Cutter relative centering precision calibration method for multi-tool-bit cutting unit

Abstract

The invention provides a tool relative centering precision calibration method for a multi-tool-bit cutting unit, relates to the field of intelligent image analysis, and judges whether the punctured part of a patient has an infection sign or not by performing time sequence analysis on high-definition images of the punctured part of the patient at different time points in a target monitoring time period. Therefore, the accuracy and timeliness of infection monitoring can be improved, and infection signs can be found in an early stage, so that measures can be taken in time, and the incidence rate and severity of infection are reduced.

Description

Technical Field

[0001] This invention relates to the field of intelligent image analysis, and more specifically, to a method for calibrating the relative centering accuracy of tools for multi-head cutting units. Background Technology

[0002] With the rapid development of modern manufacturing, multi-head cutting units have been widely used in the efficient machining of complex parts. A multi-head cutting unit is an integrated, high-precision machining device that achieves efficient cutting of complex shapes through the collaborative work of multiple cutting tools. This machining method not only improves production efficiency but also demonstrates significant application value in aerospace, automotive manufacturing, and precision instruments. However, the performance of multi-head cutting units is highly dependent on the relative centering accuracy of the cutting tools, which directly affects machining quality and equipment stability. In existing technologies, the centering accuracy of the cutting tools is often calibrated using static methods, such as adjusting the tool position using optical measuring instruments or mechanical probes. However, these methods are typically only applicable when the machine tool is stationary or operating at low speeds, and cannot reflect the tool offset caused by dynamic vibrations during actual cutting, thus having significant limitations.

[0003] Currently, the vibration characteristics of multi-head cutting units under dynamic operating conditions are receiving increasing attention. Some studies attempt to analyze the dynamic behavior of cutting tools by acquiring and processing vibration signals, such as vibration monitoring methods based on single sensors. These methods can reflect the working state of a single tool to some extent, but due to the lack of precise analysis of the coupled vibrations between tools in a multi-head system, it is often difficult to achieve dynamic calibration of the relative centering accuracy of the tools. Furthermore, existing vibration analysis methods typically only focus on vibration amplitude or frequency characteristics, lacking a comprehensive consideration of vibration phase information. This makes it impossible to accurately describe the vibration coupling relationship between tools, thus limiting the accuracy and real-time performance of compensation strategies. Therefore, how to accurately acquire vibration signals of multi-head cutting units under dynamic operating conditions and, based on this, achieve real-time dynamic calibration of the relative centering accuracy between tools has become a key technical problem urgently needing to be solved in this field. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method for calibrating the relative centering accuracy of cutting tools in multi-head cutting units.

[0005] According to one aspect of the present invention, a method for calibrating the relative centering accuracy of a tool in a multi-head cutting unit is provided, comprising: Radial vibration signals of each tool in the multi-head cutting unit are collected by an accelerometer array, and the instantaneous amplitude envelope and phase characteristics of each tool are extracted to establish a tool dynamic feature fingerprint database. Based on the dynamic feature fingerprint database, wavelet coherence analysis is used to calculate the vibration coupling coefficient matrix between adjacent tools; Based on the coupling coefficient matrix, a neural vibration compensation network is constructed to output the relative centering compensation amount of each tool in real time. Based on the displacement command, the fine-tuning mechanism of each tool is driven to perform dynamic compensation until the root mean square error of the vibration coherence spectrum of adjacent tools in the main frequency band converges to a stable value.

[0006] Furthermore, the instantaneous amplitude envelope and phase characteristics of each tool are extracted, and the analytical signal is obtained by performing a Hilbert transform on the acquired radial vibration signal; The instantaneous amplitude envelope is obtained by taking the square root of the sum of the squares of the real and imaginary parts of the analytic signal, and the instantaneous phase is obtained by calculating the arctangent of the imaginary and real parts of the signal. The instantaneous amplitude envelope and instantaneous phase are denoised, and segmented spectrum analysis, phase unwinding and frequency smoothing are performed. Finally, amplitude statistical features and phase synchronization features are extracted to obtain the feature fingerprint vector.

[0007] Further, calculating the vibration coupling coefficient matrix between the adjacent tools includes: Continuous wavelet transform and cross-spectral analysis are performed on the characteristic fingerprint vectors of adjacent tools. Calculate the wavelet coherence coefficients on the time-frequency plane and extract the coherence intensity vectors in different frequency and time regions; Based on the coherence intensity vector, a coupling coefficient matrix is ​​constructed, and the phase difference sequence of each pair of tools is statistically analyzed to calculate the phase synchronization index and integrate it into the coupling coefficient matrix. The vibration coupling coefficient matrix is ​​obtained, and finally the system synergy features are extracted through singular value decomposition.

