiding inaccurate classification even if the distribution is aligned.

[0047] After the model is trained, real-time collected target domain data is input, and the clamping state classification results (normal clamping, too tight, too loose, contact wear, drive failure) are output through the Softmax function. The purpose of the Softmax function is to transform the "raw score" (a value without a range) output by the fully connected layer into a probability value between 0 and 1, and the sum of the probabilities of all categories is 1, which makes it easy to directly determine the clamping state of the sample.

[0048] For example, the output probability of a real-time sample after passing through Softmax is: [0.98, 0.01, 0.005, 0.003, 0.002] (corresponding to normal, too tight, too loose, worn, and faulty), is judged as "normal clamping"; If the output probability is [0.02, 0.01, 0.95, 0.01, 0.01], then it is judged as "too loose"; The diagnostic accuracy rate for all samples must be ≥97%, meaning the number of correctly classified samples / the total number of samples must be ≥97%.

[0049] Step 160: Using the trained improved alternating transfer learning model, adaptively adjust the clamping force of the tool holder according to the material, specifications and clamping status of the tool holder.

[0050] Please refer to Figure 3 , Figure 3 The comparison chart shows the accuracy of different algorithms in diagnosing clamping states. The comparison results show that the accuracy of this embodiment in identifying clamping states by improving the alternating transfer learning model is as high as 99.89%, which is significantly better than other algorithms in the prior art.

[0051] Furthermore, a machine tool clip adaptive clamping device constructed based on the above method includes a tool holder, an elastic clip for holding the tool holder, a piezoelectric ceramic actuator, and a wedge block transmission mechanism; the piezoelectric ceramic actuator is driven to the elastic jaw through the wedge block transmission mechanism; the piezoelectric ceramic actuator realizes the identification of the clamping state and the adaptive adjustment of the clamping force of the clip through an adaptive clamping method for machine tool clips based on improved alternating transfer learning as described above.

[0052] Based on the diagnostic / identification results of the improved alternating transfer learning model and the output of the dynamic model, the control module dynamically adjusts the output force of the piezoelectric ceramic actuator used to control the clamping force of the tool holder through a PID algorithm.

[0053] This device is compatible with various tool holders such as BT, HSK, and CAT, and is suitable for different machining conditions such as milling, drilling, and grinding. It can directly replace the traditional tool holders of existing machine tools without requiring major modifications to the machine tool spindle, and has strong compatibility. In practical applications, the dynamic model parameters, CNN network structure, and algorithm iteration number can be adjusted according to the specific machine tool model and processing requirements to further optimize clamping accuracy and diagnostic efficiency. The algorithm can be directly ported to embedded systems for engineering applications. The key charts used all serve the core principles and effect verification of the technical solution, with no redundant information, and have good industrialization prospects.

[0054] For example, when the diagnostic result is "too tight", the output voltage is reduced by 5% to 8%; when the diagnostic result is "too loose", the output voltage is increased by 8% to 12%; when contact wear is detected, the contact stiffness compensation coefficient is dynamically adjusted (adjustment range 1.0 to 1.15) to maintain clamping stability; for different tool holder materials and machining loads, the clamping parameters are optimized to achieve adaptive clamping with "rigidity and flexibility".

[0055] In this PID controller, the three components each perform their specific functions, working together to achieve "rapid response, precise stability, and predictive adjustment." The adjustment strategies for different faults / operating conditions are as follows: 1. Proportional element (P): Quickly responds to deviations and determines the adjustment range; Based on the magnitude of the deviation, a proportional control quantity is output, which is the "main force" for adjustment, enabling rapid correction of the deviation.

[0056] Operating condition adaptation: Diagnosed as "too tight": Actual clamping force > target value → negative deviation → P circuit immediately outputs negative adjustment, reducing driver voltage by 5%~8% (voltage reduction → output force reduction → clamping force drop); Diagnosed as "too loose": Actual clamping force < target value → deviation is positive → P circuit immediately outputs positive adjustment, increasing driver voltage by 8%~12% (voltage increase → output force increase → clamping force increase); Features: Fast response, but pure proportional adjustment may leave a small "steady-state error" (such as clamping force slightly lower / higher than the target value).

[0057] 2. Integral component (I): Eliminates steady-state error and provides long-term compensation; By accumulating the total deviation over a period of time, the control quantity is slowly adjusted to eliminate the residual steady-state error of the proportional element, ensuring that the clamping force ultimately and accurately matches the target value.

