A method and system for optimizing grinding parameters of intelligent cutting tools

Abstract

This invention discloses a method and system for optimizing grinding parameters of an intelligent cutting tool. The invention relates to the technical field of parameter optimization. Based on the dynamic identification of key regions, multiple current parameters in different dimensions are determined. A dynamic parameter model of the intelligent cutting tool is constructed using its current 3D model. This dynamic parameter model is then optimized against multiple objectives. Multiple optimization objectives are determined based on the application scenario of the intelligent cutting tool. Each optimization objective is input into its corresponding optimization space. A pre-defined neural network is used to establish a nonlinear relationship between the grinding parameters and machining quality of the intelligent cutting tool. Furthermore, the optimized grinding parameters are determined based on the grinding constraints of the intelligent cutting tool. The optimized grinding parameters, combined with the dynamic parameter model of the intelligent cutting tool, trigger a corresponding dynamic simulation. During the dynamic simulation, multiple simulation results are output, improving the accuracy of the final simulation results.

Description

Technical Field

[0001] This invention relates to the technical field of parameter optimization, and in particular to a method and system for optimizing grinding parameters of an intelligent cutting tool. Background Technology

[0002] Grinding CNC inserts is an extreme manufacturing process involving multi-physics coupling, high precision requirements, and complex micro-morphology shaping. Intelligent inserts are a type of CNC insert. With the increasing demands for cutting performance in high-end manufacturing fields such as aerospace and new energy vehicles, the control of the grinding process of intelligent inserts has shifted from simply dimensional accuracy to surface integrity and physical properties.

[0003] Existing grinding monitoring systems mostly use single-dimensional sensing, such as monitoring only spindle power or grinding force in a single direction. Even if some technologies introduce multiple sensors, they are merely simple parallel displays of data. Due to the lack of a joint registration mechanism between high-frequency dynamic sensing and machine vision in a five-dimensional spatiotemporal coordinate system, geometric parameters and working parameters such as force and heat are severely misaligned in spatial position and timestamp. This fragmentation of heterogeneous data makes it impossible for the system to discover the coupling mutation points between parameters to form key parameter combinations with causal relationships, affecting the accuracy of multiple optimization targets. This leads to the simulation process of intelligent cutting tools during optimization, resulting in lower accuracy of the final simulation results. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art. This invention provides a method and system for optimizing the grinding parameters of an intelligent cutting tool.

[0005] This invention provides a method for optimizing grinding parameters of an intelligent cutting tool, comprising: For CNC insert grinding centers, multiple morphological parameters of intelligent inserts are collected and combined with the working parameters of the intelligent inserts during the grinding process for spatiotemporal alignment. During the alignment process, corresponding key parameter combinations are formed, and multiple key areas of the intelligent inserts are determined by multi-level iteration along the key parameter combinations. Based on the dynamic identification of each key area, multiple current parameters in different dimensions are determined, and a dynamic parameter model of the intelligent blade is constructed by combining the current three-dimensional model of the intelligent blade. The dynamic parameter model is then optimized in multiple objectives, and multiple optimization objectives are determined in combination with the application scenario of the intelligent blade. Each target to be optimized is input into the corresponding optimization space, and multi-level training is performed in combination with the preset neural network. During the training process, the nonlinear relationship between the grinding parameters and the machining quality of the intelligent cutting tool is determined. The optimized grinding parameters are further determined by combining the grinding constraint relationship of the intelligent cutting tool. The optimized grinding parameters are loaded into the simulation space of the intelligent cutting tool, and the corresponding dynamic simulation is triggered in combination with the dynamic parameter model of the intelligent cutting tool. During the dynamic simulation, multiple simulation results are output, and the final simulation result is determined by combining the adaptive optimization mechanism. The grinding parameters corresponding to the final simulation result are marked.

[0006] This invention provides a grinding parameter optimization system for intelligent cutting tools, which is applied to the aforementioned grinding parameter optimization method for intelligent cutting tools; the grinding parameter optimization system for intelligent cutting tools includes: The key area module is used to collect multiple morphological parameters of the intelligent cutting tool for the CNC cutting tool grinding center, and perform spatiotemporal alignment with the working parameters of the intelligent cutting tool during the grinding process. During the alignment process, a corresponding key parameter combination is formed, and multiple key areas of the intelligent cutting tool are determined by multi-level iteration along the key parameter combination. The optimization target module is used to determine multiple current parameters in different dimensions based on the dynamic identification of each key area, and to construct the dynamic parameter model of the intelligent blade by combining the current three-dimensional model of the intelligent blade. The dynamic parameter model is then optimized in multiple objectives, and multiple optimization targets are determined in combination with the application scenario of the intelligent blade. The optimization module is used to input each target to be optimized into the corresponding optimization space, and perform multi-level training in combination with the preset neural network. During the training process, the nonlinear relationship between the grinding parameters and the processing quality of the intelligent cutting tool is determined, and the optimized grinding parameters are further determined by combining the grinding constraint relationship of the intelligent cutting tool. The simulation module is used to load the optimized grinding parameters into the simulation space of the intelligent cutting tool, and trigger the corresponding dynamic simulation in combination with the dynamic parameter model of the intelligent cutting tool. During the dynamic simulation, multiple simulation results are output, and the final simulation result is determined by combining the adaptive optimization mechanism, and the grinding parameters corresponding to the final simulation result are marked.

[0007] Compared with the prior art, the beneficial effects of the present invention are: (1) For CNC insert grinding center, multiple morphological parameters of intelligent insert are collected and combined with the working parameters of intelligent insert in the grinding process for spatiotemporal alignment. During the alignment process, a corresponding key parameter combination is formed. Multiple key areas of intelligent insert are determined by multi-level iteration along the key parameter combination. Multiple current parameters in different dimensions are determined based on the dynamic identification of each key area. The dynamic parameter model of intelligent insert is constructed by combining the current three-dimensional model of intelligent insert. The dynamic parameter model is optimized by multiple objectives. Multiple optimization objectives are determined by combining the application scenario of intelligent insert. The current three-dimensional model and dynamic parameter model of intelligent insert are introduced to further control the application scenario of intelligent insert and improve the accuracy of multiple optimization objectives.

[0008] (2) Input each target to be optimized into the corresponding optimization space, and perform multi-level training in combination with the preset neural network. During the training process, determine the nonlinear relationship between the grinding parameters and the processing quality of the intelligent cutting tool. Further combine the grinding constraint relationship of the intelligent cutting tool to determine the optimized grinding parameters. Load the optimized grinding parameters into the simulation space of the intelligent cutting tool, and trigger the corresponding dynamic simulation in combination with the dynamic parameter model of the intelligent cutting tool. Output multiple simulation results during the dynamic simulation process, and determine the final simulation result in combination with the adaptive optimization mechanism. Mark the grinding parameters corresponding to the final simulation result. The optimized grinding parameters are introduced, and the dynamic parameter model of the intelligent cutting tool and the simulation space of the intelligent cutting tool are fully considered, which improves the accuracy of the final simulation result. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the method for optimizing grinding parameters of an intelligent cutting tool according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating step S11 in the method for optimizing grinding parameters of an intelligent cutting tool according to an embodiment of the present invention. Figure 3 This is a flowchart illustrating step S12 in the method for optimizing grinding parameters of an intelligent cutting tool according to an embodiment of the present invention. Figure 4 This is a flowchart illustrating step S13 in the method for optimizing grinding parameters of an intelligent cutting tool according to an embodiment of the present invention. Figure 5 This is a flowchart illustrating step S14 in the method for optimizing grinding parameters of an intelligent cutting tool according to an embodiment of the present invention. Figure 6 This is a schematic diagram of the structure of the intelligent cutting tool grinding parameter optimization system in an embodiment of the present invention. Detailed Implementation

[0010] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0011] Please see Figures 1 to 6 A method for optimizing grinding parameters of intelligent cutting tools, applied to parameter optimization scenarios; the method for optimizing grinding parameters of intelligent cutting tools includes: Step S11: For the CNC insert grinding center, collect multiple morphological parameters of the intelligent insert and perform spatiotemporal alignment with the working parameters of the intelligent insert during the grinding process. During the alignment process, a corresponding key parameter combination is formed, and multiple key areas of the intelligent insert are determined by multi-level iteration along the key parameter combination. Step S12: Based on the dynamic identification of each key area, determine multiple current parameters in different dimensions, and construct the dynamic parameter model of the intelligent blade by combining the current three-dimensional model of the intelligent blade. Perform multi-objective optimization on the dynamic parameter model, and determine multiple optimization objectives based on the application scenario of the intelligent blade. Step S13: Input each target to be optimized into the corresponding optimization space, and perform multi-level training in combination with the preset neural network. During the training process, determine the nonlinear relationship between the grinding parameters and processing quality of the intelligent cutting tool, and further determine the optimized grinding parameters in combination with the grinding constraint relationship of the intelligent cutting tool. Step S14: Load the optimized grinding parameters into the simulation space of the intelligent cutting tool, and trigger the corresponding dynamic simulation in combination with the dynamic parameter model of the intelligent cutting tool. During the dynamic simulation, output multiple simulation results, and determine the final simulation result in combination with the adaptive optimization mechanism, and mark the grinding parameters corresponding to the final simulation result.

