Milling cutter temperature dynamic regulation method and system based on micro-channel internal cooling monitoring

Abstract

The application relates to the technical field of numerical control machining temperature monitoring, and discloses a milling cutter temperature dynamic regulation and control method and system based on micro-channel internal cooling monitoring, the method comprising the following steps: acquiring a discrete temperature data sequence collected by a micro-channel inside a milling cutter in a cutting process; performing spatial feature extraction and time sequence migration feature extraction on the discrete temperature data sequence to obtain a heat space-time evolution sequence; performing cubic spline interpolation processing and phase compensation processing on the heat space-time evolution sequence to obtain a space-time continuous temperature matrix; through local extreme value extraction, classification screening and time sequence deduction, a predicted temperature peak value and an aggregation position are obtained, and a local cooling intervention instruction is generated before the peak value arrives to perform pre-control action. The method can solve the problem of tool local overheating damage caused by cutting temperature peak value response lag.

Description

Technical Field

[0001] This invention relates to the field of CNC machining temperature monitoring technology, and in particular to a method and system for dynamic control of milling cutter temperature based on microchannel internal cooling monitoring. Background Technology

[0002] In high-end manufacturing, milling is a core process that determines the precision and surface quality of parts, and it is widely used in aerospace, precision molds, automotive power systems, and other applications where high reliability is required. In industrial control systems, precise control of the milling cutter's cutting temperature is considered a key factor in ensuring machining quality and tool life.

[0003] Currently, most end mill temperature control relies on fixed-flow coolant spraying or intervention only after the temperature exceeds the limit. These approaches are reactive and reactive, with a significant lag between the temperature rise and the coolant actually reaching the heat source. By the time cooling is complete, the cutting edge has often suffered irreversible damage. Furthermore, due to a lack of advance assessment of temperature trends, cooling resources are often evenly distributed across the entire tool surface, leaving high-temperature areas untreated. During cutting, heat is not uniformly distributed across the entire cutting edge; instead, it forms distinct temperature gradients at different locations along the edge depending on the cutting angle, feed rate, and material hardness. Small areas can accumulate significantly more heat than their surroundings, becoming potential failure points. However, there is a complex dynamic relationship between discrete temperature readings and the actual spatiotemporal evolution of heat. Reconstructing the inherent rhythm of temperature evolution from a continuously changing data sequence is difficult, making it challenging to determine the location and magnitude of the next high-temperature peak before the critical temperature is reached.

[0004] In summary, the existing technology suffers from insufficient cooling control accuracy due to lag in the peak response of cutting temperature. Summary of the Invention

[0005] This invention provides a method and system for dynamic control of milling cutter temperature based on microchannel internal cooling monitoring, in order to solve the problem of insufficient cooling control accuracy due to lag in peak cutting temperature response.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a method for dynamic temperature control of a milling cutter based on microchannel internal cooling monitoring, comprising: A discrete temperature data sequence is acquired by the microchannel inside the milling cutter during the cutting process. Spatial feature extraction and temporal migration feature extraction are performed on the discrete temperature data sequence to obtain a thermal spatiotemporal evolution sequence. The spatiotemporal evolution sequence of heat is subjected to cubic spline interpolation and phase compensation to obtain a spatiotemporal continuous temperature matrix; Local temperature extrema are extracted from the spatiotemporal continuous temperature matrix to obtain a candidate peak set. The candidate peak set is then classified, filtered, and subjected to time-series deduction to obtain the predicted temperature peak. Based on the predicted temperature peak, coordinate mapping is performed to obtain the clustering location. If the predicted temperature peak is greater than the preset critical temperature threshold, a local cooling intervention command is generated based on the aggregation location, and the multi-channel coolant spray device is controlled to perform pre-deployment actions on the aggregation location based on the local cooling intervention command.

[0007] Secondly, the present invention provides a dynamic temperature control system for milling cutters based on microchannel internal cooling monitoring, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the dynamic temperature control method for milling cutters based on microchannel internal cooling monitoring as described above.

[0008] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for dynamic control of milling cutter temperature based on microchannel internal cooling monitoring.

[0009] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention restores the continuous temperature evolution law and eliminates the sensor response hysteresis error by performing cubic spline interpolation and phase compensation on the thermal spatiotemporal evolution sequence, and obtains a spatiotemporal continuous temperature matrix. This solves the problem that discrete temperature data cannot accurately reflect the actual heating state of the tool, and provides a reliable data basis for subsequent peak prediction.

[0010] (2) By classifying and screening the candidate peak set and performing time-series deduction, the present invention filters out the false peaks generated by sensor noise and chip splash, and predicts the temperature peak in the future time period before the actual formation of the cutting heat peak, thus realizing the transformation from post-detection to pre-prediction and avoiding the lag of cooling response.

[0011] (3) This invention generates local cooling intervention instructions based on the predicted temperature peak and the location of accumulation, controls the multi-channel coolant spray device to perform pre-deployment actions on the location of accumulation, and accurately deploys cooling resources to the small area where the high temperature peak is about to occur, thus solving the problem that the high temperature area cannot be cooled in the traditional uniform cooling method. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating a method for dynamic temperature control of a milling cutter based on microchannel internal cooling monitoring, provided in an embodiment of the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] Reference Figure 1 The first embodiment of the present invention provides a method for dynamic control of milling cutter temperature based on microchannel internal cooling monitoring, comprising the following steps: S11, acquire discrete temperature data sequence collected by the microchannel inside the milling cutter during the cutting process, and perform spatial feature extraction and temporal migration feature extraction on the discrete temperature data sequence to obtain the thermal spatiotemporal evolution sequence; S12, perform cubic spline interpolation and phase compensation processing on the heat spatiotemporal evolution sequence to obtain a spatiotemporal continuous temperature matrix; S13, Local temperature extrema are extracted from the spatiotemporal continuous temperature matrix to obtain a candidate peak set. The candidate peak set is classified, filtered, and time-series deduced to obtain the predicted temperature peak. The predicted temperature peak is then mapped to coordinates to obtain the clustering location. S14, if the predicted temperature peak is greater than the preset critical temperature threshold, a local cooling intervention command is generated according to the gathering location, and the multi-channel coolant spray device is controlled to perform pre-deployment actions on the gathering location according to the local cooling intervention command.

[0015] In step S11, a discrete temperature data sequence collected by the microchannel inside the milling cutter during the cutting process is obtained. Spatial feature extraction and temporal migration feature extraction are performed on the discrete temperature data sequence to obtain a thermal spatiotemporal evolution sequence.

[0016] In one implementation, the step of extracting spatial features and temporal migration features from the discrete temperature data sequence to obtain a thermal spatiotemporal evolution sequence includes: Spatial features are extracted from the discrete temperature data sequence to obtain a spatial temperature distribution matrix. Temporal migration features are then extracted from the spatial temperature distribution matrix to obtain dynamic migration features. The dynamic migration characteristics are approximated by a polynomial to obtain the spatiotemporal evolution sequence of heat.