[0058] Operating condition adaptation: Contact wear condition: Wear leads to a decrease in contact stiffness, and the clamping force will slowly and continuously decrease → The proportional circuit can only quickly compensate for the voltage once, while the integral circuit will continuously accumulate the deviation of "insufficient clamping force", slowly increase the voltage (while adjusting the contact stiffness compensation coefficient to 1.0~1.15), and maintain clamping stability in the long term. Normal working conditions: Slight fluctuations in machining load cause slight deviations in clamping force → The integral stage gradually eliminates the deviation, avoiding "fluctuations in clamping force".

[0059] 3. Differential component (D): Predicts deviation trends and suppresses overshoot; Adjust the control quantity according to the rate of change of the deviation (how fast the deviation changes over time), predict the development trend of the deviation, and "brake" in advance to avoid over-adjustment (overshoot).

[0060] Fault condition adaptation: When adjusting for excessive looseness: If the voltage is increased rapidly in the P stage, the clamping force may rise rapidly and exceed the target value (becoming "excessive tightness") → The D stage detects that "the clamping force rises too quickly" and reduces the voltage adjustment amplitude in advance to suppress overshoot; Sudden changes in machining load: For example, a sudden loading of heavy-duty cutting causes a rapid decrease in clamping force. The D stage immediately responds to the rapid change in deviation, assisting the P stage in quickly compensating for the pressure, while preventing the voltage from rising too quickly and causing subsequent overshoot.

[0061] To facilitate understanding, specific application examples are provided below: The clamping body inner diameter is compatible with various tool holders such as BT30 / BT40 / HSK-A63, and the elastic jaws open at an angle of 0°~15°, with the contact surface curvature of the jaws matching the outer radius of the tool holder; the wedge block drive mechanism has an inclination angle of 18° and is rigidly connected to the piezoelectric ceramic actuator with a clearance of ≤0.02mm. The force sensor is fixed to the connection between the elastic gripper and the transmission mechanism via a threaded connection, and the direction of the force is consistent with the direction of movement of the elastic gripper; the displacement sensor is installed in three mounting holes evenly distributed around the inner wall of the fixture, with the measuring point 2mm away from the surface of the tool holder; the vibration sensor is fixed to the middle of the outer side of the fixture by adhesive bonding, and the temperature sensor is embedded in a groove in the inner wall of the fixture, with the distance between the temperature sensor and the contact area with the tool holder ≤1mm; The STM32H743 microcontroller is responsible for data reception and algorithm execution, the FPGA is responsible for signal preprocessing, the dynamic model is implemented by C language programming, and the improved alternating transfer learning algorithm is developed based on the Python TensorFlow framework. The model training iterations are 100, the learning rate is 0.001, and the batch size is 64. The piezoelectric ceramic drive controller communicates with the data processing module via a CAN bus. The PID parameters (proportional coefficient Kp=0.8, integral coefficient Ki=0.1, derivative coefficient Kd=0.05) of the feedback adjustment unit can be adaptively adjusted according to the actual working conditions.

[0062] The implementation steps of the method are as follows: For the BT40 toolholder, three elastic jaws were set, with a contact point diameter of 5mm and a pitch diameter of 60mm. The contact stiffness was calculated based on Hertz contact theory, and the radial displacement excitation function and impact force parameters were corrected. The dynamic equations were solved using the fourth-order Runge-Kutta method to generate simulation data for five states: normal clamping, over-tight (clamping force exceeding the threshold by 20%), over-loose (clamping force below the threshold by 30%), contact wear (contact stiffness decreased by 15%), and driver failure (output force fluctuation ±10%). A source domain dataset (1200 samples, sample length 1024) was constructed. Measured data of three tool holders (BT30, BT40, and HSK-A63) under different machining loads (500N, 1000N, 1500N) and different rotational speeds (1000r / min, 1500r / min, 2000r / min) were collected to construct a target domain dataset. The CS-GWO algorithm was used to optimize the VMD parameters (modal number K=6, quadratic penalty factor α=10000) to decompose and reconstruct the vibration signal, and extract characteristic parameters such as amplitude spectrum entropy and peak factor. The source and target domain data are converted into 32×32 two-dimensional grayscale images and input into the CNN network; the CORAL loss (convolutional layer 1) and MMD loss + classification loss (fully connected layer 1) are calculated alternately, and the network weights are updated by backpropagation. After training, the clamping state of the actual test data is diagnosed. The accuracy rate of normal state recognition is ≥99.2%, and the average accuracy rate of fault state recognition is ≥97.5%. The clamping force is dynamically adjusted based on the diagnostic results. For example, the clamping force is set to 3kN for BT40 tool holders made of cemented carbide and 2kN for BT30 tool holders made of high-speed steel. When an abnormal vibration signal is detected during cutting, the clamping force is automatically increased by 8% to maintain machining stability.

