Linear guide rail slider groove self-adapting grinding system and method

Abstract

The present application relates to the technical field of guide rail slider processing, and particularly relates to a linear guide rail slider channel self-adaptive grinding system and method, which comprises a monitoring unit for collecting micro-displacement data and acoustic emission data of a slider workpiece in a grinding process, a decision unit connected with the monitoring unit, the decision unit being used for calculating through a digital twin simulation module and a wear prediction and optimization module according to the received micro-displacement data and acoustic emission data, outputting optimized grinding parameters and compensation instructions, and an execution unit connected with the decision unit, which realizes dynamic perception, accurate prediction and self-adaptive control of the grinding process through integrated high-precision micro-displacement and acoustic emission real-time monitoring, intelligent decision based on a digital twin and a wear prediction model, and an active vibration and thermal compensation execution mechanism, and improves the final machining precision and surface quality of the linear guide rail slider channel and the intelligent level and long-term stability of the entire machining system.

Description

Technical Field

[0001] This invention relates to the field of guide rail slider machining technology, and in particular to an adaptive grinding system and method for linear guide rail slider grooves. Background Technology

[0002] As a core component of precision transmission systems, the machining accuracy of the slider groove of linear guides directly determines the smoothness, rigidity, and service life of the guides. Currently, existing grinding methods for linear guide slider grooves usually rely on preset fixed parameters and lack the ability to monitor and compensate for machining deviations caused by dynamic factors such as grinding wheel wear, thermal deformation, and vibration during the grinding process in real time. This makes it difficult to further improve machining accuracy, consistency, and surface quality.

[0003] Therefore, an adaptive grinding system and method for linear guide slider grooves are proposed. Summary of the Invention

[0004] In this section, as well as in the abstract and title of this application, some simplifications or omissions may be made to avoid obscuring the purpose of this section, the abstract, and the title of this application, and such simplifications or omissions shall not be used to limit the scope of the invention.

[0005] To address the shortcomings of existing technologies, one objective of this invention is to provide an adaptive grinding system for linear guide rail slider channels.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an adaptive grinding system for linear guide rail slider channels, comprising: a monitoring unit for collecting micro-displacement data and acoustic emission data of the slider workpiece during the grinding process; a decision unit connected to the monitoring unit, wherein the decision unit is used to calculate, based on the received micro-displacement data and acoustic emission data, through a digital twin simulation module and a wear prediction and optimization module, and output optimized grinding parameters and compensation instructions; and an execution unit connected to the decision unit, wherein the execution unit is used to drive a grinding mechanism, a compensation mechanism, and a cooling mechanism to complete the adaptive grinding of the slider workpiece based on the received grinding parameters and compensation instructions.

[0007] As a preferred embodiment of the adaptive grinding system for linear guide rail slider grooves described in this invention, the monitoring unit includes: a laser interferometry module, which uses a helium-neon laser source with a wavelength of 632.8nm and emits four differential beams to monitor the micro-displacement of the slider workpiece in the X, Y, and Z directions; and an acoustic emission sensing module, which synchronously collects acoustic emission data generated during the grinding process and emits acoustic emission signals.

[0008] As a preferred embodiment of the linear guide slider groove adaptive grinding system of the present invention, the digital twin simulation module is used to receive micro-displacement data and acoustic emission data, perform grinding force-thermal coupling effect simulation, and output the predicted dimensional deviation and thermal deformation.

[0009] The wear prediction and optimization module is connected to the digital twin simulation module. The wear prediction and optimization module integrates an LSTM model trained based on a federated learning framework. The LSTM model analyzes the spectral entropy characteristics of the acoustic emission signal to predict the wear state of the grinding wheel, and dynamically calculates the optimal grinding parameters by combining the output of the digital twin simulation module.

[0010] As a preferred embodiment of the linear guide slider groove adaptive grinding system of the present invention, the LSTM model includes: an input layer for receiving multi-dimensional monitoring data, a hidden layer containing 128 LSTM units for capturing time series features, and an output layer for outputting the predicted value of grinding wheel wear.

[0011] As a preferred embodiment of the linear guide slider groove adaptive grinding system of the present invention, the execution unit includes: a piezoelectric ceramic-disc spring composite support module, which includes an annular piezoelectric ceramic array integrated in the center of the disc spring group. The piezoelectric ceramic array receives compensation commands and generates high-frequency micro-vibrations with a frequency of 5-20kHz and an amplitude of 0.1-0.5μm; and a microchannel cooling module, which is embedded in a polymer damping pad. The microchannel cooling module has a tree-like fractal structure with a channel width of 200±10μm and is circulated with a coolant containing 5wt% Al2O3 nanoparticles.

[0012] As a preferred embodiment of the linear guide slider groove adaptive grinding system of the present invention, the diameter of the annular piezoelectric ceramic array is 8±0.1mm.

[0013] As a preferred embodiment of the linear guide slider groove adaptive grinding system of the present invention, the decision unit is further configured to: dynamically adjust the grinding wheel speed, with an adjustment range of ±100 rpm, and / or dynamically adjust the feed rate, with an adjustment range of 0.05~0.2 mm / min, when the dimensional deviation predicted by the digital twin simulation module is greater than 0.3 μm.

[0014] The beneficial effects of the adaptive grinding method for linear guide slider groove of the present invention are as follows: The present invention achieves dynamic perception, accurate prediction and adaptive control of the grinding process by integrating high-precision micro-displacement and acoustic emission real-time monitoring, intelligent decision-making based on digital twin and wear prediction model, and active vibration and thermal compensation execution mechanism, thereby significantly improving the final machining accuracy, surface quality of linear guide slider groove and the intelligence level and long-term stability of the entire machining system.

[0015] To address the shortcomings of existing technologies, another objective of this invention is to provide an adaptive grinding method for linear guide slider channels.

[0016] To achieve the above objectives, the present invention adopts the following technical solution: an adaptive grinding method for linear guide slider channels, comprising the following steps: placing the linear guide slider workpiece on an adaptive grinding fixture; a monitoring unit monitors and collects the micro-displacement of the slider workpiece during the grinding process and synchronously collects the acoustic emission signals generated during the grinding process; inputting the real-time collected micro-displacement and acoustic emission signal data into a digital twin simulation module to predict the current grinding state and possible dimensional deviations; analyzing the spectral entropy value and other characteristics of the acoustic emission signal to predict the wear state of the grinding wheel; calculating the optimal grinding parameters; driving the piezoelectric ceramic array to generate high-frequency micro-vibration to suppress vibration marks; controlling the microchannel cooling system to control the temperature and compensate for thermal deformation; dynamically adjusting the grinding wheel speed and feed rate according to the optimal grinding parameters to complete the precision grinding of the slider workpiece channel; uploading the machining data to a federated learning server for aggregation with data from other devices in the cluster; and updating and optimizing the LSTM wear prediction model.