[0012] refer to Figure 2 In step S11, the specific steps are as follows: S111: Real-time monitoring of the CNC insert grinding center's work process and marking the position of the intelligent insert. By deploying a high-frequency dynamic sensor and machine vision fusion system, data is collected along the position of the intelligent insert, and multiple morphological parameters of the intelligent insert are obtained. These multiple morphological parameters cover the geometric morphological parameters, surface morphology parameters, and grinding wheel state parameters of the intelligent insert. S112: Obtain the working parameters of the intelligent cutting tool during the grinding process, introduce a spatiotemporal joint registration mechanism, dynamically align multiple morphological parameters with the working parameters of the intelligent cutting tool during the grinding process in a five-dimensional spatiotemporal coordinate system, and automatically cluster to generate key parameter combinations with causal relationships by mining coupling mutation points between parameters during the alignment process. The working parameters of the intelligent cutting tool during the grinding process cover the instantaneous power of the spindle, the dynamic component of the grinding force, and the coolant flow field distribution. S113: The combination of key parameters is dynamically evolved, and multi-level iterations are performed along the evolution trajectory of the combination of key parameters. In this process, multiple key regions of the intelligent cutting tool under the action of grinding thermo-mechanical coupling are determined. These multiple key regions cover the stress concentration zone, the crack sensitive zone, and the cutting edge passivation zone.

[0013] In the embodiments of this application, the working process of the CNC insert grinding center is monitored in real time, and the position of the intelligent insert is marked. By deploying a high-frequency dynamic sensor and machine vision fusion system, data is collected along the position of the intelligent insert, and multiple morphological parameters of the intelligent insert are obtained. These multiple morphological parameters cover the geometric morphological parameters, surface morphology parameters, and grinding wheel state parameters of the intelligent insert.

[0014] At this point, the system's bottom layer first takes over the underlying bus protocol of the CNC insert grinding center, reads and parses the machine tool's feed axis coordinates and spindle status information in real time, thereby realizing high-frequency cyclic monitoring of the entire grinding process. In this monitoring closed loop, the system reads the machine tool's zero point and the offset matrix of the insert fixture, and accurately marks the three-dimensional spatial position of the intelligent insert to be processed in the machine tool's absolute coordinate system. Using this spatial position as a static reference, combined with the interpolation command trajectory issued by the CNC system, the system calculates the dynamic tracking coordinates of the theoretical contact area between the grinding wheel and the insert in real time, thereby establishing a unified spatial mapping anchor point for subsequent multi-source sensor data streams.

[0015] At the physical hardware level, a high-frequency dynamic sensor array consisting of acoustic emission sensors and a three-dimensional force gauge is rigidly deployed near the grinding spindle, and a high-resolution industrial camera and a structured light projection module are deployed at a specific downward angle around the grinding wheel to jointly form a multimodal fusion sensing system. In terms of the data acquisition triggering mechanism, the system does not adopt indiscriminate all-time blind sampling, but sets a spatial trigger threshold based on the dynamic tracking coordinates generated in the first stage. When the machine tool interpolation trajectory runs into the preset spatial envelope of the area to be measured by the cutting tool, the system synchronously triggers the data sampling card of the high-frequency dynamic sensor and the exposure acquisition of the machine vision system with a hardware-level interrupt signal to ensure that the acquired multi-source data maintains strict spatial consistency at the microsecond-level timestamp.

[0016] After acquiring the raw sensor signals, the system immediately initiates the parallel processing channel for heterogeneous data. For high-frequency dynamic sensor signals, wavelet packet decomposition and envelope spectrum analysis are used to filter out low-frequency mechanical vibration noise from the machine tool and extract high-frequency features reflecting the transient response of the material removal. For machine vision images, the three-dimensional morphology of the measured area is reconstructed through stereo matching and point cloud registration. The system calculates and outputs three core morphological parameter sets from the above fused data stream: the first is the geometric morphological parameters that present the macroscopic contour and groove structure of the blade; the second is the surface morphology parameters that characterize the micro-serration of the cutting edge, surface micro-cracks, and roughness distribution; and the third is the grinding wheel state parameters indirectly derived by analyzing the high-frequency vibration frequency shift and the porosity of the grinding wheel surface in the visual image.

[0017] Specifically, when the CNC system controls the diamond grinding wheel to begin grinding the intelligent insert, the system not only monitors the wheel's path but also precisely marks the complex three-dimensional spatial position at the intersection of the "chip breaker groove" and the "straight main cutting edge" on the rake face of the intelligent insert. It also calculates in real time the dynamic spatial coordinates of the grinding wheel's R-circle as it approaches this intersection. Simultaneously, when the grinding wheel enters the aforementioned intersection area, a spatial trigger threshold is activated. At this time, an acoustic emission sensor fixed to the grinding head captures the elastic stress wave released when the grinding wheel abrasive grains cut the cemented carbide matrix of the intelligent insert at a sampling rate of 1MHz. Simultaneously, a synchronously triggered line scan vision system uses structured light of a specific wavelength to penetrate the grinding fluid mist and instantly capture the microscopic contour image of the intersection area before it is obscured by the grinding wheel, achieving a perfect fusion of dynamic force signals and static visual images at the key node of the chip breaker groove.

[0018] The system performs specific parameter calculations for intelligent cutting inserts: First, it extracts geometric parameters such as the width and depth of the chip breaker groove and the radius of the transition arc between it and the rake face from the visual point cloud; Second, it extracts surface morphology parameters such as the micro-chipping size and surface waviness of the main cutting edge of the intelligent cutting insert at the root of the chip breaker groove due to the interference of the grinding trajectory, by combining the high-frequency components of the force signal.

[0019] The system compares the energy attenuation ratio of acoustic emission signals when the machine is unloaded and when it is cutting into the intelligent insert. Combined with the change in reflectivity of the grinding wheel working surface captured by the vision system, it accurately deduce the flattening wear area and micro-clogging degree of the diamond grinding wheel at the top of the abrasive grains after grinding the complex groove of the intelligent insert. This outputs crucial grinding wheel state parameters, providing the most direct data support for subsequent judgment on whether dressing parameters need to be adjusted or the feed rate reduced.

[0020] Furthermore, the working parameters of the intelligent cutting tool during the grinding process are obtained, and a spatiotemporal joint registration mechanism is introduced to dynamically align multiple morphological parameters with the working parameters of the intelligent cutting tool during the grinding process in a five-dimensional spatiotemporal coordinate system. During the alignment process, coupling mutation points between parameters are discovered, and key parameter combinations with causal relationships are automatically clustered to generate them. The working parameters of the intelligent cutting tool during the grinding process include the instantaneous power of the spindle, the dynamic component of the grinding force, and the coolant flow field distribution.

[0021] At this time, the system comprehensively acquires the multi-dimensional working parameters of the intelligent cutting tool during the grinding process through the high-speed data bus inside the machine tool and the external high-precision sensor network. Specifically, the system captures the transient values ​​of the current and voltage of the servo motor driving the spindle at a microsecond sampling rate, and calculates the instantaneous power of the spindle reflecting the energy consumption of material removal through the power calibration model. Simultaneously, it acquires the three-dimensional high-frequency signal output by the force gauge, and separates the dynamic component of the grinding force after removing the self-excited vibration of the machine tool through Kalman filtering. At the same time, it integrates the pre-built cooling nozzle model based on fluid mechanics, and combines the flow meter and pressure sensor installed at the inlet of the grinding zone to reconstruct and acquire the coolant flow field distribution characteristics covering the contact arc area between the cutting tool and the grinding wheel in real time. The coolant flow field distribution characteristics include fluid dynamic pressure, flow velocity vector and effective wetting area ratio.

[0022] To address the significant differences in sampling frequency and physical dimension between heterogeneous data, the system constructs a five-dimensional spatiotemporal coordinate system comprising three-dimensional spatial coordinates (x, y, z), a time dimension t, and a processing feature dimension ϕ. Within this framework, the system introduces a spatiotemporal joint registration mechanism: on the time axis, an interpolation method based on dynamic time warping is used to align the low-frequency flow field distribution characteristics with the high-frequency power and force dynamic components using timestamps; on the spatial and feature axes, the instantaneous spatial position of the contact line between the grinding wheel and the cutting tool is used as a reference to map the working parameters to the corresponding processing feature dimension ϕ, such as the current stage of rough grinding, fine grinding, or rounding. Through this multi-scale joint registration, the absolute binding and dynamic alignment of each set of morphological parameters with the physical field working parameters at the time of occurrence are achieved in the five-dimensional spatial grid.

[0023] Based on the five-dimensional aligned data stream, the system initiates a sliding window mechanism to calculate the mutual information matrix and partial correlation coefficients between parameters in each dimension in real time. When the correlation between parameters deviates from the steady-state baseline and exceeds the preset mutation threshold, the system immediately marks the "coupling mutation point". To eliminate spurious correlation interference, the system further introduces a causal inference method based on physical constraints to analyze the sequential logic and energy transfer direction of mutation points in the time series. Using causal strength as the weight, density clustering is used to automatically cluster these verified mutation points in the five-dimensional space, thereby generating a series of key parameter combinations with clear physical causal logic.

[0024] Specifically, when the grinding wheel is grinding the chip breaker groove with drastic curvature changes on the rake face of the intelligent insert, the system keenly detects: high-frequency, tiny sawtooth-like fluctuations in the instantaneous power of the spindle; the force gauge detects that high-frequency harmonics are suddenly superimposed on the dynamic component of the normal grinding force; at the same time, due to the geometric blocking effect of the chip breaker groove pit, the system monitors that the coolant flow field distribution changes drastically—the originally gentle laminar flow turns into turbulent flow at the bottom of the groove, and the effective wetting area shrinks sharply, forming a local "dry friction" micro-region.

[0025] The system immediately pulls the above-mentioned sudden situation into a five-dimensional spatiotemporal coordinate system; at this time, the spatial coordinates (x,y,z) are determined to be the center of the deepest depression in the chip breaker groove of the intelligent cutting tool, the time t is determined to be in the millisecond interval of the grinding wheel cutting into the bottom of the depression, and the feature dimension ϕ is marked as "the forming stage of the bottom of the chip breaker groove". Through spatiotemporal joint registration, the system forcibly compresses and maps the macroscopic low-frequency image of the flow field contraction, the mid-frequency signal of the power fluctuation, and the high-frequency signal of the grinding force harmonics onto the same five-dimensional feature node, realizing the accurate presentation of "bottom position of the groove, instantaneous time, flow field anomaly, and force energy anomaly".