[0017] It should be noted that by embedding multiple sets of fiber Bragg grating sensors within the microchannels of the milling cutter, the temperature changes at various positions of the cutting edge are collected in real time during the cutting process, forming a discrete temperature data sequence. Since the sensor positions are fixed while the milling cutter is rotating at high speed, the collected discrete temperature data sequence is recorded in time series form. This data needs to be matched with the spatial coordinate system of the milling cutter to reconstruct the actual heating conditions of the cutting edge at different spatial positions. Specifically, the discrete temperature data sequence and real-time angle data provided by the spindle encoder are input, and a coordinate transformation algorithm is used to map the time-dimensional discrete temperature data to the three-dimensional spatial coordinate system of the milling cutter, outputting a spatial temperature set containing three-dimensional spatial coordinates and corresponding temperature values.

[0018] For example, the spindle speed is 3000 r / min, corresponding to a single rotation cycle of 20 ms. The sampling frequency of the fiber Bragg grating sensor is 200 Hz, and four temperature sampling points are collected in each rotation cycle. By transforming the coordinates, the temperature value of each sampling point is associated with the corresponding angular position of the cutting edge, resulting in a spatial temperature set distributed in the rake and flank face regions of the tool.

[0019] It should be noted that due to the limited number of sensors deployed within the microchannel, the data points in the spatial temperature dataset are spatially unevenly distributed. Therefore, interpolation is needed to reconstruct a regular temperature field grid. Specifically, the spatial temperature dataset is input, and the Kriging interpolation algorithm is used to estimate the temperature in areas without sensors, outputting a 50×50 resolution two-dimensional temperature field grid. The Kriging interpolation algorithm, based on spatial autocorrelation, uses the temperature values ​​of known sampling points to perform optimal unbiased estimation of unknown locations, effectively compensating for the insufficient spatial resolution of microchannel sensors.

[0020] Specifically, the interpolation region is set to cover a 10mm × 10mm area on the tool rake face, with a grid resolution of 50 × 50 and a single grid edge length of 0.2mm. An exponential variogram model is selected as the spatial correlation model for Kriging interpolation, and the variogram parameters are calibrated by fitting temperature data from the locations of the deployed sensors. Each grid node in the interpolated temperature field grid data contains a temperature estimate and its corresponding interpolation variance.

[0021] For example, in the machining of titanium alloy workpieces, eight fiber Bragg grating sensors were arranged at different depths from the cutting edge, ranging from 0.5 mm to 4.0 mm, and measured discrete temperature ranges from 230°C to 420°C. After Kriging interpolation, complete temperature values ​​were obtained for 2500 grid nodes, with the mean interpolation variance controlled within 15°C. 2 Within this range, the accuracy requirements for subsequent feature extraction are met.

[0022] It should be noted that during the cutting process, chip splashing or high-frequency vibration can cause abrupt noise data collected by some sensors, necessitating quality inspection and smoothing of the temperature field grid data. Specifically, the temperature field grid data is traversed, and the number of outlier values ​​exceeding 50°C of the average of adjacent grids is counted. If this number exceeds a preset threshold of 5, the current grid data is considered to have severe noise interference, and a Gaussian filtering algorithm is used for smoothing, outputting smooth temperature field grid data. The preset threshold of 5 is set based on the upper limit of the number of outlier values ​​in a single frame of the temperature field obtained through repeated experiments under normal cutting conditions. The experiment used a carbide end mill to cut a 45 steel workpiece, repeatedly collecting 500 frames of temperature field data within the working conditions of a spindle speed of 2000 r / min to 4000 r / min and a feed rate of 200 mm / min to 600 mm / min. The 99th percentile of the number of outlier values ​​was found to be 4.2, which was rounded up and set to 5. The kernel size of the Gaussian filter is 3×3, and the standard deviation σ is set to 1.0, effectively suppressing noise while maintaining the spatial distribution characteristics of the temperature gradient.

[0023] It should be noted that the smoothed temperature field grid data needs to undergo deep feature extraction to obtain a spatial distribution representation that condenses the core features of the blade's heating state. Specifically, the smoothed temperature field grid data is input into a pre-trained convolutional neural network model, which contains three convolutional layers and two pooling layers. The convolutional kernel size is 3×3, and the number of channels in each convolutional layer is 32, 64, and 128, respectively. The pooling layers use 2×2 max pooling with ReLU activation function. Through layer-by-layer convolution and pooling operations, the model extracts the spatial distribution features of the temperature gradient and outputs a 16×16 spatial temperature distribution matrix. Each element of this matrix corresponds to the feature value of a 0.625mm×0.625mm grid region in the original temperature field, highly condensing the core features of the initial heating state of the blade.

[0024] For example, 50×50 temperature field grid data is input, and after passing through the first convolutional layer (32 channels, 3×3 convolutional kernel, stride 1), a 48×48×32 feature map is output. After passing through the first pooling layer (2×2, stride 2), it is reduced to 24×24×32. After passing through the second convolutional layer (64 channels), a 22×22×64 feature map is output. After passing through the second pooling layer, it is reduced to 11×11×64. After passing through the third convolutional layer (128 channels), a 9×9×128 feature map is output. Finally, it is mapped to a 16×16 spatial temperature distribution matrix through a fully connected layer. The training of the convolutional neural network model employs a two-stage strategy: The first stage uses temperature field reconstruction as the pre-training objective. This involves setting up an upsampling decoding network symmetrical to the encoding structure after the encoding network, reconstructing a 50×50 temperature field from a 16×16 feature map. 10,000 sets of temperature field grid data, collected through cutting experiments with carbide tools under different cutting parameters and reconstructed using Kriging interpolation, are used as the unsupervised training set. The cutting parameters cover a working condition range of spindle speed from 2000 r / min to 4000 r / min and feed rate from 200 mm / min to 600 mm / min. The mean squared error is used as the loss function, and the network weights are optimized using the backpropagation algorithm. The second stage uses the measured tool wear after cutting as the supervision label for supervised fine-tuning of the regression head after the fully connected layer. During the inference stage, the regression head and decoding network are removed, and the 16×16 output of the third convolutional layer mapped by the fully connected layer is taken as the spatial temperature distribution matrix. After training convergence, the model's temperature distribution reconstruction error on the validation set is less than 3%.

[0025] It should be noted that after obtaining the spatial temperature distribution matrix, it is necessary to perform time-domain analysis in conjunction with the rotational conditions of the milling cutter to capture the flow and migration patterns of heat on the milling cutter cutting edge over time. Specifically, using the spindle rotation cycle as the time window, the continuously generated spatial temperature distribution matrix is ​​truncated along the time axis, outputting a set of matrix slices containing 100 consecutive time steps.

[0026] For example, with a spindle speed of 3000 r / min and a rotation cycle of 20 ms, the spatial temperature distribution matrix generated within 100 adjacent rotation cycles is extracted to obtain a set of 100 matrix slices with a size of 16×16, covering a cutting time span of 2000 ms.

