A deep learning-based performance evaluation and analysis method for an electric spindle of a numerical control machine tool

Abstract

The application discloses a kind of based on deep learning's numerical control machine tool electric main shaft performance evaluation analysis method, belong to new generation information technology and high-end manufacturing equipment technical field, including: while considering the high-speed work characteristic of electric main shaft, proposed the static model simplification rule of lightweight to carry out finite element simulation analysis process, and corresponding carried out the stiffness numerical fitting calculation of fusion deep learning network, further on the basis of static model, the orthogonal constraint model for electric main shaft dynamic performance analysis is constructed, and the electric main shaft modal frequency under multiple order vibration modes is obtained by the extended modal analysis under free grid division, accurately analyze the key static performance index and dynamic performance index of electric main shaft, provide quantitative support for subsequent rough machining capacity analysis and working precision improvement of numerical control machine tool.

Description

Technical Field

[0001] This invention relates to the field of next-generation information technology and high-end manufacturing equipment technology, specifically a method for evaluating and analyzing the performance of CNC machine tool electric spindles based on deep learning networks. Background Technology

[0002] As a core component of high-value CNC machine tools, the electric spindle is the main power drive unit supporting the machine tool in completing roughing and finishing processes. Its actual performance directly determines the machining accuracy of the CNC machine tool and the finished quality of the workpiece. Therefore, the efficiency and accuracy of the calculation for performance evaluation and analysis of the electric spindle will directly affect the assembly and debugging effect of the entire machine tool. At the same time, the electric spindle not only has a complex internal structure, but is also affected by the static indeterminate nature of multiple sets of precision bearings, making it difficult for the analysis and evaluation results of the static and dynamic performance of the electric spindle to truly reflect its working characteristics. Therefore, this invention focuses on designing a deep learning-based method for evaluating and analyzing the performance of CNC machine tool electric spindles. While considering the high-speed working characteristics of the electric spindle, it proposes a lightweight static model simplification rule to carry out the finite element simulation analysis process, and correspondingly conducts stiffness numerical fitting calculations fused with deep learning networks. Furthermore, based on the static model, an orthogonal constraint model for dynamic performance analysis of the electric spindle is constructed, and the modal frequencies of the electric spindle under multiple vibration modes are obtained through extended modal analysis under free mesh division. The key static and dynamic performance indicators of the electric spindle are accurately analyzed, providing quantitative support for subsequent roughing capability analysis and working accuracy improvement of CNC machine tools.

[0003] Currently, performance evaluation and analysis methods for CNC machine tool electric spindles are mainly divided into two categories: one is the numerical calculation method based on physical mechanism formulas, which can evaluate key performance indicators through directly explicit formulas; the other is the simulation calculation method using finite element analysis software, which obtains dynamic and static stiffness information through detailed simulation of the electric spindle's geometry, material composition, and functional components. In their high-level paper "Dynamic and Static Performance Analysis of High-Speed ​​Grinding Electric Spindle Based on ANSYS" (Mechatronics Information, 2023, Vol. 16, pp. 37-41), Chi Yunfei et al. used the domestically produced 120MD60Y6 high-speed grinding electric spindle as the research object and implemented dynamic and static performance analysis of the high-speed electric spindle based on ANSYS Workbench.

[0004] However, existing numerical calculation methods cannot meet the performance evaluation requirements of electric spindles with complex structures in terms of the accuracy of physical property analysis and evaluation. Performance evaluation methods based on finite element analysis software also face challenges due to the complex structure of electric spindles, including a large modeling workload and difficulty in quickly adapting to actual industrial scenarios. Therefore, there is an urgent need to invent an efficient performance evaluation method suitable for CNC machine tool electric spindles. This method should comprehensively utilize the analytical efficiency advantages of numerical calculation methods and the modeling accuracy advantages of finite element analysis methods, and combine the function fitting and representation capabilities of deep learning networks to form a static model solution and analysis process for electric spindles oriented towards static stiffness. This process should output the modal frequencies of the electric spindle under different order vibration modes, ensuring the accurate analysis and quantitative output of the electric spindle's dynamic performance indicators. This will provide reliable parameter support for subsequent CNC system process optimization and CNC machine tool structural improvement. Summary of the Invention