[0026] The system performs causal analysis within the five-dimensional nodes: it discovers that the change is not initially caused by the morphological parameters of the grinding wheel wear, but rather by the obstruction of the coolant flow field distribution due to the deep cavity structure of the chip breaker groove, which leads to local heat dissipation failure. This directly causes the carbide material of the intelligent insert to undergo plastic yielding at high temperatures, manifested as high-frequency oscillations in the dynamic components of the grinding force, ultimately dissipating more spindle power and potentially causing microcracks on the groove bottom surface. After capturing this causal chain, the system generates a specific combination of key parameters in the five-dimensional space through clustering: [deep cavity location of the chip breaker groove + lack of coolant turbulence + high-frequency surge of normal force + risk morphology of microcracks at the groove bottom]. This causal combination will be directly fed into subsequent steps to specifically optimize the grinding feed rate or adjust the coolant spray angle at this specific groove location.

[0027] Therefore, the combination of key parameters is dynamically evolved, and multiple iterations are performed along the evolution trajectory of the combination of key parameters. In the process of iteration, multiple key regions of the intelligent cutting tool under the action of grinding thermo-mechanical coupling are determined. These multiple key regions cover the stress concentration area, the crack sensitive area, and the cutting edge blunting area. This takes into account the overall evolution trajectory of the combination of key parameters, ensuring the accuracy of the intelligent cutting tool under the action of grinding thermo-mechanical coupling in multiple key regions.

[0028] At this point, after obtaining the key parameter combination with causal relationship, the system abandons the static cross-sectional analysis method and regards the combination as a combination that continuously flows with the processing process, driving it to evolve dynamically in the five-dimensional spatiotemporal coordinate system. The system takes the spatial movement path of the grinding wheel relative to the intelligent cutting tool as the geometric reference and the time series as the driving axis. It performs temporal splicing and state transition deduction on the key parameter combination captured at each discrete moment. By introducing a long short-term memory network, the system not only tracks the translation trajectory of the parameter combination in spatial position, but also deeply presents the attenuation, superposition or nonlinear divergence characteristics of its internal physical parameters (such as the magnitude of force, the accumulation of temperature, and the deformation of the flow field) in the evolution process, thereby generating a high-dimensional dynamic evolution trajectory containing the evolution logic of the complete grinding physical mechanism.

[0029] Following the generated dynamic evolution trajectory, the system initiates a multi-level iterative mechanism driven by physical information. In the initial iteration, the system takes the instantaneous grinding force dynamic components in the evolution trajectory as input, combines them with the constitutive equation of the tool material, and performs point-by-point mapping of the mechanical response on the three-dimensional mesh model of the cutting tool to calculate the first-level residual stress field distribution. In the intermediate iteration, the system converts the instantaneous spindle power into heat flux density input, combines it with the convective heat transfer coefficient boundary conditions of the coolant flow field distribution, solves the transient heat conduction partial differential equation, and obtains the second-level temperature gradient field distribution. In the advanced iteration, the system performs bidirectional strong coupling iteration between the stress field and temperature field of the first two levels, calculates the thermal stress generated due to the limitation of thermal expansion, and vector superimposes it with the mechanical stress, finally outputting a high-fidelity grinding thermo-mechanical coupling field covering the entire evolution trajectory.

[0030] After obtaining the convergent thermo-coupling field, the system extracts the extreme value characteristics of the multi-physics composite gradient for region delineation; for the stress field, the system calculates the extreme value envelope of the partial derivative of the equivalent stress, defining the continuous space where the material reaches the critical state of plastic yielding as the stress concentration region; for the thermo-coupling field, the system extracts the overlapping region of tensile stress and transient extremely high temperature gradient, defining the region where the risk of material phase transformation or microstructure tearing due to thermal shock exceeds the threshold as the crack sensitive region; for the end of the geometric and mechanical evolution trajectory, the system extracts the ratio distribution of the elastic-plastic springback amount to the micromaterial removal rate at the moment the grinding wheel leaves the workpiece, defining the edge contour where obvious material lateral flow or chipping occurs as the cutting edge blunting region, and finally completes the absolute spatial calibration of multiple key regions of the intelligent cutting tool.

[0031] Specifically, following the key parameter combination generated by S112, which is "force surge caused by deep cavity cooling failure of chip breaker", the system begins to track the entire process of the grinding wheel climbing from the bottom of the chip breaker and preparing to transition to the straight main cutting edge of the intelligent insert. The system finds that when the grinding wheel walks out of the depression at the bottom of the chip breaker and contacts the "raised ridge" on the back of the chip breaker, the contact arc length increases sharply and the original coolant turbulence is further disrupted. This key parameter combination is manifested in the evolution trajectory as follows: the spatial coordinates climb from the bottom of the chip breaker to the ridge, and the internal state diverges nonlinearly from "local dry friction heat accumulation" to a deteriorated state of "large area high load compression".

[0032] Following the aforementioned extreme deterioration trajectory, the system unfolds a multi-level iterative process involving thermo-mechanical coupling. The primary mechanical iteration reveals that when the grinding wheel compresses the chip-breaking groove to form a "ridge," the extremely large dynamic component of the normal force causes a plastic compaction zone with a depth of tens of micrometers to be generated inside the cemented carbide matrix of the intelligent cutting tool. The intermediate thermal iteration reveals that, since the coolant cannot adhere to the ridge at all, the heat generated by the instantaneous power conversion of the spindle forms a "heat trap" locally, and the temperature gradient increases exponentially. The advanced coupling iteration superimposes the two and calculates that, due to the huge temperature difference between the groove bottom and the ridge, coupled with the high compression of the ridge, an extremely dangerous alternating zone of tension and compression stress is generated at the interface between the two.

[0033] The system delineates a region with a sudden increase in equivalent stress on the sand-facing surface of the chip breaker's "raised ridge," clearly defining it as a stress concentration area. At this point, collapse is highly likely to occur during subsequent cutting. Secondly, at the curvature inflection point where the groove bottom transitions to the ridge, the system finds that the tensile stress completely coincides with an extremely high temperature gradient. This area has the highest risk of grain thermal tearing, and is precisely defined as a crack-sensitive area. When the grinding wheel completes the chip breaker grinding and the trajectory evolves to the outermost straight main cutting edge of the intelligent insert, the system captures the microscopic lateral flow phenomenon of the material caused by thermal release, defining this extremely narrow edge band as the cutting edge blunting zone. The size of the blunt radius of this area will directly determine the surface finish and cutting life of the machined surface when the intelligent insert is machining aerospace aluminum alloys. Through this series of definitions, the most vulnerable and most differentiated optimization-required three-dimensional space on the intelligent insert is specifically presented.

[0034] refer to Figure 3 In step S12, the specific steps are as follows: S121: Dynamically identify each key region and introduce dynamic weights for each key region in the dynamic identification process, thereby extracting multiple current parameters in different dimensions in real time. The different dimensions cover the material removal rate dimension, local thermal damage dimension, and residual stress distribution state dimension. S122: Obtain the current 3D model of the intelligent cutting tool, and dynamically load multiple current parameters as constraint nodes into the current 3D model of the intelligent cutting tool to construct the dynamic parameter model of the intelligent cutting tool. The dynamic parameter model can evolve in real time with the grinding process. For the dynamic parameter model, a multi-objective optimization mechanism based on dynamic game is adopted, and multiple optimization objectives are adaptively deconstructed by combining the usage requirements of the intelligent cutting tool in the corresponding application scenario. In the embodiments of this application, each key region is dynamically identified, and a dynamic weight is introduced for each key region in the dynamic identification process, so as to extract multiple current parameters in different dimensions in real time. The different dimensions cover the material removal rate dimension, the local thermal damage dimension, and the residual stress distribution state dimension.

[0035] At this point, after identifying several key areas of the intelligent cutting tool, the system establishes a dynamic recognition model based on spatial envelope and physical field gradient. As the grinding wheel continues to advance, the geometric boundaries and physical states of the key areas are not static, but rather undergo spatiotemporal migration as material is continuously removed. The system uses high-frequency comparison of the thermo-coupling field distribution in the current five-dimensional spatiotemporal coordinate system with a preset critical threshold template to determine in real time whether the current contact point of the grinding wheel has fallen into or is about to enter the stress concentration area, crack-sensitive area, or cutting edge blunting area. Simultaneously, the system uses a sliding window to track the spatial offset of the centroid of these areas and the field intensity diffusion rate, achieving highly sensitive dynamic recognition and real-time boundary refresh of each key area under complex grinding conditions.

[0036] During the synchronous process of dynamic identification, the system assigns non-static dynamic weights to each key region in real time. This weight allocation abandons the equalization process and instead dynamically calculates the weights by deeply binding them to the application scenario logic and physical degradation mechanism of the intelligent blade. Specifically, the system assesses in real time the degree of harm and irreversibility to the final performance of the blade if the currently identified key region fails. For example, when a crack-sensitive area is identified and its internal thermal tensile stress increases exponentially, the system determines that the irreversible risk of micro-tears is extremely high, and thus raises the weight of that area to the highest level in a very short time. Conversely, if a stress concentration area is identified and the stress level is within the safety margin of the material's yield limit, the system appropriately lowers its weight. This dynamic weighting mechanism ensures that subsequent parameter extraction can accurately focus on the physical region with the greatest risk of damage at present.

