Tool holder position collaborative control method based on industrial big data

Abstract

This invention relates to program-controlled direction in general control systems, specifically to a collaborative control method for tool holder position based on industrial big data. This method collects multi-dimensional operating status data of a CNC lathe and retrieves historical machining feature data. This data is input into an eccentricity prediction model, which outputs a predicted eccentricity vector. Based on this predicted eccentricity vector and the kinematic constraints of the tool holder auxiliary axis, the target displacement is calculated. A collaborative control command is generated based on the target displacement and sent to the tool holder auxiliary axis driver, controlling the tool to move along a direction perpendicular to the spindle axis to a position concentric with the spindle axis. This method transforms offline trial-cut physical compensation into online closed-loop displacement control based on multi-source data fusion and model prediction. It performs eccentricity prediction and automatic displacement compensation based on the real-time state of the machine tool, eliminating the interference of discrete errors from manual measurement and shim thickness grading limitations on the correction process.

Description

Technical Field

[0001] This invention relates to program-controlled direction in general control systems, and more specifically to a tool holder position collaborative control method based on industrial big data. Background Technology

[0002] In CNC turning processes in the mechanical manufacturing field, drilling, reaming, and tapping operations require the tool axis to maintain strict concentricity with the spindle axis. Currently, most commercially available CNC lathes only have transverse and longitudinal feed axes; the tool holder itself lacks independent adjustment functionality perpendicular to the spindle axis. To achieve this concentricity requirement, existing technologies typically employ a combination of manual physical adjustment and trial cutting correction. The specific process is as follows: After the operator installs the tool in the tool holder, they start the machine for a trial cut. Then, the workpiece is removed, and a measuring instrument is used to obtain the hole offset. Based on this offset, the operator selects a metal shim of a specific thickness and inserts it between the tool holder and the tool, or manually rotates the fine-tuning screw on the tool holder to adjust the position. After one physical adjustment, another trial cut and measurement are performed. This process of repeated trial cuts, measurements, and shim adjustments continues until the measured offset falls within the allowable tolerance zone, thus completing the concentricity correction.

[0003] The core problem with the aforementioned existing technical solutions lies in the fact that the correction method relying on manual trial cutting combined with physical shims cannot predict eccentricity and automatically compensate for displacement based on the real-time dynamic operating status of the machine tool. Trial cutting occurs in the dynamic environment of the machine tool, and factors such as spindle deformation and tool vibration cause the offset measured during trial cutting to contain dynamic errors. Using this static measurement result to guide the physical adjustment of the shim thickness is essentially a delayed and distorted compensation for the dynamic eccentricity. Without an independent drive axis perpendicular to the spindle axis, operators cannot correct the tool position in real time based on state signals reflecting dynamic eccentricity, such as spindle vibration and motor current, during machine operation. This results in the entire correction process being in an open-loop, offline adjustment state. The correction results are limited by the graded limitations of the shim thickness and the discrete errors of manual measurement, failing to achieve continuous displacement compensation that matches the dynamic operating characteristics of the machine tool. Summary of the Invention

[0004] The purpose of this invention is to provide a collaborative control method for tool holder position based on industrial big data, which can effectively solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A collaborative control method for tool holder position based on industrial big data includes: Collect multi-dimensional operating status data of the CNC lathe under the current machining task, and retrieve historical machining feature data that matches the current machining task from the industrial big data platform; The multidimensional operating status data and the historical machining feature data are input into the pre-constructed eccentricity prediction model, and the predicted eccentricity vector of the tool relative to the spindle axis is output. Based on the predicted eccentricity vector and combined with the kinematic constraints of the tool holder auxiliary Y-axis, the target displacement of the Y-axis is calculated. Based on the target displacement of the Y-axis, a collaborative control command is generated and sent to the tool holder auxiliary Y-axis driver to control the tool holder auxiliary Y-axis driver to drive the tool to move along the direction perpendicular to the spindle axis to a position concentric with the spindle axis.

[0007] Preferably, the multidimensional operating status data includes spindle radial vibration signal, spindle motor three-phase current signal, and ambient sound signal in the tool holder area; During the acquisition of the multidimensional operating status data, the radial vibration signal of the spindle is synchronously sampled, the three-phase current signal of the spindle motor is subjected to fast Fourier transform to extract the current harmonic components, and the ambient sound signal of the tool holder area is subjected to Mel frequency cepstral coefficient feature extraction. The extracted current harmonic components and the Mel frequency cepstral coefficient features are timestamped with the synchronously sampled radial vibration signal of the spindle to generate a fused time-series feature matrix as the multidimensional operating status data.

[0008] Preferably, the eccentricity prediction model includes a spatial feature extraction branch and a temporal feature extraction branch; In the process of inputting the multidimensional operating status data and the historical processing feature data into the eccentricity prediction model, the multidimensional operating status data is processed by two-dimensional convolution through the spatial feature extraction branch, and the historical processing feature data is processed by long short-term memory network through the temporal feature extraction branch. The output results of the spatial feature extraction branch and the output results of the temporal feature extraction branch are concatenated and then input into the fully connected layer. The predicted eccentricity vector is output through the fully connected layer.

[0009] Preferably, the kinematic constraints of the tool holder auxiliary Y-axis include the lead screw backlash compensation amount and the lead screw pitch error compensation amount; In the process of calculating the target displacement of the Y-axis based on the predicted eccentricity vector and the kinematic constraints of the tool holder auxiliary Y-axis, the current rotation angle of the tool holder auxiliary Y-axis driver is obtained. Based on the current rotation angle, the backlash compensation amount of the leadscrew is obtained by querying the pre-established backlash compensation mapping table. Based on the projection component of the predicted eccentricity vector in the moving direction of the tool holder auxiliary Y-axis, the backlash compensation amount of the leadscrew and the leadcrew pitch error compensation amount are added to calculate the target displacement of the Y-axis.

[0010] Preferably, during the process of generating the collaborative control command based on the target displacement of the Y-axis, the current interpolation motion trajectory of the CNC lathe X-axis and Z-axis is obtained, and it is determined whether the current interpolation motion trajectory is in a linear interpolation state or a circular interpolation state. If the current interpolation motion trajectory is in the linear interpolation state, the target displacement of the Y-axis is allocated to the non-cutting reversal period of the current interpolation motion trajectory to generate the collaborative control command. If the current interpolation trajectory is in the circular interpolation state, the target displacement of the Y-axis is divided into multiple sub-displacements according to a preset ratio, and the multiple sub-displacements are respectively assigned to each interpolation cycle of the circular interpolation trajectory to generate the cooperative control command.

