Magnetic core intelligent cutting parameter self-adaptive optimization system based on multi-mode sensing

Abstract

The invention provides a magnetic core intelligent cutting parameter self-adaptive optimization system based on multi-mode perception, and relates to the technical field of data processing.The method comprises the steps that a multi-mode sensor module is integrated on magnetic core cutting equipment, and the module comprises a force sensor, a visual sensor and a temperature sensor; the acquisition units are respectively used for acquiring cutting force dynamic signals, cutting track image sequences and cutter temperature time sequence data in real time; magnetic core surface texture features and three-dimensional contour data are captured through a visual sensor, and an initial cutting parameter set is generated in combination with a magnetic core material type recognition result, associated parameters in a historical process database and preset process constraint conditions; and first workpiece trial cutting is executed based on the initial cutting parameter set, multi-modal data fusion collection is synchronously started, cutting force frequency domain feature vectors, a tool temperature change rate curve and cutting surface defect image features are obtained, and multi-modal data are obtained. According to the invention, multi-objective collaborative optimization of processing efficiency and energy consumption is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent manufacturing and intelligent equipment, and in particular to a magnetic core intelligent cutting parameter adaptive optimization system based on multimodal perception. Background Art

[0002] In the field of magnetic core manufacturing, the accuracy and efficiency of the core cutting process directly affect the performance and production cost of electronic components. Traditional magnetic core cutting mostly relies on manual experience to set cutting parameters, which has problems such as parameter adjustment lag and unstable processing quality. Some automated cutting systems use a single sensor (such as relying only on force sensors or visual sensors). Because they cannot fully perceive complex factors such as fluctuations in material properties, tool wear, and environmental changes during the cutting process, their parameter optimization capabilities are limited and it is difficult to meet the production needs of high precision and high consistency. In addition, the existing cutting system lacks a dynamic closed-loop feedback mechanism, making it difficult to correct parameters in real time during continuous processing, which can easily lead to defects such as chipping and cracking, resulting in material waste and loss of production capacity. It is urgent to achieve intelligent adaptive optimization of cutting parameters through the integration of multiple technologies. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a magnetic core intelligent cutting parameter adaptive optimization system based on multimodal perception to achieve multi-objective collaborative optimization of processing efficiency and energy consumption.

[0004] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0005] In a first aspect, a method for adaptively optimizing magnetic core intelligent cutting parameters based on multimodal perception is provided, the method comprising:

[0006] S1: A multimodal sensor module is integrated on the core cutting equipment. The module includes a force sensor, a visual sensor, and a temperature sensor, which are used to respectively collect cutting force dynamic signals, cutting trajectory image sequences, and tool temperature time series data in real time;

[0007] S2: Capture the core surface texture features and 3D contour data through a visual sensor, combine the core material type identification results, associated parameters in the historical process database, and preset process constraints to generate an initial cutting parameter set;

[0008] S3: Perform the first trial cutting based on the initial cutting parameter set, and simultaneously start multimodal data fusion acquisition to obtain the cutting force frequency domain feature vector, tool temperature change rate curve and cutting surface defect image features to obtain multimodal data;

[0009] S4: Perform feature-level fusion processing on the collected multimodal data, including time-frequency joint analysis of cutting force signals, calculation of fractal dimension of cutting surface images, and derivation of tool thermal deformation compensation coefficients to generate a multidimensional process state representation matrix;

[0010] S5: Inputting the multi-dimensional process state representation matrix into a pre-trained cutting parameter optimization model, setting a first monitoring point and a second monitoring point in the tool-material contact area, respectively obtaining cutting force distribution difference characteristics and tool temperature gradient change data in real time, combining the historical optimal parameter mapping relationship in the process knowledge base, and performing a comparative analysis of the dynamic characteristics of the two monitoring points to correct the tool-material interaction coefficient in the model, and outputting an adaptive parameter group including a speed compensation amount, a feed correction factor, and a tool angle optimization value;

[0011] S6: Adjust the operating status of the cutting equipment in real time according to the adaptive parameter group, and establish a closed-loop feedback mechanism during the continuous processing process, and realize iterative optimization of cutting parameters and process stability control through the multimodal data dynamic monitoring window.

[0012] Furthermore, a multimodal sensor module is integrated on the core cutting equipment. The module includes a force sensor, a visual sensor, and a temperature sensor, which are respectively used to collect cutting force dynamic signals, cutting trajectory image sequences, and tool temperature time series data in real time, including:

[0013] S1.1: Configure the data bus of the multimodal sensor based on a time-sharing multiplexing strategy. Ensure that the conflict rate of concurrent acquisition of cutting force signals, image sequences, and temperature data is less than 0.5% through the data transmission timing of the force sensor, vision sensor, and temperature sensor.

[0014] S1.2: Deploy an FPGA coprocessor at the sensor end, build a double-buffer queue structure to cache asynchronous data, and trigger a synchronization signal based on the tool displacement encoder to achieve microsecond-level timing alignment of the cutting force sampling time, image frame exposure time, and temperature acquisition point.

[0015] Furthermore, the core surface texture features and 3D contour data are captured by a visual sensor. Combined with the core material type identification results, associated parameters in the historical process database, and preset process constraints, an initial cutting parameter set is generated, including:

[0016] S2.1: Based on the stereo image sequences collected by the vision sensor, the 3D point cloud of the magnetic core is reconstructed using adaptive structured light coding technology. An improved iterative closest point algorithm is used to match the preset process reference model to achieve a contour positioning accuracy of ±0.03mm.

[0017] S2.2: Based on the contour positioning accuracy, the grayscale distribution matrix of the core surface is extracted. The texture quality evaluation index is constructed by combining the wavelet transform and the local binary pattern fusion feature to dynamically identify abnormal areas with grain orientation deviation exceeding 5°.

[0018] S2.3: Based on the abnormal area, a lightweight material recognition model is deployed to receive the texture feature vector after edge preprocessing. The model then fuses the historical material map features through knowledge distillation technology to output the material type and the corresponding process parameter confidence level.

[0019] S2.4: Based on the material identification results, the historical process database is called and a dynamic time warping algorithm is used to match the 3D contour characteristic curves of similar working conditions. Combined with the current tool wear compensation coefficient, the initial cutting speed, feed rate, and tool inclination angle parameters are generated.

[0020] S2.5: Based on the initial cutting speed, feed rate, and tool inclination parameters, a cutting parameter feasibility verification model is established. The cutting stress distribution is predicted based on finite element simulation data. The parameter set is dynamically adjusted to meet the preset constraints and an initial parameter set with a safety margin greater than 15% is output.

[0021] Furthermore, the first trial cutting is performed based on the initial cutting parameter set, and multimodal data fusion acquisition is started simultaneously to obtain the cutting force frequency domain feature vector, tool temperature change rate curve and cutting surface defect image features, and obtain multimodal data, including:

[0022] S3.1: Based on the execution sequence of the initial cutting parameter set, a synchronous trigger pulse signal is generated through the tool spindle encoder to jointly control the force sensor sampling window, the visual sensor exposure time, and the temperature sensor acquisition cycle to ensure the spatiotemporal consistency of multimodal data;

[0023] S3.2: Based on the spatiotemporal consistency of the data, a parallel computing channel was constructed to perform fast Fourier transform and short-time energy analysis on the cutting force signal, extracting the energy proportion of the 0.5-1.5kHz characteristic frequency band. A sliding window difference method was used on the tool temperature time series data to calculate the temperature change rate gradient. An improved U-Net segmentation network was applied to the cut surface image to detect the edge chipping defect area and crack extension direction in real time.