[0037] Driven and guided by dynamic weights, the system initiates a multi-dimensional feature calculation channel to extract current parameters that reflect the physical essence of key areas in real time. In the dimension of material removal rate, the system combines the current interpolation speed and dynamic normal grinding force to calculate the material volume removal rate and specific grinding energy per unit time. In the dimension of local thermal damage, the system extracts the transient maximum temperature, temperature gradient change rate, and duration of action exceeding the phase transition temperature threshold in the current region. In the dimension of residual stress distribution, the system extracts the peak stress and stress layer depth parameters characterizing the surface compressive or tensile stress state by analyzing the stress tensor of the coupled field. The system weights and fuses the parameter vectors of the above three dimensions with the corresponding dynamic weights to output a multi-dimensional weighted current parameter set that can truly reflect the current grinding microstate.

[0038] Specifically, as the grinding wheel's path advances at high speed from the bottom of the chip breaker groove on the rake face of the intelligent insert towards the main cutting edge, the system tracks the previously defined "crack-sensitive zone" (located at the curvature inflection point where the groove bottom transitions to the ridge) in real time. Since the grinding wheel's envelope is constantly removing material, the three-dimensional spatial envelope of this crack-sensitive zone is not fixed but "shifts" towards the cutting edge as the grinding wheel's position changes. By frequently refreshing the thermal field grid, the system accurately and dynamically identifies the actual spatial occupancy of this sensitive zone and its real-time changes in physical intensity every millisecond.

[0039] The system dynamically assigns weights to specific working conditions of intelligent cutting tools. At a certain moment, the system dynamically identifies that the grinding wheel completely covers the "crack-sensitive area," and due to the poor chip removal of the chip breaker structure, the local thermal damage index at this location is extremely deteriorated. Based on the physical logic that "once a crack occurs, it is an irreversible and fatal defect," the system instantly and dynamically jumps the weight of this crack-sensitive area from the baseline value of 0.3 to 0.8. At the same time, the adjacent "stress concentration area" has a dynamic weight that is automatically suppressed to 0.1 because the material removal is smoother and the stress release is stable. This transient weight redistribution completely changes the focus of attention in parameter extraction.

[0040] Under the absolute high weighting of 0.8 in the crack-sensitive zone, the system extracts multi-dimensional current parameters directly related to the region in real time: in the dimension of material removal rate, the system extracts the parameter of abnormally soaring specific grinding energy caused by the grinding wheel squeezing into the corner of the chip breaker groove; in the dimension of local thermal damage, the system extracts the parameter of excessive temperature gradient and high-temperature residence time caused by coolant obstruction at the corner; in the dimension of residual stress distribution, the system extracts the critical parameter of the surface of the region reversing from a beneficial compressive stress state to an extremely dangerous tensile stress state. The parameters of these three dimensions are fused with high weights to form a set of "high-dimensional alarm-level current parameters" characterizing the imminent thermal crack at the corner of the chip breaker groove of the intelligent cutting tool, providing the most accurate quantitative input for the instantaneous intervention of the subsequent model.

[0041] Furthermore, the current 3D model of the intelligent cutting tool is obtained, and multiple current parameters are dynamically loaded into the current 3D model of the intelligent cutting tool as constraint nodes, thereby constructing a dynamic parameter model of the intelligent cutting tool. This dynamic parameter model can evolve in real time with the grinding process. For this dynamic parameter model, a multi-objective optimization mechanism based on dynamic game theory is adopted. In addition, combined with the usage requirements of the intelligent cutting tool in the corresponding application scenario, multiple optimization objectives are adaptively deconstructed. Multiple optimization objectives are introduced, and at the same time, the current 3D model and dynamic parameter model of the intelligent cutting tool are introduced, which further controls the application scenario of the intelligent cutting tool and improves the accuracy of multiple optimization objectives.

[0042] At this point, the system directly retrieves the initial 3D model of the intelligent cutting tool at the current machining moment from the model database of the CNC grinding center, and discretizes it into a set of hexahedral micro-mesh with topological association. The system converts the multi-dimensional weighted current parameter set output by S121, which covers the material removal rate, local thermal damage, and residual stress distribution, into physical boundary conditions, and dynamically attaches them as "constraint nodes" to the corresponding mesh nodes of the 3D model. Under the continuous drive of the grinding process, as the spatial position of the grinding wheel shifts, the spatial coordinates, numerical values, and influence domain radii of these constraint nodes are overwritten and iterated in milliseconds with the input of newly acquired parameters. Thus, the originally static geometric 3D model is endowed with dynamic life characteristics that interact with time, space, and physical fields in real time, constructing a dynamic parameter model that can completely map the current microscopic physical state of grinding.

[0043] For the constructed dynamic parameter model, the system abandons the traditional weighted summation single optimization method and introduces a multi-objective optimization mechanism based on dynamic game theory. Under this mechanism, the mutually constraining physical quantities in the dynamic parameter model are defined as different game players, such as the "efficiency player" that pursues the ultimate material removal rate and the "quality player" that suppresses local thermal damage and tensile stress. With Nash equilibrium or Pareto optimality as the theoretical core, the system simulates the strategic confrontation of each game player under given grinding resources (such as the upper limit of spindle power and the limit of coolant) at the moment of each constraint node update. By constructing the payoff function matrix of non-cooperative game, the system deduces the optimal state boundary after compromise of each party in the grid space of the dynamic parameter model, ensuring that the parameter solution found under extreme working conditions will not lead to the overall grinding failure due to the deterioration of a certain objective.

[0044] The system further integrates a database of terminal application scenarios for intelligent cutting tools, adaptively deconstructing macroscopic process requirements into microscopic, executable optimization objectives. By analyzing the current cutting tool's model identifier and material grade, the system matches its core requirements in actual cutting processes, such as fatigue resistance and chipping resistance in the aerospace field, or high cycle life requirements in the automotive production line. Combining the current focus area of ​​the game in the dynamic parameter model, the system reduces these macroscopic requirements to a set of optimization objectives with clear directionality and numerical thresholds. For example, if the current model is in a high-risk area of ​​thermo-coupling and the application scenario requires a long lifespan, the system adaptively deconstructs the objective into "maximizing residual compressive stress under the maximum tolerable thermal damage threshold" and "smooth transition of local material removal rate," thereby outputting a targeted objective group that accurately guides subsequent optimization.

[0045] Specifically, the system extracts a complex three-dimensional chip breaker groove mesh model of the intelligent cutting tool. At this time, the system precisely maps the high-weight "current parameters" extracted from S121 for the "critical area of ​​chip breaker corner crack" onto dozens of micro-mesh nodes at the bottom of the chip breaker groove, forming extremely strong local "constraint nodes". As the grinding wheel moves at high speed along the chip breaker groove curve, these constraint nodes spread in the mesh model, forcing the three-dimensional model of the intelligent cutting tool to no longer be a fixed geometry, but to become a "digital twin" filled with intense thermal stress and evolving in real time with the position of the grinding wheel, that is, the dynamic parameter model of the intelligent cutting tool.

[0046] Intense interference erupts within the dynamic parameter model of the intelligent cutting tool: In order to efficiently grind the complex curvature of the chip breaker groove, it is necessary to increase the feed rate as an efficiency-oriented game strategy; however, high-weight constraint nodes show that once the feed rate is increased, the local thermal damage at the corner of the chip breaker groove will instantly break through the phase transformation critical point of the cemented carbide as a quality-oriented game strategy; Based on the dynamic game mechanism, the system performs rapid adversarial deduction within the model and finally determines a "Nash equilibrium point" - that is, reducing the feed rate 5 milliseconds before the grinding wheel enters the corner, using inertia to smooth the transition, which not only ensures the overall groove shape machining cycle, but also forcibly suppresses the thermal tensile stress at the corner within the safety grid.

[0047] The system identified the application scenario of this intelligent cutting tool as "aerospace titanium alloy precision machining." Titanium alloy machining is prone to built-up edge formation, requiring the cutting tool to have an extremely sharp cutting edge and extremely strong surface compressive stress resistance. Based on this, the system performs adaptive deconstruction, combining the macroscopic "aerospace-grade fatigue resistance requirements" with the aforementioned game equilibrium point, and precisely transforming them into two highly specific optimization objectives: Objective 1, for the corner of the chip breaker groove, is deconstructed as "forcibly reducing the local thermal damage gradient by 30% to avoid microscopic phase transformation"; Objective 2, for the area transitioning to the main cutting edge, is deconstructed as "achieving an increase in the residual compressive stress distribution depth of the cutting edge passivation zone to more than 15 micrometers while ensuring that the material removal rate does not decrease by 15%". These two objectives are directly used as bullseye and thrown into the subsequent neural network optimization space.

[0048] refer to Figure 4 In step S13, the specific steps are as follows: S131: Label the corresponding optimization space, map each target to be optimized to its own independent optimization space, and perform multi-level training in combination with the preset neural network. At this time, the preset neural network covers the dual driving capabilities of deep reinforcement learning and physical information, and trains each target to be optimized. S132: In the initial stage of training, the neural network establishes a preliminary response surface through massive historical data; in the deep training stage, the neural network embeds the physical laws of the grinding process as a penalty term into the loss function, thereby determining the nonlinear relationship between the grinding parameters of the intelligent cutting tool and the processing quality of the intelligent cutting tool. S133: The nonlinear relationship is analyzed, and a grinding constraint relationship processing mechanism that combines multidimensional convex hull mechanism and tabu search is introduced in the analysis process. This allows for multi-factor matching of the nonlinear relationship and the grinding constraint relationship of the intelligent cutting tool. After matching, the optimized grinding parameters with anti-interference robustness are determined by combining the corresponding optimal solution set.