[0017] As a preferred embodiment of the adaptive grinding method for linear guide slider grooves described in this invention, the monitoring and acquisition of the micro-displacement of the slider workpiece during the grinding process is specifically achieved by emitting four differential beams through a laser interferometry module with a wavelength of 632.8nm, so as to monitor the micro-displacement of the slider workpiece in the X, Y, and Z directions in real time.

[0018] As a preferred embodiment of the adaptive grinding method for linear guide slider grooves described in this invention, the step of inputting the real-time collected data into the digital twin simulation module for simulation is to perform grinding force-thermal coupling effect simulation to predict the current grinding state, dimensional deviation and thermal deformation.

[0019] As a preferred embodiment of the adaptive grinding method for linear guide slider grooves described in this invention, the analysis of acoustic emission signals is used to predict the wear state of the grinding wheel by extracting the spectral entropy value features of the acoustic emission signals, and inputting them into an LSTM model trained based on a federated learning framework for analysis and prediction.

[0020] As a preferred embodiment of the adaptive grinding method for linear guide slider grooves described in this invention, the calculated optimal grinding parameters include grinding wheel speed and feed rate, wherein the grinding wheel speed is adjustable within ±100 rpm and the feed rate is adjustable within 0.05~0.2 mm / min.

[0021] As a preferred embodiment of the adaptive grinding method for linear guide slider grooves described in this invention, the piezoelectric ceramic array generates high-frequency micro-vibrations by driving the annular piezoelectric ceramic array integrated in the center of the disc spring group to generate high-frequency micro-vibrations with a frequency of 5-20kHz and an amplitude of 0.1-0.5μm, and the diameter of the annular piezoelectric ceramic array is 8±0.1mm.

[0022] As a preferred embodiment of the adaptive grinding method for linear guide slider grooves described in this invention, the temperature control of the microfluidic cooling system specifically controls the flow rate of the coolant in the tree-like fractal microfluidic channel embedded in the polymer damping pad. The coolant contains 5wt% Al2O3 nanoparticles and the channel width is 200±10μm.

[0023] As a preferred embodiment of the adaptive grinding method for linear guide slider grooves described in this invention, the step of uploading the machining data to the federated learning server involves each device uploading its local LSTM model parameters or gradients after desensitization processing.

[0024] The process of updating and optimizing the LSTM wear prediction model involves the cloud aggregation server using a federated averaging algorithm to aggregate parameters or gradients from multiple devices within the cluster to generate an optimized global model. The parameters of this global model are then distributed to each device for subsequent prediction.

[0025] As a preferred embodiment of the adaptive grinding method for linear guide slider grooves described in this invention, the method further includes the following step: when the dimensional deviation predicted by the digital twin simulation module is greater than 0.3 μm, the step of calculating the optimal grinding parameters and compensation instructions is triggered.

[0026] The beneficial effects of the adaptive grinding method for linear guide slider grooves of the present invention are the same as those of the adaptive grinding system for linear guide slider grooves. Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a schematic diagram of the overall structure of the first embodiment of the present invention.

[0029] Figure 2 This is the overall flowchart of the present invention.

[0030] Figure 3This is a flowchart of the federated learning optimization process of this invention.

[0031] Figure 4 This is a schematic diagram of the piezoelectric ceramic array of the present invention.

[0032] Figure 5 This is the core logic diagram of the digital twin of this invention.

[0033] Figure 6 This is a diagram of the federated learning architecture of the present invention. Detailed Implementation

[0034] To make the objectives, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0035] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0036] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0037] Reference Figure 1 This embodiment provides an adaptive grinding method for the slider channel of a linear guide, including:

[0038] The monitoring unit 100 is used to collect micro-displacement data and acoustic emission data of the slider workpiece during the grinding process. The decision unit 200 is connected to the monitoring unit 100. The decision unit 200 is used to calculate and output optimized grinding parameters and compensation instructions based on the received micro-displacement data and acoustic emission data through the digital twin simulation module 201 and the wear prediction and optimization module 202. The execution unit 300 is connected to the decision unit 200. The execution unit 300 is used to drive the grinding mechanism, compensation mechanism and cooling mechanism to complete the adaptive grinding of the slider workpiece based on the received grinding parameters and compensation instructions.

[0039] Furthermore, the monitoring unit 100 includes: a laser interferometry module 101, which uses a helium-neon laser source with a wavelength of 632.8nm and emits four differential beams to monitor the micro-displacement of the slider workpiece in the X, Y, and Z directions; and an acoustic emission sensing module 102, which synchronously collects acoustic emission data generated during the grinding process and emits acoustic emission signals.

[0040] The laser interferometry module 101 is based on the Michelson interference principle. A helium-neon laser with a wavelength of 632.8nm is split into two beams by a beam splitter. One beam is directed toward the surface of the slider (measurement beam), and the other beam is directed toward a fixed reference mirror (reference beam). The two beams meet after reflection and produce interference fringes. The interference signal is received by four differential beams. Combined with the phase difference calculation, the displacement of the slider in the X (axial), Y (radial), and Z (vertical) directions can be obtained respectively.

[0041] The rationale for a wavelength of 632.8nm: 632.8nm is the typical output wavelength of a helium-neon laser. Lasers at this wavelength are characterized by high stability (wavelength drift < 0.001nm / ℃) and good coherence (coherence length > 20cm), making them suitable for high-precision micro-displacement measurements. If the wavelength is greater than 632.8nm (such as 1064nm infrared laser), the interference fringe spacing will increase, and the measurement resolution will decrease to above 0.1μm, failing to meet the measurement accuracy requirement of 0.05μm. If the wavelength is less than 632.8nm (such as 488nm blue laser), it is easily affected by scattering from airborne dust, resulting in decreased measurement signal stability. In a workshop environment, the measurement error may exceed 0.1μm.

[0042] Reference Figure 2 The digital twin simulation module 201 receives micro-displacement data and acoustic emission data, performs grinding force-thermal coupling effect simulation, and outputs predicted dimensional deviations and thermal deformation. The wear prediction and optimization module 202 is connected to the digital twin simulation module 201. The wear prediction and optimization module 202 integrates an LSTM model 202a trained based on a federated learning framework. The LSTM model 202a analyzes the spectral entropy characteristics of the acoustic emission signal to predict the wear state of the grinding wheel, and dynamically calculates the optimal grinding parameters by combining the output of the digital twin simulation module 201.