[0024] S3.3: By detecting the chipping defect area and crack extension direction, a dynamic mapping relationship between multimodal data and process parameters is established. Based on the deviation value between the cutting force frequency domain characteristics and the initial parameter set, the similarity between the tool temperature change rate and the preset thermal stability curve, and the matching degree between the cutting surface defect distribution and the three-dimensional contour prediction model, a process state evaluation matrix is ​​generated to obtain multimodal data.

[0025] Furthermore, the collected multimodal data is subjected to feature-level fusion processing, including time-frequency joint analysis of cutting force signals, calculation of the fractal dimension of the cutting surface image, and derivation of the tool thermal deformation compensation coefficient, to generate a multidimensional process state representation matrix, including:

[0026] S4.1: Perform a modified Hilbert-Huang transform on the cutting force signal, extract 6-8 intrinsic mode functions through adaptive noise-assisted empirical mode decomposition, and construct a time-frequency energy density spectrum based on the instantaneous frequency calculation to quantify the energy entropy distribution characteristics in the 0.5-2 kHz frequency band;

[0027] S4.2: Based on the energy entropy distribution characteristics, an improved multi-scale box counting method is used to analyze the cutting surface image. The mesh density is adaptively adjusted based on the tool feed direction. The anisotropic fractal index is calculated by combining the grayscale gradient co-occurrence matrix, and the fractal dimension calculation threshold is dynamically set.

[0028] S4.3: Based on the fractal dimension, the threshold is calculated and the tool temperature gradient data is integrated with the finite element heat conduction simulation results to establish a dynamic prediction model for the axial / radial thermal expansion coefficient. Through the regression analysis of the temperature change rate and cutting parameters, the tool thermal deformation compensation coefficient is derived.

[0029] S4.4: Based on the tool thermal deformation compensation coefficient, a feature weighted fusion network based on the attention mechanism is designed to dynamically normalize the time-frequency energy entropy, fractal dimension, and thermal compensation coefficient to generate a 16-dimensional process state feature vector.

[0030] S4.5: Map the eigenvectors to the process parameter space, and construct a three-dimensional process state representation matrix including the cutting stability index, surface quality score, and thermal-mechanical coupling coefficient through joint optimization of principal component analysis and kernel density estimation.

[0031] Furthermore, the multi-dimensional process state representation matrix is ​​input into a pre-trained cutting parameter optimization model, and a first monitoring point and a second monitoring point are set in the tool-material contact area to respectively obtain the cutting force distribution difference characteristics and tool temperature gradient change data in real time. Combined with the historical optimal parameter mapping relationship in the process knowledge base, the tool-material interaction coefficient in the model is corrected through dynamic feature comparison and analysis of the two monitoring points, and an adaptive parameter group including speed compensation, feed correction factor and tool angle optimization value is output, including:

[0032] S5.1: Set the first and second monitoring points at the leading and trailing edges of the tool-material contact area, respectively. Use a high-density strain gauge array to capture the spatial distribution of cutting forces in real time. Combined with an infrared thermal imager, simultaneously obtain the temperature gradient change rate at the two monitoring points.

[0033] S5.2: Based on the temperature gradient change rate and the multidimensional process state representation matrix, an improved grey correlation analysis method is used to calculate the difference in cutting force distribution between two monitoring points, a dynamic correlation coefficient matrix of the temperature gradient change rate is constructed, and the weight coefficient of the comparative analysis is optimized using a genetic algorithm;

[0034] S5.3: By comparing and analyzing the weight coefficients, a dynamic feedback layer is embedded in the pre-trained cutting parameter optimization model. The tool-material interaction coefficient is updated online based on the quantified results of the feature differences between the two monitoring points and the historical optimal parameter mapping relationship in the process knowledge base.

[0035] S5.4: Based on the interaction coefficient, design a multi-objective optimizer based on the NSGA-II algorithm. With cutting efficiency, surface quality, and tool life as optimization objectives, and combined with the modified interaction coefficient, iteratively solve the Pareto optimal solution set containing speed compensation, feed correction factor, and tool angle optimization value.

[0036] S5.5: Based on the Pareto optimal solution set, verify the feasibility of the parameter group through digital twin system simulation, dynamically adjust the parameter weight distribution based on real-time working condition data, and output an adaptive parameter group that meets the process constraints.

[0037] Furthermore, the operating status of the cutting equipment is adjusted in real time according to the adaptive parameter group, and a closed-loop feedback mechanism is established during the continuous processing process. The iterative optimization of cutting parameters and process stability control are achieved through the multimodal data dynamic monitoring window, including:

[0038] S6.1: Generates a device control instruction stream based on the adaptive parameter set, and synchronously sends the speed compensation, feed correction factor, and tool angle optimization value to the cutting device controller via the time-sensitive network protocol, achieving precise matching of parameter adjustment and processing timing;

[0039] S6.2: By precisely matching parameter adjustments with machining timing, the length of the multimodal data acquisition window is adaptively adjusted according to the characteristics of the machining stage. A sliding window algorithm is used to extract the cutting force standard deviation, temperature fluctuation rate, and defect density gradient as process stability evaluation indicators.

[0040] S6.3: Construct a fuzzy comprehensive evaluation model for process stability, integrating the real-time cutting vibration spectrum, tool wear characteristic vector, and environmental disturbance factors. Dynamically predict process deviation trends through a cloud-based digital twin system to generate a stability index.

[0041] S6.4: Design a three-level fault-tolerant mechanism. When the stability index is detected to be lower than 0.7, parameter fine-tuning, machining path optimization, and emergency rollback are triggered in sequence to ensure seamless connection of continuous machining processes.

[0042] S6.5: The multimodal data, parameter adjustment records, and processing effect evaluations generated during the closed-loop optimization process are synchronously transmitted back to the process knowledge base. By comparing the simulation predictions with the actual results, the finite element model boundary conditions and optimization algorithm weight coefficients are automatically corrected.

[0043] S6.6: An energy efficiency evaluation module is embedded in the parameter iterative optimization process. Based on the real-time monitoring data of cutting power and the equipment energy consumption model, the processing efficiency and energy consumption are dynamically balanced to achieve the optimization goal of reducing the energy consumption per unit workpiece by ≥15%.

[0044] Secondly, the magnetic core intelligent cutting parameter adaptive optimization system based on multimodal perception includes:

[0045] An acquisition module is used to integrate a multimodal sensor module on the magnetic core cutting equipment. The module includes a force sensor, a visual sensor, and a temperature sensor, which are respectively used to collect real-time cutting force dynamic signals, cutting trajectory image sequences, and tool temperature time series data. The visual sensor captures the surface texture features and three-dimensional contour data of the magnetic core, and combines the results of the magnetic core material type identification, the associated parameters in the historical process database, and the preset process constraints to generate an initial cutting parameter set.