[0049] In the embodiments of this application, the corresponding optimization space is marked, each target to be optimized is mapped to its own independent optimization space, and multi-level training is performed in combination with a preset neural network. At this time, the preset neural network covers the dual driving capabilities of deep reinforcement learning and physical information, and trains each target to be optimized.

[0050] At this point, the system first delineates and marks mutually decoupled independent optimization spaces in the global decision domain based on the differences in the physical attributes of the targets to be optimized, so as to avoid the curse of dimensionality and mutual exclusion interference of targets in the high-dimensional optimization process. For each target to be optimized deconstructed from the application scenario, the system performs mathematical modeling of its state space and action space together with the corresponding boundary conditions, so as to achieve accurate mapping from physical semantics to high-dimensional feature space. The independent optimization spaces are kept in parallel in logic, but are bridged at the data level through a unified grinding state representation vector, so as to ensure that each space can independently cultivate a specific target and maintain state synchronization under global constraints during subsequent optimization.

[0051] For the high-dimensional optimization space after mapping, the system activates a pre-defined composite neural network architecture that deeply integrates the dual capabilities of physical information-driven and deep reinforcement learning. In the forward propagation backbone of the network, a deep reinforcement learning module is constructed, such as using the SoftActor-Critic framework, where grinding parameters are defined as agent actions and processing quality indicators are defined as reward functions, enabling it to explore and experiment under unknown or highly nonlinear conditions. At the same time, a physical information constraint module based on the grinding mechanism is rigidly embedded in the hidden layers and loss function of the network, and the discretized residuals of the grinding specific energy equation and the heat conduction partial differential equation are added as penalty terms to the backpropagation process of the network, thereby forcibly constraining the output of the neural network to follow the objective physical laws of thermodynamics and materials mechanics.

[0052] After the dual-drive network architecture is ready, the system initiates a multi-level training mechanism from shallow to deep for each independent optimization space. In the initial training stage, it mainly relies on the physical information constraint module for supervised pre-training, using historical grinding data and analytical solutions to quickly converge and establish a coarse-grained physical baseline for grinding parameters and processing results. In the intermediate training stage, the deep reinforcement learning module is activated, and the agent, guided by the physical baseline, conducts large-scale interactive exploration for specific optimization targets in the simulation environment, maximizing cumulative rewards to approach the Pareto front. In the advanced training stage, the system introduces dynamic noise perturbation to simulate sensor drift and material inhomogeneity in the intelligent cutting tool grinding process, prompting the dual-drive network to perform robust fine-tuning under extreme conditions approaching the physical constraint boundary, and finally outputs policy network weights that converge in each optimization space and have strong generalization ability.

[0053] Specifically, taking into account the two optimization objectives deconstructed by S122, the system delineates two independent optimization regions in the global decision domain. For the objective of "forcibly reducing the local thermal damage gradient at the corner of the chip breaker by 30%", the system marks the "thermal control optimization space" and uses the heat flux density boundary and coolant flow rate as the core state vectors of this space. For the objective of "increasing the residual compressive stress in the edge passivation zone to 15 micrometers", the system marks the "stress shape optimization space" and uses the specific grinding energy and normal force dynamic components as the core state vectors of this space. The two spaces operate independently, but are implicitly linked through the spatial position connection parameters of the intelligent insert from the chip breaker to the main cutting edge.

[0054] The system deploys dual-drive neural networks simultaneously in these two spaces. In the "thermal control optimization space," if reinforcement learning is relied upon alone, only a shortcut solution of "reducing the grinding wheel linear speed to an extremely low level" can be obtained. However, at this time, the physical information driving module comes into play. It writes the "critical heat flux density equation" of the intelligent cutting tool's cemented carbide matrix as a hard constraint into the loss function. When the agent attempts to give an action that violates the minimum energy threshold for material removal, the physical residual penalty term will instantly lower its reward value, forcing the agent to find a cooling strategy within the parameter range that conforms to the laws of heat transfer. This cooling strategy is such as dynamically adjusting the feed waveform rather than simply reducing the speed.

[0055] The system conducts multi-level training on the network. In the pre-training stage, the network absorbs a large amount of data from past grinding of ordinary two-dimensional grooves, quickly mastering the basic physical laws of abrasive cutting of cemented carbide by grinding wheels. After entering the intermediate reinforcement training, for the complex three-dimensional curvature of the chip breaker groove of the intelligent insert, the agent continuously tries different combinations of coolant injection delay and grinding wheel feed acceleration within the "thermal control space". Once it finds that a certain set of parameters can reduce the corner temperature gradient without triggering physical penalties, it strengthens the strategy. In the advanced robustness training, the system deliberately adds interference from hardness fluctuations caused by uneven titanium alloy coating to the input, forcing the dual-drive network to still be able to stably output a set of optimal grinding parameter strategies that can both prevent the chip breaker groove of the intelligent insert from burning and form a sufficiently deep compressive stress on the cutting edge under extreme perturbations.

[0056] Furthermore, in the initial training stage, the neural network establishes a preliminary response surface using massive amounts of historical data; in the deep training stage, the neural network embeds the physical laws of the grinding process as a penalty term into the loss function, thereby determining the nonlinear relationship between the grinding parameters of the intelligent cutting tool and the processing quality of the intelligent cutting tool. This is compatible with the overall consideration of embedding the physical laws of the grinding process as a penalty term into the loss function of the neural network, ensuring the accuracy of the nonlinear relationship between the grinding parameters of the intelligent cutting tool and the processing quality of the intelligent cutting tool.

[0057] At this initial stage of training, the system first disconnects from the complex real-time physical calculation link and directly connects to the underlying historical database of the grinding big data center for pure data-driven pre-training. The system extracts massive historical process records covering different insert groove types, different grinding wheel states, and various combinations of process parameters, using them as high-dimensional input features. It also uses the geometric accuracy, surface roughness, and other processing quality quantitative indicators fed back by the corresponding detection process as output labels. Through a multilayer perceptron or residual network structure, the above input-output mapping is fitted with a large-scale stochastic gradient descent. The network converges rapidly in the parameter space, thereby constructing a preliminary response surface that can reflect macroscopic statistical laws. Although this response surface lacks a precise explanation of the microscopic mechanism, it successfully filters out data noise and provides a reasonable parameter search neighborhood and initial weight distribution for subsequent deep training.

[0058] Once the pre-trained response surface is solidified, the system immediately enters the deep training phase, reconstructing and modifying the core loss function of the neural network. The system discretizes the underlying physical laws of the grinding process, such as the law of conservation of grinding specific energy, the maximum shear stress criterion, and the transient heat conduction boundary conditions, into differentiable algebraic expressions through partial differential equations. The system then transforms these physical algebraic expressions into regularization penalty terms, which are embedded with strict weighting coefficients into the basic loss function that originally only contained the root mean square of the data error. Under this mechanism, after the neural network predicts the processing quality in each forward propagation, it not only calculates the data deviation from the true label but also simultaneously calculates the "physical residual" of the predicted state under given input parameters, which violates the physical conservation laws. The data error and the physical residual are then jointly differentiated.

[0059] During the deep training phase, the system performs rigorous backpropagation and parameter updates. When the hidden layers of the neural network attempt to reduce data errors by fitting abnormal noise in historical data, the physical law penalty term generates a huge gradient backpropagation, forcibly blocking this "overfitting" tendency that violates the common sense of thermodynamics or materials mechanics. Under the dual pressure of data fidelity and physical compliance, the update trajectory of network weights is forcibly restricted to a high-dimensional manifold that conforms to the material removal mechanism. As the iteration rounds advance, the network is no longer just a black box data mapping, but evolves into a gray box model that integrates prior physical knowledge. Finally, a strong nonlinear relationship with extremely high extrapolation and generalization ability and clear physical meaning is established between the independent variable of grinding parameters and the dependent variable of processing quality.

[0060] Specifically, the system retrieved hundreds of thousands of historical data points from the past three years of CNC grinding centers processing various intelligent inserts and similar groove inserts. By absorbing this data, the neural network quickly established a preliminary response surface between "grinding wheel linear speed - feed rate - coolant pressure" and "surface roughness - groove width error". At this point, the network only knew the macroscopic statistical probability that "when the feed rate is reduced to a certain threshold, the roughness of the chip breaker groove of the intelligent insert usually improves", but it was completely unaware of the underlying thermodynamic reasons and was easily misled by noise in the historical data, such as "abnormal force signals caused by untimely grinding wheel dressing".

[0061] To address the issue of easy burning in the chip breaker groove of intelligent cutting tools, the system forcibly embeds a discretization penalty term for the "Jaeger moving heat source model" for cemented carbide materials into the loss function of deep training. This penalty term explicitly stipulates that the temperature field distribution of the grinding zone predicted by the network must satisfy the energy conservation between the input heat, the heat removed by the coolant, and the latent heat of phase transformation of the material. If the network outputs an absurd prediction result that "the chip breaker groove temperature is still very low at extremely high grinding wheel linear speed" in order to cater to a few biased historical data, this physical penalty term will instantly calculate a huge residual of energy non-conservation.

[0062] Furthermore, the neural network corrected its internal weights, abandoning the idea of ​​taking shortcuts to fit outlier data. Instead, it learned the deep physical logic of how the complex chip breaker groove curvature of the intelligent insert affects the contact arc length of the grinding wheel, and thus how it changes the heat flux density distribution. The nonlinear relationship finally established is no longer a simple linear interpolation, but accurately presents a deep nonlinear mapping relationship with great physical insight, such as "when the grinding wheel grinds the chip breaker groove of the intelligent insert at the inflection point of the variable cross section, a small step in the feed rate will cause the normal grinding force to increase exponentially and nonlinearly, thus inducing a sudden change in surface tensile stress." This lays an absolutely reliable model foundation for the subsequent solution of the true optimal parameters.