[0005] This invention provides a method for evaluating and analyzing the performance of CNC machine tool electric spindles based on deep learning networks, targeting CNC machine tools and their electric spindle components. The method includes the following steps:

[0006] S1: Obtain the geometric parameters of the CNC machine tool electric spindle. The geometric parameters include the geometric dimensions of the stepped shaft, the hollow hole, and the support joint. The geometric dimensions include the diameter of each segment of the stepped shaft, the inner diameter of the electric spindle, the cylindricity of the electric spindle, and the rotational accuracy of the electric spindle.

[0007] S2: Construct a static model of the electric spindle in finite element analysis software; set the mechanism of the electric spindle as a spatial elastic beam;

[0008] S3: Adjust the assembly position of the angular contact bearing so that the fulcrum of the radial surface of the angular contact bearing and the electric spindle is located at the intersection of the contact line between the electric spindle and the angular contact bearing;

[0009] S4: Use Hertzian contact theory to evaluate the effect of bearing load on the stiffness of angular contact bearings;

[0010] S5: Utilize the deep learning network training dataset to construct a deep learning convolutional network and obtain... and Fitting Characterization nonlinear functions, This indicates the effect of bearing speed on the stiffness of angular contact bearings. This indicates the specific numerical value of the effect of bearing load on the rigidity of the angular contact bearing. This indicates the radial stiffness of the angular contact bearing after static preload.

[0011] S6: Adjust the overall structure of the electric spindle to an interference fit, and mesh the static model. Input the static model and apply full constraints, while simultaneously applying radial load to the static model and calculating the static stiffness value of the electric spindle. ;

[0012] S7: Construct the static model into a three-dimensional finite element dynamic model. While keeping the two-dimensional boundary conditions unchanged, retain the stepped shaft, hollow hole, and support joint, and remove the small structures of chamfer and positioning hole to complete the simplification of the three-dimensional finite element dynamic model.

[0013] S8: The angular contact bearing is simulated as a support unit of four orthogonal springs, and the angle difference between each angular contact bearing is ninety degrees, which is used to characterize the stiffness characteristics of the angular contact bearing.

[0014] S9: Restrict the axial motion degrees of freedom of the electric spindle in the three-dimensional finite element dynamic model, apply full constraints to the outer diameter circle of the angular contact bearing, perform finite element mesh generation on the three-dimensional finite element dynamic model, and obtain the natural frequency and mode shape of the electric spindle through the modal extraction method of Subspace.

[0015] S10: Using the obtained natural frequency and mode shape of the electric spindle, evaluate the vibration characteristics of the electric spindle under different modes, and provide input information for improved methods of elastic support, span optimization, and overhang optimization.

[0016] Furthermore, step S1 also includes performing coordinate scale alignment, null value filling, and interpolation calculation on the geometric parameters to complete data cleaning; and further performing outlier removal, data dimensionality reduction, and feature extraction operations to improve the data effectiveness of subsequent processing.

[0017] Furthermore, in step S2, the angular contact bearing is assembled in groups, and in the finite element model, the model of the assembled angular contact bearing is simplified to a spring mass element with only radial stiffness, and the electric spindle becomes an elastic simulation element under statically indeterminate structural constraints.