[0008] 2. Further, the phase synchronization index is calculated as shown in the following formula: , in, To achieve the overall phase synchronization index, The Shannon entropy of the phase difference sequence within the current time window. The maximum entropy value under ideal uniform distribution. The cyclic variance of the phase difference sequence. and The weighting coefficients are adaptively adjusted according to the cutting conditions and satisfy the following conditions: This represents the overlap deviation between adjacent time windows. The standard time window length, is the root mean square value of the phase difference sequence.

[0009] Furthermore, the neural vibration compensation network adopts a multilayer perceptron structure. It receives the coupling coefficient matrix features through the input layer, performs feature extraction and optimization through three hidden layers, and finally outputs the relative centering compensation amount between adjacent tools in the radial, axial and tangential directions after hierarchical output structure and Kalman filtering.

[0010] Furthermore, the feature extraction and optimization include adaptive weighting of amplitude and phase information through a phase attention mechanism, dynamic adjustment of the learning rate by a cutting condition adaptive module, storage of historical compensation effects by a recurrent memory unit and dynamic adjustment of compensation gain, and calculation of the correlation between compensation amounts between tools by a cross-attention module.

[0011] Furthermore, the adaptive weighting of amplitude and phase information by the phase attention mechanism includes: A multi-head attention mechanism is employed to perform high-dimensional feature projection on amplitude and phase coupling information and to calculate the attention weight distribution. An adaptive modulation function is constructed by combining the temporal rate of change of the phase synchronization index; By modulating the attention response speed through the dynamic triggering mechanism of phase synchronization exponential gradient and the cutting condition stability index, adaptive weight allocation of amplitude and phase characteristics is ultimately achieved.

[0012] Furthermore, the cutting condition adaptive module dynamically adjusts the learning rate by standardizing the cutting parameters and calculating the deviation rate, and constructs a condition change index based on the deviation rate. The operating condition change index is input into the dynamic regulator to generate a sensitivity adjustment factor, and finally the network response characteristics are dynamically adjusted according to the acceleration characteristics of the operating condition change.

[0013] Furthermore, the dynamic regulator includes a response speed adjustment module, a feature extraction adjustment module, and an output adjustment module; The response speed adjustment module determines the adjustment range of the learning rate by analyzing the first and second derivatives of parameter changes; The feature extraction and adjustment module adjusts the temperature parameters of the attention mechanism in the network according to the time-varying nature of the operating conditions; The output adjustment module uses an adaptive slope adjustment algorithm to achieve smooth control of the compensated output.

[0014] Furthermore, the dynamic compensation, based on the compensation instructions output by the neural network, performs orthogonal component compensation in the X and Y directions through a piezoelectric ceramic actuator, and dynamically adjusts the compensation parameters based on real-time vibration coherence spectrum analysis to achieve closed-loop relative centering compensation control.

[0015] Compared with existing technologies, the tool relative centering accuracy calibration method for multi-head cutting units provided by this invention collects the radial vibration signals of each tool in the multi-head cutting unit through an accelerometer array, extracts the instantaneous amplitude envelope and phase features, establishes a tool dynamic feature fingerprint database, and then uses wavelet coherence analysis to calculate the vibration coupling coefficient matrix between adjacent tools, constructing a neural vibration compensation network to output the relative centering compensation amount of each tool in real time. This allows for accurate analysis of the vibration coupling characteristics between multiple tools and enables real-time calibration of the tool relative centering accuracy under dynamic operating conditions, helping to improve the machining accuracy and stability of multi-head cutting units, thereby reducing machining errors and equipment wear, and significantly improving machining efficiency and product quality. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 This is a system block diagram of a tool relative centering accuracy calibration method for a multi-head cutting unit according to an embodiment of the present invention.

[0017] Figure 2 This is a comparison diagram of the relative centering errors of adjacent tools in the tool relative centering accuracy calibration method for multi-head cutting units according to an embodiment of the present invention. Detailed Implementation

[0018] Hereinafter, exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein.