[0063] Therefore, this nonlinear relationship is analyzed, and a grinding constraint relationship processing mechanism that integrates multidimensional convex hull mechanism and tabu search is introduced in the analysis process. This allows for multi-factor matching of the nonlinear relationship and the grinding constraint relationship of the intelligent cutting tool. After matching, the optimized grinding parameters with anti-interference robustness are determined by combining the corresponding optimal solution set. This approach takes into account the overall consideration of matching and combining the corresponding optimal solution set, ensuring the accuracy of the optimized grinding parameters with anti-interference robustness.

[0064] At this point, after establishing the strong nonlinear relationship between grinding parameters and machining quality, the system first performs dimensionality reduction and convex hull analysis on the high-dimensional nonlinear space. The system extracts the Pareto front scatter points formed by multiple optimization objectives deconstructed from S122 in the parameter space. Using a fast convex hull-based approach or a support vector data description method, one or more tightly wrapped nonlinear convex hull geometries are constructed outside these high-dimensional scatter points. These convex hulls mathematically represent the maximum physical boundary allowed by the combination of grinding parameters without causing quality collapse. This transforms the complex nonlinear relationship into a feasible optimization domain with a clear geometric contour, eliminating redundant computational space far from the optimal solution set.

[0065] Within the feasible region constructed by the convex hull, the system further introduces a mechanism that integrates multi-dimensional convex hull and tabu search to handle stringent grinding constraints. The system sets constraints such as the maximum power of the machine tool spindle, the maximum acceleration of the feed axis, the limit linear velocity of the grinding wheel, and the critical threshold of thermal damage as dynamic "hyperplane barriers" interspersed within the convex hull. During the optimization process, when the tabu search method performs random neighborhood jumps within the convex hull, once a candidate solution touches these constraint hyperplanes, it violates the constraint relationship. The system not only immediately applies a "taboo" penalty to the coordinate point and adds it to the taboo list to prevent loops, but also uses the geometric normal vector characteristics of the convex hull to force the search direction to bounce back to the safe area inside the convex hull, thus achieving deep multi-factor matching of nonlinear relationships and physical constraints.

[0066] After completing multi-factor matching, the system outputs a candidate optimal solution set that satisfies all constraints in the safe and feasible region within the convex hull. To cope with the unavoidable dynamic disturbances in the actual grinding environment, such as sensor drift, uneven material microhardness, and grinding wheel wear, the system performs an anti-interference robustness evaluation on the candidate solution set. Centered on the current optimal solution, the system applies Gaussian white noise perturbation to the neighborhood of its multidimensional grinding parameters to simulate extreme working conditions, and uses the established physical information neural network to predict the machining quality after the perturbation again. The system selects the solution whose quality index decreases the least after suffering the maximum parameter perturbation, i.e., the solution that is least sensitive to boundary constraints, and determines this solution as the final optimized grinding parameters to ensure that it has extremely strong working condition tolerance after being sent to the machine tool.

[0067] Specifically, for the complex nonlinear relationship between "feed speed - grinding wheel linear speed - thermal stress - residual compressive stress" established by S132 for intelligent cutting tools, the system extracts all Pareto scatter points that can satisfy "the chip breaker groove is not burned and the cutting edge has compressive stress". The system uses a multi-dimensional convex hull method to wrap an irregular high-dimensional geometric convex hull around these scatter points like blowing up a balloon. The interior of this convex hull represents the range of all "qualified" parameter combinations when the intelligent cutting tool grinds the chip breaker groove, such as the safe zone where the grinding wheel speed may fluctuate between 25-35m / s.

[0068] The system begins searching for the true optimal solution within this convex hull, but must simultaneously satisfy stringent constraints. For example, the machine tool's Z-axis for intelligent cutting tools has a physical constraint that "the maximum instantaneous acceleration cannot exceed 0.5g." When the tabu search method attempts to jump towards "higher feed rates" within the convex hull, it calculates that the required acceleration for this solution reaches 0.6g. At this point, the fusion mechanism is triggered: the system not only pulls this 0.6g parameter point into the "taboo list" to prevent it from being accessed again, but also uses the normal vector of the convex hull boundary to "bounce" the search path back to the region within the convex hull where the acceleration is less than 0.5g, like hitting a billiard ball. This achieves a perfect match between nonlinear quality requirements and the rigid constraints of the machine tool.

[0069] The system found several "optimal solution sets" within the convex hull that satisfied the constraints, such as solution A: rotational speed 28 m / s, feed rate 800 mm / min; and solution B: rotational speed 32 m / s, feed rate 750 mm / min. At this point, the system introduced a robustness assessment: it simulated the potential uneven distribution of tungsten carbide grains within the cemented carbide matrix of the intelligent cutting tool. The assessment revealed that solution A, when encountering slightly harder micro-regions of material, was prone to falling below the thermal damage constraint line due to its low rotational speed. In contrast, solution B, even when encountering fluctuations in material hardness or slight blockage of the grinding wheel, still calculated thermal stress within the safe threshold. Ultimately, the system decisively discarded solution A and officially marked and output solution B, which possessed strong anti-interference capabilities, preparing to send it into the simulation space for final verification.

[0070] refer to Figure 5 In step S14, the specific steps are as follows: S141: Load the optimized grinding parameters into the simulation space based on the digital twin architecture, synchronously call the dynamic parameter model of the intelligent cutting tool, use the dynamic parameter model as the boundary condition, trigger dynamic simulation covering multiple physics fields, and output multiple simulation results during the dynamic simulation process. The multiple simulation results cover the phase transformation trend of the microstructure, grinding burn, and edge defect morphology. S142: The embedded adaptive optimization mechanism is activated by the dynamic simulation of the intelligent cutting tool. The adaptive optimization mechanism adopts a strategy that combines variable step size hill climbing method with Monte Carlo tree search. It performs multi-dimensional edge perturbation test and sensitivity analysis on the simulation results, automatically eliminates pseudo-optimal solutions, and finally determines the ultimate simulation result with the highest machining determinism. The grinding parameters corresponding to the result are digitally verified, and the grinding parameters of the intelligent cutting tool are determined by tracing back the final simulation result. In the embodiments of this application, the optimized grinding parameters are loaded into the simulation space based on the digital twin architecture, and the dynamic parameter model of the intelligent cutting tool is called synchronously. The dynamic parameter model is used as the boundary condition to trigger dynamic simulation covering multiple physics fields. During the dynamic simulation process, multiple simulation results are output. The multiple simulation results cover the microstructure phase transformation trend, grinding burn, and cutting edge defect morphology, and introduce the microstructure phase transformation trend, grinding burn, and cutting edge defect morphology.

[0071] At this point, after receiving the optimized grinding parameters with anti-interference robustness, the system first sends them down to the virtual simulation space based on the digital twin architecture through the underlying communication bus, completing the parameter mapping from physical entity to digital virtual entity. Simultaneously, the system calls the intelligent cutting tool dynamic parameter model built in the early stage across processes. At this stage, the dynamic parameter model is no longer the observed object, but is deconstructed into the initial boundary condition matrix of the finite element simulation solver. Specifically, the stress field distribution, temperature field gradient and material removal rate characteristics that vary with spatial location and time contained in the dynamic parameter model are accurately mapped to the initial state of dynamic displacement constraints, heat flux density input load and material constitutive equation on the simulation mesh nodes, so as to completely reproduce the initial state of the physical field at the moment of grinding in the virtual space.

[0072] After the boundary conditions are precisely coupled, the system officially triggers a multi-physics dynamic simulation covering thermo-mechanical coupling, fluid dynamics, and microstructure evolution. The simulation engine adopts an explicit time integration method to simulate the microscopic processes of the abrasive cutting tool material with a very small time increment step. Within each increment step, the system simultaneously solves the elastoplastic mechanical equations of the mechanical deformation field, the transient heat conduction equations caused by frictional work and material shear, and the fluid-structure interaction heat transfer equations of the coolant in the complex three-dimensional chip breaker groove structure. By continuously iterating the stress update and energy conservation of the mesh elements at the microscale, the system reproduces the entire process of dynamic planing and reshaping of the internal structure and surface morphology of the intelligent cutting tool material under complex constraints with high fidelity through continuous deduction.

[0073] As the dynamic simulation is completed, the system performs dimensionality reduction and feature extraction on massive spatiotemporal field data in the post-simulation processing module. Based on different physical characterization mechanisms of grinding defects, it outputs simulation results in multiple dimensions. First, the system analyzes the transient temperature history curves of the micro-grid units and, combined with the material's thermodynamic phase transition model, outputs a phase transition trend map to determine whether the material's lattice has undergone carbonization or oxidation. Second, the system extracts the peak temperature field envelope and duration of the cutting tool surface, spatially compares it with the material's thermal damage threshold, and outputs a grinding burn cloud map calibrating high-risk areas. Finally, the system removes the failed units and distorted / collapsed grid nodes from the simulation model, reconstructs the microscopic geometry at the end of the grinding trajectory, and outputs a precise image of the cutting edge defect morphology, showing chipping, tearing, or plastic bulging.

[0074] Specifically, the system inputs the robust parameters finally determined by S133 into the digital twin simulation platform. The robust parameters include the grinding wheel linear velocity of 32m / s, the feed rate of 750mm / min, and a specific coolant pressure. At the same time, the system retrieves the dynamic parameter model of the intelligent cutting tool at the corner of the chip breaker groove. At this time, the physical characteristics of "the corner of the chip breaker groove is prone to heat accumulation and stress concentration" in the dynamic model are transformed into "ultra-high initial heat flux density input" and "local triaxial compressive stress initial boundary" of the mesh node in the corner area in the simulation software. Before the calculation starts, the intelligent cutting tool in the virtual space has been given the same characteristics and initial stress state as the actual machining.