[0063] The simulation and experimental verification results are as follows: By comparing the key parameters (contact stiffness, radial displacement, and failure frequency) of the simulation data and the measured data, the errors were all within 0.64%, which verified the accuracy of the model. The CS-GWO optimized VMD algorithm was used to process noisy vibration signals, achieving a fault characteristic coefficient (FFC) of 5.28% and a signal-to-noise ratio (SNR) of 12.94, which is superior to optimized algorithms such as BA, SSA, and WOA, demonstrating the effectiveness of feature extraction. Under conditions of small sample size (5% of the target domain samples), cross-working conditions, and cross-tool holder models, the diagnostic accuracy of the improved alternating transfer learning algorithm reached 84.65%, 92.89%, and 89.82%, respectively, which is better than the comparison algorithms such as TCA, CNN, and CNN-MMD. Milling tests using this device showed that the workpiece dimensional error was ≤ ±0.005mm, which is 60% lower than that of traditional hydraulic tool holders and extends tool life by more than 30%.

[0064] In summary, this embodiment provides a method for adaptive clamping of machine tool holders based on improved alternating transfer learning. It constructs a nonlinear dynamic model of the tool holder-tool shank based on Hertz contact theory and combines it with the CS-GWO optimized VMD algorithm for feature extraction, effectively improving the anti-interference capability of feature extraction and solving the problem of cutting noise masking key signals. The error between the model simulation data and the actual working conditions does not exceed 0.64%. An improved alternating transfer learning (IATL) algorithm is adopted. By alternating the calculation of CORAL loss and MMD loss, feature transfer adaptation between source domain simulation data and target domain measured data is achieved. The diagnostic accuracy is over 84.65% under small sample conditions, and the generalization ability across working conditions and tool holder models is strong. It achieves closed-loop adaptive control of clamping force, and can dynamically adjust parameters according to the material, specifications and machining conditions of the tool holder to avoid problems of being too tight or too loose, thereby improving machining accuracy and tool life. It integrates multi-sensor monitoring and intelligent fault diagnosis functions, which can identify potential faults such as clamping failure and contact wear in real time, with a diagnosis response time of ≤1ms, thus improving clamping reliability and production safety. The device has a compact structure and can directly replace the traditional tool holder of existing machine tools without major modifications to the machine tool spindle. It has strong compatibility and a wide range of applications.

[0065] Based on the same concept, embodiments of the present invention also provide a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement a method for constructing a target domain dataset or a method for adaptive clamping of machine tool clips based on improved alternating transfer learning provided in embodiments of the present invention.

[0066] Any combination of one or more computer-readable media may be used. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device.

[0067] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0068] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including—but not limited to—wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0069] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0070] Based on the same concept, embodiments of the present invention also provide a computer program product, including a computer program / instruction, which, when executed by a processor, implements a method for constructing a target domain dataset or a method for adaptive clamping of machine tool clips based on improved alternating transfer learning provided in embodiments of the present invention.

[0071] Computer program products may be loaded onto computer devices, and the components of computer devices may include, but are not limited to: one or more processors or processing units, system memory, and buses connecting different system components (including system memory and processing units).

[0072] Computer devices typically include a variety of computer system-readable media. These media can be any available media that can be accessed by a computer device, including volatile and non-volatile media, and removable and non-removable media.

[0073] System memory may include computer system readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory. The computer device may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system may be used to read and write non-removable, non-volatile magnetic media. The computer program product has a set (e.g., at least one) of program modules configured to perform the functions of the various embodiments of the present invention.

[0074] A program / utility having a set (at least one) of program modules can be stored, for example, in memory. Such program modules include, but are not limited to, an operating system, one or more applications, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. The program modules typically perform the functions and / or methods described in the embodiments of this invention.

[0075] Computer devices can also communicate with one or more external devices (such as keyboards, pointing devices, monitors, etc.), one or more devices that enable users to interact with the computer device, and / or any device that enables the computer device to communicate with one or more other computing devices (such as network interface cards, modems, etc.). This communication can be performed through input / output (I / O) interfaces. Furthermore, computer devices can communicate with one or more networks (such as local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via network adapters. As shown in the figure, the network adapter communicates with other modules of the computer device via a bus. It should be understood that other hardware and / or software modules can be used in conjunction with the computer device, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0076] The processing unit executes various functional applications and data processing by running programs stored in the system memory, such as implementing a method for constructing a target domain dataset or a method for adaptive clamping of machine tool clips based on improved alternating transfer learning provided in the embodiments of the present invention.