[0019] Figure 1 This is a system block diagram of a tool relative centering accuracy calibration method for multi-head cutting units according to an embodiment of the present invention. Figure 1 As shown, the tool relative centering accuracy calibration method for multi-head cutting units includes: S1: The radial vibration signals of each tool in the multi-head cutting unit are collected by an accelerometer array. The instantaneous amplitude envelope and phase characteristics of each tool are extracted by Hilbert transform to establish a tool dynamic feature fingerprint database. A 3×N (N being the number of tools) sensor array network is formed by arranging high-precision accelerometers along the X, Y, and Z directions on each tool base of the multi-tool cutting unit. Each sensor has a sampling frequency of 20kHz and a signal resolution of 0.001g. Radial vibration acceleration signals of each tool during the cutting process are collected. These acceleration signals are preprocessed using an 8th-order Butterworth bandpass filter to remove machine noise below 50Hz and high-frequency interference above 5kHz. The filtered acceleration signals are then subjected to Hilbert transform processing to extract the values ​​of each tool. The instantaneous amplitude envelope curve and instantaneous phase curve of the tool are extracted. The extracted instantaneous amplitude envelope curve and instantaneous phase curve are segmented in the time domain, with each time window having a length of 0.1s and a window overlap rate of 50%. Statistical characteristic parameters within each time window are calculated, including root mean square value, peak factor, waveform factor, and margin factor. The above statistical characteristic parameters are organized according to the time series to construct a dynamic feature fingerprint library of the tool. Each element in the dynamic feature fingerprint library contains a timestamp, amplitude feature vector, and phase feature vector, which are used to characterize the dynamic motion state of the tool at different times.

[0020] In this process, after obtaining the original radial vibration acceleration signal of the tool after bandpass filtering preprocessing, a Hilbert transform is first performed on the original signal to obtain an analytic signal, which is composed of the original real signal and its Hilbert-transformed imaginary signal. Then, based on the analytic signal, the instantaneous amplitude envelope is obtained by calculating the square root of the sum of the squares of the real and imaginary parts, and the instantaneous phase is obtained by calculating the arctangent of the imaginary and real parts. Next, wavelet denoising is applied to the instantaneous amplitude envelope to remove high-frequency noise, and the denoised amplitude envelope is then processed according to a set time length. The amplitude is segmented, with a certain overlap between adjacent segments. Then, spectral analysis is performed on the envelope of each segment to obtain the envelope spectrum. The amplitude and phase information corresponding to the characteristic frequency points with significant amplitudes in the envelope spectrum are extracted. For the instantaneous phase, phase unwrapping is first performed to eliminate phase abrupt changes. Then, the phase change rate is calculated to obtain the instantaneous frequency, and the instantaneous frequency is smoothed to reduce fluctuations. Finally, the obtained instantaneous amplitude features and phase features are combined to extract the statistical feature values ​​of the amplitude and the synchronous feature values ​​of the phase, forming a feature fingerprint vector that can characterize the dynamic cutting characteristics of the tool.

[0021] It should be noted that the instantaneous amplitude envelope and phase characteristics are of great significance in characterizing dynamic cutting states. When the tool is in a normal cutting state, the instantaneous amplitude envelope exhibits regular fluctuation characteristics, and the statistical distribution of the amplitude follows a Gaussian distribution. At this time, the stability of the cutting process can be assessed by the peak density and waveform factor of the amplitude envelope. However, when the tool experiences a slight deviation, resulting in uneven cutting depth, the instantaneous amplitude envelope will exhibit modulation, manifested as sideband frequencies near the fundamental frequency component. The degree of cutting deviation can be judged by analyzing the intensity of the sideband frequencies. Regarding phase characteristics, when multiple tools are cutting simultaneously, the phase difference between adjacent tools should remain constant. If phase difference drift occurs, it indicates a problem with the tools. Deviations in relative centering accuracy can occur. By calculating the phase synchronization index, the degree of collaborative cutting among multiple tools can be assessed. A decrease in the synchronization index indicates an abnormality in the relative motion between tools. In addition, the phase modulation characteristics can reflect the periodic motion changes of the tools. When the cutting load fluctuates, the phase modulation index will change accordingly, which allows for real-time monitoring of dynamic changes in the cutting state. By combining the analysis of instantaneous amplitude envelope and phase characteristics, not only can abnormal states in the cutting process be identified, but also a basis for dynamic compensation of the relative centering accuracy of the tools can be provided, thereby ensuring the machining accuracy of the multi-head cutting unit.