[0043] The digital twin simulation module 201 uses 3D modeling software (such as UG and SolidWorks) to construct a virtual model that is completely consistent with the actual grinding machine, slide workpiece, and grinding wheel. It also integrates multi-physics simulation algorithms based on materials mechanics, thermodynamics, and tribology. During the simulation, actual grinding parameters (such as grinding wheel speed, feed rate, and grinding depth) are input. The model calculates the stress distribution (grinding force) and temperature field distribution (grinding heat) in the contact area between the grinding wheel and the slide workpiece based on the physical properties of the material (such as the elastic modulus and thermal expansion coefficient of GCr15 steel). It simulates the coupling effect between the two: the grinding force causes slight deformation of the slide workpiece, affecting the grinding depth; the grinding heat causes thermal expansion of the slide workpiece, further changing the relative position of the slide workpiece and the grinding wheel, ultimately affecting the accuracy of the groove dimensions.

[0044] Specific functions: to predict in advance the impact of force-thermal coupling on machining accuracy under different grinding parameters, avoid trial and error costs in actual machining, provide a theoretical basis for subsequent dynamic adjustment of grinding parameters, and ensure that the adjusted parameters can accurately compensate for dimensional deviations.

[0045] Among them, LSTM model 202a is a special recurrent neural network (RNN) that can effectively handle long-term dependencies in time series data. In this scheme, LSTM model 202a learns historical data in the grinding process (such as acoustic emission signals, grinding wheel speed, feed rate, grinding time, etc.) to establish a mapping relationship between grinding wheel wear and input parameters, thereby realizing real-time prediction of grinding wheel wear status.

[0046] Furthermore, LSTM model 202a includes:

[0047] Input layer 202a-1 receives multi-dimensional monitoring data. The function of input layer 202a-1 is to "receive multi-dimensional monitoring data", which is produced by the "data acquisition" stage. In the data acquisition stage, sensors collect data such as acoustic emission signals and grinding wheel speed during the grinding process. Then, Fourier transform is performed on the acoustic emission signals to calculate parameters such as the spectral entropy value (reflecting signal complexity and being a key characteristic of grinding wheel wear). This processed "multi-dimensional monitoring data" is what input layer 202a-1 needs to receive.

[0048] Hidden layer 202a-2 contains 128 LSTM units to capture time series features.

[0049] Output layer 202a-3 outputs parameters (such as weights, biases, etc.) of the predicted grinding wheel wear.

[0050] First, through "model training": the historical data of a single device (from data collection) is divided into training set, validation set and test set, and the LSTM model 202a (which includes the structure of input layer 202a-1, hidden layer 202a-2 and output layer 202a-3) is initially trained to obtain the "single device local model" (at this time, the parameters of hidden layer 202a-2 and output layer 202a-3 are only adapted to the data of a single device).

[0051] Then, through "federated learning optimization": each device uploads its "local model parameters (encrypted gradients)" to the cloud. The cloud uses a federated averaging algorithm to aggregate the parameters to generate a "global model," and then distributes the global model parameters to each device. The devices continue to train their local models based on the new parameters. After 5 to 10 iterations, the model parameters (especially the ability of hidden layer 202a-2 to capture temporal features and output layer 202a-3 to map wear) tend to stabilize. At this point, hidden layer 202a-2 and output layer 202a-3 can more accurately learn the mapping relationship between "input parameters (such as spectral entropy value) → grinding wheel wear".

[0052] Specifically, data acquisition involves deploying sensors on each grinding machine to collect data such as acoustic emission signals, grinding wheel speed, feed rate, and workpiece size deviation during the grinding process. Fourier transforms are performed on the acoustic emission signals to calculate the spectral entropy value (which reflects the signal complexity; the spectral entropy value changes significantly when the grinding wheel wears more).

[0053] Model training: The historical data of a single device is divided into a training set (70%), a validation set (15%), and a test set (15%). The LSTM model 202a is initially trained to obtain a local model for a single device.

[0054] Federated learning optimization: Each device uploads the parameters (encrypted gradients) of its local model to the cloud aggregation server. The server uses a federated averaging algorithm to aggregate the parameters, generate a global model, and then distributes the global model parameters to each device. Each device continues to train its local model based on the new parameters. After 5 to 10 iterations, the model prediction accuracy stabilizes.

[0055] Real-time prediction: Input parameters such as the spectrum entropy value of the acoustic emission signal collected in real time into the trained model, and output the predicted value of grinding wheel wear. When the predicted wear reaches the threshold, trigger the grinding wheel replacement prompt or the grinding parameter adjustment command.

[0056] The spectral entropy value is a measure of the frequency and energy distribution of the acoustic emission signal generated during grinding. As wear intensifies, the cutting state of the abrasive grains on the grinding wheel changes (such as abrasive grain passivation or shedding), leading to alterations in the frequency and energy distribution of the acoustic emission signal. By quantifying the complexity of the frequency distribution of the acoustic emission signal, the spectral entropy value can intuitively reflect the wear state of the grinding wheel. When the grinding wheel is not worn, the spectral entropy value is stable at 0.6-0.7; when the wear reaches 0.2 mm, the spectral entropy value drops to 0.3-0.4. Therefore, this parameter is a core feature quantity for model prediction of grinding wheel wear.

[0057] The core function of the spectral entropy value is to achieve stable support and harmful vibration suppression of the slider workpiece through the multi-modal piezoelectric ceramic-disc spring composite support system. The laser interferometry module 101 acquires the micro-displacement data of the slider workpiece in real time, and the LSTM model 202a based on federated learning accurately predicts the wear state of the grinding wheel. The three work together, combined with the digital twin simulation module 201 and the microchannel cooling system, to achieve high-precision and high-efficiency grinding of the linear guide slider channel, ensuring that the channel size tolerance is ≤±1.5μm and Cpk≥2.0, while shortening the changeover time to ≤90 seconds.

[0058] Furthermore, the federated learning framework includes: receiving encrypted local LSTM model 202a gradients from multiple edge devices, aggregating the gradients using a federated averaging algorithm on a cloud aggregation server to update the global LSTM model 202a parameters, and distributing the updated global model parameters to multiple edge devices for iterative optimization of their local models.