[0027] Specifically, the above matrix slice set is input, and the central difference algorithm is used to calculate the rate of temperature change between adjacent spatial grid points within the spatial temperature distribution matrix of each frame. The output is a temperature gradient tensor containing spatial dimensional gradient information. The spatial dimension of this tensor is 16×16, and the value of each element is the temperature difference between the corresponding coordinate point and its four neighboring grid points divided by the spatial distance (grid side length 0.625mm), with units of ℃ / mm.

[0028] It should be noted that the heat distribution during the cutting process is not uniform; intense local friction can lead to a sudden increase in heat. By iterating through each element in the aforementioned temperature gradient tensor and calculating the L2 norm of the gradient components in both the X and Y directions as the comprehensive gradient amplitude, a preset gradient threshold of 15℃ / mm is set. If the comprehensive gradient amplitude of a certain region exceeds this threshold, that region is considered to be in an active state of abnormal heating. The preset gradient threshold of 15℃ / mm was determined through orthogonal cutting experiments using carbide tools to cut 45 steel. Temperature gradient data were collected within a cutting depth range of 0.5mm to 2.0mm and a feed rate of 0.1mm / r to 0.3mm / r. Tool flank wear reaching 0.3mm was used as the failure criterion. Analysis of the subject's working characteristic curve determined 15℃ / mm to be the optimal classification threshold, corresponding to a true positive rate of 92.3% and a false positive rate of 7.7%.

[0029] Specifically, for the temperature gradient tensor elements identified as active regions, the maximum-minimum normalization algorithm is used to linearly map their values ​​to the interval between 0 and 1, and output a standard temperature gradient sequence that eliminates dimensional differences.

[0030] It should be noted that, in order to capture the flow pattern of heat in the milling cutter cutting edge over time, a standard temperature gradient sequence was input into a Long Short-Term Memory (LSTM) network for temporal modeling. This network contains two layers of LTM units, each with 128 hidden units. Through internal forgetting and input gate mechanisms, it selectively memorizes and filters historical heating states from long-sequence data, outputting dynamic migration characteristics of the heat center in the three-dimensional spatial coordinate sequence at each time step. The training data for the LTM network comes from the marked heat center movement trajectories in historical cutting experiments. The heat center coordinates are extracted from the positions of the grid nodes with the highest temperatures in the temperature field at each time step. The training samples include 500 cutting processes, each covering 100 rotation cycles. The marked heat center three-dimensional coordinates are used as the supervision signal, and the mean squared error loss function is used for training. After training convergence, the model's heat center position prediction error on the test set is less than 0.15 mm.

[0031] Specifically, the dynamic migration characteristics of the Long Short-Term Memory (LSTM) network are input, and the least squares method is used to approximate the movement path of the heat center in a three-dimensional spatial coordinate system using a polynomial. The polynomial order is set to 3, and cubic polynomials are constructed for the X, Y, and Z coordinates of the heat center, with time as the independent variable. The optimization objective of the least squares method is to minimize the sum of squared residuals between the predicted coordinates of the fitted polynomial at each time point and the known three-dimensional coordinates at each time step in the dynamic migration characteristics. The three coordinate trajectory polynomials obtained from the fitting method completely describe the continuous movement path of the heat center over the past 100 rotation cycles, outputting the spatiotemporal evolution sequence of the heat and the corresponding original temperature field grid data for subsequent steps of full-surface temperature field interpolation reconstruction.

[0032] For example, under the conditions of a cutting depth of 1.0 mm and a feed rate of 400 mm / min, the dynamic migration characteristics of the Long Short-Term Memory (LSTM) network output show that the heat center migrates along the tool rake face from the tool tip to the interior of the tool body at an average speed of 0.8 mm / s. After approximation with a third-order polynomial, we obtain X(t) = 5.2 + 0.8t - 0.003t. 2 +0.0001t 3 Y(t) = 8.7 + 0.3t + 0.001t 2 The coordinate trajectory equation is Z(t) = 2.1 + 0.1t, where t is the time variable in seconds and the coordinate unit is mm. The spatiotemporal evolution sequence of heat covers a cutting process of 2000 ms, including the three-dimensional spatial coordinates of the heat center (i.e., the grid node with the highest temperature in the temperature field at each moment) within 100 rotation cycles and the corresponding temperature values ​​extracted from the original temperature field grid data, while retaining the corresponding 50×50 original temperature field grid data after Kriging interpolation.

[0033] In step S12, cubic spline interpolation and phase compensation are performed on the spatiotemporal evolution sequence of heat to obtain a spatiotemporally continuous temperature matrix, including: Discrete time nodes and spatial coordinates are extracted from the spatiotemporal evolution sequence of heat to construct a three-dimensional coordinate grid; The three-dimensional coordinate grid is subjected to cubic spline interpolation to obtain a continuous temperature field. The continuous temperature field is then subjected to phase compensation to obtain a spatiotemporal continuous temperature matrix.

[0034] Discrete time nodes and spatial coordinates are extracted from the spatiotemporal evolution sequence of heat to construct a three-dimensional coordinate grid.

[0035] It should be noted that the thermal spatiotemporal evolution sequence records discrete spatiotemporal temperature data according to the sensor sampling period, and structured coordinate node information needs to be extracted from it for subsequent continuous processing. Specifically, time nodes recorded every 5ms are extracted from the original temperature field grid data accompanying the thermal spatiotemporal evolution sequence, along with the corresponding transverse and longitudinal spatial coordinates of the tool surface. The preset extraction interval of 5ms is consistent with the 200Hz sampling frequency of the fiber optic grating sensor, ensuring that the temperature value at each sampling moment is included in subsequent processing. The discrete time nodes and two-dimensional spatial coordinates are orthogonally combined to construct a three-dimensional coordinate grid containing the time dimension and the two-dimensional spatial dimension, where each node in the grid corresponds to a discrete temperature value.

[0036] For example, for a cutting process of 100 rotation cycles (2000ms), 400 time nodes are extracted at 5ms intervals. The spatial dimension covers the tool rake face area of ​​10mm×10mm, and the spatial node resolution is 0.2mm. The constructed three-dimensional coordinate mesh contains 400×50×50=1,000,000 nodes, of which about 3,200 nodes actually have temperature measurement values ​​(8 sensors × 400 sampling times).

[0037] It should be noted that due to limitations in sensor sampling frequency and spatial arrangement density, the directly acquired data exhibits a discontinuous spatiotemporal distribution, making it unsuitable for precise extraction of temperature peaks. Specifically, the constructed three-dimensional coordinate grid and its corresponding discrete temperature values ​​are input, and a cubic spline interpolation algorithm is used to construct piecewise cubic polynomials between adjacent discrete coordinate points for smooth transition. A cubic polynomial S(x) = a + b(xx) is constructed between each pair of adjacent nodes. i )+c(xx i ) 2 +e(xx i ) 3 By using boundary condition constraints to ensure the continuity of function values, first derivatives, and second derivatives of adjacent polynomials at the nodes, C can be obtained over the entire temperature field. 2 A continuous, smooth surface, where e is the coefficient of the cubic term.