[0011] Preferably, after controlling the tool holder auxiliary Y-axis driver to move the tool along a direction perpendicular to the spindle axis to a position concentric with the spindle axis, the actual displacement data fed back by the micro grating ruler set on the tool holder auxiliary Y-axis is obtained, the displacement deviation value between the actual displacement data and the Y-axis target displacement is calculated, and it is determined whether the displacement deviation value is within the preset deviation range. If the displacement deviation value is not within the preset deviation range, the displacement deviation value is superimposed on the predicted eccentricity vector to recalculate the updated Y-axis target displacement, and an updated cooperative control command is generated based on the updated Y-axis target displacement and sent to the tool holder auxiliary Y-axis driver.

[0012] Preferably, during the synchronous sampling of the spindle radial vibration signal, the synchronously sampled spindle radial vibration signal is subjected to three-level wavelet packet decomposition to extract eight frequency band signals obtained after the three-level wavelet packet decomposition. The energy proportion of each frequency band signal in the eight frequency band signals is calculated, and target frequency band signals with energy proportions greater than a preset energy threshold are selected. The wavelet packet coefficients of the target frequency band signals are reconstructed to obtain a reconstructed vibration signal. The reconstructed vibration signal is used to replace the synchronously sampled spindle radial vibration signal to participate in the generation of the fused temporal feature matrix.

[0013] Preferably, before inputting the output of the spatial feature extraction branch and the output of the temporal feature extraction branch into the fully connected layer after feature concatenation, the concatenated result is input into a multi-head self-attention layer. In the multi-head self-attention layer, the query vector, key vector, and value vector of the concatenated result are mapped according to the tool model code in the current machining task. An attention weight matrix is ​​generated based on the mapping calculation result. The attention weight matrix is ​​used to weight the concatenated result, and the weighted result is input into the fully connected layer.

[0014] Preferably, the kinematic constraints further include the lead screw thermal deformation compensation amount; In the process of calculating the target displacement of the Y-axis, real-time temperature data collected by a temperature sensor located at the auxiliary Y-axis ball screw bearing seat of the tool holder is obtained. The real-time temperature data is input into a pre-constructed thermal error mapping matrix. The thermal error mapping matrix is ​​a polynomial matrix generated by fitting the ball screw thermal elongation at different temperature nodes in the industrial big data platform. The ball screw thermal deformation compensation amount is calculated through the thermal error mapping matrix and then superimposed on the target displacement of the Y-axis.

[0015] Preferably, in the process of acquiring the actual displacement data fed back by the micro grating ruler set on the auxiliary Y-axis of the tool holder, the real-time humidity and real-time air pressure of the environment where the micro grating ruler is located are acquired, an air refractive index correction coefficient is calculated based on the real-time humidity and real-time air pressure, the grating pitch parameter of the micro grating ruler is dynamically updated based on the air refractive index correction coefficient, the original pulse signal output by the micro grating ruler is interpolated and subdivided using the dynamically updated grating pitch parameter, and the result of the interpolation and subdivision calculation is used as the actual displacement data.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention collects multi-dimensional operating status data and retrieves historical machining feature data. Both are input into an eccentricity prediction model to output a predicted eccentricity vector. Based on this vector and the kinematic constraints of the tool holder auxiliary axis, the target displacement is calculated, and a coordinated control command is generated. This drives the tool to move to a concentric position along a direction perpendicular to the spindle axis. This invention transforms traditional offline trial-cutting physical shim compensation into online closed-loop displacement control based on multi-source data fusion and model prediction. It utilizes operating status data and historical feature data to extract the dynamic eccentricity characteristics of the machine tool under specific machining tasks. The calculated target displacement is directly executed through the auxiliary axis driver, achieving eccentricity prediction and automatic displacement compensation based on the real-time state of the machine tool. This eliminates the interference of discrete errors from manual trial-cutting measurements and shim thickness grading limitations on the correction process.

[0017] 2. Based on the above scheme, by aligning the timestamps of the spindle radial vibration signal, the current harmonic components of the three-phase current of the spindle motor, and the cepstral coefficients of the ambient sound signal to generate a fused time-series feature matrix, the dimensional richness of the input data of the eccentric prediction model is increased. By introducing the lead screw backlash compensation, lead screw pitch error compensation, and lead screw thermal deformation compensation calculated based on real-time temperature data through a thermal error mapping matrix when calculating the target displacement, the backlash, pitch manufacturing deviation, and thermal elongation in the auxiliary shaft mechanical transmission chain are numerically offset, reducing the interference of mechanical transmission errors on the final displacement execution result. By acquiring the current interpolation motion trajectory of the CNC lathe's multi-axis, the target displacement is allocated to the non-cutting reversal period in linear interpolation mode, and the target displacement is divided into multiple sub-displacements and allocated to each interpolation cycle in circular interpolation mode, avoiding the disruption of the continuity of the original cutting interpolation trajectory by the auxiliary shaft displacement action. Attached Figure Description

[0018] Figure 1 This is a flowchart of the overall method for collaborative control of tool holder position based on industrial big data according to the present invention; Figure 2 The flowchart for the preprocessing and fusion of multidimensional operational status data and generation of time-series feature matrices in this invention is as follows; Figure 3 This is a flowchart illustrating the network structure and data processing of the eccentric prediction model of the present invention. Figure 4 This is a flowchart of the Y-axis target displacement calculation based on kinematic constraints according to the present invention. Figure 5 This is a flowchart illustrating the collaborative control command generation and interpolation trajectory coordination of the present invention. Figure 6 This is a flowchart of the closed-loop displacement deviation correction and grating ruler measurement correction of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] Please refer to Figure 1This embodiment provides a tool holder position collaborative control method based on industrial big data, applied to CNC turning equipment equipped with a tool holder auxiliary Y-axis. It is used to achieve online concentricity correction and automatic position control between the tool axis and the spindle axis during drilling, reaming, and tapping processes. The movement direction of the tool holder auxiliary Y-axis is perpendicular to the axis of the CNC lathe spindle, which is defined as the Z-axis. The original transverse feed axis of the CNC lathe is the X-axis. The tool holder auxiliary Y-axis is in the same plane as the X-axis and perpendicular to it. The tool is mounted on the slide of the tool holder auxiliary Y-axis, which is driven by an independent servo driver. This allows the tool to move linearly along the Y-axis to adjust its axis position and maintain concentricity with the spindle axis.