[0075] The simulation engine began to simulate the process of a diamond grinding wheel traversing the three-dimensional curved surface of the chip breaker groove of an intelligent cutting tool at a speed of 32 m / s in microsecond increments. During the simulation, the system not only calculated the huge cutting force generated by the grinding wheel extruding the cemented carbide matrix, but also simultaneously calculated the "air film insulation effect" caused by the complex curvature of the chip breaker groove, which prevented the coolant from effectively entering the bottom groove. This coupled calculation of thermal and flow fields accurately simulated the transient microscopic physical process of heat accumulation at the bottom of the chip breaker groove, violent plastic flow of the material, and eventual shearing removal at a feed rate of 750 mm / min.

[0076] After the simulation, the system output three highly targeted simulation results for the specific working condition of the intelligent insert. Regarding the phase transformation trend, the system output chart shows that the WC grain structure is stable on the windward side of the chip breaker groove due to sufficient cooling, but a slight diffusion trend of the Co binder phase due to high temperature was detected at the corner of the leeward side. Regarding grinding burns, the system output a thermal cloud map, clearly marking a crescent-shaped area with a radius of 2 mm at the bottom of the groove, where the transient temperature is close to the oxidation critical point of cemented carbide, posing a potential risk of secondary tempering burns. Regarding the edge defect morphology, the system clearly outputs a three-dimensional morphology map by reconstructing the deformation trajectory of the micro-mesh, showing that at the end of the transition from the chip breaker groove to the main cutting edge, due to the release of residual stress, a microscopic "serrated chipping" morphology with a width of about 3 micrometers and a depth of about 1 micrometer appears at the edge. These three results will directly determine whether this set of parameters needs to be further fine-tuned.

[0077] Furthermore, the embedded adaptive optimization mechanism is activated through the dynamic simulation of the intelligent cutting tool. This adaptive optimization mechanism employs a strategy that combines variable step size hill climbing with Monte Carlo tree search to perform multi-dimensional edge perturbation testing and sensitivity analysis on the simulation results. It automatically eliminates pseudo-optimal solutions and ultimately determines the final simulation result with the highest machining determinism. The grinding parameters corresponding to this result are then digitally verified. The grinding parameters of the intelligent cutting tool are determined based on the traceability of this final simulation result, taking into account the overall consideration of the final simulation result and ensuring the accuracy of the grinding parameters of the intelligent cutting tool. At the same time, optimized grinding parameters are introduced, fully considering the dynamic parameter model and simulation space of the intelligent cutting tool, thereby improving the accuracy of the final simulation result.

[0078] At this point, after the initial multi-dimensional simulation results are output in the digital twin space, the system immediately activates the adaptive optimization mechanism embedded in the simulation engine. This mechanism no longer regards the simulation results as a static endpoint, but rather as the starting point for a new round of exploration. It automatically performs multi-dimensional edge disturbance tests around the grinding parameter coordinate points where the current simulation results are located. By injecting tiny Gaussian random disturbances along each coordinate axis in the parameter vector space, the system simulates extreme working conditions such as limit tolerances, micro-vibrations, or material property fluctuations, forcing the digital twin model to re-solve for dimensionality reduction at the "dangerous edge" of the current parameter neighborhood, in order to explore the sensitivity of the current solution to disturbance factors and the rate of quality deterioration.

[0079] Based on the large amount of neighborhood feedback data generated by edge perturbations, the system initiates a strategy that integrates variable step size hill climbing and Monte Carlo Tree Search (MCTS) for in-depth analysis. The variable step size hill climbing mechanism is responsible for fine-grained gradient climbing and descent detection in the local neighborhood. When a perturbation is found to cause a precipitous nonlinear deterioration in a simulation result, the step size is automatically reduced to accurately locate the critical perturbation threshold that triggers the deterioration. Meanwhile, Monte Carlo Tree Search constructs a search tree in a broader global parameter space. By balancing exploration and utilization through the UCB (Upper Confidence Interval) formula, it evaluates whether there are other undiscovered parameter combinations that can maintain better robustness under the same perturbation. Under the cross-validation of the dual strategies, the system calculates the sensitivity Jacobian matrix eigenvalues ​​of each parameter solution and strictly judges those parameter combinations that seem optimal locally but whose quality indicators quickly exceed the limits under edge perturbations as "pseudo-optimal solutions" and physically removes them.

[0080] After systematically cleaning up pseudo-optimal solutions, the system identifies the parameter combination with the lowest sensitivity eigenvalues, i.e., the one that exhibits the strongest immunity to multidimensional disturbances, from the remaining candidate solution set. This combination is then established as the final simulation result with the highest machining determinism. The system calls a high-fidelity grid to rigorously digitally verify this final result. By applying an extreme composite disturbance envelope, it confirms that all quality indicators are within the safe range of absolute convergence. Once the digital verification is successful, the system immediately performs reverse mapping tracing, tracing back along the data flow to the initial input boundary that generated the final simulation result. This allows the system to extract and confirm the physical input without deviation as the intelligent cutting tool grinding parameters that are ultimately sent to the physical machine tool.

[0081] Specifically, the adaptive optimization mechanism is activated for the seemingly perfect set of robust parameters output by S141. The system adjusts these parameters in the digital twin space: it fine-tunes the grinding wheel linear velocity to 32.2 m / s, simulating slight fluctuations in the linear velocity after grinding wheel dressing, and at the same time reduces the hardness of the carbide material at the corner of the chip breaker of the intelligent cutting tool by 2%, simulating material inhomogeneity between batches. With these "toxic" edge disturbance parameters, the system re-triggers the transient thermo-mechanical coupling simulation of the chip breaker part of the intelligent cutting tool to observe whether this small change will cause the originally safe working condition to be damaged instantly.

[0082] Simulation results show that under disturbance, the burn area at the corner of the leeward side of the chip breaker groove of the intelligent cutting tool suddenly increased by 3 times from the safe value, exhibiting extremely high state sensitivity. At this time, the variable step size climbing method immediately intervened, gradually adjusting the linear velocity with a preset step size, and accurately determined that as long as the linear velocity is lower than 31.5 m / s, the burn area in this area will instantly converge to zero. At the same time, the Monte Carlo tree search was carried out globally, and it was found that if the feed rate is finely adjusted to 780 mm / min, the efficiency loss caused by the speed reduction can be offset. Moreover, in hundreds of random disturbances simulated by the tree nodes, the residual compressive stress of the chip breaker groove has always remained stable in the qualified range of more than 15 micrometers. Through this fusion analysis, the system decisively marked the initial robust parameter as a "pseudo-optimal solution" and removed it.

[0083] The system determined a new set of parameters, refined through rigorous testing, as the final simulation result. These parameters included a linear velocity of 31.5 m / s, a feed rate of 780 mm / min, and a specific coolant delay strategy. In the final digital verification stage, the system applied a triple extreme composite disturbance—grind wheel clogging, spindle axial movement, and coolant pressure fluctuation—to the three-dimensional chip breaker model of the intelligent insert. The simulation cloud map confirmed that no chipping occurred on the main cutting edge, no phase transformation trend was observed at the bottom of the chip breaker, and the grinding burn area was absolutely zero. The digital verification passed perfectly. The system traced back along the data lineage and clearly output the final grinding parameters for the intelligent insert.

[0084] Please see Figure 6 , Figure 6 This is a schematic diagram of the structure of the intelligent cutting tool grinding parameter optimization system in an embodiment of the present invention; the intelligent cutting tool grinding parameter optimization system is applied to the above-mentioned intelligent cutting tool grinding parameter optimization method; the intelligent cutting tool grinding parameter optimization system includes: The key area module 21 is used to collect multiple morphological parameters of the intelligent cutting tool for the CNC cutting tool grinding center, and perform spatiotemporal alignment with the working parameters of the intelligent cutting tool during the grinding process. During the alignment process, a corresponding key parameter combination is formed, and multiple key areas of the intelligent cutting tool are determined by multi-level iteration along the key parameter combination. The optimization target module 22 is used to determine multiple current parameters in different dimensions based on the dynamic identification of each key area, and to construct the dynamic parameter model of the intelligent blade in combination with the current three-dimensional model of the intelligent blade. The dynamic parameter model is then optimized in multiple objectives, and multiple optimization targets are determined in combination with the application scenario of the intelligent blade. The optimization module 23 is used to input each target to be optimized into the corresponding optimization space, and perform multi-level training in combination with the preset neural network. During the training process, it determines the nonlinear relationship between the grinding parameters and the processing quality of the intelligent cutting tool, and further determines the optimized grinding parameters in combination with the grinding constraint relationship of the intelligent cutting tool. The simulation module 24 is used to load the optimized grinding parameters into the simulation space of the intelligent cutting tool, and trigger the corresponding dynamic simulation in combination with the dynamic parameter model of the intelligent cutting tool. During the dynamic simulation, multiple simulation results are output, and the final simulation result is determined by combining the adaptive optimization mechanism, and the grinding parameters corresponding to the final simulation result are marked.

[0085] It should be noted that although multiple modules are mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or modules described above can be embodied in one module. Conversely, the features and functions of one module described above can be further divided and embodied by multiple modules.

[0086] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only.