[0038] For example, cubic spline interpolation uses natural boundary conditions (the second derivative is zero at the endpoints), and the coefficients of each piecewise polynomial are obtained by solving a system of tridiagonal linear equations. The interpolated continuous temperature field has a spatial resolution of 1 mm and a temporal resolution of 1 ms. For a 10 mm × 10 mm tool rake face region, the continuous temperature field contains 10 × 10 = 100 spatial grid nodes, and a total of 2000 frames of continuous temperature field snapshots are output within a 2000 ms cutting time span.

[0039] It is worth noting that the spatial resolution was adjusted from 0.2 mm before interpolation to 1 mm after interpolation because cubic spline interpolation significantly increases the data density in the time dimension (from 5 ms interval to 1 ms). In the spatial dimension, through the construction of continuous polynomial surfaces, 1 mm resolution is sufficient to capture the spatial variation characteristics of the temperature gradient on the cutting edge surface. At the same time, the number of grid nodes in a single time frame is reduced from 2500 to 100, reducing the computational overhead of subsequent peak search.

[0040] It should be noted that the inherent thermal conductivity impedance of the tool material and the inherent response delay of the temperature sensing element during the cutting process cause a time lag between the temperature values ​​in the continuous temperature field and the actual heat generated during cutting. When the sensor is no more than 1 mm away from the cutting edge, the main effect of heat diffusion in the frequency domain can be approximated as a phase delay. To eliminate this response lag error, phase compensation processing of the continuous temperature field is required.

[0041] Specifically, the cutting command data issued by the CNC machine tool is acquired, and the initial time of cutting excitation generated when the cutting teeth contact the workpiece is extracted. Assuming the detected cutting excitation time is 10ms, the continuous temperature field data is traversed to extract the peak time when the temperature value at the same spatial coordinate point reaches its maximum value. Assuming this peak time is detected at 18ms, the time difference between the peak time and the cutting excitation time is calculated to be 8ms. If this time difference is greater than a preset judgment threshold of 3ms, then the current temperature data is determined to have a response lag error. The preset judgment threshold of 3ms is set based on the superposition of the thermal diffusivity of the tool material (carbide YG8) and the sensor response time: the thermal diffusivity of YG8 is approximately 2.5 × 10⁻⁶. -5 m 2 / s, the thermal diffusion characteristic time from the cutting edge to the nearest microchannel sensor (distance about 0.5mm) in the tool body is about 10ms, and the intrinsic response time of the fiber optic grating sensor is about 0.5ms. After superposition, the theoretical hysteresis is on the order of milliseconds. 3ms lower than the theoretical hysteresis is taken as the judgment threshold to ensure that the actual response hysteresis can be detected even if it is lower than the theoretical value, thus preserving the detection sensitivity margin.

[0042] The continuous temperature field data in the time domain is converted to the frequency domain using a Fast Fourier Transform (FFT) algorithm, and the phase information of each frequency component is extracted. To meet real-time control requirements, phase compensation uses the current time as the endpoint of the sliding window, and performs FFT on historical data segments within the window. The window length is set to 256 sampling points (corresponding to 256 ms), and a Hanning window function is used at the window boundaries to suppress the Gibbs effect. Based on the calculated 8 ms time difference, a lead phase compensation angle equal to the lag time is applied to each frequency component in the frequency domain. The relationship between the phase compensation angle Δφ and the frequency f and the lag time Δt is Δφ = 2π × f × Δt. For the component with frequency f, its phase angle is increased by 2π × f × 8 × 10⁻⁶. -3 Radius. Then, the inverse fast Fourier transform algorithm is used to restore the compensated frequency domain data to the time domain, and the compensation result at the end of the window is taken as the correction value at the current moment. The spatiotemporal continuous temperature matrix with the time axis shifted forward by 8ms is output.

[0043] For example, the time series of a continuous temperature field at spatial coordinates (15, 20) contains 2000 data points (1ms interval, 2000ms duration). Using the current time as the end of the window, the first 256 data points are truncated and subjected to a Fast Fourier Transform (FFT) to obtain 256 frequency components with a frequency resolution of 3.9Hz. For the 100Hz frequency component, the applied phase compensation angle is 2π×100×8×10⁻⁶. -3 =5.03 radians. After compensation, the value is restored by inverse fast Fourier transform, and the compensation value of the last point in the window is taken. The temperature peak time of this spatial coordinate point is corrected from 18ms to 10ms, which is aligned with the cutting excitation time. The compensated spatiotemporal continuous temperature matrix retains the same spatiotemporal resolution (1mm, 1ms) as the continuous temperature field, eliminating the response hysteresis caused by heat conduction and sensor delay.

[0044] In step S13, local temperature extrema are extracted from the spatiotemporal continuous temperature matrix to obtain a candidate peak set. The candidate peak set is then classified, filtered, and subjected to time-series deduction to obtain a predicted temperature peak. Based on the predicted temperature peak, coordinate mapping is performed to obtain the clustering location.

[0045] In one implementation, the step of extracting local temperature extrema from the spatiotemporal continuous temperature matrix to obtain a candidate peak set includes: The spatiotemporal continuous temperature matrix is ​​divided into spatial grids to obtain a set of microgrids. Temperature gradients are extracted from the set of microgrids to obtain heat accumulation characteristic values. A neighborhood extremum search is performed on the heat accumulation feature value to obtain a set of candidate peak values.

[0046] It should be noted that the spatiotemporal continuous temperature matrix provides the continuous temperature distribution of the tool cutting area at any given time. To locate local high-temperature hotspots, it needs to be divided into finer spatial units for analysis. Specifically, spatiotemporal continuous temperature matrix data at a specific time is extracted along the time slice dimension, and the data is meshed according to a preset spatial step size to construct a micro-mesh set. With a mesh edge length of 1mm, the tool rake face (10mm × 10mm) is divided into 100 independent micro-mesh units, each corresponding to a temperature value.

[0047] For example, the spatiotemporal continuous temperature matrix data at time 20ms is extracted to obtain a 10×10 microgrid temperature matrix containing 100 microgrids. The temperature value of each grid is the average value of the continuous temperature surface within the 1mm×1mm region.

[0048] It should be noted that during the cutting process, when the cutting teeth encounter hard points inside the workpiece, a momentary high heat release occurs, causing a sharp rise in local temperature. The region of this temperature abrupt change is a potential candidate location for the temperature peak. Specifically, the spatial temperature distribution matrix is ​​input, and the temperature change rate between each microgrid and its neighboring grids is calculated. For the grid located at (i,j), its heat accumulation characteristic value is G(i,j)=|T(i,j)-T avg | / l, where T avg The value is the average temperature of its four neighboring grids, where l is the grid side length of 1 mm. The gradient calculation uses four neighboring grids instead of eight neighboring grids because the four neighboring grids correspond to the main direction of heat conduction on the tool surface, while the diagonal direction of the eight neighboring grids is affected by the chip flow direction during actual cutting, resulting in non-symmetrical heat transfer.