[0022] S2: Based on the dynamic feature fingerprint database, wavelet coherence analysis is used to calculate the vibration coupling coefficient matrix between adjacent tools. The coupling coefficient matrix reflects the degree of dynamic interference and relative centering error between tools. Based on the established dynamic feature fingerprint database, firstly, feature fingerprint vector pairs of adjacent tools are extracted. Continuous wavelet transform is then performed on each pair of feature fingerprint vectors, using complex-valued Morlet wavelets as the mother wavelet function during the transform process. The wavelet scale range is set within an interval that covers the main cutting frequencies of the tools. Subsequently, the cross-spectrum of the two sets of wavelet coefficients is calculated, and the wavelet coherence coefficients on the time-frequency plane are obtained through normalization. These wavelet coherence coefficients reflect the correlation strength of the two tool vibration signals at different times and frequencies. Next, multiple regions of interest are divided on the time-frequency plane, each corresponding to a specific frequency band and time window. The mean value of wavelet coherence coefficients in each region is calculated, and these mean values ​​are organized into a coherence intensity vector. Then, based on the coherence intensity vector, a coefficient matrix reflecting the vibration coupling relationship between tools is constructed. Each element of this matrix represents the degree of dynamic coupling between the corresponding two tools. At the same time, the phase difference sequence of each pair of tools is statistically analyzed, the phase synchronization index is calculated, and it is integrated into the coupling coefficient matrix, thus forming a vibration coupling coefficient matrix containing both amplitude coupling and phase coupling characteristics. Finally, the vibration coupling coefficient matrix is ​​subjected to singular value decomposition to extract the main eigenvalues, which are used to characterize the coordination level of the entire multi-head cutting system.

[0023] More specifically, after obtaining the feature fingerprint vectors of a pair of adjacent tools, the instantaneous amplitude sequence and phase sequence in the feature fingerprint vectors are first normalized. Cubic spline interpolation is used to ensure that the data of the two tools have the same time step and length. Then, the complex-valued Morlet wavelet is selected as the mother wavelet function, which has good time-frequency localization characteristics. The center frequency of the wavelet function is set to 6 rad / s, and the bandwidth parameter is set to 1. Next, the wavelet scale parameter is set to a range of 1 to 128, and 32 scale points are selected according to a logarithmic uniform distribution, corresponding to frequencies... The range covers the main frequency bands of tool cutting; then, continuous wavelet transform is performed on each normalized feature fingerprint vector, and wavelet coefficients reflecting the local time-frequency characteristics of the signal are obtained by scaling and translation operations of the mother wavelet function at different scales and time positions; after the wavelet coefficients are calculated, their complex modulus is calculated to obtain the time-frequency energy distribution map, and the main frequency components of the signal are determined according to the energy distribution; in the final wavelet coefficient matrix, each row represents the time-domain evolution characteristics at a certain scale, and each column represents the frequency distribution characteristics at a certain moment, thus realizing the time-frequency decomposition expression of the feature fingerprint vector.

[0024] Furthermore, after obtaining the phase difference sequence of a pair of adjacent tools, the phase difference sequence is first time-aligned and segmented. The length of each time window is set to an integer multiple of the cutting cycle, and adjacent windows maintain an overlap of half a window length. Then, the probability distribution function of the phase difference is calculated within each time window. When the phase difference sequence exhibits a clear concentration trend, it indicates that the movements of the two tools have strong synchronicity. Next, the Shannon entropy value of the phase difference is calculated based on the probability distribution function, and this entropy value is compared with the maximum entropy value of an ideal uniform distribution to obtain a normalized phase synchronization strength index. Then, the cyclic variance of the phase difference sequence is calculated. By performing a complex mapping on the phase difference sequence and calculating the modulus of the complex vector, an index characterizing the phase synchronization stability is obtained. After obtaining the phase synchronization strength index and stability index, the two indices are combined using a weighted average to obtain a comprehensive phase synchronization index, where the weighting coefficients are adaptively adjusted according to the cutting conditions. The calculation of the phase synchronization index is shown in the following formula: , in, To achieve the overall phase synchronization index, The Shannon entropy of the phase difference sequence within the current time window. The maximum entropy value for an ideal uniform distribution is 8 bits. The cyclic variance of the phase difference sequence. and The weighting coefficients are adaptively adjusted according to the cutting conditions and satisfy ξ+η=1. This represents the overlap deviation between adjacent time windows. It is the standard time window length (an integer multiple of the cutting cycle). is the root mean square value of the phase difference sequence.