[0021] In one embodiment, the CNC lathe receives the current machining task, which includes the material properties of the workpiece to be machined, the machining process type, the tool model, the rated spindle speed, the cutting feed parameters, and the machining tolerance requirements. The local data acquisition unit of the CNC lathe collects multi-dimensional operating status data of the CNC lathe under the current machining task through corresponding sensors. Specifically, the radial vibration signal of the spindle is collected by a two-axis accelerometer installed at the bearing housing at the front end of the spindle, with the collection direction parallel to the X-axis and Y-axis directions, respectively, to obtain the radial vibration time-domain signal of the spindle in a plane perpendicular to the axis; the three-phase current signal of the spindle motor is collected by a current transformer installed at the output end of the spindle servo driver to obtain the real-time three-phase current time-domain signal of the spindle motor during the machining process; the ambient sound signal of the tool holder area is collected by a sound pickup sensor installed on the tool holder body, with the pickup surface of the sound pickup sensor facing the contact area between the tool and the workpiece, to obtain the airborne sound signal of the tool holder area during the machining process.

[0022] Furthermore, while collecting multi-dimensional operational status data, the CNC lathe's industrial Ethernet communication unit establishes a data connection with the industrial big data platform, retrieving historical machining feature data matching the current machining task from the platform. Specifically, the industrial big data platform stores full machining data generated during historical machining processes for this model of CNC lathe and other CNC lathes of the same model and batch. Each set of historical machining data corresponds to a unique machining task tag, which contains feature parameters of the same dimension as the current machining task. When retrieving historical machining feature data, the feature parameters of the current machining task are used as search conditions. The feature parameters of the current machining task are converted into multi-dimensional feature vectors, and each dimension of the feature vector is normalized to eliminate the dimensional differences between different dimensions. The cosine similarity between the feature vector of the current machining task and the feature vectors of each historical machining data in the industrial big data platform is calculated. Historical machining datasets with a cosine similarity greater than a preset matching threshold are selected. The preset matching threshold is pre-set based on the matching accuracy requirements of the machining conditions to ensure that the selected historical machining data has a high degree of consistency with the current machining task's conditions. Historical feature data related to tool eccentricity were extracted from the filtered historical machining dataset, including the actual tool eccentricity under different working conditions during historical machining, the corresponding spindle operating state parameters, tool seat displacement compensation data, correlation data between tool wear and eccentricity, and eccentricity change trend data under different cutting parameters.

[0023] In this embodiment, the collected multidimensional operating status data and retrieved historical processing feature data are input into a pre-constructed eccentricity prediction model. The eccentricity prediction model is pre-trained using a historical processing dataset from an industrial big data platform. The input to the training dataset consists of multidimensional operating status data and historical processing feature data from the historical processing process. The label of the training dataset is the actual eccentricity vector of the tool relative to the spindle axis, measured by a laser tool setter at the corresponding processing moment. During model training, mean squared error is used as the loss function, calculated as the mean squared error between the predicted and measured eccentricity vectors. The Adam optimizer iteratively optimizes the model's weight parameters. The batch size and iteration number are pre-set based on the size of the training dataset. When the loss function value on the validation set no longer decreases after several consecutive iterations, training is terminated early, and the trained model weight parameters are saved to ensure the model has good generalization ability and prediction accuracy. The output of the eccentricity prediction model is the predicted eccentricity vector of the tool relative to the spindle axis. The predicted eccentricity vector is a two-dimensional vector. The two dimensions of the vector correspond to the eccentricity components in the X-axis and Y-axis directions, respectively. The magnitude of the vector is the straight-line distance between the tool axis and the spindle axis, and the direction of the vector is the offset direction of the tool axis relative to the spindle axis.

[0024] Furthermore, based on the predicted eccentricity vector obtained from the output, and combined with the kinematic constraints of the tool holder-assisted Y-axis, the target displacement of the Y-axis is calculated. Specifically, the kinematic constraints of the tool holder-assisted Y-axis are a displacement compensation constraint model pre-established based on the mechanical transmission structure parameters of the tool holder-assisted Y-axis. The kinematic constraints include the lead screw backlash compensation and the lead screw pitch error compensation, which are used to compensate for the errors in the mechanical transmission chain of the tool holder-assisted Y-axis, ensuring the consistency between the actual displacement and the commanded displacement of the Y-axis. When calculating the target displacement of the Y-axis, the projection component of the predicted eccentricity vector in the Y-axis direction is first extracted. This projection component is the basic displacement required to eliminate tool eccentricity. Then, the basic displacement is corrected by combining the compensation amount in the kinematic constraints, and finally the target displacement of the Y-axis is obtained.

[0025] In this embodiment, a collaborative control command is generated based on the calculated target Y-axis displacement. This command is a pulse control command or bus control command conforming to the communication protocol of the tool holder auxiliary Y-axis driver, and includes displacement, speed, and acceleration commands corresponding to the target displacement. The generated collaborative control command is sent to the tool holder auxiliary Y-axis driver via an industrial fieldbus. Upon receiving the command, the driver drives a servo motor, which in turn moves the slide of the tool holder auxiliary Y-axis along the Y-axis direction via a lead screw transmission mechanism. This, in turn, moves the tool mounted on the slide along the Y-axis direction perpendicular to the spindle axis until the tool is concentric with the spindle axis.

[0026] Table 1. Correspondence between processing task matching dimensions and historical feature data

[0027] Specifically, the table defines the matching dimensions between the current processing task and historical processing data. Each matching dimension corresponds to a unique historical feature data extraction item. When matching historical processing data, similarity is calculated sequentially according to the matching dimensions listed in the table. The similarity weight of each dimension is pre-set based on the degree of influence of the processing procedure on the eccentricity, ensuring that the matching degree between the retrieved historical processing feature data and the current processing task meets the requirements of the model input.

[0028] In this embodiment, by collecting multi-dimensional operating status data of the CNC lathe under the current machining task, retrieving the matching historical machining feature data in the industrial big data platform, outputting the predicted eccentricity vector of the tool through the eccentricity prediction model, and calculating the target displacement by combining the kinematic constraints of the tool holder auxiliary Y-axis, a collaborative control command is finally generated to drive the tool to move to a position concentric with the spindle axis. This fully realizes online prediction and automatic position compensation of tool eccentricity, transforming the traditional offline static correction into online dynamic control that matches the real-time operating status of the machine tool.