[0046] The fusion module is used to perform the first trial cutting based on the initial cutting parameter set and simultaneously start the multimodal data fusion acquisition to obtain the cutting force frequency domain feature vector, the tool temperature change rate curve and the cutting surface defect image characteristics to obtain multimodal data. The collected multimodal data is subjected to feature-level fusion processing, including time-frequency joint analysis of the cutting force signal, calculation of the fractal dimension of the cutting surface image and derivation of the tool thermal deformation compensation coefficient, to generate a multidimensional process state representation matrix.

[0047] A processing module is used to input the multidimensional process state characterization matrix into a pre-trained cutting parameter optimization model, set a first monitoring point and a second monitoring point in the tool-material contact area, respectively obtain the cutting force distribution difference characteristics and tool temperature gradient change data in real time, combine the historical optimal parameter mapping relationship in the process knowledge base, and compare and analyze the dynamic characteristics of the two monitoring points to correct the tool-material interaction coefficient in the model, and output an adaptive parameter group including a speed compensation amount, a feed correction factor and a tool angle optimization value; adjust the cutting equipment operation state in real time according to the adaptive parameter group, and establish a closed-loop feedback mechanism during the continuous processing process, so as to realize iterative optimization of cutting parameters and process stability control through a multimodal data dynamic monitoring window.

[0048] According to a third aspect, a computing device includes:

[0049] one or more processors;

[0050] The storage device is used to store one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement the method.

[0051] In a fourth aspect, a computer-readable storage medium stores a program, which implements the method when executed by a processor.

[0052] The above solution of the present invention includes at least the following beneficial effects:

[0053] Multimodal sensors collect multidimensional data in real time and deeply integrate it, achieving comprehensive and accurate perception of the magnetic core cutting process. Combined with advanced algorithms and a process knowledge base, this system dynamically generates and optimizes cutting parameters, effectively resolving the issues of lag in parameter adjustment and unstable processing quality associated with traditional cutting. The system's closed-loop feedback mechanism not only improves cutting efficiency and surface quality, but also reduces material loss and energy consumption through a three-level fault tolerance and energy efficiency assessment module. This achieves breakthroughs in both production efficiency and energy conservation, providing an innovative solution for the intelligent upgrade of magnetic core manufacturing. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flow chart of a method for adaptive optimization of magnetic core intelligent cutting parameters based on multimodal perception provided by an embodiment of the present invention.

[0055] Figure 2 Schematic diagram of a magnetic core intelligent cutting parameter adaptive optimization system based on multimodal sensing provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0056] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0057] like Figure 1 As shown, an embodiment of the present invention proposes a method for adaptively optimizing magnetic core intelligent cutting parameters based on multimodal perception, the method comprising the following steps:

[0058] Step S1: integrating a multimodal sensor module on the magnetic core cutting equipment, wherein the module includes a force sensor, a visual sensor, and a temperature sensor, which are respectively used to collect cutting force dynamic signals, cutting trajectory image sequences, and tool temperature time series data in real time;

[0059] Step S2: Capturing the surface texture features and three-dimensional contour data of the core through a visual sensor, and combining the core material type identification results, associated parameters in the historical process database, and preset process constraints to generate an initial cutting parameter set;

[0060] Step S3: Perform the first trial cutting based on the initial cutting parameter set, and simultaneously start multimodal data fusion acquisition to obtain the cutting force frequency domain feature vector, tool temperature change rate curve and cutting surface defect image features to obtain multimodal data;

[0061] Step S4: performing feature-level fusion processing on the collected multimodal data, including time-frequency joint analysis of the cutting force signal, calculation of the fractal dimension of the cutting surface image, and derivation of the tool thermal deformation compensation coefficient, to generate a multidimensional process state representation matrix;

[0062] Step S5: Inputting the multidimensional process state representation matrix into a pre-trained cutting parameter optimization model, setting a first monitoring point and a second monitoring point in the tool-material contact area, respectively acquiring cutting force distribution difference characteristics and tool temperature gradient change data in real time, combining the historical optimal parameter mapping relationship in the process knowledge base, and performing a comparative analysis of the dynamic characteristics of the two monitoring points to correct the tool-material interaction coefficient in the model, and outputting an adaptive parameter group including a speed compensation amount, a feed correction factor, and a tool angle optimization value;

[0063] Step S6: Adjust the operating status of the cutting equipment in real time according to the adaptive parameter group, establish a closed-loop feedback mechanism during the continuous processing, and realize iterative optimization of cutting parameters and process stability control through the multimodal data dynamic monitoring window.

[0064] In an embodiment of the present invention, a multimodal sensor module is integrated to realize real-time multidimensional data collection of cutting force, cutting trajectory image and tool temperature, and high-precision initial cutting parameters are generated by combining visual perception and material recognition. Then, through multimodal data fusion analysis during the first trial cutting process, a multidimensional process state representation matrix is ​​constructed to drive the dynamic correction of the cutting parameter optimization model. This method realizes adaptive adjustment of cutting parameters by comparing the dynamic characteristics of two monitoring points in the tool-material contact area, and establishes a closed-loop feedback mechanism in continuous processing, effectively improving the real-time and accuracy of cutting parameter adjustment, solving the problems of parameter lag and unstable processing quality in traditional methods, significantly improving the surface quality, processing efficiency and process stability of magnetic core cutting, and achieving synergistic improvement of tool life and energy consumption through multi-objective optimization.

[0065] In a preferred embodiment of the present invention, the above step S1 may include:

[0066] S1.1: Configure the data bus of the multimodal sensor based on a time-sharing multiplexing strategy. Ensure that the conflict rate of concurrent acquisition of cutting force signals, image sequences, and temperature data is less than 0.5% through the data transmission timing of the force sensor, vision sensor, and temperature sensor.

[0067] S1.2: Deploy an FPGA coprocessor at the sensor end, build a double-buffer queue structure to cache asynchronous data, and trigger a synchronization signal based on the tool displacement encoder to achieve microsecond-level timing alignment of the cutting force sampling time, image frame exposure time, and temperature acquisition point.

[0068] In an embodiment of the present invention, a time-sharing multiplexing strategy is used to configure the data bus of multimodal sensors, combined with an FPGA coprocessor and a double-buffered queue structure, to achieve efficient collaborative acquisition of multi-source heterogeneous data. Specifically, the time-sharing multiplexing strategy controls the concurrent acquisition conflict rate to below 0.5% by accurately planning the data transmission timing of force, vision, and temperature sensors, significantly improving the reliability of data acquisition and system stability. The deployment of the FPGA coprocessor and double-buffered queue not only achieves real-time caching and scheduling of asynchronous data through hardware acceleration, but also controls the timing error of cutting force sampling, image exposure, and temperature acquisition to the microsecond level based on the synchronous signal trigger mechanism of the tool displacement encoder. This ensures the precise alignment of the force, vision, and thermal multi-dimensional features during the cutting process, avoids misjudgment of process status due to data asynchrony, and provides high-precision data support for initial parameter generation and dynamic optimization.