[0025] The weighting coefficients are set according to the following rules: , in, The coefficient of variation of the entropy sequence. Let be the coefficient of variation of the cyclic variance sequence. Let the standard deviation of the entropy values ​​be the number of consecutive time windows (n). The mean of the entropy values ​​over n consecutive time windows. Let be the standard deviation of the cyclic variance over n consecutive time windows. Let be the mean of the cyclic variance over n consecutive time windows. The reference coefficient of variation (usually taken as 0.1) This is the normalization factor.

[0026] It should be noted that when fusing amplitude coupling and phase coupling characteristics for analysis, firstly, an amplitude coupling matrix is ​​constructed based on wavelet coherence coefficients. This matrix reflects the degree of mutual influence between tool vibration amplitudes. If the vibration amplitudes of two tools show consistent trends, it indicates that they may be subjected to the same cutting force disturbance. Simultaneously, a phase coupling matrix is ​​constructed based on the phase synchronization index. This matrix reflects the temporal coordination of tool motion. A higher phase coupling degree indicates a smaller relative centering error between tools. Subsequently, a tensor fusion method is used to combine the amplitude coupling matrix and the phase coupling matrix. By setting different weighting coefficients, the focus is on regions with abnormal amplitude coupling or abnormal phase coupling. The practical significance of this dual-coupling analysis method lies in the fact that when a tool has a relative centering error, it often manifests simultaneously as incoordination of vibration amplitude and desynchronization of phase. Amplitude coupling reflects the uniformity of cutting load distribution, while phase coupling reflects the consistency of cutting trajectory. The combination of the two can more comprehensively evaluate the dynamic performance of a multi-tool cutting system. If a pair of tools has a low amplitude coupling degree but a high phase coupling degree, it may be due to uneven tool wear. Conversely, if the amplitude coupling degree is high but the phase coupling degree is low, it may be due to tool installation deviation. Through the comprehensive analysis of this coupling characteristic, the specific cause of the relative centering error can be accurately identified.

[0027] S3: Based on the coupling coefficient matrix, a neural vibration compensation network is constructed to output the relative centering compensation amount of each tool in real time. The compensation amount is converted into displacement commands for the servo motor through the sliding mode controller. The neural vibration compensation network employs a multilayer perceptron structure. The input layer receives the feature vector of the vibration coupling coefficient matrix, including amplitude coupling degree and phase synchronization index. A phase attention mechanism module is designed after the input layer to adaptively weight the amplitude and phase information in the coupling matrix. The backbone of the network contains three hidden layers. The first hidden layer uses 128 neurons configured with a batch normalization layer and ReLU activation function, and embeds a cutting condition adaptive module. This module receives real-time cutting parameters and outputs a modulation factor to dynamically adjust the learning rate of each layer. The second hidden layer uses 64 gated recurrent memory units to record historical compensation effects using an LSTM structure and dynamically adjusts the compensation gain in different directions based on the evaluation results. The third hidden layer uses 32 neurons to construct a cross-attention module, which calculates... The network establishes a correlation matrix between the compensation amount and the tool compensation amount, and adjusts the compensation amount collaboratively based on the correlation strength. The output of the network adopts a hierarchical structure. First, a 16-dimensional fully connected layer predicts the main direction vector of compensation. This vector is normalized by the Softmax function to obtain the weight distribution of each direction. Then, the weight distribution is multiplied by the baseline compensation amount to obtain the preliminary compensation vector. Next, three parallel 8-dimensional fully connected layers calculate the accurate compensation components in the radial, axial, and tangential directions, respectively. Each component is modulated by an activation function based on local response normalization to ensure that the compensation amount is within a reasonable range. Finally, the compensation components in the three directions are combined into a complete compensation vector and smoothed by a Kalman filter to obtain the final relative centering compensation amount. The compensation amount includes the relative displacement compensation vector C_{ij} between adjacent tools, where i,j represent the numbers of adjacent tools, and the components of the compensation vector are [Δr_{ij}, Δa_{ij}, Δt_{ij}], corresponding to the compensation amounts in the radial, axial, and tangential directions, respectively.