[0049] A preset abrupt change threshold of 50℃ / mm was set. The calculation results of all grids were iterated. If the heat accumulation characteristic value of a certain grid was greater than 50℃ / mm, it was determined that a temperature abrupt change occurred in the region where that grid was located. The preset abrupt change threshold of 50℃ / mm was determined through orthogonal cutting experiments on carbide tools cutting nickel-based superalloy Inconel 718. During the process of normal tool wear to the dull standard (0.3mm of flank wear), the temperature gradient distribution under normal cutting conditions was statistically analyzed. The mean of the temperature gradient plus three times the standard deviation was 47.3℃ / mm, which was rounded up and set as 50℃ / mm.

[0050] It should be noted that after determining the existence of a temperature abrupt change, a local extremum search needs to be performed within the abrupt change region to extract candidate temperature peaks. Specifically, for abrupt change regions where the heat accumulation characteristic value exceeds a preset abrupt change threshold, the temperature value of the central grid cell is compared one by one with the temperature values ​​of its eight surrounding adjacent grid cells, using each microgrid within that region as the center. If the temperature value of the central grid cell is strictly greater than the temperature values ​​of all its surrounding adjacent grid cells, then the temperature value of the central grid cell and its corresponding two-dimensional spatial coordinates are extracted as local temperature extrema.

[0051] For example, in the spatial temperature distribution matrix at time 20ms, the microgrid temperature at coordinates (5,3) is 505℃. The temperatures of its eight neighboring grids are 493℃, 497℃, 500℃, 495℃, 502℃, 498℃, 494℃, and 496℃, respectively. The central grid temperature of 505℃ is significantly higher than all the neighboring grid temperatures. Therefore, (5,3,505℃) is extracted as a local temperature extremum. Following the same logic, another local temperature extremum is extracted at coordinates (7,4) as 462℃. The combination of these two local temperature extrema forms a candidate peak set, containing a total of two candidate peaks.

[0052] In one implementation, the step of classifying, filtering, and performing time-series deduction on the candidate peak set to obtain the predicted temperature peak includes: For each candidate peak in the candidate peak set, extract the temperature gradient and heating rate, construct a feature vector, and classify and filter the feature vectors to obtain the true peak set. By performing time-series analysis on the actual peak value set, the predicted temperature peak value is obtained.

[0053] It should be noted that the candidate peak set may include spurious peaks caused by instantaneous fluctuations in the sensor or splashing of cutting fluid. Although these spurious peaks have high temperature values, they do not represent the actual local high temperature accumulation at the cutting edge and need to be removed through classification and screening.

[0054] Specifically, for each candidate peak in the candidate peak set, the temperature difference between its microgrid and its four neighboring grids is extracted as the temperature gradient. Simultaneously, the temperature change of that grid between the current and previous times (1 ms interval) is divided by the time interval to obtain the heating rate. The absolute temperature values, temperature gradients, and heating rates of the candidate peaks are then concatenated in a fixed order to construct a three-dimensional feature vector.

[0055] For example, the temperature gradient of the candidate peak (5,3,505℃) is 8℃ / mm, the temperature at the previous moment is 502℃, the temperature at the current moment is 505℃, and the heating rate is (505-502) / 1ms=3℃ / ms. The constructed three-dimensional feature vector is [505,8,3].

[0056] It should be noted that the constructed 3D feature vector is input into a pre-trained classification model. This model has been trained using a large amount of historical cutting temperature data, constructing a classification boundary in the 3D feature space to distinguish between real cutting temperature peaks and measured outliers. The historical training data comes from cutting experiments on carbide tools under different cutting parameters, collecting a total of 5000 labeled temperature peak samples, including 3200 real peak samples and 1800 pseudo-peak samples. The labeling method involves microscopic observation of the tool surface after cutting, recording the spatial coordinates of the wear location, and geometrically matching the wear location with the peak location detected during cutting. Peaks with matching locations (Euclidean distance less than 0.5 mm) are labeled as real peaks, while peaks with no corresponding wear marks are labeled as pseudo-peaks. Training uses a radial basis function kernel, and the kernel parameter γ and penalty parameter C are validated using 5-fold cross-validation. 0.1,1,10,100 The optimal value is determined by searching within the grid.

[0057] Specifically, the vertical geometric distance from the input 3D feature vector to the classification boundary in the feature space is calculated as the classification margin. A preset margin threshold of 2 is set. If the calculated classification margin of a feature vector is greater than 2, the candidate peak corresponding to that feature vector is determined not to be a spurious peak, but rather a genuine local high-temperature accumulation at the cutting edge, and is thus confirmed as a true peak. The preset margin threshold of 2 is determined based on the optimal operating point obtained through precision and recall curve analysis on the validation set, corresponding to a precision of 94.6% and a recall of 91.2%.

[0058] For example, the classification interval of the candidate peak [505,8,3] is calculated to be 3.3, which is greater than the preset interval threshold of 2, and is therefore determined to be the true peak.

[0059] It should be noted that after a temperature peak is identified, its historical temperature change trend needs to be used to extrapolate its temperature evolution trajectory in the short term, in order to predict the actual formation of the temperature peak. Specifically, continuous temperature records of the microgrid identified as the true peak over the past 50ms are obtained and arranged in chronological order to form time series data. This time series data is then input into an autoregressive model AR(p) to fit the temporal evolution pattern. The general form of the autoregressive model is T(t) = c + Σφ i ×T(ti×Δt)+ε t Where c is a constant term, φ iHere, is the autoregressive coefficient, p is the model order, and Δt = 5ms is the sampling interval. The model order p is determined by searching within the range of 1, 2, 3, 4, 5 using the Akaike information content criterion, and the order with the smallest Akaike information content value is selected.

[0060] For example, the temperature time series of the true peak at coordinate (5,3) over the past 50ms is [438,442,447,453,460,468,477,487,498,505] (sampled at 5ms intervals, totaling 10 data points). After fitting the autoregressive model AR(2), the temperature change trend equation is obtained. Based on the fitted temperature change trend, the temperature evolution trajectory of the cutting edge in the next 10ms is deduced, and the temperature peak at time 30ms is predicted to be 520℃.

[0061] It is worth noting that the preset 10ms prediction time matches the rotation cycle (20ms) corresponding to a spindle speed of 3000r / min. 10ms is about half a rotation cycle. Within this time window, it is sufficient to complete the generation, transmission and execution of cooling commands. Moreover, the prediction error of the autoregressive model increases with the increase of prediction time, and the prediction error is controlled within ±5℃ within 10ms.