[0018] Further, step S4 is performed according to... The effect of bearing load on the stiffness of angular contact bearings is evaluated, among which... This indicates the specific numerical value of the effect of bearing load on its stiffness. The elastic modulus of the bearing material. Poisson's ratio, This refers to the number of rolling elements in an angular contact bearing. The diameter of the rolling element, This is an empirical coefficient determined by the bearing structure and load distribution; according to Evaluate the effect of bearing speed on the stiffness of angular contact bearings, among which... The dynamic viscosity of the oil film formed by the lubricating oil. The angular velocity of the angular contact bearing is determined by its rotational speed. The length of the angular contact bearing. Where is the journal diameter of the angular contact bearing. This refers to the radial clearance during the assembly of angular contact bearings; according to The radial stiffness of the angular contact bearing after static preload is evaluated, wherein... An empirical coefficient determined by the material of the angular contact bearing. The diameter of the rolling element, This refers to the number of rolling elements in an angular contact bearing. For bearing contact angle, The radial preload; step S4 calculates the angular contact bearing under different static preloads. Radial stiffness values ​​below The numerical effect of bearing speed on the stiffness of angular contact bearings Numerical influence of bearing load on the stiffness of angular contact bearings This forms a value encompassing radial stiffness. Influencing values Influence values The deep learning network training dataset.

[0019] Furthermore, the deep learning network training dataset in step S5 includes historical measurement data and corresponding performance evaluation results, so that the performance evaluation results can successively enhance the coverage of the training data.

[0020] Furthermore, step S5 determines the parameters of convolution kernel size, number of convolution kernels, activation function, and pooling layer size through cross-validation and mesh optimization, and sets the radial stiffness value accordingly. As a label value, it affects the numerical value. And the impact value As input values, construct a description and Fitting Characterization Numerical functions make them easy to measure and calculate. and Capable of performing difficult assessments The equivalent representation of .

[0021] Further, step S6 applies an application to the static model. Given the radial load, solve for the static deformation value at the front end of the electric spindle. ,pass Calculate the static stiffness of the electric spindle .

[0022] Furthermore, step S10 provides characteristic parameter input information for the vibration amplitude, resonant frequency, and vibration transmissibility under the first, second, third, and fourth modes for elastic support improvement; step S10 provides input information for the natural frequencies, mode shapes, and bending vibration deformation of the first, second, third, and fourth modes for span optimization; and step S10 provides input information for the peak vibration displacement of the overhang end of the electric spindle for overhang optimization.

[0023] Compared with existing technologies, this invention offers the following advantages: it ensures the efficiency of performance evaluation and analysis of the electric spindle as a key component of CNC machine tools. By fully utilizing the respective advantages of numerical simulation and finite element analysis methods, and combining the function representation and fitting capabilities of deep learning networks, it effectively quantifies the influence of angular contact bearings on the static stiffness of the electric spindle. This provides an accurate and reliable extended foundation for the analysis of three-dimensional finite element dynamic models. Furthermore, it innovatively designs an efficient and reliable simplified three-dimensional model process, reducing the computational complexity of the overall performance evaluation process while ensuring the accuracy of the evaluation of the dynamic and static performance indicators of the electric spindle under analysis. This provides a basis for the optimization of the overall structure and improvement of the dynamic performance of CNC machine tools, and has significant engineering practical value. Attached Figure Description

[0024] Figure 1 This is a flowchart of a method for evaluating and analyzing the performance of CNC machine tool electric spindles based on deep learning networks. Detailed Implementation

[0025] The present invention will now be further described with reference to the accompanying drawings and embodiments.

[0026] This embodiment proposes a CNC machine tool electric spindle performance evaluation and analysis method based on deep learning networks, which includes the following steps:

[0027] S1: Based on the structural characteristics of the CNC machine tool electric spindle, the main geometric dimensions of key parts such as the stepped shaft, hollow hole, and support joint of the CNC machine tool electric spindle are accurately measured using tools such as an outside micrometer, an inside micrometer, and a radial runout instrument. This ensures an accurate reflection of the physical shape parameters of the CNC machine tool electric spindle. The outside micrometer is used to measure the diameter of each section of the stepped shaft, the inside micrometer is used to measure the inner diameter of the electric spindle, and the radial runout instrument is used to measure the cylindricity and rotational accuracy of the electric spindle.

[0028] S2: Based on the dimensional measurement results and actual structural characteristics of the CNC machine tool electric spindle, a lightweight static model of the electric spindle is constructed in the finite element analysis software. The overall electric spindle structure is set as a spatial elastic beam by processing three-dimensional linear beam elements. The model of the three-dimensional linear beam element is set to BEAM188 to reduce the computational complexity in the finite element simulation process.