[0029] In a preferred embodiment, reference Figure 2 During the acquisition of multi-dimensional operational status data, various acquired signals undergo preprocessing and feature extraction to generate a fused temporal feature matrix, which serves as the input to the eccentricity prediction model. Specifically, the multi-dimensional operational status data includes spindle radial vibration signals, spindle motor three-phase current signals, and ambient sound signals from the tool holder area. Corresponding preprocessing and feature extraction operations are performed on each of these three types of signals.

[0030] The radial vibration signal of the spindle is synchronously sampled. The trigger source for synchronous sampling is the Z-phase pulse signal output by the spindle encoder. For each revolution of the spindle, the encoder outputs a Z-phase pulse, which serves as the synchronous trigger signal for sampling. Simultaneously, the A-phase and B-phase quadrature pulses output by the encoder are used as counting pulses to sample the radial vibration signal at equal angular intervals. The configuration of the encoder lines and sampling points satisfies the requirement of equal angular interval sampling, ensuring that the sampled vibration signal is strictly synchronized with the spindle rotation angle. This eliminates interference caused by spindle speed fluctuations on the time-domain correspondence of the sampled signal. Even during spindle acceleration and deceleration, the correspondence between the sampling points and the spindle rotation angle remains unchanged, ensuring that the characteristics of the vibration signal correspond to the spindle's rotational position, providing an accurate time-domain data foundation for subsequent eccentricity feature extraction. The sampling frequency is set based on the spindle's maximum rated speed, ensuring that the number of sampling points per revolution is greater than the preset minimum number of sampling points, satisfying the Nyquist sampling theorem.

[0031] Furthermore, a Fast Fourier Transform (FFT) is performed on the three-phase current signal of the spindle motor to extract the current harmonic components. Specifically, the synchronously sampled three-phase current signal is segmented into frames, with the length of each frame matching the number of sampling points per cycle of the spindle radial vibration signal. After windowing each frame, a FFT is performed to convert the time-domain current signal into a frequency-domain spectral signal. The fundamental component and harmonic components of a preset order, including odd harmonic components, are extracted from the spectral signal. The amplitude and phase of each harmonic component are used as the extracted current harmonic features. For discrete current time-domain signal sequences... , ,in The number of sampling points in a single frame signal is given by the formula for calculating its Fast Fourier Transform:

[0032] in, For the first The complex spectral values ​​corresponding to each frequency point Frequency point number, The imaginary unit. (Through) The amplitude and phase corresponding to each frequency point are calculated, and the amplitude and phase of the corresponding harmonic order are extracted as current harmonic components.

[0033] Mel-frequency cepstral coefficient features are extracted from the ambient sound signal in the tool holder area. Specifically, after pre-emphasis, framing, and windowing of the acquired sound signal, a Fast Fourier Transform (FFT) is performed to obtain the sound signal spectrum. The spectrum is then filtered through a Mel-filter bank to obtain the energy value output by each Mel filter. The logarithm of the energy value is then used to perform a Discrete Cosine Transform (DCT) to extract the Mel-frequency cepstral coefficient features of a preset order. The conversion relationship between Mel frequency and linear frequency is as follows:

[0034] in, Mel frequency, measured in melons; It is a linear frequency, and the unit is Hz.

[0035] For the log-energy sequence output of the Mel filter bank , ,in Given the number of Mel filters, the Mel frequency cepstral coefficients obtained from their discrete cosine transform are:

[0036] in, For the first Mehr frequency cepstral coefficients, This represents the total order of the extracted Mel frequency cepstral coefficients.

[0037] Furthermore, the extracted current harmonic components and Mel-frequency cepstral coefficients are timestamped with the synchronously sampled spindle radial vibration signal to generate a fused time-series feature matrix. Specifically, a unique timestamp is assigned to each sampling point of the synchronously sampled spindle radial vibration signal, with the timestamp precision consistent with the sampling period; a corresponding timestamp is assigned to each frame of data for the current harmonic components and Mel-frequency cepstral coefficients, with the timestamp of each frame being the timestamp of the starting sampling point of that frame; using the timestamp of the spindle radial vibration signal as a reference, the current harmonic components and Mel-frequency cepstral coefficients are interpolated and aligned according to the timestamp, so that the three types of feature data have corresponding feature values ​​at the same time, ultimately generating a two-dimensional fused time-series feature matrix, where the row dimension is the time series and the column dimension is the feature dimension of different types. Fusion Time-Series Feature Matrix The formula for calculation is:

[0038] in, This represents the length of the time series, corresponding to the total number of synchronous sampling points; The total number of feature dimensions includes the X and Y sampled values ​​of the spindle radial vibration signal, the amplitude and phase values ​​of each harmonic of the current harmonic component, and the characteristic values ​​of each order of the Mel frequency cepstral coefficients. For the first The moment, the first The feature values ​​corresponding to each feature dimension are all timestamp aligned to ensure that the feature values ​​in the same row correspond to the same sampling time.

[0039] After synchronously sampling the radial vibration signal of the main shaft, a three-level wavelet packet decomposition is performed on the sampled signal. The db4 wavelet basis function is used as the mother wavelet, and the decomposition is performed at three levels, resulting in eight non-overlapping frequency bands. Each frequency band corresponds to a wavelet packet coefficient node. Single-branch reconstruction is performed on the wavelet packet coefficients corresponding to each frequency band to obtain the time-domain reconstructed signal for each band. The energy value of each reconstructed signal and the proportion of each band's energy value to the total energy value are calculated. Target frequency bands with an energy proportion greater than a preset energy threshold are selected. The preset energy threshold is pre-set based on the signal-to-noise ratio requirements of the vibration signal. The wavelet packet coefficients corresponding to the target frequency bands are merged and reconstructed to obtain the final reconstructed vibration signal. This reconstructed vibration signal replaces the synchronously sampled radial vibration signal of the main shaft and participates in the generation of the fused time-series feature matrix. For the first... Reconstructed time-domain signal of each frequency band , ,in The number of sampling points for the signal, and the energy value of this frequency band. The formula for calculation is:

[0040] The energy proportion of this frequency band The formula for calculation is:

[0041] The total energy value is the sum of the energy values ​​of the eight frequency bands. The corresponding 8 frequency band numbers are obtained from the three-layer wavelet packet decomposition.

[0042] Table 2. Statistics on the energy proportion of each frequency band in three-level wavelet packet decomposition.