[0087] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for optimizing grinding parameters of an intelligent cutting tool, characterized in that, include: For CNC insert grinding centers, multiple morphological parameters of intelligent inserts are collected and combined with the working parameters of the intelligent inserts during the grinding process for spatiotemporal alignment. During the alignment process, corresponding key parameter combinations are formed, and multiple key areas of the intelligent inserts are determined by multi-level iteration along the key parameter combinations. Based on the dynamic identification of each key area, multiple current parameters in different dimensions are determined, and a dynamic parameter model of the intelligent blade is constructed by combining the current three-dimensional model of the intelligent blade. The dynamic parameter model is then optimized in multiple objectives, and multiple optimization objectives are determined in combination with the application scenario of the intelligent blade. Each target to be optimized is input into the corresponding optimization space, and multi-level training is performed in combination with the preset neural network. During the training process, the nonlinear relationship between the grinding parameters and the machining quality of the intelligent cutting tool is determined. The optimized grinding parameters are further determined by combining the grinding constraint relationship of the intelligent cutting tool. The optimized grinding parameters are loaded into the simulation space of the intelligent cutting tool, and the corresponding dynamic simulation is triggered in combination with the dynamic parameter model of the intelligent cutting tool. During the dynamic simulation, multiple simulation results are output, and the final simulation result is determined by combining the adaptive optimization mechanism. The grinding parameters corresponding to the final simulation result are marked.

2. The method for optimizing grinding parameters of an intelligent cutting tool according to claim 1, characterized in that, For the CNC insert grinding center, multiple morphological parameters of the intelligent insert are collected and combined with the working parameters of the intelligent insert during the grinding process for spatiotemporal alignment. During the alignment process, corresponding key parameter combinations are formed. Through multi-level iterations along these key parameter combinations, multiple key regions of the intelligent insert are determined, including: The system monitors the operation of the CNC insert grinding center in real time and marks the position of the intelligent insert. By deploying a high-frequency dynamic sensor and machine vision fusion system, data is collected along the position of the intelligent insert, and multiple morphological parameters of the intelligent insert are obtained. These multiple morphological parameters cover the geometric morphological parameters, surface morphology parameters, and grinding wheel state parameters of the intelligent insert. The working parameters of the intelligent cutting tool during the grinding process are obtained, and a spatiotemporal joint registration mechanism is introduced to dynamically align multiple morphological parameters with the working parameters of the intelligent cutting tool during the grinding process in a five-dimensional spatiotemporal coordinate system. During the alignment process, coupling mutation points between parameters are discovered, and key parameter combinations with causal relationships are automatically clustered to generate them. The working parameters of the intelligent cutting tool during the grinding process cover the instantaneous power of the spindle, the dynamic component of the grinding force, and the coolant flow field distribution.

3. The method for optimizing grinding parameters of an intelligent cutting tool according to claim 2, characterized in that, The method for a CNC insert grinding center involves collecting multiple morphological parameters of an intelligent insert and combining them with the working parameters of the intelligent insert during the grinding process for spatiotemporal alignment. During this alignment process, corresponding key parameter combinations are formed. Multiple key regions of the intelligent insert are determined through multi-level iterations along these key parameter combinations. The method also includes: The combination of key parameters is dynamically evolved, and multiple iterations are performed along the evolution trajectory of the combination of key parameters. In this process, multiple key regions of the intelligent cutting tool under the action of grinding thermo-mechanical coupling are determined. These multiple key regions cover the stress concentration zone, the crack sensitive zone, and the cutting edge passivation zone.

4. The method for optimizing grinding parameters of an intelligent cutting tool according to claim 1, characterized in that, The process involves determining multiple current parameters in different dimensions based on the dynamic identification of each key region, constructing a dynamic parameter model of the intelligent blade using the current 3D model of the blade, performing multi-objective optimization on this dynamic parameter model, and determining multiple optimization objectives based on the application scenario of the intelligent blade, including: Each key region is dynamically identified, and a dynamic weight is introduced for each key region during the dynamic identification process, thereby extracting multiple current parameters in different dimensions in real time. These different dimensions cover the material removal rate dimension, the local thermal damage dimension, and the residual stress distribution state dimension.

5. The method for optimizing grinding parameters of an intelligent cutting tool according to claim 4, characterized in that, The process of determining multiple current parameters in different dimensions based on the dynamic identification of each key region, constructing a dynamic parameter model of the intelligent blade based on the current 3D model of the intelligent blade, performing multi-objective optimization on the dynamic parameter model, and determining multiple optimization objectives based on the application scenario of the intelligent blade, also includes: The current 3D model of the intelligent cutting tool is obtained, and multiple current parameters are dynamically loaded into the current 3D model of the intelligent cutting tool as constraint nodes, thereby constructing a dynamic parameter model of the intelligent cutting tool. This dynamic parameter model can evolve in real time with the grinding process. For this dynamic parameter model, a multi-objective optimization mechanism based on dynamic game is adopted, and multiple optimization objectives are adaptively deconstructed by combining the usage requirements of the intelligent cutting tool in the corresponding application scenario.

6. The method for optimizing grinding parameters of an intelligent cutting tool according to claim 1, characterized in that, The process involves inputting each target to be optimized into the corresponding optimization space, performing multi-level training in conjunction with a pre-set neural network, and establishing a nonlinear relationship between the grinding parameters and machining quality of the intelligent cutting tool during the training process. Further, the optimized grinding parameters are determined by combining the grinding constraints of the intelligent cutting tool. This includes: The corresponding optimization space is labeled, and each target to be optimized is mapped to its own independent optimization space. Then, multi-level training is carried out in combination with the preset neural network. At this time, the preset neural network covers the dual driving capabilities of deep reinforcement learning and physical information, and trains each target to be optimized. In the initial training phase, the neural network establishes a preliminary response surface using massive amounts of historical data. In the deep training phase, the neural network embeds the physical laws of the grinding process as a penalty term into the loss function, thereby determining the nonlinear relationship between the grinding parameters of the intelligent cutting tool and the processing quality of the intelligent cutting tool.

7. The method for optimizing grinding parameters of an intelligent cutting tool according to claim 6, characterized in that, The process of inputting each target to be optimized into the corresponding optimization space, performing multi-level training in conjunction with a pre-set neural network, and establishing a nonlinear relationship between the grinding parameters and machining quality of the intelligent cutting tool during the training process, further combining the grinding constraint relationship of the intelligent cutting tool to determine the optimized grinding parameters, also includes: The nonlinear relationship is analyzed, and a grinding constraint relationship processing mechanism that combines multidimensional convex hull mechanism and tabu search is introduced in the analysis process. This allows for multi-factor matching of the nonlinear relationship and the grinding constraint relationship of the intelligent cutting tool. After matching, the optimized grinding parameters with anti-interference robustness are determined by combining the corresponding optimal solution set.

8. The method for optimizing grinding parameters of an intelligent cutting tool according to claim 1, characterized in that, The optimized grinding parameters are loaded into the simulation space of the intelligent cutting tool, and the corresponding dynamic simulation is triggered in conjunction with the dynamic parameter model of the intelligent cutting tool. During the dynamic simulation, multiple simulation results are output, and the final simulation result is determined by combining an adaptive optimization mechanism. The grinding parameters corresponding to the final simulation result are then marked, including: The optimized grinding parameters are loaded into the simulation space based on the digital twin architecture. The dynamic parameter model of the intelligent cutting tool is called synchronously. The dynamic parameter model is used as the boundary condition to trigger dynamic simulation covering multiple physics fields. During the dynamic simulation process, multiple simulation results are output, which cover the phase transformation trend of the microstructure, grinding burns, and edge defect morphology.

9. The method for optimizing grinding parameters of an intelligent cutting tool according to claim 8, characterized in that, The process of loading optimized grinding parameters into the simulation space of the intelligent cutting tool, triggering corresponding dynamic simulations based on the dynamic parameter model of the intelligent cutting tool, outputting multiple simulation results during the dynamic simulation, determining the final simulation result using an adaptive optimization mechanism, and marking the grinding parameters corresponding to the final simulation result, also includes: The embedded adaptive optimization mechanism is activated by the dynamic simulation of the intelligent cutting tool. This adaptive optimization mechanism adopts a strategy that combines variable step size hill climbing with Monte Carlo tree search to perform multi-dimensional edge perturbation testing and sensitivity analysis on the simulation results, automatically eliminate pseudo-optimal solutions, and finally determine the ultimate simulation result with the highest machining determinism. The grinding parameters corresponding to this result are then digitally verified, and the grinding parameters of the intelligent cutting tool are determined by tracing back to the final simulation result.

10. A smart cutting tool grinding parameter optimization system, characterized in that, The intelligent cutting tool grinding parameter optimization system is applied to the intelligent cutting tool grinding parameter optimization method as described in any one of claims 1-9; the intelligent cutting tool grinding parameter optimization system includes: The key area module is used to collect multiple morphological parameters of the intelligent cutting tool for the CNC cutting tool grinding center, and perform spatiotemporal alignment with the working parameters of the intelligent cutting tool during the grinding process. During the alignment process, a corresponding key parameter combination is formed, and multiple key areas of the intelligent cutting tool are determined by multi-level iteration along the key parameter combination. The optimization target module is used to determine multiple current parameters in different dimensions based on the dynamic identification of each key area, and to construct the dynamic parameter model of the intelligent blade by combining the current three-dimensional model of the intelligent blade. The dynamic parameter model is then optimized in multiple objectives, and multiple optimization targets are determined in combination with the application scenario of the intelligent blade. The optimization module is used to input each target to be optimized into the corresponding optimization space, and perform multi-level training in combination with the preset neural network. During the training process, the nonlinear relationship between the grinding parameters and the processing quality of the intelligent cutting tool is determined, and the optimized grinding parameters are further determined by combining the grinding constraint relationship of the intelligent cutting tool. The simulation module is used to load the optimized grinding parameters into the simulation space of the intelligent cutting tool, and trigger the corresponding dynamic simulation in combination with the dynamic parameter model of the intelligent cutting tool. During the dynamic simulation, multiple simulation results are output, and the final simulation result is determined by combining the adaptive optimization mechanism, and the grinding parameters corresponding to the final simulation result are marked.

Images