[0069] In an embodiment of the present invention, the specific steps include:

[0070] Step S1.1: Time-division multiplexing strategy: By pre-planning the data transmission timing of the three types of sensors (such as allocating independent time slices for each sensor), multiple sensors are prevented from sending data to the data bus at the same time, and the signal conflict rate is controlled below 0.5%.

[0071] Data bus coordination: Establish a sensor priority mechanism (such as real-time transmission of cutting force signals and compressed transmission of image sequences in batches), and ensure that different types of data (analog signals, image data, digital signals) are transmitted efficiently and in sequence through time-slice polling or dynamic scheduling algorithms.

[0072] Step S1.2: FPGA hardware acceleration: Deploy a field programmable gate array (FPGA) coprocessor at the sensor end and use its parallel computing capability to build a double buffer queue (such as a ping-pong buffer structure) to cache the asynchronous data of each sensor in real time to avoid data loss or disorder.

[0073] Synchronous signal triggering: Using the pulse signal of the tool displacement encoder as the global synchronization benchmark, when the encoder detects that the tool has reached the preset position (such as the cutting starting point), it sends a trigger signal to all sensors, forcing the unification of the cutting force sampling time, visual sensor exposure time and temperature collection point, to achieve microsecond-level timing alignment accuracy.

[0074] Asynchronous data calibration: The cached data is timestamped through a double-buffered queue, and the data across time slices is calibrated using an interpolation algorithm.

[0075] In a preferred embodiment of the present invention, the above step S2 may include:

[0076] S2.1: Based on the stereo image sequences collected by the vision sensor, the 3D point cloud of the magnetic core is reconstructed using adaptive structured light coding technology. An improved iterative closest point algorithm is used to match the preset process reference model to achieve a contour positioning accuracy of ±0.03mm.

[0077] S2.2: Based on the contour positioning accuracy, the grayscale distribution matrix of the core surface is extracted. The texture quality evaluation index is constructed by combining the wavelet transform and the local binary pattern fusion feature to dynamically identify abnormal areas with grain orientation deviation exceeding 5°.

[0078] S2.3: Based on the abnormal area, a lightweight material recognition model is deployed to receive the texture feature vector after edge preprocessing. The model then fuses the historical material map features through knowledge distillation technology to output the material type and the corresponding process parameter confidence level.

[0079] S2.4: Based on the material identification results, the historical process database is called and a dynamic time warping algorithm is used to match the 3D contour characteristic curves of similar working conditions. Combined with the current tool wear compensation coefficient, the initial cutting speed, feed rate, and tool inclination angle parameters are generated.

[0080] S2.5: Based on the initial cutting speed, feed rate, and tool inclination parameters, a cutting parameter feasibility verification model is established. The cutting stress distribution is predicted based on finite element simulation data. The parameter set is dynamically adjusted to meet the preset constraints and an initial parameter set with a safety margin greater than 15% is output.

[0081] In this embodiment of the present invention, intelligent and precise generation of initial parameters for core cutting is achieved through the integration of multiple technologies. Adaptive structured light encoding and an improved iterative closest point algorithm are used to achieve ±0.03mm-level 3D core contour positioning, providing a high-precision geometric benchmark for parameter generation. Texture analysis based on wavelet transform and local binary patterns dynamically identifies abnormal regions with grain orientation deviations exceeding 5°. A lightweight material identification model and knowledge distillation technology are combined to rapidly output material type and process parameter confidence. A dynamic time warping algorithm is used to match historical process data, and tool wear compensation and finite element simulation verification are introduced to ensure that the initial parameter set not only matches the core material properties and contour characteristics, but also passes a safety margin of ≥15%, effectively avoiding defects such as edge chipping and stress concentration caused by rough parameter settings. This process deeply couples geometric feature analysis, intelligent material identification, reuse of historical experience, and physical simulation verification, significantly improving the rationality and reliability of initial cutting parameters. This lays a high-precision data foundation for subsequent adaptive optimization, significantly reducing manual trial-and-error costs and increasing the success rate of first-part processing.

[0082] In an embodiment of the present invention, the specific steps include:

[0083] Step S2.1: Stereo image acquisition: A multi-angle stereo image sequence of the magnetic core is acquired through a visual sensor (such as a structured light camera), and a structured light pattern is projected onto the surface of the magnetic core using adaptive structured light coding technology (such as dynamic fringe projection) to obtain a deformed fringe image containing depth information.

[0084] 3D point cloud reconstruction: Decode the deformed fringe image to generate 3D point cloud data of the core surface.

[0085] Benchmark model matching: The improved iterative closest point (ICP) algorithm is used to match the reconstructed 3D point cloud with the preset process benchmark model (such as the standard core CAD model). By iteratively optimizing the spatial position relationship between the point cloud and the model, the contour positioning accuracy of ±0.03mm is achieved.

[0086] Step S2.2: Grayscale distribution matrix extraction: Based on the contour positioning results, the core surface ROI (region of interest) is intercepted from the visual image, converted into a grayscale image and a grayscale distribution matrix is ​​constructed.

[0087] Texture feature fusion:

[0088] Wavelet transform: Perform multi-scale wavelet decomposition on the grayscale matrix to extract texture detail features (such as edges and roughness) at different frequencies.

[0089] Local Binary Pattern (LBP): By Calculate the LBP value of each pixel, where g c Indicates the grayscale value of the current center pixel, gi It represents the grayscale value of 8 surrounding pixels in a 3×3 neighborhood centered on the central pixel. s is the sign function, which generates a texture pattern histogram to characterize the local texture structure.

[0090] Abnormal area identification: Wavelet and LBP features are integrated to construct texture quality evaluation indicators (such as texture consistency index), set the grain orientation deviation threshold (such as 5°), and dynamically mark abnormal areas (such as texture disorder and irregular grain arrangement areas) through threshold comparison.

[0091] Step S2.3: Texture feature preprocessing: perform edge detection (such as Canny operator) and normalization processing on the texture image of the abnormal area to extract the edge contour feature vector.

[0092] Lightweight model deployment: Use a lightweight convolutional neural network (such as MobileNet) or a machine learning model (such as random forest) as the material recognition model and input the preprocessed feature vector.

[0093] Knowledge distillation and fusion: Through knowledge distillation technology, the prior knowledge of historical material maps (such as the texture feature database of known materials) is integrated into model training, so that the model outputs the material type (such as ferrite, silicon steel, etc.) and the confidence level of the corresponding process parameters (such as the credible probability of the recommended cutting speed value).

[0094] Step S2.4: Historical data retrieval: Based on the material identification results, the cutting records of the same or similar materials are retrieved from the historical process database, and the corresponding three-dimensional contour feature curves (such as contour complexity and curvature distribution) are extracted.

[0095] Dynamic Time Warping (DTW) matching: Pass Among them, d(q i ,c j ) represents the local distance between the i-th point in Q and the j-th point in C, i is the i-th point of the current core contour characteristic curve Q, j is the j-th point of the contour characteristic curve C in the historical data, and the three-dimensional contour characteristic curve of the current core is matched with the curve in the historical data for similarity, and the Top-N similar working conditions are screened out.