[0029] S3: Check the precision of the assembly between the electric spindle and the angular contact bearing. Use a contact angle measuring instrument to determine if the fulcrum position exists only at the intersection of the electric spindle axis and the bearing contact line. If this condition is met, simplify the assembled angular contact bearing into a spring mass unit with only radial stiffness, supporting the electric spindle to form an elastic simulation unit under statically indeterminate structural constraints. If this condition is not met, adjust the assembly position of the angular contact bearing and ensure the uniqueness of the intersection position using a contact angle measuring instrument. The measurement deviation of the intersection point by the contact angle measuring instrument should be less than [value missing]. .

[0030] S4: The specific impact of bearing load on the stiffness of angular contact bearings is evaluated using numerical simulation methods. Based on Hertzian contact theory, the impact of bearing load on the stiffness of angular contact bearings can be characterized by the following formula. , in the formula This indicates the specific numerical value of the effect of bearing load on its stiffness. The elastic modulus of the bearing material. Poisson's ratio, This refers to the number of rolling elements in an angular contact bearing. The diameter of the rolling element, This is an empirical coefficient determined by the bearing structure and load distribution.

[0031] The influence of bearing speed on the stiffness of angular contact bearings is evaluated using numerical simulation, taking into account variations in bearing oil film stiffness. The effect of bearing speed on the stiffness of angular contact bearings can be characterized by the following formula: To characterize, in the formula The dynamic viscosity of the oil film formed by the lubricating oil. The angular velocity of the angular contact bearing is determined by its rotational speed. The length of the angular contact bearing. Where is the journal diameter of the angular contact bearing. This refers to the radial clearance during the assembly of angular contact bearings.

[0032] The radial stiffness of angular contact bearings after static preload was evaluated using numerical simulation. ,in, An empirical coefficient determined by the material of the angular contact bearing. The diameter of the rolling element, This refers to the number of rolling elements in an angular contact bearing. For bearing contact angle, This is the radial preload.

[0033] S5: Through multiple sets of experimental designs, the angular contact bearings under different static preloads were calculated. Radial stiffness values ​​below Simultaneously calculate the numerical influence of bearing speed on the stiffness of angular contact bearings. The influence of bearing load on the stiffness of angular contact bearings This forms a value encompassing radial stiffness. Influence values Influence values Load, speed and combined characteristics , , A deep learning network training dataset was used to convert bearing physical parameters into network input tensors. Table 1 lists the specific composition of the dataset. All input features and target values ​​were mapped to the [0,1] interval using a normalization method to eliminate differences between different units. The dataset was divided into training and test sets in an 8:2 ratio, and a fixed random seed was set to ensure the reproducibility of experimental results. The training set was randomly shuffled during loading, while the test set was loaded sequentially to improve training efficiency and result stability. There were no fewer than 20 experimental groups, with different groups corresponding to different static preload conditions.

[0034] Table 1. Specific composition of the deep learning network training dataset

[0035] Column type Column names meaning unit Data range Original feature columns preload_force Static preload N - Original feature columns load bearing load N - Original feature columns speed bearing speed RPM - Derived feature columns Load stiffness Pa / m 1.18e+12 ~ 1.26e+12 Derived feature columns Rotational stiffness Pa / m 2.58e+07 ~ 2.58e+08 Extended feature columns Load-speed stiffness interaction characteristics (Pa / m)² - Extended feature columns Secondary characteristic of load stiffness (Pa / m)² - Extended feature columns Rotational stiffness quadratic characteristic (Pa / m)² - Target column Preload stiffness