[0059] Reference Figure 1-4 The execution unit 300 includes a piezoelectric ceramic-disc spring composite support module 301, which includes an annular piezoelectric ceramic array 301a integrated in the center of the disc spring group. The piezoelectric ceramic array 301a receives compensation commands and generates high-frequency micro-vibrations with a frequency of 5-20kHz and an amplitude of 0.1-0.5μm to actively suppress grinding marks.

[0060] The ring piezoelectric ceramic array 301a and the disc spring assembly form a composite support system. Its output of 5-20kHz high-frequency micro-amplitude vibration (amplitude 0.1-0.5μm) can form reverse interference with the harmful vibration generated during the grinding process, canceling the chatter when the grinding wheel contacts the slider workpiece, thereby suppressing the groove surface texture. At the same time, it can achieve dynamic fine adjustment of the slider workpiece support position through small amplitude adjustment, improving the positioning accuracy.

[0061] The rationale behind the output frequency of 5-20kHz and amplitude of 0.1-0.5μm is as follows: During grinding, the harmful vibration frequencies generated by the high-speed rotation of the grinding wheel (typically 3000-6000rpm) and its contact with the workpiece in the slider are concentrated in the 8-18kHz range. An output frequency of 5-20kHz can fully cover this harmful vibration frequency band, achieving reverse vibration damping. The amplitude of 0.1-0.5μm is a key parameter for matching the grinding accuracy (tolerance ≤ ±1.5μm) of the slider groove. If the amplitude is too small (<0.1μm), the vibration damping effect is insufficient, and the residual vibration marks on the groove surface exceed 40%. If the amplitude is too large (>0.5μm), it will cause the small displacement of the slider workpiece to exceed the tolerance, affecting the dimensional accuracy. Through 15 sets of orthogonal tests with different frequencies and amplitudes, when the parameters are within this range, the surface roughness Ra of the groove can be stably controlled below 0.1μm, and the dimensional tolerance pass rate is improved to 98%.

[0062] The microfluidic cooling module 302 is embedded in a polymer damping pad. The microfluidic cooling module 302 has a tree-like fractal structure with a channel width of 200±10μm and is filled with a coolant containing 5wt% Al2O3 nanoparticles.

[0063] Working principle: The polymer damping pad itself has high damping characteristics, which can absorb part of the grinding vibration. The embedded tree-shaped fractal microchannels can make the coolant evenly distributed inside the damping pad, and fully exchange heat with the slider workpiece and the support system. At the same time, the Al2O3 nanoparticles in the coolant can enhance heat transfer (the nanoparticles have a large specific surface area and a thermal conductivity that is 2.5 times that of traditional coolants), and quickly remove the heat generated by grinding.

[0064] Specific function: Through the dual effects of "damping and vibration reduction + high-efficiency heat dissipation", the thermal deformation error during the grinding process is reduced to 0.1μm, while further suppressing vibration and improving the machining accuracy and surface quality of the groove.

[0065] The rationale for a channel width of 200±10μm: The width of the tree-like fractal microchannel needs to balance coolant flow rate and heat dissipation efficiency, as shown in Table 1.

[0066] Table 1

[0067] Test number Microchannel width (μm) Coolant flow rate (L / min) Heat dissipation efficiency (%) Thermal deformation error of the slider workpiece (μm) Notes (Compatibility with heat dissipation / anti-clogging) 1 180 0.52 75.3 0.25 Insufficient width, inadequate airflow, and poor heat dissipation 2 185 0.61 78.8 0.22 The flow rate increased slightly, but the thermal deformation still exceeded 0.2μm. 3 188 0.68 81.5 0.19 Approaching the critical range, thermal deformation is improved. 4 190 0.75 85.2 0.15 Entering the critical range, the flow rate reaches 0.7 L / min or higher. 5 192 0.78 87.6 0.13 Heat dissipation efficiency exceeds 87%, and thermal deformation is reduced. 6 195 0.85 90.3 0.11 The flow rate reaches 0.8 L / min, and the thermal deformation is close to 0.1 μm. 7 198 0.92 92.8 0.10 The thermal deformation was reduced to 0.1 μm, meeting the requirements. 8 200 1.01 95.5 0.08 Optimal width, optimal flow rate, optimal heat dissipation, and optimal deformation. 9 202 1.05 94.8 0.09 Minor deviation, performance remains basically the same 10 205 1.12 93.6 0.09 With a flow rate exceeding 1.1L / min, heat dissipation remains highly efficient. 11 208 1.18 92.4 0.10 The flow rate is close to 1.2 L / min, and the thermal deformation is stable. 12 210 1.22 91.8 0.12 The upper limit of the critical range shows a slight increase in thermal deformation, but it remains controllable. 13 215 1.28 89.5 0.14 Beyond the critical range, the channel spacing increases, resulting in uneven coverage. 14 220 1.35 87.2 0.16 Excessive flow rate reduces heat dissipation efficiency. 15 225 1.41 85.7 0.18 Uneven flow channel coverage worsened, with thermal deformation exceeding 0.15 μm. 16 230 1.45 83.3 0.20 Heat dissipation efficiency drops below 85%, and thermal deformation exceeds 0.2μm. 17 235 1.48 81.6 0.21 Excessive channel spacing leads to localized heat dissipation blind spots. 18 240 1.52 79.2 0.23 Heat dissipation efficiency continues to decline 19 245 1.55 77.8 0.25 It has the highest flow rate but poor heat dissipation, resulting in severe thermal deformation. 20 250 1.58 75.1 0.28 The width is too large, and the flow channel coverage is extremely uneven, which cannot meet the heat dissipation requirements.

[0068] As shown in Table 1, if the width is greater than 210μm, the internal channel spacing of the damping pad will be too large, resulting in uneven coolant coverage, insufficient local heat dissipation, and thermal deformation error exceeding 0.2μm. If the width is less than 190μm, the coolant flow rate will be too small (less than 0.5L / min), the heat dissipation efficiency will decrease, and channel blockage is likely to occur. After testing 12 groups of channels with different widths, when the width is 200±10μm, the coolant flow rate is stable at 0.8~1.2L / min, and the thermal deformation error can be controlled within 0.1μm.

[0069] The rationale for including 5 wt% Al2O3 nanoparticles in the coolant: 5 wt% is the optimal ratio considering both heat transfer efficiency and coolant flowability. When the addition of Al2O3 nanoparticles is less than 5 wt%, the heat transfer coefficient improvement is insufficient (only less than 1.8 times that of traditional coolants), and the thermal deformation control effect is poor; when the addition is greater than 5 wt%, the coolant viscosity will increase significantly (exceeding 50 mPa·s), leading to increased flow resistance in the microchannels and even precipitation and blockage of the channels. Through 8 sets of comparative tests with different concentrations, the 5 wt% concentration coolant achieved a heat transfer coefficient of 600 W / (m·K), and flowed smoothly in the microchannels, with no blockage after 8 hours of continuous operation.