[0096] Parameter compensation adjustment: Obtain the initial cutting parameters of similar working conditions, combine them with the current tool wear status (such as obtaining the wear amount through tool visual inspection), introduce a wear compensation coefficient (such as for every 0.01mm increase in tool wear, the feed rate is reduced by 2%), and generate the initial cutting speed, feed rate and tool inclination angle parameters.

[0097] Step S2.5: Finite element simulation modeling: Based on the initial parameters, a finite element model of the core cutting process is established, material properties (such as hardness, thermal conductivity) and tool parameters are input, and physical quantities such as cutting stress distribution and temperature field are simulated and predicted.

[0098] Parameter feasibility check: Compare the simulation results with the preset constraints (such as maximum allowable stress and temperature threshold). If stress concentration or temperature exceeds the limit, adjust the parameters (such as reducing the feed rate and increasing the tool inclination angle).

[0099] Safety margin assessment: The safety margin is equal to the ratio of (allowable stress minus actual stress) to the allowable stress, multiplied by 100%, to calculate the safety margin after parameter adjustment (such as the ratio of the difference between the actual stress and the allowable stress), ensuring that the safety margin is greater than 15%, and finally outputting the initial parameter set that meets the requirements.

[0100] In a preferred embodiment of the present invention, the above step S3 may include:

[0101] S3.1: Based on the execution sequence of the initial cutting parameter set, a synchronous trigger pulse signal is generated through the tool spindle encoder to jointly control the force sensor sampling window, the visual sensor exposure time, and the temperature sensor acquisition cycle to ensure the spatiotemporal consistency of multimodal data;

[0102] S3.2: Based on the spatiotemporal consistency of the data, a parallel computing channel was constructed to perform fast Fourier transform and short-time energy analysis on the cutting force signal, extracting the energy proportion of the 0.5-1.5kHz characteristic frequency band. A sliding window difference method was used on the tool temperature time series data to calculate the temperature change rate gradient. An improved U-Net segmentation network was applied to the cut surface image to detect the edge chipping defect area and crack extension direction in real time.

[0103] S3.3: By detecting the chipping defect area and crack extension direction, a dynamic mapping relationship between multimodal data and process parameters is established. Based on the deviation value between the cutting force frequency domain characteristics and the initial parameter set, the similarity between the tool temperature change rate and the preset thermal stability curve, and the matching degree between the cutting surface defect distribution and the three-dimensional contour prediction model, a process state evaluation matrix is ​​generated to obtain multimodal data.

[0104] In an embodiment of the present invention, the precise collection and in-depth analysis of multimodal data during the first trial cutting process are achieved through the collaboration of multiple technologies: the synchronous trigger mechanism of the tool spindle encoder is utilized to strictly align the acquisition timing of the force, vision, and temperature sensors with the cutting action to ensure the temporal and spatial consistency of the data and avoid feature dislocation caused by asynchronous acquisition; multi-source data are differentially processed through parallel computing channels, such as using fast Fourier transform to extract cutting force frequency domain characteristics, sliding window difference method to analyze temperature change rate, and improved U-Net network to detect cutting defects in real time, thereby achieving efficient conversion from raw signals to process state characteristics; a process state evaluation matrix constructed based on cutting force deviation, temperature similarity, and defect matching degree dynamically associates multimodal data with initial parameters, which not only provides a multi-dimensional feedback basis for subsequent parameter optimization, but also identifies potential process risks (such as stress concentration and thermal deformation trend) in advance through quantitative analysis, significantly improving the process verification efficiency and data support capabilities of the first trial cutting, and laying a precise state perception foundation for adaptive parameter optimization.

[0105] In an embodiment of the present invention, the specific steps include:

[0106] Step S3.1: Synchronous trigger mechanism: The tool position and speed are monitored in real time through the tool spindle encoder (installed on the tool rotation axis). When the encoder detects that the tool reaches the preset cutting starting point or a specific angle, a synchronous trigger pulse signal (such as a TTL level signal) is generated.

[0107] Sensor linkage control:

[0108] Force sensor: After receiving the trigger signal, the sampling window is opened to collect the dynamic signal of the cutting force at a fixed frequency (such as 10kHz) until the trigger signal ends.

[0109] Vision sensor: controls the exposure time according to the trigger signal to ensure that the collected cutting trajectory image corresponds to the current position of the tool (such as exposing once per revolution, or exposing at a specific displacement interval during feed motion).

[0110] Temperature sensor: Synchronously adjusts the acquisition cycle according to the trigger signal (such as switching from low-frequency sampling to high-frequency sampling) to ensure that the tool temperature data accurately matches the cutting stage.

[0111] Temporal and spatial consistency assurance: The trigger signal is used to force the unified starting point of cutting force sampling, image exposure, and temperature acquisition, and combined with the position data of the tool displacement encoder (such as X / Y / Z axis coordinates), temporal and spatial labels (time stamp plus spatial coordinates) are added to all data to ensure that the data can be traced back to the specific cutting position and time during subsequent analysis.

[0112] Step S3.2: Cutting force signal processing Fast Fourier Transform (FFT):

[0113] The time domain cutting force signal is converted into the frequency domain to obtain the energy distribution (such as power spectrum density) in the 0.5-1.5kHz frequency band, and the proportion of the energy in this frequency band to the total energy is extracted (reflecting the vibration characteristics of the cutting process, such as tool-material resonance).

[0114] Short-term energy analysis:

[0115] The signal is divided into overlapping time windows (e.g., 50 ms / window), and the signal energy in each window is obtained to detect sudden changes in cutting force (e.g., the moment when the tool cuts in / out).

[0116] Tool temperature data analysis sliding window difference method:

[0117] Apply a sliding window (e.g., a window size of 10 sampling points) to the temperature time series data and calculate the temperature change rate at adjacent time points: the temperature change rate is equal to the difference between the temperature value at the current moment and the temperature value at the previous moment, divided by the time interval, and the maximum and average temperature gradients are extracted to characterize the thermal response speed and thermal stability of the tool.

[0118] Step S3.3: Dynamic mapping relationship establishment:

[0119] Cutting force deviation: Calculate the absolute or relative deviation (e.g., deviation rate = |measured value - preset value| / preset value) between the measured frequency domain characteristics (e.g., 0.5-1.5kHz energy ratio) and the preset value of the initial parameter set.

[0120] Temperature similarity: Dynamic time warping (DTW) is performed on the measured temperature rate of change curve and the preset thermal stability curve (such as the temperature curve under the historical optimal operating conditions). The similarity score is calculated by subtracting the ratio of the total DTW distance to the maximum possible distance from 1, and then multiplying the result by 100. (The score ranges from 0 to 100, with higher scores indicating closer stability.)

[0121] Defect matching: Compare the degree of agreement between the defect distribution of the cut surface (such as chipping location and crack direction) and the 3D contour prediction model (based on the finite element simulation results of the initial parameters), and quantify the matching degree using the intersection over union (IoU) or Hausdorff distance.