[0036] In terms of network structure design, a one-dimensional convolutional neural network is used as the feature extraction module, and a fully connected network is superimposed as the prediction module. The input layer receives a tensor of shape [batch size, 1, 8], corresponding to 8 input features. Then, four convolutional layers are set: the first convolutional layer has 32 output channels and a kernel size of 3, with padding to maintain the same input and output size. This layer is followed by a batch normalization layer, a LeakyReLU activation function with a negative slope of 0.1, and a max pooling layer with a kernel size of 2 and a stride of 2. After this processing, the feature size is reduced from 8 to 4. The second convolutional layer has 64 output channels and a kernel size of 3, with the same padding, followed by batch normalization and a LeakyReLU activation function. The third convolutional layer has 128 output channels and a kernel size of 3, followed by batch normalization and LeakyReLU. The fourth convolutional layer has 256 output channels and a kernel size of 3, followed by batch normalization and LeakyReLU. The tensor output of the convolutional part has a size of [batch size, 256, 4], which is flattened into a two-dimensional structure of [batch size, 1024] and used as the input to the fully connected layer. The fully connected part consists of four layers: the first fully connected layer maps the input dimension of 1024 to 256, followed by the LeakyReLU activation function and a random inactivation rate of 0.4; the second fully connected layer maps the 256 dimension to 128, followed by LeakyReLU and a random inactivation rate of 0.3; the third fully connected layer maps the 128 dimension to 64, followed by LeakyReLU and a random inactivation rate of 0.2; and the final output layer maps the 64 dimension to a single output for predicting radial preload stiffness. The overall network structure is a cascaded combination of four convolutional layers and four fully connected layers. For the training strategy, a smooth L1 loss function is chosen. This function is similar to mean squared error in the small error range and similar to mean absolute error in the large error range, thus balancing convergence smoothness and robustness to outliers. The Adam optimizer was used for optimization, with an initial learning rate of 0.001. The exponential weighting coefficients for the first and second moments were set to 0.9 and 0.999, respectively. The numerical stability factor was set to 10^(-8), and the weight decay coefficient was set to 10^(-4) to avoid overfitting. A periodic decay method based on cosine annealing was adopted, with an initial period of 10 training epochs, which subsequently doubled, with a minimum learning rate of 1×10^(-6). This method maintains a relatively fast convergence speed in the early stages of training and enables refined search in the later stages. An early stopping mechanism was implemented during training, automatically terminating training when the validation set loss showed no significant improvement within 20 consecutive training epochs to avoid invalid iterations and overfitting. The maximum number of training epochs was 300, but in practice, training often ended prematurely due to the early stopping mechanism. The batch size was 64 samples. During training, the GPU was prioritized for acceleration; if no GPU was available, training reverted to the CPU environment.

[0037] Training completed to determine the radial stiffness values ​​of diagonal contact bearings After training the deep learning network for fitting calculations, it is ensured that the overall structure of the electric spindle to be analyzed is in an interference fit. If the interference fit is not met, the assembly process is adjusted to meet the interference fit. If the interference fit is met, the simulation of the two-dimensional finite element static model of the electric spindle is further completed. Based on the two-dimensional static model, finite element elements are divided to complete the mesh generation. For the 400H type electric spindle, a total of 368 mesh elements are generated.

[0038] S6: Based on the fitting results of the deep learning network, output the radial stiffness value of the angular contact bearing. The data is input into a two-dimensional finite element static model to apply full constraints to the angular contact bearing elements. Simultaneously, based on the main operating conditions of the electric spindle to be analyzed, forces under different radial loads are applied to the front section of the electric spindle. Solve for the static deformation value at the front end of the electric spindle. ,pass The formula definition yields the static stiffness values ​​at the component level of the electric spindle. The final static stiffness performance index of the electric spindle under analysis was determined by averaging the static stiffness values ​​under different radial loads. The force range of the radial load applied to the front section of the electric spindle was as follows: .