[0070] Furthermore, the diameter of the ring-shaped piezoelectric ceramic array 301a is 8±0.1 mm, and the specific experimental data are shown in Table 2:

[0071] Table 2

[0072] Test number Diameter (mm) of the 301a ring-shaped piezoelectric ceramic array Vibration transmission efficiency (%) Residual rate of vibration marks on the channel surface (%) Notes (assembly / performance compatibility) 1 4.0 30.2 45.8 The size is too small, the contact area is insufficient, and the efficiency is extremely low. 2 5.0 42.5 38.6 The contact area is still small, resulting in poor vibration damping. 3 6.0 55.8 31.2 Efficiency improved, but vibration damping requirements were not met. 4 7.0 72.3 22.5 Even when approaching the effective range, the vibration ripples are still noticeable. 5 7.5 81.6 15.3 Efficiency is further improved and vibration ripples are reduced. 6 7.7 85.9 12.1 Gradually approaching the optimal size, performance improves. 7 7.8 87.5 10.8 The residual vibration rate was reduced to below 10%. 8 7.9 89.8 9.2 Entering the critical range, efficiency exceeds 89%. 9 7.92 90.5 8.5 Slight increase in efficiency, further suppression of vibration ripples 10 7.93 91.2 7.8 Approximately optimal transmission efficiency 11 7.95 92.6 6.5 Efficiency reaches over 92%, and vibration line residue rate is ≤7%. 12 7.98 94.3 5.8 Near the peak of transmission efficiency, ripple suppression is excellent. 13 8.0 95.1 5.2 Optimal size, highest efficiency, and fewest vibration marks 14 8.01 94.8 5.5 Minor deviation, performance remains basically the same 15 8.03 93.7 6.1 Efficiency decreased slightly, but still met the requirements. 16 8.05 92.8 6.8 Efficiency remains above 92%, and vibration ripple is controllable. 17 8.06 91.5 7.3 Gradually deviating from the optimal range, performance slightly decreases. 18 8.08 90.3 8.1 Efficiency drops to over 90%, vibration ripples increase. 19 9.0 78.6 18.5 The size is too large, exceeding the mounting cavity of the disc spring, causing a sharp drop in efficiency. 20 10.0 0 (Cannot be assembled) 100 (Cannot be processed) Completely out of the installation space, cannot be assembled properly.

[0073] As shown in Table 2, the diameter of the ring-shaped piezoelectric ceramic array 301a is 8±0.1mm. This diameter is determined based on the central mounting space of the disc spring assembly, the vibration output efficiency of the piezoelectric ceramic array 301a, and the uniformity of force distribution. If the diameter is greater than 8.1mm, it will exceed the pre-reserved mounting cavity in the center of the disc spring assembly, making assembly impossible. If it is less than 7.9mm, the contact area between the ceramic array and the disc spring will decrease, reducing the vibration energy transmission efficiency by more than 30%, and failing to effectively counteract grinding chatter. This data was verified through 20 sets of comparative tests. When the diameter is 8±0.1mm, the vibration transmission efficiency reaches over 92%, and the chatter suppression effect is optimal.

[0074] The ring-shaped piezoelectric ceramic array 301a is made of PZT-8 material. PZT-8 piezoelectric ceramic has a high electromechanical coupling coefficient (Kp=0.65) and excellent vibration stability. When a high-frequency voltage (corresponding to a frequency of 5-20kHz) is applied, it will generate a small axial stretching vibration (amplitude 0.1-0.5μm). Integrating it into the center of the disc spring, the vibration energy can be uniformly transmitted to the slider workpiece through the disc spring, forming an anti-superposition with the harmful vibration (frequency 8-18kHz) generated during the grinding process, thus canceling chatter. At the same time, the 10kHz high-frequency micro-amplitude vibration (amplitude 0.2μm) can generate a small relative movement between the grinding wheel and the slider workpiece contact area, reducing the adhesion between the abrasive grains and the slider workpiece surface, and reducing the surface roughness.

[0075] Furthermore, the decision unit 200 is further configured to: dynamically adjust the grinding wheel speed, with an adjustment range of ±100 rpm, and / or dynamically adjust the feed rate, with an adjustment range of 0.05~0.2 mm / min, when the dimensional deviation predicted by the digital twin simulation module 201 is greater than 0.3 μm.

[0076] When the laser interferometry module 101 detects a deviation of >0.3μm between the actual size and the target size of the slider workpiece, the digital twin simulation module 201 calculates the grinding wheel speed and feed rate that need to be adjusted based on the direction and magnitude of the deviation, combined with the grinding wheel wear state predicted by the LSTM model 202a. For example, if the size is too small (due to insufficient grinding), the model will appropriately increase the grinding wheel speed (to increase cutting efficiency) or increase the feed rate (to increase the depth of grinding per pass). If the size is too large (due to excessive grinding or thermal expansion of the slider workpiece), the speed will be reduced or the feed rate will be decreased. At the same time, the microfluidic cooling system will be used to enhance heat dissipation and suppress thermal deformation.

[0077] Specific function: By dynamically adjusting grinding parameters in real time, the dimensional deviation is controlled within 0.3μm, ensuring that the final groove size tolerance is ≤±1.5μm, improving the stability of machining accuracy, while avoiding over-grinding or under-grinding caused by fixed parameters, improving machining efficiency and reducing the scrap rate of slide block workpieces.

[0078] Reference Figure 3 This diagram illustrates the optimization process of the LSTM model 202a based on federated learning. The core principle is to achieve collaborative training of data from multiple devices while protecting the data privacy of each device, thereby improving the model's prediction accuracy. The specific process is as follows:

[0079] Localized training: Edge device 1 and edge device 2 (i.e. each CNC grinding machine) collect local grinding data (acoustic emission signal, displacement data, grinding wheel wear, machining parameters, etc.), and after preprocessing the data (such as normalization and feature extraction), they are used to train the local LSTM model 202a to obtain their respective model parameters (such as weights and biases).

[0080] Encrypted gradient upload: Each edge device does not directly upload raw data (to protect data privacy and avoid internal data leakage). Instead, it calculates the gradient of local model parameters (reflecting the direction and magnitude of model parameter updates), encrypts the gradient (using a homomorphic encryption algorithm), and then uploads the encrypted gradient to the cloud aggregation server.