[0122] Evaluation matrix generation:

[0123] After the above three indicators (deviation rate, similarity, and matching degree) are normalized (such as mapping to the [0,1] interval), they are arranged in rows or columns to form a process status evaluation matrix.

[0124] Multimodal data fusion:

[0125] The evaluation matrix is ​​combined with the original multimodal data (such as cutting force time domain waveform, temperature curve, defect image) to generate a comprehensive data set containing time domain, frequency domain, and image domain features.

[0126] In a preferred embodiment of the present invention, the above step S4 may include:

[0127] S4.1: Perform a modified Hilbert-Huang transform on the cutting force signal, extract 6-8 intrinsic mode functions through adaptive noise-assisted empirical mode decomposition, and construct a time-frequency energy density spectrum based on the instantaneous frequency calculation to quantify the energy entropy distribution characteristics in the 0.5-2 kHz frequency band;

[0128] S4.2: Based on the energy entropy distribution characteristics, an improved multi-scale box counting method is used to analyze the cutting surface image. The mesh density is adaptively adjusted based on the tool feed direction. The anisotropic fractal index is calculated by combining the grayscale gradient co-occurrence matrix, and the fractal dimension calculation threshold is dynamically set.

[0129] S4.3: Based on the fractal dimension, the threshold is calculated and the tool temperature gradient data is integrated with the finite element heat conduction simulation results to establish a dynamic prediction model for the axial / radial thermal expansion coefficient. Through the regression analysis of the temperature change rate and cutting parameters, the tool thermal deformation compensation coefficient is derived.

[0130] S4.4: Based on the tool thermal deformation compensation coefficient, a feature weighted fusion network based on the attention mechanism is designed to dynamically normalize the time-frequency energy entropy, fractal dimension, and thermal compensation coefficient to generate a 16-dimensional process state feature vector.

[0131] S4.5: Map the eigenvectors to the process parameter space, and construct a three-dimensional process state representation matrix including the cutting stability index, surface quality score, and thermal-mechanical coupling coefficient through joint optimization of principal component analysis and kernel density estimation.

[0132] In an embodiment of the present invention, deep feature mining and efficient fusion of multimodal data are achieved through the integration of multiple technologies and innovative algorithms. The improved Hilbert-Huang transform and adaptive noise-assisted empirical mode decomposition accurately extract the time-frequency energy characteristics of the cutting force signal, quantify the energy entropy of the key frequency band, and provide a dynamic basis for cutting state evaluation; the improved multi-scale box counting method is combined with the grayscale gradient co-occurrence matrix to adaptively analyze the cutting surface image according to the tool feed direction, realize the dynamic calculation of the fractal dimension, and accurately capture the surface quality characteristics; the thermal expansion coefficient prediction model established by integrating temperature and simulation data realizes the scientific derivation of the tool thermal deformation compensation coefficient. On this basis, the weighted fusion network based on the attention mechanism dynamically integrates multi-source features to generate high-dimensional feature vectors, and after principal component analysis and kernel density estimation optimization, a three-dimensional process state characterization matrix including cutting stability, surface quality and thermal-mechanical coupling is constructed. It not only comprehensively and accurately characterizes the cutting process state, but also provides high-quality input for the subsequent parameter optimization model, effectively improving the intelligent analysis and decision-making capabilities of the core cutting process.

[0133] In an embodiment of the present invention, the specific steps include:

[0134] Step S4.1: Improved Hilbert-Huang transform (HHT): Adaptive noise-assisted empirical mode decomposition (EEMD) is performed on the cutting force time domain signal. By adding white noise to suppress modal aliasing, 6-8 intrinsic mode functions (IMFs) are decomposed. Each IMF represents a signal with different frequency components. The instantaneous frequency is calculated for each IMF (such as the derivative of the instantaneous phase obtained through Hilbert transform), and a time-frequency energy density spectrum is constructed (the horizontal axis is time, the vertical axis is frequency, and color / grayscale represents energy density).

[0135] Energy entropy quantification:

[0136] Focusing on the 0.5-2kHz frequency band, the time-frequency energy density spectrum is divided into multiple frequency sub-bands, and the energy proportion of each sub-band is calculated using the above formula.

[0137] Based on the information entropy theory, the energy entropy of this frequency band is calculated: Among them, p i It represents the energy proportion of the ith frequency subband, n represents the total number of frequency subbands, and ln represents the natural logarithm.

[0138] Step S4.2: Improved multi-scale box counting method:

[0139] The cutting surface image is converted into a grayscale image and divided into multiple rectangular ROIs (regions of interest) based on the tool feed direction (e.g., X-axis). A smaller grid density (e.g., 10 × 10 pixels / box) is used in the feed direction and a larger density (e.g., 20 × 20 pixels / box) is used in the vertical direction to adapt to the directionality of the cutting texture.

[0140] For each scale of the grid, the minimum number of boxes N(r) required to cover the non-zero pixels of the image is calculated, and the fractal dimension D (reflecting the surface roughness, the larger the D, the rougher the surface) is obtained by logarithmic fitting of the slope of lnN(r) to ln(1 / r).

[0141] Step S4.3: Fusion of heat conduction simulation and measured data:

[0142] A tool heat conduction model is established through finite element simulation (such as ANSYS), and cutting parameters (such as feed rate and cutting speed) and material properties are input to predict the tool's axial (feed direction) and radial (vertical direction) thermal expansion coefficients. Combined with the actual measured data of the tool temperature gradient (such as infrared thermal imager or thermocouple measurement values), the boundary conditions of the simulation model (such as the convection heat transfer coefficient) are corrected to establish a dynamic prediction model: α(t) = f(T(t), v, f), where T(t) is the real-time temperature, v is the cutting speed, and f is the feed rate.

[0143] Regression analysis and compensation coefficient calculation: Through multiple linear regression or machine learning algorithms (such as random forest), the correlation between temperature change rate and cutting parameters is analyzed to derive the thermal deformation compensation coefficient k c (For example, the change in tool size for every 1°C increase): Where ΔL is the change in tool size, L is the original length of the tool, and ΔT is the temperature change.

[0144] Step S4.4: Attention Mechanism Network Design: Construct a three-layer neural network (input layer, attention layer, output layer), with inputs consisting of three features: time-frequency energy entropy, fractal dimension, and thermal compensation coefficient (a total of three dimensions). In the attention layer, a self-attention mechanism is used to calculate the weight of each feature, w1, w2, w3. The weight reflects the importance of the feature under the current working conditions (e.g., when cutting vibration is severe, the time-frequency energy entropy weight is higher).

[0145] Dynamic normalization and feature expansion: Normalize the original features (such as Z-score normalization) and calculate the weighted feature value based on the attention weight: f w =w1×E e +w2×D+w3×k c , through feature combination (such as square, product, difference), the 3-dimensional features are expanded to 16 dimensions (such as including the linear terms, quadratic terms, and cross terms of each feature), thereby enhancing the expressive power of the feature space.