[0039] S7: Extend the two-dimensional finite element static model of the electric spindle to be analyzed into a three-dimensional finite element dynamic model. While keeping the boundary conditions corresponding to the two-dimensional finite element static model unchanged, retain the key geometric features of the electric spindle and remove the small structures that do not affect the overall dynamic performance analysis results. In this way, while ensuring that all constraints are satisfied, the three-dimensional finite element dynamic model is simplified efficiently. The boundary conditions corresponding to the two-dimensional finite element static model include, but are not limited to, statically indeterminate constraints, fixed size constraints, and bearing radial loads. The key geometric features of the electric spindle that are retained include, but are not limited to, the electric spindle diameter, the position of the angular contact bearing, and the size of the angular contact bearing. The small structures of the electric spindle that are removed include, but are not limited to, transition arcs, small holes, and chamfers.

[0040] S8: For the 3D extended angular contact bearing, it is simulated as a dynamic support unit corresponding to four orthogonal springs. For each group of angular contact bearings, the angle difference between each spring is 90 degrees, which are used to characterize the radial stiffness and axial stiffness characteristics of the angular contact bearing. The four orthogonal springs corresponding to each group of angular contact bearings are simulated by the COMBIN14 spring element.

[0041] S9: While restricting the axial motion degrees of freedom of the three-dimensional finite element dynamic model of the electric spindle to be analyzed, full constraints are set on the outer diameter circle of the angular contact bearing, and the radial motion degrees of freedom constraints of the bearing simulated by the orthogonal spring are improved. Finite element meshing is performed on the three-dimensional finite element dynamic model of the electric spindle. Then, the natural frequencies and mode shapes of the electric spindle are analyzed by the modal extraction method. While determining the modal order of the extracted mode shape, the influence of the low-order modes on the dynamic performance of the electric spindle to be analyzed is focused. The modal extraction method is set to the Subspace method, and the modal order of the extracted electric spindle mode shape is 4.

[0042] S10: By modal analysis and extended extraction of the three-dimensional finite element dynamic model of the electric spindle, the vibration characteristics of the electric spindle under different modes are obtained, and the dynamic performance index is rapidly evaluated and analyzed, thereby effectively ensuring subsequent improvements to the overall structure such as elastic support, span optimization, and overhang optimization.

[0043] The following is an example of a CNC machining center Taking the performance evaluation and analysis results of the built-in high-speed electric spindle (spindle structure material is 20MnCr5) as an example, we can further illustrate the specific implementation and effect of this invention.

[0044] First, for the established two-dimensional finite element static model of the electric spindle, a value range of is asymptotically applied at the front end point of the electric spindle. A radial load was applied, with a progressive adjustment interval of 300 N. The corresponding displacement dimension of the electric spindle tip was determined using a finite element analysis model. Simultaneously, a corresponding radial load was applied to the physical entity of the high-speed electric spindle using a hydraulic loading device, and the corresponding displacement dimension of the electric spindle tip was measured using a laser displacement sensor. The experimental results are shown in Table 2. The average radial static stiffness of the electric spindle tip obtained from the two-dimensional finite element static model was... The average true radial static stiffness of the electric spindle tip, obtained through actual physical experiments, is: The proposed method achieves rapid evaluation and analysis with high accuracy in solving static performance indices. Table 2 provides the displacement dimensions and static stiffness results of the electric spindle front end under different radial loads.

[0045] Table 2. Displacement dimensions and static stiffness results of the electric spindle front end under different radial loads.

[0046] Displacement ( Radial load ( ) 3000 3300 3600 3900 4200 4500 Solving displacement using a two-dimensional finite element model 198.6 220.1 243.2 261.7 280.0 296.1 Laser sensor measures displacement 199.2 222.2 245.3 263.5 281.9 298.3 Displacement calculation error 0.30% 0.95% 0.86% 0.68% 0.67% 0.74%

[0047] Using finite element analysis software and the Subspace modal extraction method, the natural frequencies and mode shapes of the three-dimensional finite element dynamic model of the electric spindle at different orders are shown in Table 2. Simultaneously, using a modal analysis testing system and a field experimental method of single-point excitation and multi-point response picking, the actual natural frequencies and mode shapes of the electric spindle under analysis at each order were obtained, as shown in Table 3. Furthermore, comparing the dynamic performance indicators evaluated by the three-dimensional finite element dynamic model with those measured by actual physical experiments, it can be found that the dynamic mode shapes obtained by the proposed method are consistent with the actual mode shapes, and the error of the natural frequencies solved at each order is less than 7%, effectively proving the accuracy of the proposed method in dynamic performance evaluation. Table 3 provides the solved values ​​and actual experimental values ​​of the dynamic performance indicators of the electric spindle under analysis at different orders.