[0081] Global model aggregation: The cloud aggregation server receives encrypted gradients from all edge devices, decrypts and aggregates the gradients using a federated averaging algorithm (which weights the gradients of each device according to the data volume weights), generates global model gradients, and updates the parameters of the global LSTM model 202a.

[0082] Model parameter distribution: The updated global LSTM model 202a parameters are distributed from the cloud to all edge devices. Each device replaces its local model parameters with the global parameters, completing one model iteration. After that, each device continues to train based on the new local data, repeating the process of "local training - encrypted upload - cloud aggregation - parameter distribution". After 5 to 10 iterations, the global model prediction accuracy stabilizes (grind wheel wear prediction error < 0.05 mm).

[0083] Core functions: Solve the problem of data silos among multiple devices, realize cross-device data sharing training, enable the LSTM model 202a to learn more comprehensive grinding condition features, and improve prediction accuracy by more than 40% compared with single-device training models; avoid uploading raw data and protect enterprise data privacy; quickly adapt to the processing conditions of different devices through global model, and do not need to retrain the model when changing models, only need to fine-tune parameters, reducing the changeover time from the traditional 15 minutes to 88 seconds.

[0084] Reference Figure 1-6 An adaptive grinding method for the groove of a linear guide slider includes the following steps:

[0085] S1: Place the linear guide slider workpiece on the adaptive grinding fixture.

[0086] The system is started, and the piezoelectric ceramic array 301a and the disc spring group form a composite support system. The laser interferometry module 101 and the microchannel cooling module 302 perform self-testing and initialization.

[0087] S2: The laser interferometry module 101 monitors and collects the micro-displacement of the slider workpiece during the grinding process, and the acoustic emission sensing module 102 synchronously collects the acoustic emission signals generated during the grinding process.

[0088] S3: Input the real-time acquired micro-displacement and acoustic emission signal data into the digital twin simulation module 201. The model performs grinding force-thermal coupling effect simulation to predict the current grinding state and possible dimensional deviations.

[0089] S4: A federated learning-based LSTM wear prediction model analyzes features such as the spectral entropy of acoustic emission signals to predict the wear state of the grinding wheel. The optimal grinding parameters are dynamically calculated by combining the prediction results from the digital twin.

[0090] S5: Drives the piezoelectric ceramic array 301a to generate high-frequency micro-vibrations (5-20kHz, 0.1-0.5μm), actively suppresses vibration ripples, controls the flow rate of the microchannel cooling module 302 (containing 5wt% Al2O3 nanoparticles), precisely controls the temperature, and compensates for thermal deformation.

[0091] S6: Based on the received optimized parameter instructions, the grinding wheel speed (±100rpm) and feed rate (0.05~0.2mm / min) are dynamically adjusted in real time. With the cooperation of the piezoelectric ceramic array 301a and the disc spring group forming a composite support system and the microchannel cooling module 302, the precision grinding of the slider workpiece groove is completed, and the operator removes the processed slider workpiece.

[0092] S7: Upload the processing data (after anonymization) to the federated learning server, aggregate it with data from other devices in the cluster, and update and optimize the LSTM wear prediction model.

[0093] By integrating high-precision real-time monitoring of micro-displacement and acoustic emission, intelligent decision-making based on digital twin and wear prediction models, and active vibration and thermal compensation execution mechanisms, dynamic perception, accurate prediction and adaptive control of the grinding process are achieved, thereby significantly improving the final machining accuracy and surface quality of the linear guide slider groove, as well as the intelligence level and long-term stability of the entire machining system.

[0094] Furthermore, the micro-displacement of the slider workpiece during the grinding process is monitored and acquired. Specifically, this is achieved by emitting four differential beams through a laser interferometry module 101 with a wavelength of 632.8 nm. This allows for real-time monitoring of the micro-displacement of the slider workpiece in the X, Y, and Z directions, with a measurement accuracy of 0.05 μm and a sampling rate of 10 kHz. The laser interferometry module 101 is based on the Michelson interference principle. The 632.8 nm helium-neon laser is split into two beams by a beam splitter. One beam is directed towards the slider surface (measurement beam), and the other beam is directed towards a fixed reference mirror (reference beam). The two beams meet after reflection, generating interference fringes. The interference signals are received through the four differential beams, and combined with the phase difference calculation, the displacement of the slider in the X (axial), Y (radial), and Z (vertical) directions can be obtained respectively.

[0095] 632.8 nm is a typical output wavelength for helium-neon lasers. Lasers at this wavelength are characterized by high stability (wavelength drift < 0.001 nm / ℃) and good coherence (coherence length > 20 cm), making them suitable for high-precision micro-displacement measurements. If the wavelength is greater than 632.8 nm (such as 1064 nm infrared lasers), the interference fringe spacing will increase, and the measurement resolution will decrease to above 0.1 μm, failing to meet the measurement accuracy requirement of 0.05 μm. If the wavelength is less than 632.8 nm (such as 488 nm blue lasers), it is easily affected by scattering from airborne dust, resulting in decreased measurement signal stability. In workshop environments, the measurement error may exceed 0.1 μm.

[0096] Furthermore, the real-time collected data is input into the digital twin simulation module 201 for simulation. This involves simulating the grinding force-thermal coupling effect to predict the current grinding state, dimensional deviations, and thermal deformation. The digital twin simulation module 201 constructs a virtual model that is completely consistent with the actual grinding machine, slide workpiece, and grinding wheel using 3D modeling software (such as UG or SolidWorks). It also integrates multi-physics simulation algorithms based on materials mechanics, thermodynamics, and tribology. During the simulation, actual grinding parameters (such as grinding wheel speed, feed rate, and grinding depth) are input. The model calculates the stress distribution (grinding force) and temperature field distribution (grinding heat) in the contact area between the grinding wheel and the slide workpiece based on the material's physical properties (such as the elastic modulus and thermal expansion coefficient of GCr15 steel). It then simulates the coupling effect between the two: the grinding force causes slight deformation of the slide workpiece, affecting the grinding depth; the grinding heat causes thermal expansion of the slide workpiece, further altering the relative position of the slide workpiece and the grinding wheel, ultimately affecting the dimensional accuracy of the groove.

[0097] Specific functions: to predict in advance the impact of force-thermal coupling on machining accuracy under different grinding parameters, avoid trial and error costs in actual machining, provide a theoretical basis for subsequent dynamic adjustment of grinding parameters, and ensure that the adjusted parameters can accurately compensate for dimensional deviations.