[0043] Specifically, this table presents the energy statistics of each frequency band of the spindle radial vibration signal after three-level wavelet packet decomposition at a certain sampling time. The sampling frequency is set according to the maximum operating speed of the spindle. The bandwidth of each frequency band is consistent after the three-level decomposition. The preset energy threshold is preset according to the feature extraction requirements of the vibration signal. The cumulative energy ratio of the selected target frequency band meets the preset effective feature ratio requirements. It contains the main feature information related to tool eccentricity in the spindle radial vibration signal. By reconstructing the wavelet packet coefficients of the target frequency band, noise components and irrelevant interference components in the vibration signal can be effectively removed, thereby improving the feature effectiveness of the fused temporal feature matrix.

[0044] In this embodiment, by synchronously sampling the radial vibration signal of the spindle, extracting the current harmonic components from the three-phase current signal of the spindle motor, extracting the Mel frequency cepstral coefficient features from the ambient sound signal, and aligning the three types of features with timestamps to generate a fused temporal feature matrix, and simultaneously using wavelet packet decomposition to denoise and reconstruct the vibration signal, the feature richness and anti-interference capability of the multidimensional operating state data input to the eccentricity prediction model are improved, providing a high-quality input data foundation for the accurate output of the eccentricity prediction model.

[0045] In a preferred embodiment, reference Figure 3 The eccentricity prediction model includes a spatial feature extraction branch and a temporal feature extraction branch. The input data for the two branches are multidimensional running state data and historical processing feature data, respectively. The output results of the two branches are concatenated and then input into the subsequent network layer, and finally output the predicted eccentricity vector.

[0046] Specifically, the input to the spatial feature extraction branch is the fused temporal feature matrix generated in the aforementioned embodiment. The spatial feature extraction branch consists of multiple cascaded two-dimensional convolutional layers, pooling layers, and batch normalization layers, used to extract spatial correlation features between different feature dimensions in the fused temporal feature matrix. The convolutional kernel size of the two-dimensional convolutional layers is a preset two-dimensional size. The convolutional kernel performs sliding convolution operations along the time and feature dimensions of the fused temporal feature matrix to extract spatial correlation features in local regions. The pooling layer uses max pooling to downsample the feature map output by the convolutional layer, reducing the feature dimension while retaining key feature information. The batch normalization layer normalizes the output features of each layer, accelerating the model's training convergence speed and improving its generalization ability. For the input feature map... The size is ,in The height of the feature map corresponds to the length of the time series. The width of the feature map corresponds to the number of feature dimensions; Number of input channels; convolution kernel The size is ,in The height of the convolution kernel. The width of the convolution kernel. The number of output channels; the output feature map of the convolution operation. In the middle, position The formula for calculating the eigenvalue at a given location is:

[0047] in, To output the height coordinates of the feature map, This is the width coordinate of the output feature map. This is the output channel number. For the first The bias term corresponding to each output channel.

[0048] Furthermore, the input to the time feature extraction branch is historical machining feature data retrieved from an industrial big data platform. This historical machining feature data consists of multi-dimensional historical feature parameters arranged in a time series. The time feature extraction branch is composed of cascaded multi-layer long short-term memory (LSTM) networks, used to extract time-series dependent features related to tool eccentricity from the historical machining feature data. Each unit of the LSM network includes a forget gate, an input gate, a cell state, and an output gate. This gating structure controls the transmission and forgetting of historical information, effectively solving the gradient vanishing and gradient exploding problems of traditional recurrent neural networks in long-sequence processing, and accurately extracting feature dependencies from long-time-series historical data. For a single unit of the LSM network… The input at time step 1 is the feature vector at the current time step 2. The hidden state of the previous moment The cell state at the previous moment The calculation formulas for each part are as follows:

[0049]

[0050]

[0051]

[0052]

[0053]

[0054] in, It is the sigmoid activation function. It is the hyperbolic tangent activation function; These are the weight matrices corresponding to the forget gate, input gate, cell state, and output gate, respectively. These are the corresponding bias terms; For the output of the forget gate, The output of the input gate, Candidate cell state, This represents the current state of the cell. For the output of the output gate, This represents the hidden state at the current moment, which is the output feature of the Long Short-Term Memory network.

[0055] In this embodiment, the output results of the spatial feature extraction branch and the output results of the temporal feature extraction branch are concatenated. Specifically, the feature vectors output by the two branches are concatenated along the feature dimension to obtain the concatenated joint feature vector. The joint feature vector contains both the spatial correlation features of the current processing state and the temporal series dependency features of the historical processing data.

[0056] Furthermore, before inputting the concatenated feature result into the fully connected layer, the concatenated result is input into a multi-head self-attention layer. In this layer, the query vector, key vector, and value vector of the concatenated feature result are mapped based on the tool model code in the current machining task. An attention weight matrix is ​​generated based on the mapping calculation results, and this matrix is ​​used to weight the concatenated feature result before inputting the weighted result into the fully connected layer. The number of heads in the multi-head self-attention layer is pre-set based on the number of feature dimensions. The key dimension and value dimension of each head are consistent, and the total dimension matches the dimension of the concatenated joint feature vector. Through this multi-head self-attention mechanism, the model can simultaneously learn the correlation between the joint feature vector and the tool model code from different feature subspaces, assigning higher attention weights to feature dimensions more correlated with tool eccentricity and reducing the interference of irrelevant feature dimensions on the model's prediction results. The tool model encoding vector is generated using a one-hot encoding method. The encoding dimension is consistent with the total number of tool models stored in the industrial big data platform. Each tool model corresponds to a unique one-hot encoded vector, ensuring that the model can accurately identify the tool model for the current machining task, extract the historical eccentricity features of the corresponding tool model, and improve the accuracy of the prediction results. For the concatenated joint feature vector... and the tool model encoding vector corresponding to the current machining task. First, query vectors are generated through a linear mapping layer. Key vector Value vector :

[0057]

[0058]

[0059] in, These are the linear mapping weight matrices corresponding to the query vector, key vector, and value vector, respectively.

[0060] query vector Key vector Value vector Split along the feature dimension There are 3 parallel heads, and the scaling dot product attention formula for each head is:

[0061] in, The first The corresponding query vector, key vector, and value vector for each head. The dimension of the key vector corresponding to each head. This is a scaling factor used to prevent the gradient of the softmax function from vanishing due to an excessively large dot product result.