[0062] In one implementation, the step of performing coordinate mapping based on the predicted temperature peak to obtain the aggregation location includes: Extract the timestamp and spatial location identifier corresponding to the predicted temperature peak, and perform index matching in the spatiotemporal continuous temperature matrix to obtain the matrix element index; The neighborhood density distribution of the matrix element indices is statistically analyzed to obtain high-density regions. The geometric center of the high-density regions is then extracted to obtain the clustering locations.

[0063] It should be noted that the predicted temperature peak, having determined the temperature value and corresponding microgrid spatial coordinates during the extrapolation phase, needs to be mapped back into the spatiotemporal continuous temperature matrix data structure. Specifically, the timestamp and spatial location identifier corresponding to the predicted temperature peak are extracted and input into the spatiotemporal continuous temperature matrix for matching. The spatiotemporal continuous temperature matrix is ​​a data structure containing a three-dimensional coordinate system, where the horizontal and vertical axes represent the spatial distribution of the microgrid, and the depth axis represents the temporal evolution sequence.

[0064] For example, the predicted temperature peak is 520℃, corresponding to the timestamp of 30ms, and the spatial location is identified as 5mm horizontally and 3mm vertically. In the spatiotemporal continuous temperature matrix, the time axis scale is matched to 30, and the spatial axis scales are 5 and 3. The three-dimensional index of the predicted temperature peak in the matrix is ​​calculated to be (5,3,30).

[0065] It should be noted that the cutting heat peak is often not an isolated point, but rather forms a high-temperature accumulation zone in space around the peak core. To ensure that cooling intervention can cover all areas that may be damaged by heat, statistical analysis of the temperature field density distribution around the predicted peak is required. Specifically, with the matrix element index (5,3,30) as the center, the spatial search radius is set to two microgrids (i.e., 2 mm), and the time search window is 5 ms. Using the timestamp corresponding to the predicted temperature peak (30 ms) as the benchmark, an autoregressive model is used to perform time-series extrapolation on each microgrid in the neighborhood to obtain the temperature data of the neighborhood at the prediction time. The number of high-temperature nodes with temperatures exceeding 500℃ is counted, and their proportion to the total number of nodes in the neighborhood is calculated. The preset temperature screening value of 500℃ is set based on the recommended upper limit of the operating temperature of carbide tool materials: the hardness of YG8 carbide begins to decrease significantly above 500℃ (from 89 HRA at room temperature to about 82 HRA), and the risk of plastic deformation of the cutting edge increases sharply. Therefore, 500℃ is used as the criterion for judging high-temperature nodes.

[0066] For example, a spatial search radius of 2mm covers 5×5=25 spatial grids centered at (5,3), and a time window of 5ms covers the prediction time and the four time points preceding it (from 26ms to 30ms, with a 1ms interval). The three-dimensional neighborhood contains a total of 25×5=125 nodes. Among them, the number of nodes with a temperature exceeding 500℃ is 106, and the neighborhood density distribution value is 106 / 125×100%=85%.

[0067] A preset density threshold of 70% was set. Since 85% of the calculated neighborhood density distribution values ​​are greater than 70%, the region is determined to be a high-density region with high concentration of cutting heat. The preset density threshold of 70% was determined based on statistical analysis of real cutting heat accumulation regions under 100 sets of different cutting parameters. The mean neighborhood density of the real accumulation regions was 82.5%, and the standard deviation was 8.3%. The threshold was set as the mean minus 1.5 times the standard deviation (82.5% - 12.5% ​​= 70%), ensuring that the recall rate for identifying real accumulation regions is higher than 95%.

[0068] Specifically, for all high-temperature nodes (nodes with temperatures greater than 500℃) within the area identified as a high-density region, a weighted centroid algorithm is used to calculate their geometric center coordinates. The algorithm uses the temperature value of each node as a weight, performing a temperature-weighted average on the X and Y coordinates of the space. The calculated weighted geometric center coordinates are (5.2, 3.1), which are used as the final aggregation location corresponding to the predicted temperature peak on the cutting edge. For example, the weighted centroid calculation process is as follows: X... c =Σ(T i ×X i ) / ΣT i Ti Let X be the temperature value of the i-th high-temperature node. i The X-coordinate is used as the average value. The average X-coordinate of the 106 high-temperature nodes is 5.2 mm, and the average Y-coordinate is 3.1 mm.

[0069] In step S14, if the predicted temperature peak value is greater than a preset critical temperature threshold, a local cooling intervention command is generated based on the aggregation location, and the multi-channel coolant spray device is controlled to perform pre-deployment actions on the aggregation location based on the local cooling intervention command. If the predicted temperature peak value is not greater than the preset critical temperature threshold, the process returns to step S11 to continue acquiring the discrete temperature data sequence for the next cycle.

[0070] In one implementation, generating the local cooling intervention command based on the aggregation location includes: Extract the temperature gradient at the aggregation location, calculate the heat diffusion radius based on the preset thermal diffusivity of the tool material and the temperature gradient, and obtain the intervention boundary; The intervention boundary is spatially divided to obtain cooling zones; Extract the partition temperature data of each cooling partition from the spatiotemporal continuous temperature matrix, calculate the difference between the partition temperature data and the preset cooling temperature, and obtain the partition temperature difference set; The cooling medium flow rate of each zone is calculated based on the temperature difference set of the zones to obtain the zone flow rate set, and a local cooling intervention command is generated based on the zone flow rate set.

[0071] It should be noted that before generating a cooling intervention command, it is necessary to determine whether the predicted temperature peak has reached the level requiring intervention. A preset critical temperature threshold is set at 500℃. The intervention process is triggered when the detected predicted temperature peak exceeds 500℃. The preset critical temperature threshold of 500℃ is set based on the same criteria as the aforementioned high-temperature node determination standard, representing the upper limit of the recommended operating temperature for carbide tool materials.

[0072] Specifically, after triggering the intervention process, the temperature gradient value at the aggregation location (5.2, 3.1) is extracted. An autoregressive model is used to perform time-series extrapolation of the microgrid at this location, yielding a temperature gradient of 48℃ / mm at the predicted time. The heat diffusion radius r is calculated using the transient heat diffusion characteristic length formula: r = √(α × t). lead ), where α is the thermal diffusivity of the tool material, which is approximately 2.5 × 10⁻⁶ for YG8 cemented carbide. -5 m 2 / s,t lead =10ms is the cooldown advance trigger time. The calculated value is r = √(2.5 × 10⁻⁶). -5 ×0.01)=5.0×10 -4 m=500μm.

[0073] It should be noted that the distance from different locations within the intervention boundary to the heat center varies, resulting in different required cooling intensities. Therefore, the area within the intervention boundary needs to be divided to implement differentiated cooling. Specifically, the circular intervention boundary is divided into four equal fan-shaped regions along the circumference, centered on the accumulation location, with each fan-shaped region having a central angle of 90°. These four cooling zones are labeled as quadrants one through four.

[0074] For example, each of the four cooling zones has a coverage area of ​​approximately 0.20 mm. 2 Each cooling zone corresponds to an independent micro-nozzle for coolant spraying.