[0098] Furthermore, the acoustic emission signal is analyzed to predict the wear state of the grinding wheel by extracting the spectral entropy value feature of the acoustic emission signal, and then input into the LSTM model 202a trained based on the federated learning framework for analysis and prediction.

[0099] Furthermore, the calculated optimal grinding parameters include the grinding wheel speed and feed rate, wherein the grinding wheel speed can be adjusted within ±100 rpm, and the feed rate can be adjusted within 0.05 to 0.2 mm / min.

[0100] Furthermore, the piezoelectric ceramic array 301a generates high-frequency micro-vibrations to drive the annular piezoelectric ceramic array 301a integrated in the center of the disc spring assembly, generating high-frequency micro-vibrations with a frequency of 5-20kHz and an amplitude of 0.1-0.5μm, in order to actively suppress grinding marks. The diameter of the annular piezoelectric ceramic array 301a is 8±0.1mm (see Table 2 for specific experimental data).

[0101] The ring piezoelectric ceramic array 301a and the disc spring assembly form a composite support system. Its output of 5-20kHz high-frequency micro-amplitude vibration (amplitude 0.1-0.5μm) can form reverse interference with the harmful vibration generated during the grinding process, canceling the chatter when the grinding wheel contacts the slider workpiece, thereby suppressing the groove surface texture. At the same time, it can achieve dynamic fine adjustment of the slider workpiece support position through small amplitude adjustment, improving the positioning accuracy.

[0102] The rationale behind the output frequency of 5-20kHz and amplitude of 0.1-0.5μm is as follows: During grinding, the harmful vibration frequencies generated by the high-speed rotation of the grinding wheel (typically 3000-6000rpm) and its contact with the workpiece in the slider are concentrated in the 8-18kHz range. An output frequency of 5-20kHz can fully cover this harmful vibration frequency band, achieving reverse vibration damping. The amplitude of 0.1-0.5μm is a key parameter for matching the grinding accuracy (tolerance ≤ ±1.5μm) of the slider groove. If the amplitude is too small (<0.1μm), the vibration damping effect is insufficient, and the residual vibration marks on the groove surface exceed 40%. If the amplitude is too large (>0.5μm), it will cause the small displacement of the slider workpiece to exceed the tolerance, affecting the dimensional accuracy. Through 15 sets of orthogonal tests with different frequencies and amplitudes, when the parameters are within this range, the surface roughness Ra of the groove can be stably controlled below 0.1μm, and the dimensional tolerance pass rate is improved to 98%.

[0103] Furthermore, the temperature of the microfluidic cooling system is controlled by controlling the flow rate of the coolant in the tree-like fractal microfluidic channels embedded in the polymer damping pad; the coolant contains 5wt% Al2O3 nanoparticles and the channel width is 200±10μm (experimental data refer to Table 1) to achieve precise temperature control and thermal deformation compensation.

[0104] Working principle: The polymer damping pad itself has high damping characteristics, which can absorb part of the grinding vibration. The embedded tree-shaped fractal microchannels can make the coolant evenly distributed inside the damping pad, and fully exchange heat with the slider workpiece and the support system. At the same time, the Al2O3 nanoparticles in the coolant can enhance heat transfer (the nanoparticles have a large specific surface area and a thermal conductivity that is 2.5 times that of traditional coolants), and quickly remove the heat generated by grinding.

[0105] Specific function: Through the dual effects of "damping and vibration reduction + high-efficiency heat dissipation", the thermal deformation error during the grinding process is reduced to 0.1μm, while further suppressing vibration and improving the machining accuracy and surface quality of the groove.

[0106] Reasonableness of channel width of 200±10μm: The width of the tree-like fractal microchannel needs to balance the coolant flow rate and heat dissipation efficiency. If the width is greater than 210μm, it will lead to excessive spacing between the internal channels of the damping pad, uneven coolant coverage, insufficient local heat dissipation, and thermal deformation error exceeding 0.2μm. If the width is less than 190μm, the coolant flow rate is too small (less than 0.5L / min), the heat dissipation efficiency decreases, and channel blockage is likely to occur. After testing 12 groups of channels with different widths, when the width is 200±10μm, the coolant flow rate is stable at 0.8~1.2L / min, and the thermal deformation error can be controlled within 0.1μm.

[0107] The rationale for including 5 wt% Al2O3 nanoparticles in the coolant: 5 wt% is the optimal ratio considering both heat transfer efficiency and coolant flowability. When the addition of Al2O3 nanoparticles is less than 5 wt%, the heat transfer coefficient improvement is insufficient (only less than 1.8 times that of traditional coolants), and the thermal deformation control effect is poor; when the addition is greater than 5 wt%, the coolant viscosity will increase significantly (exceeding 50 mPa·s), leading to increased flow resistance in the microchannels and even precipitation and blockage of the channels. Through 8 sets of comparative tests with different concentrations, the 5 wt% concentration coolant achieved a heat transfer coefficient of 600 W / (m·K), and flowed smoothly in the microchannels, with no blockage after 8 hours of continuous operation.

[0108] Furthermore, the step of uploading the processed data to the federated learning server involves each device uploading its local LSTM model 202a parameters or gradients after desensitization processing, updating and optimizing the LSTM wear prediction model. Specifically, the cloud aggregation server uses a federated averaging algorithm to aggregate the parameters or gradients from multiple devices in the cluster to generate an optimized global model, and then distributes the global model parameters to each device for subsequent prediction.

[0109] Furthermore, the process includes the following steps: When the dimensional deviation predicted by the digital twin simulation module 201 is greater than 0.3 μm, the step of calculating the optimal grinding parameters and compensation instructions is triggered. When the laser interferometry module 101 detects that the deviation between the actual size of the slider workpiece and the target size is greater than 0.3 μm, the digital twin simulation module 201 will calculate the grinding wheel speed and feed rate that need to be adjusted based on the direction and magnitude of the deviation, combined with the grinding wheel wear state predicted by the LSTM model 202a. For example, if the size is too small (due to insufficient grinding), the model will appropriately increase the grinding wheel speed (increase cutting efficiency) or increase the feed rate (increase the depth of grinding per pass). If the size is too large (due to excessive grinding or thermal expansion of the slider workpiece), the speed will be reduced or the feed rate will be decreased. At the same time, the microfluidic cooling system will be used to enhance heat dissipation and suppress thermal deformation.