[0028] In this phase attention mechanism, after obtaining the coupling coefficient matrix, the amplitude coupling and phase coupling information are first projected onto a high-dimensional feature space, respectively, and a query vector, key vector, and value vector are generated through linear transformation. Then, the dot product of the query vector and key vector is calculated, and the attention weight distribution is obtained by normalization using the softmax function. This weight distribution reflects the correlation strength between different coupling features. Next, based on the temporal change rate of the phase synchronization index, an adaptive modulation function is constructed. This function uses the ratio of the standard deviation to the mean of the phase synchronization index as the adjustment criterion. When this ratio increases, the attention weight of the phase feature is dynamically strengthened, while the weight of the amplitude feature is weakened. Finally, when calculating the final attention output... A multi-head attention mechanism is introduced, dividing the feature space into multiple subspaces for parallel processing. Each subspace independently calculates its attention weights, and the outputs of each subspace are fused through residual connections. Simultaneously, a dynamic triggering mechanism based on the phase synchronization index gradient is used. When the gradient of the phase synchronization index exceeds the upper quartile of its historical statistical distribution, the weight contribution of phase features in the attention calculation is automatically increased. Finally, the response speed of the attention mechanism is modulated by a stability index of the cutting condition. This stability index is calculated from the sliding variance of parameters such as cutting force fluctuations and spindle speed fluctuations, thereby achieving an adaptive trade-off between amplitude coupling and phase coupling information, resulting in the weight allocation of amplitude and phase features, as shown in the following equation: , in, , in, The dynamic weights for the amplitude characteristics, The dynamic weights for phase features, To focus on the number of heads, For the adaptive weights of the i-th attention head, For the output of the i-th attention head, For amplitude coupling feature mapping, For phase coupling feature mapping, The standard deviation of the phase synchronization index. The mean of the phase synchronization index. The time gradient of the phase synchronization exponent. This is the phase gradient gain coefficient. The amplitude gradient attenuation coefficient, For attention temperature parameters, Let be the L2 norm gradient of the output of the i-th attention head.

[0029] On the other hand, after the cutting condition adaptive module receives the real-time cutting parameter signal, it first standardizes parameters such as cutting speed, feed rate, and depth of cut, and calculates the deviation rate of these parameters relative to the nominal working condition. Then, based on the deviation rate, it constructs a working condition change index. This index adaptively combines the influence of each parameter through a nonlinear mapping function. The index calculation employs a dynamic weight allocation strategy, with weight values ​​determined by both the time-domain and frequency-domain characteristics of parameter changes. Next, the working condition change index is input into a dynamic regulator, which contains three functional modules: a response speed adjustment module dynamically adjusts the network's learning rate based on the trend characteristics of working condition changes, determining the adjustment range by analyzing the first and second derivatives of parameter changes; and a feature extraction adjustment module adjusts the attention mechanism in the network based on the time-varying nature of the working condition characteristics. Temperature parameters are used to adaptively capture features at different time scales. The output adjustment module adjusts the slope of the activation function of the network output layer according to the dynamic characteristics of the operating conditions, and uses an adaptive slope adjustment algorithm to achieve smooth control of the compensated output. Then, the adjustment signals of these three modules are fused to generate the sensitivity adjustment factor of each layer of the network. This factor achieves dynamic adjustment of sensitivity by scaling the feature map amplitude transmitted between layers. During the dynamic changes in the operating conditions, the system calculates the acceleration characteristics of the changes in operating conditions in real time and dynamically adjusts the network's response characteristics according to the magnitude of the acceleration. Finally, the adjustment results are fed back to the operating condition evaluation unit, which continuously monitors the compensation effect, establishes a compensation effect evaluation index library, evaluates the effectiveness of the current compensation strategy in real time, and optimizes the adjustment parameters through the feedback loop, thereby forming a closed-loop adaptive adjustment mechanism. The sensitivity adjustment factor of each layer of the network is shown in the following formula: , in, , in, The sensitivity modulator for layer l. For response speed adjustment, The first derivative of the cutting parameters, The second derivative of the cutting parameters. For feature extraction adjustment term, Let be the weight for the i-th time scale. Let i be the feature value at the i-th time scale. The standard deviation of the time characteristic, For output adjustment terms, This is the slope adjustment coefficient. The slope of the current activation function. As the baseline slope, For the acceleration due to changes in operating conditions, This is a reference acceleration value.

[0030] S4: Based on the displacement command, drive the fine-tuning mechanism of each tool to perform dynamic compensation until the root mean square error of the vibration coherence spectrum of adjacent tools in the main frequency band converges to a stable value.