[0075] It should be noted that the temperature distribution varies within each cooling zone, requiring accurate calculation of the required cooling capacity using zone temperature data. Specifically, an autoregressive model is used to perform time-series extrapolation on each microgrid within each cooling zone, obtaining the temperature data for each zone at the predicted time (30ms). For each cooling zone, the extrapolated temperature values ​​of all microgrid nodes within the zone are extracted, and the average predicted temperature for the zone is calculated. The preset cooling temperature is set to 25℃, corresponding to the measured temperature of the coolant (water-based emulsion) at the nozzle outlet.

[0076] For example, the average predicted temperature of the first quadrant is 475°C, and the difference from the preset cooling temperature is 475-25=450°C. The temperature difference between the second and fourth quadrants is calculated in the same way, resulting in a set of temperature differences for the quadrants as [450°C, 465°C, 430°C, 420°C].

[0077] It should be noted that the cooling medium flow rate for each zone needs to be quantitatively calculated based on the temperature difference and the thermophysical properties of the tool material. The temperature difference set of the input zones is fed into a pre-calibrated fluid heat exchange model, which describes the mapping relationship between coolant flow rate and tool surface cooling rate. The model calibration process involves monitoring the tool surface temperature in real time under static heating conditions using thermocouples, spraying the tool surface heated to a known temperature with water-based emulsion at different flow rates (0.5 mL / s to 5 mL / s, with a step size of 0.5 mL / s), recording the time it takes for the temperature to drop to the preset cooling temperature, and establishing a temperature difference-flow rate mapping curve. During actual inference, the corresponding coolant flow rate is found on the calibration curve through linear interpolation based on the input temperature difference.

[0078] Specifically, the fluid heat exchange model outputs the required coolant flow rate for each cooling zone. A temperature difference of 450℃ corresponds to a required flow rate of 2.1 mL / s, 465℃ to 2.4 mL / s, 430℃ to 1.8 mL / s, and 420℃ to 1.7 mL / s, resulting in a zone flow rate set of [2.1, 2.4, 1.8, 1.7] mL / s. A piezoelectric valve response mapping algorithm is used to convert the flow rate requirement into a driving voltage signal to control the opening of the micro-valve. The relationship between the piezoelectric valve's flow rate and driving voltage is described by a pre-calibrated curve. For example, 2.1 mL / s corresponds to a driving voltage of 3.6V, 2.4 mL / s to 4.0V, 1.8 mL / s to 3.2V, and 1.7 mL / s to 3.0V. The target coolant flow rate, corresponding spatial location identifier, and micro-nozzle hardware address for each cooling zone are encapsulated into a data frame to generate a local cooling intervention command. After receiving the command, the local piezoelectric valve controller of each nozzle converts the target flow rate into a drive voltage signal according to the pre-calibrated curve.

[0079] In one implementation, controlling a multi-channel coolant spray device to perform pre-deployment actions on the accumulation location according to the local cooling intervention command includes: The local cooling intervention command is parsed to obtain the spatial coordinates of the aggregation location and the target coolant flow rate; Based on the spatial coordinates of the aggregation location, distance matching is performed in a preset cooling channel topology mapping model to obtain the target spray channel; The valve opening is calculated based on the target coolant flow rate and the geometric parameters of the target spray channel. A pulse width modulation drive control signal is generated to control the multi-channel coolant spray device to perform pre-deployment actions at the gathering position.

[0080] It should be noted that the local cooling intervention command is transmitted in the control system in data frame format, and the actuator needs to extract key physical control parameters from it. Specifically, the payload segment of the command data packet is read through a preset communication protocol unpacking algorithm, and parsed segment by segment in the order of frame header identifier, address field, data payload, and check field to separate the spatial coordinates of the aggregation location, the target coolant flow rate corresponding to each cooling zone, and the hardware address of the target nozzle.

[0081] For example, the spatial coordinates of the aggregation location are extracted as (5.2mm, 3.1mm), and the target coolant flow rates of each cooling zone are 2.1mL / s in the first quadrant, 2.4mL / s in the second quadrant, 1.8mL / s in the third quadrant, and 1.7mL / s in the fourth quadrant. The corresponding target nozzle hardware addresses are 2, 4, 6, and 8.

[0082] It should be noted that the tool has multiple pre-designed coolant channels, each connected to a micro-nozzle at its end. Each cooling zone needs to be matched with the closest physically located nozzle. The pre-defined coolant channel topology mapping model is a three-dimensional geometric model of the internal coolant channel network of the tool, recording the spatial direction, outlet coordinates, and corresponding micro-nozzle number of each channel. This model, derived from the tool's computer-aided design model, contains eight independent channels, with nozzle outlets distributed at different positions above the tool's rake face.

[0083] Specifically, the spatial coordinates of each cooling zone are input into a preset cooling channel topology mapping model. The spatial Euclidean distance between the zone and the outlet coordinates of each candidate nozzle is calculated, and the nozzle with the smallest distance is selected as the target spray channel for each zone. For example, for the first quadrant zone (flow rate 2.1 mL / s), the outlet coordinates of nozzle No. 2 among the eight nozzle outlets are (5.5 mm, 3.3 mm). The Euclidean distance between nozzle No. 2 and the center coordinates of the zone (5.2 mm, 3.1 mm) is 0.36 mm, which is the smallest distance. Therefore, channel No. 2 is selected as the target spray channel for the first quadrant zone. Similarly, nozzles No. 4, No. 6, and No. 8 are matched for the second to fourth quadrant zones, respectively, with corresponding flow rates of 2.4 mL / s, 1.8 mL / s, and 1.7 mL / s.

[0084] It should be noted that after determining the target spray channel, the precise opening of the channel valves needs to be calculated based on the target coolant flow rate. Specifically, taking the first quadrant as an example, the target coolant flow rate of 2.1 mL / s and the dimensional parameters of the inner diameter of channel 2 (2 mm) are input into a pre-set fluid dynamics mapping table to find the corresponding valve cross-sectional area percentage. The fluid dynamics mapping table is established through a combination of computational fluid dynamics simulation and experimental calibration. The simulation uses a k-ε turbulence model to solve the internal flow field of the channel under different valve openings, and the experiment verifies this by measuring the actual flow rate with an electromagnetic flowmeter. The table shows that the valve opening must reach 65% to meet the conditions of a flow rate of 2.1 mL / s and a pipe diameter of 2 mm.

[0085] Specifically, the rotational speed parameters of the current machine tool spindle are obtained, and the estimated arrival time of the cutting edge contacting the workpiece is calculated, thereby setting the advance amount for pre-triggered cooling. For example, with a spindle speed of 3000 r / min and a rotation cycle of 20 ms, the arc angle that the cutting edge needs to traverse from its current position to the corresponding angle of the convergence position is approximately 180°, corresponding to a time of 10 ms. A 65% valve opening degree is encoded and synthesized with the 10 ms advance opening timing to generate a pulse width modulation (PWM) drive control signal with a duty cycle of 65% and a frequency set to 1 kHz. Similarly, PWM signals corresponding to the valve opening degrees are generated for the nozzles in the other three zones. These PWM drive control signals are converted into drive voltage signals by the local piezoelectric valve controllers of each nozzle according to a pre-calibrated curve, directly acting on the piezoelectric valves of the multi-channel coolant spray device. Each micro-nozzle opens synchronously 10 ms before the cutting edge reaches the corresponding convergence position, completing a precise pre-positioning action at the convergence position before a large amount of cutting heat is generated.