[0062] The attention calculation results from multiple heads are concatenated, and then passed through a linear mapping layer to obtain the final attention-weighted feature vector. , as input to the fully connected layer.

[0063] Specifically, the fully connected layer is composed of multiple fully connected sub-layers cascaded in sequence. The last layer of the fully connected layer is a linear output layer with an output dimension of 2, corresponding to the X and Y components of the predicted eccentricity vector. The activation function of the fully connected layer is the ReLU activation function, which is used to fit the nonlinear mapping relationship between the input features and the output eccentricity vector.

[0064] Table 3. Structure and parameter configuration of each network layer in the eccentricity prediction model.

[0065] Specifically, this table defines the complete network structure and core parameter configuration of each layer of the eccentricity prediction model. The input of the spatial feature extraction branch is the fused temporal feature matrix corresponding to the time step and the corresponding feature dimension. After two levels of convolution and pooling, it outputs the spatial feature vector of the corresponding dimension. The input of the temporal feature extraction branch is the historical processed feature data corresponding to the time step and the corresponding feature dimension. After two layers of LSTM network processing, it outputs the temporal feature vector of the corresponding dimension. The feature vectors of the two branches are concatenated and input into the multi-head self-attention layer for weighted processing. Finally, after passing through the fully connected layer, a 2-dimensional predicted eccentricity vector is output. The complete network structure can simultaneously extract the spatial features of the current state and the temporal features of historical data, ensuring the accuracy of the predicted eccentricity vector.

[0066] In this embodiment, the spatial correlation features of multidimensional operating status data are extracted by a two-dimensional convolutional network of spatial feature extraction branch, and the temporal series dependency features of historical processing feature data are extracted by a long short-term memory network of temporal feature extraction branch. After feature concatenation and weighted processing by a multi-head self-attention layer, the predicted eccentricity vector is output through a fully connected layer. The feature extraction and nonlinear mapping logic of the eccentricity prediction model are fully realized, ensuring that the model can accurately output the predicted eccentricity vector of the tool relative to the spindle axis.

[0067] In a preferred embodiment, reference Figure 4 The kinematic constraints of the tool holder-assisted Y-axis include the lead screw backlash compensation and the lead screw pitch error compensation. In the process of calculating the target displacement of the Y-axis based on the predicted eccentric vector and the kinematic constraints of the tool holder-assisted Y-axis, the current rotation angle of the tool holder-assisted Y-axis driver is first obtained. Based on the current rotation angle, the backlash compensation is obtained by querying the pre-established backlash compensation mapping table. Based on the projection component of the predicted eccentric vector in the moving direction of the tool holder-assisted Y-axis, the backlash compensation and the lead screw pitch error compensation are added to calculate the target displacement of the Y-axis.

[0068] Specifically, the transmission mechanism of the tool holder's auxiliary Y-axis is a ball screw transmission mechanism. The rotational motion of the servo motor is converted into the linear displacement of the slide through the screw. The backlash of the screw is the transmission clearance that exists in the screw-nut pair during the reversing motion, which will cause a deviation between the motor rotation angle and the actual displacement of the slide. The backlash compensation amount is determined based on the rotation direction of the screw and the current rotation angle. The backlash corresponding to different rotation angles within the entire stroke of the screw is measured in advance using a laser interferometer to establish a gap compensation mapping table between the rotation angle and the backlash compensation amount. When calculating the target displacement, the corresponding backlash compensation amount is obtained by looking up the gap compensation mapping table based on the current rotation angle of the motor fed back by the driver and the rotation direction corresponding to the displacement command.

[0069] Furthermore, the lead screw pitch error compensation amount is the compensation value corresponding to the cumulative pitch error and local error generated during the lead screw manufacturing process. The pitch error at different positions throughout the lead screw's full stroke is measured in advance using a laser interferometer, establishing a pitch error mapping table between the lead screw position and the pitch error compensation amount. When calculating the target displacement, the corresponding pitch error compensation amount is obtained by consulting the pitch error mapping table based on the lead screw position corresponding to the target displacement. Y-axis target displacement. The formula for calculation is:

[0070] in, To predict the projection component of the eccentric vector in the Y-axis direction, i.e. the basic displacement required to eliminate eccentricity; This is the backlash compensation amount for the lead screw; This is the compensation amount for the lead screw pitch error.

[0071] Kinematic constraints also include the lead screw thermal deformation compensation. During CNC lathe operation, friction between the lead screw and the bearing housing, as well as the heat generated by the servo motor, cause the lead screw temperature to rise, resulting in thermal elongation deformation and a deviation between the actual and commanded displacement of the slide. In calculating the target Y-axis displacement, real-time temperature data is acquired from temperature sensors located at the auxiliary Y-axis lead screw bearing housing of the tool holder. At least two temperature sensors are used, one at the front bearing housing and the other at the rear bearing housing of the lead screw, to collect real-time temperature data at both ends of the lead screw. This real-time temperature data is input into a pre-constructed thermal error mapping matrix. The thermal error mapping matrix is ​​a polynomial matrix fitted based on the lead screw thermal elongation at different temperature nodes in an industrial big data platform. The lead screw thermal deformation compensation is calculated using the thermal error mapping matrix and then superimposed onto the target Y-axis displacement. (Lead screw thermal deformation compensation) The formula for calculation is:

[0072] in, The coefficient of linear expansion of the lead screw material; This is the effective stroke length of the leadscrew. This represents the real-time average temperature of the bearing housings at both ends of the lead screw. The reference temperature for calibrating the thermal error of the lead screw.

[0073] Furthermore, based on the measured data of lead screw thermal elongation at different temperature nodes in the industrial big data platform, a thermal error mapping matrix is ​​constructed. The matrix has rows representing temperature nodes and columns representing the travel position of the leadscrew. Each element in the matrix represents the thermal deformation compensation amount at the corresponding temperature node and travel position. By querying the thermal error mapping matrix using real-time temperature data and the target position, the corresponding thermal deformation compensation amount of the leadscrew is obtained. The compensation amount obtained from the above calculation is then corrected to ensure the accuracy of thermal deformation compensation.

[0074] Table 4. Compensation Parameter Mapping Table for Different Rotation Angles of the Lead Screw

[0075] Specifically, this table is a mapping table of compensation parameters for the entire stroke of the tool holder-assisted Y-axis leadscrew. According to the rotation angle and stroke length corresponding to the leadscrew lead, the entire stroke of the leadscrew is divided into multiple continuous intervals. Each interval corresponds to a unique backlash compensation amount, pitch error compensation amount, and reference thermal deformation compensation amount. When calculating the target Y-axis displacement, the corresponding compensation parameters are obtained by looking up the table based on the leadscrew rotation angle and stroke position corresponding to the target displacement, and the basic displacement is corrected to ensure the execution accuracy of the Y-axis displacement.