[0031] After receiving the displacement command output from the neural vibration compensation network, the command signal is first smoothed using a digital filter to eliminate high-frequency components that may cause mechanical resonance. Then, based on the current working state of each tool, the displacement command is decomposed into orthogonal components in the X and Y directions, and the priority sequence for compensation is determined according to the relative positions of the tools to avoid interference during the compensation process. Next, a piezoelectric ceramic actuator is used as the execution mechanism to convert the decomposed displacement command into a driving voltage. The rate of change of the driving voltage is constrained by the dynamic response characteristics; when the displacement command changes significantly, an acceleration limiting algorithm is used for dynamic adjustment. During the compensation execution process, the vibration signals of each tool are collected in real time, and the mutual power between adjacent tools is calculated. The vibration coherence spectrum is obtained by analyzing the power spectrum and the self-power spectrum, with a focus on the coherence characteristics within the main cutting frequency band. Then, the root mean square error (RMS) of the vibration coherence spectrum is calculated. This error value reflects the similarity of vibrations between tools. When the RMS error value begins to show an oscillating convergence trend, the compensation gain is appropriately reduced to prevent overcompensation. Simultaneously, the phase coherence of adjacent tools is monitored to ensure that the compensation process does not introduce new phase misalignments. During the compensation process, if the compensation effect in a certain direction is found to be poor, the compensation weight in that direction is dynamically adjusted using an adaptive algorithm. Finally, when the change in the RMS error of the vibration coherence spectrum within multiple consecutive compensation cycles is less than a set threshold, the dynamic compensation process is completed, and the compensation parameters are recorded in the system state database for reference in the next compensation cycle.

[0032] In summary, the tool relative centering accuracy calibration method for multi-head cutting units based on embodiments of the present invention has been clarified. It acquires radial vibration signals from each tool in the multi-head cutting unit using an accelerometer array, extracts instantaneous amplitude envelopes and phase features, establishes a tool dynamic feature fingerprint database, and then uses wavelet coherence analysis to calculate the vibration coupling coefficient matrix between adjacent tools. A neural vibration compensation network is then constructed to output the relative centering compensation amount of each tool in real time. This allows for accurate analysis of the vibration coupling characteristics between multiple tools and enables real-time calibration of tool relative centering accuracy under dynamic operating conditions. This helps improve the machining accuracy and stability of multi-head cutting units, thereby reducing machining errors and equipment wear, and significantly improving machining efficiency and product quality.

[0033] Here, those skilled in the art will understand that the specific operations of each step in the above-described method for calibrating the relative centering accuracy of a multi-head cutting unit have been referenced above. Figure 1 and Figure 2The method for calibrating the relative centering accuracy of tools for multi-head cutting units has been described in detail, and therefore, its repeated description will be omitted.

[0034] In summary, the tool relative centering accuracy calibration method for multi-head cutting units based on embodiments of the present invention has been clarified. It acquires radial vibration signals from each tool in the multi-head cutting unit using an accelerometer array, extracts instantaneous amplitude envelopes and phase features, establishes a tool dynamic feature fingerprint database, and then uses wavelet coherence analysis to calculate the vibration coupling coefficient matrix between adjacent tools. A neural vibration compensation network is then constructed to output the relative centering compensation amount of each tool in real time. This allows for accurate analysis of the vibration coupling characteristics between multiple tools and enables real-time calibration of tool relative centering accuracy under dynamic operating conditions. This helps improve the machining accuracy and stability of multi-head cutting units, thereby reducing machining errors and equipment wear, and significantly improving machining efficiency and product quality.

Claims

1. A method for calibrating the relative centering accuracy of a tool in a multi-head cutting unit, characterized in that, include: Radial vibration signals of each tool in the multi-head cutting unit are collected by an accelerometer array, and the instantaneous amplitude envelope and phase characteristics of each tool are extracted to establish a tool dynamic feature fingerprint database. Based on the dynamic feature fingerprint database, wavelet coherence analysis is used to calculate the vibration coupling coefficient matrix between adjacent tools; Based on the coupling coefficient matrix, a neural vibration compensation network is constructed to output the relative centering compensation amount of each tool in real time. Based on the displacement command, the fine-tuning mechanism of each tool is driven to perform dynamic compensation until the root mean square error of the vibration coherence spectrum of adjacent tools in the main frequency band converges to a stable value.