[0086] In summary, this invention discloses a method for dynamic temperature control of milling cutters based on microchannel internal cooling monitoring. This invention acquires discrete temperature data from the microchannel and extracts the spatiotemporal evolution sequence of heat. Response lag is eliminated through cubic spline interpolation and phase compensation. Then, extreme value extraction, classification screening, and autoregressive time-series extrapolation are used to predict future temperature peaks and cluster locations. Local cooling intervention commands are generated before the peak arrives for pre-emptive control, achieving a shift from post-event remediation to precise pre-event intervention, thus improving cutting stability and tool life.

[0087] The second embodiment of the present invention provides a dynamic temperature control system for milling cutters based on microchannel internal cooling monitoring, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the dynamic temperature control method for milling cutters based on microchannel internal cooling monitoring as described above.

[0088] It should be noted that the end mill temperature dynamic control system based on microchannel internal cooling monitoring provided in this embodiment of the invention is used to execute all the process steps of the end mill temperature dynamic control method based on microchannel internal cooling monitoring in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0089] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for dynamic temperature control of a milling cutter based on microchannel internal cooling monitoring, characterized in that, include: A discrete temperature data sequence is acquired by collecting data from the microchannel inside the milling cutter during the cutting process. Spatial feature extraction and temporal migration feature extraction are performed on the discrete temperature data sequence to obtain a thermal spatiotemporal evolution sequence. The spatiotemporal evolution sequence of heat is subjected to cubic spline interpolation and phase compensation to obtain a spatiotemporal continuous temperature matrix; Local temperature extrema are extracted from the spatiotemporal continuous temperature matrix to obtain a candidate peak set. The candidate peak set is then classified, filtered, and subjected to time-series deduction to obtain the predicted temperature peak. Based on the predicted temperature peak, coordinate mapping is performed to obtain the clustering location. If the predicted temperature peak is greater than the preset critical temperature threshold, a local cooling intervention command is generated based on the aggregation location, and the multi-channel coolant spray device is controlled to perform pre-deployment actions on the aggregation location based on the local cooling intervention command.

2. The method for dynamic temperature control of milling cutters based on microchannel internal cooling monitoring according to claim 1, characterized in that, The step of extracting spatial features and temporal migration features from the discrete temperature data sequence to obtain a thermal spatiotemporal evolution sequence includes: Spatial features are extracted from the discrete temperature data sequence to obtain a spatial temperature distribution matrix. Temporal migration features are then extracted from the spatial temperature distribution matrix to obtain dynamic migration features. The dynamic migration characteristics are approximated by a polynomial to obtain the spatiotemporal evolution sequence of heat.

3. The method for dynamic temperature control of milling cutters based on microchannel internal cooling monitoring according to claim 1, characterized in that, The process of performing cubic spline interpolation and phase compensation on the spatiotemporal evolution sequence of heat to obtain a spatiotemporally continuous temperature matrix includes: Discrete time nodes and spatial coordinates are extracted from the spatiotemporal evolution sequence of heat to construct a three-dimensional coordinate grid; The three-dimensional coordinate grid is subjected to cubic spline interpolation to obtain a continuous temperature field. The continuous temperature field is then subjected to phase compensation to obtain a spatiotemporal continuous temperature matrix.

4. The method for dynamic temperature control of milling cutters based on microchannel internal cooling monitoring according to claim 1, characterized in that, The step of extracting local temperature extrema from the spatiotemporal continuous temperature matrix to obtain a candidate peak set includes: The spatiotemporal continuous temperature matrix is ​​divided into spatial grids to obtain a set of microgrids. Temperature gradients are extracted from the set of microgrids to obtain heat accumulation characteristic values. A neighborhood extremum search is performed on the heat accumulation feature value to obtain a set of candidate peak values.

5. The method for dynamic temperature control of milling cutters based on microchannel internal cooling monitoring according to claim 1, characterized in that, The process of classifying, filtering, and performing time-series analysis on the candidate peak set to obtain the predicted temperature peak includes: For each candidate peak in the candidate peak set, extract the temperature gradient and heating rate, construct a feature vector, and classify and filter the feature vectors to obtain the true peak set. By performing time-series analysis on the actual peak value set, the predicted temperature peak value is obtained.

6. The method for dynamic temperature control of milling cutters based on microchannel internal cooling monitoring according to claim 1, characterized in that, The step of mapping coordinates based on the predicted temperature peak to obtain the aggregation location includes: Extract the timestamp and spatial location identifier corresponding to the predicted temperature peak, and perform index matching in the spatiotemporal continuous temperature matrix to obtain the matrix element index; The neighborhood density distribution of the matrix element indices is statistically analyzed to obtain high-density regions. The geometric center of the high-density regions is then extracted to obtain the clustering locations.

7. The method for dynamic temperature control of milling cutters based on microchannel internal cooling monitoring according to claim 1, characterized in that, The step of generating a localized cooling intervention command based on the aggregation location includes: Extract the temperature gradient at the aggregation location, calculate the heat diffusion radius based on the preset thermal diffusivity of the tool material and the temperature gradient, and obtain the intervention boundary; The intervention boundary is spatially divided to obtain cooling zones; Extract the partition temperature data of each cooling partition from the spatiotemporal continuous temperature matrix, calculate the difference between the partition temperature data and the preset cooling temperature, and obtain the partition temperature difference set; The cooling medium flow rate of each zone is calculated based on the temperature difference set of the zones to obtain the zone flow rate set, and a local cooling intervention command is generated based on the zone flow rate set.

8. The method for dynamic temperature control of milling cutters based on microchannel internal cooling monitoring according to claim 1, characterized in that, The step of controlling the multi-channel coolant spray device to perform pre-deployment actions at the accumulation location according to the local cooling intervention command includes: The local cooling intervention command is parsed to obtain the spatial coordinates of the aggregation location and the target coolant flow rate; Based on the spatial coordinates of the aggregation location, distance matching is performed in a preset cooling channel topology mapping model to obtain the target spray channel; The valve opening is calculated based on the target coolant flow rate and the geometric parameters of the target spray channel. A pulse width modulation drive control signal is generated to control the multi-channel coolant spray device to perform pre-deployment actions at the gathering position.

9. A dynamic temperature control system for milling cutters based on microchannel internal cooling monitoring, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for dynamic control of milling cutter temperature based on microchannel internal cooling monitoring as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the method for dynamic control of milling cutter temperature based on microchannel internal cooling monitoring as described in any one of claims 1 to 8.

Images