[0146] Step S4.5: Principal Component Analysis (PCA) Dimensionality Reduction:

[0147] PCA was performed on the 16-dimensional feature vector, and the first three principal components (cumulative variance contribution rate ≥ 90%) were extracted, corresponding to the cutting stability index (reflecting vibration intensity), surface quality score (reflecting roughness and defects), and thermo-mechanical coupling coefficient (reflecting the coupling degree between thermal deformation and mechanical load).

[0148] Kernel Density Estimation (KDE) Optimization:

[0149] Kernel density estimation is used to analyze the distribution characteristics of the three principal components, correct the influence of outliers, and ensure that the physical meaning of each dimension index is clear (for example, a higher cutting stability index means less vibration).

[0150] Three-dimensional matrix construction:

[0151] The three types of indices are normalized to the interval [0,1] and arranged in columns to form a three-dimensional process state characterization matrix.

[0152] In a preferred embodiment of the present invention, the above step S5 may include:

[0153] S5.1: Set the first and second monitoring points at the leading and trailing edges of the tool-material contact area, respectively. Use a high-density strain gauge array to capture the spatial distribution of cutting forces in real time. Combined with an infrared thermal imager, simultaneously obtain the temperature gradient change rate at the two monitoring points.

[0154] S5.2: Based on the temperature gradient change rate and the multidimensional process state representation matrix, an improved grey correlation analysis method is used to calculate the difference in cutting force distribution between two monitoring points, a dynamic correlation coefficient matrix of the temperature gradient change rate is constructed, and the weight coefficient of the comparative analysis is optimized using a genetic algorithm;

[0155] S5.3: By comparing and analyzing the weight coefficients, a dynamic feedback layer is embedded in the pre-trained cutting parameter optimization model. The tool-material interaction coefficient is updated online based on the quantified results of the feature differences between the two monitoring points and the historical optimal parameter mapping relationship in the process knowledge base.

[0156] S5.4: Based on the interaction coefficient, design a multi-objective optimizer based on the NSGA-II algorithm. With cutting efficiency, surface quality, and tool life as optimization objectives, and combined with the modified interaction coefficient, iteratively solve the Pareto optimal solution set containing speed compensation, feed correction factor, and tool angle optimization value.

[0157] S5.5: Based on the Pareto optimal solution set, verify the feasibility of the parameter group through digital twin system simulation, dynamically adjust the parameter weight distribution based on real-time working condition data, and output an adaptive parameter group that meets the process constraints.

[0158] In an embodiment of the present invention, dynamic adaptive optimization of cutting parameters is achieved through multi-dimensional monitoring, intelligent algorithm fusion and digital twin verification: dual monitoring points are set in the tool-material contact area, and high-density strain gauges and infrared thermal imagers are used to capture the cutting force distribution and temperature gradient characteristics in real time, providing accurate real-time working condition data for parameter optimization; improved grey correlation analysis and genetic algorithm dynamically calculate feature differences and optimize weights to ensure the scientific nature of comparative analysis of monitoring point data and effectively capture subtle changes in tool-material interaction; by embedding a dynamic feedback layer in the pre-trained model and combining it with historical optimal parameter mapping, the interaction coefficient is updated in real time, so that the model has the ability of self-learning of working conditions; the multi-objective optimizer based on the NSGA-II algorithm generates a Pareto optimal solution with cutting efficiency, surface quality and tool life as the goals, taking into account the balance of multiple performance indicators, and then undergoing simulation verification and dynamic weight adjustment of the digital twin system. The final output adaptive parameter group not only meets the process constraints, but also can respond quickly according to real-time working conditions. This method significantly improves the real-time, accuracy and robustness of parameter optimization, effectively solves the limitations of traditional single-objective optimization, and realizes the coordinated optimization of processing quality, efficiency and tool life during the core cutting process, providing reliable support for high-precision and high-stability intelligent processing.

[0159] In an embodiment of the present invention, the specific steps include:

[0160] Step S5.1: Monitoring point layout:

[0161] The first monitoring point is set at the front edge of the tool when it cuts into the material (close to the tool feed direction), and the second monitoring point is set at the trailing edge after it is cut out, covering the complete stress-heat evolution area of ​​the cutting process.

[0162] Sensor deployment:

[0163] High-density strain gauge array: Dense strain gauges are attached to the material surface at two monitoring points to form a grid array, which collects three-dimensional stress components in real time and constructs a cutting force spatial distribution matrix (such as a stress cloud map).

[0164] Infrared thermal imager: Use a high frame rate (e.g., 500fps) infrared thermal imager to synchronously shoot two monitoring points, extract the temperature field video stream, and calculate the temperature gradient change rate (ΔT / Δt), focusing on the heat conduction characteristics of the contact interface between the tool and the material.

[0165] Data synchronization: Through hardware triggering (such as synchronous clock signal) or software timestamp alignment, the data acquisition time error between the strain gauge array and the infrared thermal imager is ensured to be less than 1ms, thus achieving spatiotemporal synchronization of force-thermal data.

[0166] Step S5.2: Improvement of grey relational analysis:

[0167] Reference sequence setting: The cutting force distribution vector F1 and temperature gradient change rate T1 of the first monitoring point are used as the reference sequence, and the corresponding data F2, T2 of the second monitoring point are used as the comparison sequence.

[0168] Correlation coefficient calculation: For each characteristic dimension (such as stress component, temperature gradient), calculate the correlation coefficient between the comparison sequence and the reference sequence:

[0169]

[0170] Where ρ is the resolution coefficient (usually 0.5), and k is the sequence number of the feature point.

[0171] Difference quantification: The difference in cutting force distribution and temperature gradient between two monitoring points is calculated by the mean or weighted sum of the correlation coefficients.

[0172] Construction of dynamic correlation coefficient matrix: The cutting force distribution difference and temperature gradient difference are combined with the multidimensional process state characterization matrix (such as cutting stability index and thermal-mechanical coupling coefficient) to form a dynamic correlation coefficient matrix R, which characterizes the degree of correlation between the force-thermal characteristics and the process state.

[0173] Genetic algorithm optimization weight: Define fitness function: Take the correlation between the difference calculation result and the actual process defects (such as edge collapse) as the optimization target, and iteratively optimize the weight vector of the correlation coefficient w=[w F ,w T ] to make the weights adaptive to the current working conditions.

[0174] Step S5.3: Dynamic feedback layer embedding: Add a dynamic feedback layer to the pre-trained cutting parameter optimization model (such as a neural network or machine learning model), and the input is the difference quantization result of the two monitoring points and the optimized weight.

[0175] Historical parameter mapping and coefficient update: query the process knowledge base, retrieve the historical optimal parameter cases that match the current difference characteristics, and extract the corresponding tool-material interaction coefficients (such as friction coefficient, heat transfer coefficient). The difference between the current working condition and the historical cases is calculated through the feedback layer, and the interaction coefficients in the model are updated online using weighted average or incremental learning algorithms. The formula is λ n =λ o +η×w×(λ h -λ o ), where η is the learning rate, λ h is the historical case coefficient.

[0176] Step S5.4: Optimization target definition:

[0177] Cutting efficiency: Material removal rate per unit time (such as mm 3 / s) is the index, and the objective function f1 = -v×f, where v is the cutting speed and f is the feed rate.