2. The tool relative centering accuracy calibration method for multi-head cutting units according to claim 1, characterized in that, The instantaneous amplitude envelope and phase characteristics of each tool are extracted, and the analytical signal is obtained by performing a Hilbert transform on the acquired radial vibration signal. The instantaneous amplitude envelope is obtained by taking the square root of the sum of the squares of the real and imaginary parts of the analytic signal, and the instantaneous phase is obtained by calculating the arctangent of the imaginary and real parts of the signal. The instantaneous amplitude envelope and instantaneous phase are denoised, and segmented spectrum analysis, phase unwinding and frequency smoothing are performed. Finally, amplitude statistical features and phase synchronization features are extracted to obtain the feature fingerprint vector.

3. The tool relative centering accuracy calibration method for multi-head cutting units according to claim 2, characterized in that, Calculating the vibration coupling coefficient matrix between adjacent tools includes: Continuous wavelet transform and cross-spectral analysis are performed on the characteristic fingerprint vectors of adjacent tools. Calculate the wavelet coherence coefficients on the time-frequency plane and extract the coherence intensity vectors in different frequency and time regions; Based on the coherence intensity vector, a coupling coefficient matrix is ​​constructed, and the phase difference sequence of each pair of tools is statistically analyzed to calculate the phase synchronization index and integrate it into the coupling coefficient matrix. The vibration coupling coefficient matrix is ​​obtained, and finally the system synergy features are extracted through singular value decomposition.

4. The tool relative centering accuracy calibration method for multi-head cutting units according to claim 3, characterized in that, The phase synchronization index is calculated as follows: ， in, To achieve the overall phase synchronization index, The Shannon entropy of the phase difference sequence within the current time window. The maximum entropy value under ideal uniform distribution. The cyclic variance of the phase difference sequence. and The weighting coefficients are adaptively adjusted according to the cutting conditions and satisfy the following conditions: This represents the overlap deviation between adjacent time windows. The standard time window length, is the root mean square value of the phase difference sequence.

5. The tool relative centering accuracy calibration method for multi-head cutting units according to claim 1, characterized in that, The neural vibration compensation network adopts a multilayer perceptron structure. It receives the coupling coefficient matrix features through the input layer, performs feature extraction and optimization through three hidden layers, and finally outputs the relative centering compensation amount between adjacent tools in the radial, axial and tangential directions after hierarchical output structure and Kalman filtering.

6. The tool relative centering accuracy calibration method for multi-head cutting units according to claim 5, characterized in that, The feature extraction and optimization include adaptive weighting of amplitude and phase information through a phase attention mechanism, dynamic adjustment of the learning rate by a cutting condition adaptive module, storage of historical compensation effects by a recurrent memory unit and dynamic adjustment of compensation gain, and calculation of the correlation between compensation amounts between tools by a cross-attention module.

7. The tool relative centering accuracy calibration method for multi-head cutting units according to claim 6, characterized in that, The phase attention mechanism includes adaptive weighting of amplitude and phase information, including: A multi-head attention mechanism is employed to perform high-dimensional feature projection on amplitude and phase coupling information and to calculate the attention weight distribution. An adaptive modulation function is constructed by combining the temporal rate of change of the phase synchronization index; By modulating the attention response speed through the dynamic triggering mechanism of phase synchronization exponential gradient and the cutting condition stability index, adaptive weight allocation of amplitude and phase characteristics is ultimately achieved.

8. The tool relative centering accuracy calibration method for multi-head cutting units according to claim 6, characterized in that, include: The cutting condition adaptive module dynamically adjusts the learning rate by standardizing the cutting parameters and calculating the deviation rate, and constructs a condition change index based on the deviation rate. The operating condition change index is input into the dynamic regulator to generate a sensitivity adjustment factor, and finally the network response characteristics are dynamically adjusted according to the acceleration characteristics of the operating condition change.

9. The tool relative centering accuracy calibration method for multi-head cutting units according to claim 8, characterized in that, The dynamic regulator includes a response speed adjustment module, a feature extraction adjustment module, and an output adjustment module; The response speed adjustment module determines the adjustment range of the learning rate by analyzing the first and second derivatives of parameter changes; The feature extraction and adjustment module adjusts the temperature parameters of the attention mechanism in the network according to the time-varying nature of the operating conditions; The output adjustment module uses an adaptive slope adjustment algorithm to achieve smooth control of the compensated output.

10. The tool relative centering accuracy calibration method for multi-head cutting units according to claim 1, characterized in that, The dynamic compensation is based on the compensation instructions output by the neural network. The piezoelectric ceramic actuator performs orthogonal component compensation in the X and Y directions and dynamically adjusts the compensation parameters based on real-time vibration coherence spectrum analysis to achieve closed-loop relative centering compensation control.

Images