[0048] Table 3. Solved and experimental values ​​of dynamic performance indicators of the electric spindle under different orders.

[0049] Electric spindle mode order 1 2 3 4 Modal order corresponds to mode shape First-order radial bending Second-order radial bending Third-order two-end bending Fourth-order radial bending Dynamic model evaluation frequency 634.8 1322.6 1732.7 1984.5 Field test true frequency 618.0 1260.0 1620.0 1870.0 Frequency evaluation error results 2.72% 4.97% 6.96% 6.12%

[0050] This invention provides a deep learning network-based method for evaluating the performance of CNC machine tool electric spindles, significantly improving the efficiency of performance evaluation for electric spindles as a core component of CNC machine tools. By integrating the advantages of numerical simulation and finite element analysis, and combining the powerful function fitting capabilities of deep learning networks, this method accurately quantifies the influence of angular contact bearings on the static stiffness of the electric spindle, providing a reliable foundation for the analysis of three-dimensional finite element dynamic models. Furthermore, this invention innovatively designs an efficient three-dimensional model simplification process, reducing overall computational complexity while ensuring the accuracy of dynamic and static performance index evaluation. This method not only provides a scientific basis for the overall structural optimization and dynamic performance improvement of CNC machine tools but also has significant engineering application value.

[0051] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications and improvements without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be defined by the claims.

Claims

1. A method for performance evaluation and analysis of CNC machine tool electric spindles based on deep learning networks, characterized in that, Includes the following steps: S1: Obtain the geometric parameters of the CNC machine tool electric spindle. The geometric parameters include the geometric dimensions of the stepped shaft, the hollow hole, and the support joint. The geometric dimensions include the diameter of each segment of the stepped shaft, the inner diameter of the electric spindle, the cylindricity of the electric spindle, and the rotational accuracy of the electric spindle. S2: Construct a static model of the electric spindle in finite element analysis software; set the mechanism of the electric spindle as a spatial elastic beam; S3: Adjust the assembly position of the angular contact bearing so that the fulcrum of the radial surface of the angular contact bearing and the electric spindle is located at the intersection of the contact line between the electric spindle and the angular contact bearing; S4: Use Hertzian contact theory to evaluate the effect of bearing load on the stiffness of angular contact bearings; S5: Utilize the training dataset for deep learning networks to construct deep learning convolutional networks and obtain... and Fitting Characterization Nonlinear functions; This indicates the effect of bearing speed on the stiffness of angular contact bearings. This indicates the specific numerical value of the effect of bearing load on its stiffness. This indicates the radial stiffness of the angular contact bearing after static preload. S6: Adjust the overall structure of the electric spindle to an interference fit, and mesh the static model. Input the static model and apply full constraints, simultaneously apply radial load to the static model, and calculate the static stiffness value of the electric spindle. ; S7: Construct the static model into a three-dimensional finite element dynamic model. While keeping the two-dimensional boundary conditions unchanged, retain the stepped shaft, hollow hole, and support joint, and remove the small structures of chamfer and positioning hole to complete the simplification of the three-dimensional finite element dynamic model. S8: The angular contact bearing is simulated as a support unit of four orthogonal springs, and the angle difference between each angular contact bearing is ninety degrees, which is used to characterize the stiffness characteristics of the angular contact bearing. S9: Restrict the axial motion degrees of freedom of the electric spindle in the three-dimensional finite element dynamic model, apply full constraints to the outer diameter circle of the angular contact bearing, perform finite element mesh generation on the three-dimensional finite element dynamic model, and obtain the natural frequency and mode shape of the electric spindle through the modal extraction method of Subspace. S10: Using the obtained natural frequency and mode shape of the electric spindle, evaluate the vibration characteristics of the electric spindle under different modes, and provide input information for improved methods of elastic support, span optimization, and overhang optimization.