[0077] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware, or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0078] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for adaptive clamping of machine tool clips based on improved alternating transfer learning, characterized in that, include: Based on Hertz contact theory, a nonlinear dynamic model of tool holder and tool clip is constructed according to the elastic contact characteristics of tool holder and tool clip. The radial displacement excitation function is modified according to the real-time position of the tool holder, and the contact stiffness and impact force parameters are modified according to the tool holder specifications and clamping state to obtain the modified dynamic equation. The modified dynamic equations are solved using the fourth-order Runge-Kutta method, generating simulation data including clamping force, radial displacement, and vibration signals under different clamping states, and constructing a source domain dataset. Clamping data of tool holders of different specifications and materials under different clamping conditions were collected to construct a measured dataset; wherein, the clamping data includes clamping force, radial displacement and vibration signal; The CS-GWO optimized VMD algorithm was used to denoise the measured dataset to obtain the target domain dataset. The improved alternating transfer learning model is trained using a training set consisting of source domain datasets and target domain datasets. Using a trained improved alternating transfer learning model, the clamping force of the tool holder is adaptively adjusted according to the material, specifications and clamping status of the tool holder.

2. The method for adaptive clamping of machine tool clips based on improved alternating transfer learning according to claim 1, characterized in that, The nonlinear dynamic model of the tool holder-tool holder based on Hertz contact theory and the elastic contact characteristics of the tool holder and tool clip is constructed, specifically including: Based on Hertz contact theory, the contact force and contact deformation between the tool holder and the tool clip exhibit a nonlinear relationship: ， ， ； in, This is the normal clamping force generated when the tool holder contacts the tool shank. For the equivalent elastic modulus, E1 and E2 are the equivalent radius of curvature, respectively, and the elastic moduli of the clip and shank materials. and R1 and R2 are the Poisson's ratios of the tool holder and tool shank materials, respectively; R1 is the radius of curvature of the tool shank cone surface, and R2 is the radius of curvature of the tool holder cone surface. This represents the normal deformation at the contact point. The original dynamic equations were: ， ； Among them, M, C K I , , and The following are, in order: mass distribution matrix of the tool holder-tool holder system, acceleration at each point, damping distribution matrix, velocity at each point, linear stiffness matrix, nonlinear contact stiffness, displacement, and external excitation; The process involves correcting the radial displacement excitation function based on the real-time position of the tool holder, and correcting the contact stiffness and impact force parameters based on the tool holder specifications and clamping state to obtain the corrected dynamic equations, specifically including: Incorporate the real-time position coordinates of the tool holder into the radial displacement excitation function: u 修正 (t)=ecos(ωt)+u0(t); where, ecos(ωt) is the periodic radial displacement caused by rotational eccentricity, ω is the rotational angular velocity, e is the eccentricity, and u0(t) is the additional displacement caused by cutting force fluctuation. Introducing a deformation correction factor for the contact area, the nonlinear contact stiffness is corrected as follows: In the formula, η≤1, and the more severe the wear, the smaller η becomes; Based on the strong correlation between impact force and tool holder specifications and clamping state, the momentum theorem is used to... Revised to: In the formula, For correction factor, This represents the change in velocity before and after the impact. The duration of the impact.

3. The method for adaptive clamping of machine tool clips based on improved alternating transfer learning according to claim 2, characterized in that, The modified dynamic equations are solved using the fourth-order Runge-Kutta method to generate simulation data including clamping force, radial displacement, and vibration signals under different clamping states, thus constructing a source domain dataset, specifically: The modified dynamic equations are solved using the fourth-order Runge-Kutta method to generate simulation data of the tool holder under five clamping states: normal clamping, over-tightening, over-loosening, contact wear, and driver failure. Source domain datasets labeled with clamping states are then constructed.

4. The method for adaptive clamping of machine tool clips based on improved alternating transfer learning according to claim 3, characterized in that, The method of using the CS-GWO optimized VMD algorithm to denoise the measured dataset to obtain the target domain dataset specifically includes: The amplitude spectrum of the clamped data in the measured dataset is selected as the fitness function of the CS-GWO algorithm, and the minimum value of the amplitude spectrum entropy is used as the fitness function to search for the optimal parameter combination in the VMD algorithm. Where K is the modality number K, It is a secondary penalty factor; The VMD algorithm with optimal parameter combination is used to decompose the clamping data in the measured dataset to obtain K modal components; Calculate the correlation between each modal component and the measured sample, retain the highly correlated modal components and sum them to reconstruct the target domain data used to construct the target domain dataset.