[0076] refer to Figure 5 During the process of generating collaborative control commands based on the target displacement of the Y-axis, the CNC system of the CNC lathe acquires the current interpolation motion trajectories of the X-axis and Z-axis, and determines whether the current interpolation motion trajectory is in a linear interpolation state or a circular interpolation state. If the current interpolation motion trajectory is in a linear interpolation state, the target displacement of the Y-axis is allocated to the non-cutting reversal period of the current interpolation motion trajectory to generate collaborative control commands. The non-cutting reversal period is the idle travel period before the X-axis and Z-axis enter the next interpolation period after completing a linear interpolation segment. During this period, the displacement adjustment of the Y-axis is performed without interfering with the continuity of the cutting trajectory. If the current interpolation trajectory is in circular interpolation mode, the target displacement of the Y-axis is divided into multiple sub-displacements according to a preset ratio. The number of sub-displacements is consistent with the total number of interpolation cycles for the circular interpolation trajectory. The sub-displacement allocated in each interpolation cycle is directly proportional to the combined feed rate of the X-axis and Z-axis in that interpolation cycle. The higher the combined feed rate of the interpolation cycle, the smaller the allocated sub-displacement, ensuring that the Y-axis displacement adjustment does not cause abrupt changes in the combined motion trajectory of the tool. The sub-displacement executed in each interpolation cycle does not exceed the preset maximum displacement threshold for a single cycle. The maximum displacement threshold for a single cycle is preset according to the machining contour accuracy requirements. The duration of the interpolation cycle is consistent with the interpolation cycle of the CNC system, ensuring that the Y-axis displacement adjustment process is smooth and synchronized with the circular interpolation motion of the X-axis and Z-axis, avoiding any impact on the contour accuracy of the circular machining.

[0077] refer to Figure 6After the tool holder auxiliary Y-axis driver moves the tool along a direction perpendicular to the spindle axis to a position concentric with the spindle axis, the actual displacement data fed back by a micro-grating ruler set on the tool holder auxiliary Y-axis is acquired. The measurement direction of the micro-grating ruler is parallel to the movement direction of the Y-axis, and the measurement accuracy is adapted to the machining tolerance requirements of the current machining task, used to provide real-time feedback on the actual linear displacement of the slide. The displacement deviation value between the actual displacement data and the target Y-axis displacement is calculated, and it is determined whether the displacement deviation value is within a preset deviation range, which is set according to the machining tolerance requirements of the current machining task. If the displacement deviation value is not within the preset deviation range, the displacement deviation value is superimposed on the predicted eccentricity vector to recalculate the updated target Y-axis displacement, and an updated collaborative control command is generated based on the updated target Y-axis displacement and sent to the tool holder auxiliary Y-axis driver to complete the closed-loop displacement deviation correction. The updated target Y-axis displacement is then used to generate an updated collaborative control command. The formula for calculation is:

[0078] in, This represents the displacement deviation between the actual displacement data and the target displacement along the Y-axis. , This is the actual displacement data fed back by the miniature grating ruler.

[0079] During the acquisition of actual displacement data from a micro-grating ruler mounted on the auxiliary Y-axis of the tool holder, real-time humidity and air pressure of the environment surrounding the micro-grating ruler are obtained. These ambient humidity and air pressure are acquired using an integrated temperature, humidity, and air pressure sensor located in the tool holder area, with the acquisition frequency matching the sampling frequency of the grating ruler. An air refractive index correction coefficient is calculated based on the real-time humidity and air pressure. This coefficient is then used to dynamically update the grating pitch parameter of the micro-grating ruler. The dynamically updated grating pitch parameter is then used to perform interpolation subdivision calculations on the original pulse signal output by the micro-grating ruler. The result of this interpolation subdivision calculation is then used as the actual displacement data. (Air refractive index correction coefficient) The calculation formula is simplified using Edlén's formula:

[0080] in, The refractive index of air under standard conditions, where standard conditions are defined as the reference temperature. Reference atmospheric pressure An environment with a relative humidity of 0; This is real-time atmospheric pressure, expressed in Pa. Real-time ambient temperature, in Kelvin (K). This is the refractive index correction factor for water vapor; This is the real-time relative humidity.

[0081] Furthermore, the standard grating pitch parameter of the micro grating ruler is: Dynamically updated gate pitch parameters The formula for calculation is:

[0082] The original pulse signal output by the grating ruler is counted using the dynamically updated grating pitch parameter, and the phase of the pulse signal is subdivided and calculated using an interpolation subdivision algorithm to obtain higher precision actual displacement data and eliminate the interference caused by changes in environmental humidity and air pressure on the measurement accuracy of the grating ruler.

[0083] In this embodiment, kinematic constraints for the tool holder's auxiliary Y-axis are constructed by introducing lead screw backlash compensation, pitch error compensation, and thermal deformation compensation to correct the target displacement of the Y-axis, reducing the impact of mechanical transmission chain errors on displacement execution accuracy. By combining the interpolation motion trajectories of the X-axis and Z-axis to generate collaborative control commands, interference of the Y-axis displacement action on the original cutting trajectory is avoided. A closed-loop control loop is constructed through the actual displacement feedback of the micro grating ruler to correct displacement deviations in real time. At the same time, the measurement results of the grating ruler are dynamically corrected through environmental parameters, improving the accuracy of displacement measurement and control and ensuring that the tool can be accurately moved to a position concentric with the spindle axis.

Claims

1. A method for collaborative control of tool holder position based on industrial big data, characterized in that, include: Collect multi-dimensional operating status data of the CNC lathe under the current machining task, and retrieve historical machining feature data that matches the current machining task from the industrial big data platform; The multidimensional operating status data and the historical machining feature data are input into the pre-constructed eccentricity prediction model, and the predicted eccentricity vector of the tool relative to the spindle axis is output. Based on the predicted eccentricity vector and combined with the kinematic constraints of the tool holder auxiliary Y-axis, the target displacement of the Y-axis is calculated. Based on the target displacement of the Y-axis, a collaborative control command is generated and sent to the tool holder auxiliary Y-axis driver to control the tool holder auxiliary Y-axis driver to drive the tool to move along the direction perpendicular to the spindle axis to a position concentric with the spindle axis.