2. The method for performance evaluation and analysis of CNC machine tool electric spindles based on deep learning networks according to claim 1, characterized in that, Step S1 further includes performing coordinate scale alignment, null value filling, and interpolation calculation on the geometric parameters to complete data cleaning; and further performing outlier removal, data dimensionality reduction, and feature extraction operations to improve the data effectiveness of subsequent processing.

3. The method for performance evaluation and analysis of CNC machine tool electric spindles based on deep learning networks according to claim 1, characterized in that, In step S2, the angular contact bearings are assembled in groups, and in the finite element model, the model of the assembled angular contact bearings is simplified to a spring mass element with only radial stiffness. The electric spindle becomes an elastic simulation element under the constraint of a statically indeterminate structure.

4. The method for performance evaluation and analysis of CNC machine tool electric spindles based on deep learning networks according to claim 1, characterized in that, The step S4 is as follows The effect of bearing load on the stiffness of angular contact bearings is evaluated, among which... This indicates the specific numerical value of the effect of bearing load on its stiffness. The elastic modulus of the bearing material. Poisson's ratio, This refers to the number of rolling elements in an angular contact bearing. The diameter of the rolling element, This is an empirical coefficient determined by the bearing structure and load distribution; according to Evaluate the effect of bearing speed on the stiffness of angular contact bearings, where... The dynamic viscosity of the oil film formed by the lubricating oil. The angular velocity of the angular contact bearing is determined by its rotational speed. The length of the angular contact bearing. Where is the journal diameter of the angular contact bearing. This refers to the radial clearance during the assembly of angular contact bearings; according to... The radial stiffness of the angular contact bearing after static preload is evaluated, wherein... An empirical coefficient determined by the material of the angular contact bearing. The diameter of the rolling element, This refers to the number of rolling elements in an angular contact bearing. For bearing contact angle, The radial preload; step S4 calculates the angular contact bearing under different static preloads. Radial stiffness values ​​below The numerical effect of bearing speed on the stiffness of angular contact bearings Numerical influence of bearing load on the stiffness of angular contact bearings This forms a value encompassing radial stiffness. Influence values Influencing values The deep learning network training dataset.

5. The method for performance evaluation and analysis of CNC machine tool electric spindles based on deep learning networks according to claim 1, characterized in that, The deep learning network training dataset in step S5 includes historical measurement data and corresponding performance evaluation results, so that the performance evaluation results can successively enhance the coverage of the training data.

6. The method for performance evaluation and analysis of CNC machine tool electric spindles based on deep learning networks according to claim 1, characterized in that, Step S5 determines the parameters of convolution kernel size, number of convolution kernels, activation function, and pooling layer size through cross-validation and mesh optimization, and sets the radial stiffness value accordingly. As a label value, it affects the numerical value. And the impact value As input values, construct a description and Fitting Characterization Numerical functions make them easy to measure and calculate. and Capable of performing difficult assessments The equivalent representation of .

7. The method for performance evaluation and analysis of CNC machine tool electric spindles based on deep learning networks according to claim 1, characterized in that, Step S6 applies a static model. Given the radial load, solve for the static deformation value at the front end of the electric spindle. ,pass Calculate the static stiffness value of the electric spindle .

8. The method for performance evaluation and analysis of CNC machine tool electric spindles based on deep learning networks according to claim 1, characterized in that, Step S10 provides characteristic parameter input information for the vibration amplitude, resonant frequency, and vibration transmissibility under the first, second, third, and fourth modes for elastic support improvement; Step S10 provides input information for the natural frequency, mode shape distribution, and bending vibration deformation of the first, second, third, and fourth modes for span optimization; Step S10 provides input information for the peak vibration displacement of the overhang end of the electric spindle for overhang optimization.

Images