[0110] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An adaptive grinding system for linear guide slider grooves, characterized in that: include, The monitoring unit (100) is used to collect micro-displacement data and acoustic emission data of the slider workpiece during the grinding process; A decision unit (200), which is connected to the monitoring unit (100), is used to calculate, based on the received micro-displacement data and acoustic emission data, through a digital twin simulation module (201) and a wear prediction and optimization module (202), and output optimized grinding parameters and compensation instructions; An execution unit (300) is connected to the decision unit (200). The execution unit (300) is used to drive the grinding mechanism, the compensation mechanism and the cooling mechanism to complete the adaptive grinding of the slider workpiece according to the received grinding parameters and compensation instructions. The digital twin simulation module (201) is used to receive micro-displacement data and acoustic emission data, perform grinding force-thermal coupling effect simulation, and output the predicted dimensional deviation and thermal deformation. The wear prediction and optimization module (202) is connected to the digital twin simulation module (201). The wear prediction and optimization module (202) integrates an LSTM model (202a) trained based on a federated learning framework. The LSTM model (202a) analyzes the spectral entropy characteristics of the acoustic emission signal to predict the wear state of the grinding wheel, and dynamically calculates the optimal grinding parameters by combining the output of the digital twin simulation module (201). The execution unit (300) includes: A piezoelectric ceramic-disc spring composite support module (301) includes an annular piezoelectric ceramic array (301a) integrated at the center of a disc spring assembly. The piezoelectric ceramic array (301a) receives compensation commands and generates high-frequency micro-vibrations with a frequency of 5-20kHz and an amplitude of 0.1-0.5μm. The microfluidic cooling module (302) is embedded in a polymer damping pad. The microfluidic cooling module (302) has a tree-like fractal structure with a channel width of 200±10μm and is filled with a coolant containing 5wt% Al2O3 nanoparticles.

2. The adaptive grinding system for linear guide slider grooves as described in claim 1, characterized in that: The monitoring unit (100) includes: The laser interferometry module (101) uses a helium-neon laser source with a wavelength of 632.8nm and emits four differential beams to monitor the micro-displacement of the slider workpiece in the X, Y and Z directions. The acoustic emission sensing module (102) synchronously collects acoustic emission data generated during the grinding process and emits acoustic emission signals.

3. The adaptive grinding system for linear guide slider grooves as described in claim 2, characterized in that: The LSTM model (202a) includes: The input layer (202a-1) receives multi-dimensional monitoring data; Hidden layer (202a-2) contains 128 LSTM units to capture time series features; Output layer (202a-3) outputs the predicted value of grinding wheel wear.

4. The linear guide slider groove adaptive grinding system as described in claim 3, characterized in that: The diameter of the ring piezoelectric ceramic array (301a) is 8±0.1mm.

5. The adaptive grinding system for linear guide slider grooves as described in claim 4, characterized in that: The decision-making unit (200) is further configured as follows: When the dimensional deviation predicted by the digital twin simulation module (201) is greater than 0.3 μm, the grinding wheel speed is dynamically adjusted within a range of ±100 rpm, and / or the feed rate is dynamically adjusted within a range of 0.05 to 0.2 mm / min.

6. An adaptive grinding method for linear guide slider grooves, applied to the adaptive grinding system for linear guide slider grooves as described in claim 5, characterized in that: Place the linear guide slider workpiece on the adaptive grinding fixture; The monitoring unit (100) monitors and collects the micro-displacement of the slide block workpiece during the grinding process and simultaneously collects the acoustic emission signals generated during the grinding process; The real-time collected micro-displacement and acoustic emission signal data are input into the digital twin simulation module (201) to predict the current grinding state and possible dimensional deviations. By analyzing the spectral entropy and other characteristics of acoustic emission signals, the wear state of the grinding wheel is predicted, and the optimal grinding parameters are calculated. The piezoelectric ceramic array (301a) is driven to generate high-frequency micro-vibrations to suppress vibration marks, and the microchannel cooling system is controlled to control the temperature and compensate for thermal deformation. The grinding wheel speed and feed rate are dynamically adjusted according to the optimal grinding parameters to complete the precision grinding of the groove of the slider workpiece. The processing data is uploaded to the federated learning server, aggregated with data from other devices in the cluster, and the LSTM wear prediction model is updated and optimized. The step of inputting the real-time collected data into the digital twin simulation module (201) for simulation is to perform grinding force-thermal coupling effect simulation in order to predict the current grinding state, dimensional deviation and thermal deformation. The analysis of acoustic emission signals to predict the wear state of grinding wheels specifically involves extracting the spectral entropy features of the acoustic emission signals and inputting them into an LSTM model (202a) trained based on a federated learning framework for analysis and prediction. The step of uploading the processed data to the federated learning server is that each device uploads the local LSTM model (202a) parameters or gradients after the data has been de-identified and encrypted. The update and optimization of the LSTM wear prediction model specifically involves the cloud aggregation server using a federated averaging algorithm to aggregate parameters or gradients from multiple devices within the cluster to generate an optimized global model, and then distributing the global model parameters to each device for subsequent prediction. It also includes the following steps: When the dimensional deviation predicted by the digital twin simulation module (201) is greater than 0.3 μm, the step of calculating the optimal grinding parameters and compensation instructions is triggered.

7. The adaptive grinding method for linear guide slider grooves as described in claim 6, characterized in that: The monitoring and acquisition of the micro-displacement of the slider workpiece during the grinding process is specifically achieved by emitting four differential beams through a laser interferometry module (101) with a wavelength of 632.8nm, so as to monitor the micro-displacement of the slider workpiece in the X, Y and Z directions in real time.

8. The adaptive grinding method for linear guide slider grooves as described in claim 7, characterized in that: The calculated optimal grinding parameters include grinding wheel speed and feed rate; The grinding wheel speed can be adjusted within ±100 rpm, and the feed rate can be adjusted within 0.05~0.2 mm / min.

9. The adaptive grinding method for linear guide slider grooves as described in claim 8, characterized in that: The piezoelectric ceramic array (301a) generates high-frequency micro-vibrations to drive the annular piezoelectric ceramic array (301a) integrated in the center of the disc spring group, generating high-frequency micro-vibrations with a frequency of 5-20kHz and an amplitude of 0.1-0.5μm. The diameter of the annular piezoelectric ceramic array (301a) is 8±0.1mm.

10. The adaptive grinding method for linear guide slider channels as described in claim 9, characterized in that: The temperature control of the microfluidic cooling system is specifically controlled by controlling the flow rate of the coolant in the tree-like fractal microfluidic channels embedded in the polymer damping pad. The coolant contains 5 wt% Al2O3 nanoparticles and has a channel width of 200 ± 10 μm.

Images