2. The method for collaborative control of tool holder position based on industrial big data according to claim 1, characterized in that, The multidimensional operating status data includes spindle radial vibration signal, spindle motor three-phase current signal, and ambient sound signal in the tool holder area; During the acquisition of the multidimensional operating status data, the radial vibration signal of the spindle is synchronously sampled, the three-phase current signal of the spindle motor is subjected to fast Fourier transform to extract the current harmonic components, and the ambient sound signal of the tool holder area is subjected to Mel frequency cepstral coefficient feature extraction. The extracted current harmonic components and the Mel frequency cepstral coefficient features are timestamped with the synchronously sampled radial vibration signal of the spindle to generate a fused time-series feature matrix as the multidimensional operating status data.

3. The method for collaborative control of tool holder position based on industrial big data according to claim 1, characterized in that, The eccentricity prediction model includes a spatial feature extraction branch and a temporal feature extraction branch; In the process of inputting the multidimensional operating status data and the historical processing feature data into the eccentricity prediction model, the multidimensional operating status data is processed by two-dimensional convolution through the spatial feature extraction branch, and the historical processing feature data is processed by long short-term memory network through the temporal feature extraction branch. The output results of the spatial feature extraction branch and the output results of the temporal feature extraction branch are concatenated and then input into the fully connected layer. The predicted eccentricity vector is output through the fully connected layer.

4. The method for collaborative control of tool holder position based on industrial big data according to claim 1, characterized in that, The kinematic constraints of the tool holder auxiliary Y-axis include the lead screw backlash compensation amount and the lead screw pitch error compensation amount. In the process of calculating the target displacement of the Y-axis based on the predicted eccentricity vector and the kinematic constraints of the tool holder auxiliary Y-axis, the current rotation angle of the tool holder auxiliary Y-axis driver is obtained. Based on the current rotation angle, the backlash compensation amount of the leadscrew is obtained by querying the pre-established backlash compensation mapping table. Based on the projection component of the predicted eccentricity vector in the moving direction of the tool holder auxiliary Y-axis, the backlash compensation amount of the leadscrew and the leadcrew pitch error compensation amount are added to calculate the target displacement of the Y-axis.

5. The method for collaborative control of tool holder position based on industrial big data according to claim 1, characterized in that, During the process of generating the collaborative control command based on the target displacement of the Y-axis, the current interpolation motion trajectory of the CNC lathe X-axis and Z-axis is obtained, and it is determined whether the current interpolation motion trajectory is in a linear interpolation state or a circular interpolation state. If the current interpolation motion trajectory is in the linear interpolation state, the target displacement of the Y-axis is allocated to the non-cutting reversal period of the current interpolation motion trajectory to generate the collaborative control command. If the current interpolation trajectory is in the circular interpolation state, the target displacement of the Y-axis is divided into multiple sub-displacements according to a preset ratio, and the multiple sub-displacements are respectively assigned to each interpolation cycle of the circular interpolation trajectory to generate the cooperative control command.

6. The method for collaborative control of tool holder position based on industrial big data according to claim 1, characterized in that, After controlling the tool holder auxiliary Y-axis driver to move the tool along a direction perpendicular to the spindle axis to a position concentric with the spindle axis, the actual displacement data fed back by the micro grating ruler set on the tool holder auxiliary Y-axis is acquired. The displacement deviation value between the actual displacement data and the target Y-axis displacement is calculated, and it is determined whether the displacement deviation value is within the preset deviation range. If the displacement deviation value is not within the preset deviation range, the displacement deviation value is superimposed on the predicted eccentricity vector to recalculate the updated target Y-axis displacement. An updated cooperative control command is generated based on the updated target Y-axis displacement and sent to the tool holder auxiliary Y-axis driver.

7. The method for collaborative control of tool holder position based on industrial big data according to claim 2, characterized in that, During the synchronous sampling of the main shaft radial vibration signal, the synchronously sampled main shaft radial vibration signal is subjected to three-level wavelet packet decomposition. Eight frequency band signals are extracted after the three-level wavelet packet decomposition. The energy proportion of each frequency band signal in the eight frequency band signals is calculated. Target frequency band signals with energy proportions greater than a preset energy threshold are selected. The wavelet packet coefficients of the target frequency band signals are reconstructed to obtain the reconstructed vibration signal. The reconstructed vibration signal is used to replace the synchronously sampled main shaft radial vibration signal to participate in the generation of the fused temporal feature matrix.

8. The method for collaborative control of tool holder position based on industrial big data according to claim 3, characterized in that, Before inputting the output of the spatial feature extraction branch and the output of the temporal feature extraction branch into the fully connected layer after feature concatenation, the concatenated result is input into the multi-head self-attention layer. In the multi-head self-attention layer, the query vector, key vector, and value vector of the concatenated result are mapped according to the tool model code in the current machining task. An attention weight matrix is ​​generated based on the mapping calculation result. The attention weight matrix is ​​used to weight the concatenated result. The weighted result is then input into the fully connected layer.

9. The method for collaborative control of tool holder position based on industrial big data according to claim 4, characterized in that, The kinematic constraints also include the thermal deformation compensation of the lead screw. In the process of calculating the target displacement of the Y-axis, real-time temperature data collected by a temperature sensor located at the auxiliary Y-axis ball screw bearing seat of the tool holder is obtained. The real-time temperature data is input into a pre-constructed thermal error mapping matrix. The thermal error mapping matrix is ​​a polynomial matrix generated by fitting the ball screw thermal elongation at different temperature nodes in the industrial big data platform. The ball screw thermal deformation compensation amount is calculated through the thermal error mapping matrix and then superimposed on the target displacement of the Y-axis.

10. The method for collaborative control of tool holder position based on industrial big data according to claim 6, characterized in that, In the process of acquiring the actual displacement data fed back by the micro grating ruler set on the auxiliary Y-axis of the tool holder, the real-time humidity and real-time air pressure of the environment where the micro grating ruler is located are acquired. An air refractive index correction coefficient is calculated based on the real-time humidity and real-time air pressure. The grating pitch parameter of the micro grating ruler is dynamically updated based on the air refractive index correction coefficient. The original pulse signal output by the micro grating ruler is interpolated and subdivided using the dynamically updated grating pitch parameter. The result of the interpolation and subdivision calculation is used as the actual displacement data.

Images