5. The method for adaptive clamping of machine tool clips based on improved alternating transfer learning according to claim 4, characterized in that, The amplitude spectrum of the clamped data in the selected measured dataset is used as the fitness function of the CS-GWO algorithm, and the minimum value of the amplitude spectrum entropy is used as the fitness function to search for the optimal parameter combination in the VMD algorithm. Specifically, it includes: Initialize the parameters of the CS-GWO algorithm and VMD; Using the minimum amplitude spectral entropy as the fitness function, the optimal parameter combination in the VMD algorithm is searched using the CS-GWO algorithm. The formula for calculating the amplitude spectral entropy is: ； In the above formula, L i For modal components u i The amplitude spectrum, H i For modal components u i The amplitude spectral entropy, where N is the length of the modal component.

6. The method for adaptive clamping of machine tool clips based on improved alternating transfer learning according to claim 5, characterized in that, The method for constructing the improved alternating transfer learning model is as follows: Construct a CNN model containing five convolutional layers, two pooling layers, and two fully connected layers. Add a batch normalization layer after each convolutional layer. Use ReLU as the activation function for the hidden layers and Softmax as the activation function for the output layer. The first pooling layer is located between the second and third convolutional layers, and the second pooling layer is located after the fifth convolutional layer and before the first fully connected layer. The CORAL loss function is calculated after the first convolutional layer of the CNN model to reduce the difference in second-order statistics between the source and target domain datasets. The sum of the MMD loss function and the classification loss is calculated in the fully connected layer, and the network weights and bias parameters are updated by alternating backpropagation.

7. The method for adaptive clamping of machine tool clips based on improved alternating transfer learning according to claim 6, characterized in that, The method for setting the loss function of the improved alternating transfer learning model is as follows: The distance between the second-order statistics of the data features in the source and target datasets is defined as the CORAL loss function L. CORAL : , is the Frobenius norm of the mean square matrix; Where d is the feature dimension of the output of the first convolutional layer, and the source domain dataset D s , and the target domain dataset D t The covariance matrices of the features are C s With C t ; MMD is used to measure the feature set D. s and D t Distributional differences in the regenerating nucleus Hilbert space: , A feature mapping function to map the output of a fully connected layer to a high-dimensional space; Where, n s n t D respectively s and D t The number of samples, , D respectively s The i-th sample and D t The features of the j-th sample; For the i-th target domain data sample, predict the probability that it belongs to category c. : ； in, This is the raw output of the fully connected layer for the i-th target domain data sample in category c. It is a natural constant; The total loss of the fully connected layer is L total =L MMD +λL CE Where λ is a preset weighting coefficient; ； Where N is the total number of samples, and c is the category of clamping state. Let be the true label of the i-th target domain data sample; The weights and biases of the fully connected layer are updated first through backpropagation, and then the parameters of the convolutional layer are updated by combining the CORAL loss of the first convolutional layer. This process is repeated iteratively.

8. A device for adaptive clamping of machine tool chucks based on improved alternating transfer learning, comprising a tool holder, an elastic tool holder for clamping the tool holder, a piezoelectric ceramic actuator, and a wedge block transmission mechanism; the piezoelectric ceramic actuator is drivenly connected to the elastic gripper via the wedge block transmission mechanism; characterized in that, The piezoelectric ceramic actuator achieves the identification of clamping state and adaptive adjustment of clamping force of the tool holder through an adaptive clamping method for machine tool holder based on improved alternating transfer learning as described in any one of claims 1-7.

9. The device for adaptive clamping of machine tool clips based on improved alternating transfer learning according to claim 8, characterized in that, When the clamping condition is identified as "too tight", the piezoelectric ceramic actuator reduces the output voltage by 5% to 8%. When the clamping state is identified as "too loose", the piezoelectric ceramic actuator is controlled to increase the output voltage by 8% to 12%. When the clamping state is identified as "too tight", the piezoelectric ceramic actuator is controlled to reduce the output voltage by 5% to 8%. When the clamping condition is identified as "contact wear", the compensation coefficient of the contact stiffness is dynamically adjusted; the adjustment range of the compensation coefficient is 1.0~1.

15.

10. A computer program product, comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements the machine tool clip adaptive clamping method based on improved alternating transfer learning as described in any one of claims 1-7.

Images