[0178] Surface quality: The cutting surface roughness Ra or chipping area rate is used as an indicator, and the objective function f2 = Ra.

[0179] Tool life: Taking tool wear Δw as the indicator, the objective function f3 = Δw.

[0180] Decision variables and constraints:

[0181] Variables: speed compensation, feed correction factor, tool angle optimization value (relative to initial parameters). Constraints: cutting force does not exceed the material allowable stress, temperature does not exceed the tool thermal stability threshold, parameter adjustment range ≤ ± 20% of the initial value.

[0182] NSGA-II algorithm iteration:

[0183] Population initialization: Randomly generate a parameter combination population containing 100-200 individuals.

[0184] Non-dominated sorting: Sort individuals according to three objective functions and assign Pareto ranks.

[0185] Crowding calculation: Maintaining population diversity and avoiding premature convergence.

[0186] Selection-Crossover-Mutation: Generates the offspring population through roulette wheel selection, simulated binary crossover (SBX), and polynomial mutation.

[0187] Termination condition: After 50-100 generations of iteration or when the Pareto front converges, the Pareto optimal solution set consisting of the top 20% of excellent individuals is output.

[0188] Step S5.5: Digital twin system simulation:

[0189] The Pareto optimal parameter group is input into the digital twin model to simulate the cutting process and predict indicators such as cutting force, temperature field, and surface roughness.

[0190] Compare the simulation results with the process constraints (such as safety margin > 15%, roughness ≤ Ra0.8μm), and eliminate the parameter groups that do not meet the conditions.

[0191] Dynamic weight adjustment:

[0192] According to real-time working condition data (such as current tool wear and material batch differences), the weights of the three optimization objectives are dynamically adjusted.

[0193] Parameter output: The linear weighted method is used to screen the parameter group with the best comprehensive performance from the feasible solution set. The formula is: the comprehensive score is equal to the sum of each item from i=1 to i=3 (the weight coefficient ωi of the i-th target multiplied by the normalized function value of the target).

[0194] In a preferred embodiment of the present invention, the above step S6 may include:

[0195] S6.1: Generates a device control instruction stream based on the adaptive parameter set, and synchronously sends the speed compensation, feed correction factor, and tool angle optimization value to the cutting device controller via the time-sensitive network protocol, achieving precise matching of parameter adjustment and processing timing;

[0196] S6.2: By precisely matching parameter adjustments with machining timing, the length of the multimodal data acquisition window is adaptively adjusted according to the characteristics of the machining stage. A sliding window algorithm is used to extract the cutting force standard deviation, temperature fluctuation rate, and defect density gradient as process stability evaluation indicators.

[0197] S6.3: Construct a fuzzy comprehensive evaluation model for process stability, integrating the real-time cutting vibration spectrum, tool wear characteristic vector, and environmental disturbance factors. Dynamically predict process deviation trends through a cloud-based digital twin system to generate a stability index.

[0198] S6.4: Design a three-level fault-tolerant mechanism. When the stability index is detected to be lower than 0.7, parameter fine-tuning, machining path optimization, and emergency rollback are triggered in sequence to ensure seamless connection of continuous machining processes.

[0199] S6.5: The multimodal data, parameter adjustment records, and processing effect evaluations generated during the closed-loop optimization process are synchronously transmitted back to the process knowledge base. By comparing the simulation predictions with the actual results, the finite element model boundary conditions and optimization algorithm weight coefficients are automatically corrected.

[0200] S6.6: An energy efficiency evaluation module is embedded in the parameter iterative optimization process. Based on the real-time monitoring data of cutting power and the equipment energy consumption model, the processing efficiency and energy consumption are dynamically balanced to achieve the optimization goal of reducing the energy consumption per unit workpiece by ≥15%.

[0201] In an embodiment of the present invention, full-process closed-loop control and intelligent optimization strategies are used to achieve a synergistic improvement in the high precision, high reliability, and high efficiency of the magnetic core cutting process. A time-sensitive network protocol is used to ensure microsecond-level precision matching of the adaptive parameter set with the processing sequence, avoiding parameter inaccuracy caused by control delays. Dynamic adjustment of the multimodal data acquisition window based on the processing stage, combined with stability assessment indicators extracted using a sliding window algorithm, enables real-time capture of vibration, thermal deformation, and defect evolution characteristics during the cutting process. Through the deep integration of a fuzzy comprehensive evaluation model and a digital twin system, advanced prediction of process deviation trends and dynamic quantification of the stability index are achieved, providing a scientific basis for active fault tolerance. The hierarchical trigger logic of the three-level fault tolerance mechanism ensures processing continuity while enabling rapid response to abnormal conditions, minimizing scrap rates. The bidirectional data feedback mechanism of the process knowledge base forms a self-evolving closed loop, enabling continuous iterative upgrades of the finite element simulation and optimization algorithms as processing experience accumulates. The embedded energy efficiency assessment module ensures green manufacturing by matching cutting power with the energy consumption model in real time, ensuring processing efficiency while achieving a 15% or greater reduction in unit energy consumption. The coordinated application of this series of technologies not only builds a complete intelligent processing closed loop of perception-analysis-decision-execution-feedback, but also significantly improves process robustness and resource utilization efficiency under complex working conditions.

[0202] In an embodiment of the present invention, the specific steps include:

[0203] Step S6.1: Control instruction stream generation: Generate an instruction sequence that complies with the device controller protocol based on the adaptive parameter group output in step S5.

[0204] Time-Sensitive Network (TSN) transmission: The instruction stream is transmitted to the cutting equipment controller through the TSN protocol. The traffic shaping and time synchronization mechanisms of TSN (such as IEEE1588) are used to ensure that the instruction transmission delay is less than 1ms and is strictly aligned with the timing signals such as spindle rotation and feed motion (such as updating parameters at a fixed phase point every time the tool rotates one circle).

[0205] Execution verification: After receiving the instruction, the controller feeds back the parameter reception status and verifies whether the parameter adjustment is effective at the specified processing position through encoder pulse counting (such as enabling new parameters at the beginning of the nth feed cycle).

[0206] Step S6.2: Dynamic adjustment of acquisition window:

[0207] Roughing stage: Use a long window (e.g., 500ms) to collect low-frequency force / heat signals and focus on the overall energy distribution;

[0208] Finishing stage: Switch to high-frequency acquisition with a short window (e.g. 50ms) to capture tiny vibrations or sudden temperature changes.

[0209] Sliding window algorithm application:

[0210] A sliding window (window length L = 100 ms, step size S = 50 ms) was applied to the cutting force signal, and the standard deviation within each window characterized the fluctuation amplitude of the force signal;

[0211] The fluctuation rate (ΔT divided by the sliding mean of Δt) of the temperature data was calculated to reflect thermal stability;

[0212] For the defect image of the cut surface, the number of chipped edges or cracks per unit area is counted, and the defect density change gradient (the difference in defect density between adjacent windows divided by time) is calculated.

[0213] Step S6.3: Construction of fuzzy comprehensive evaluation model:

[0214] Input variables:

[0215] Cutting vibration spectrum characteristics (such as 0.5-2kHz energy entropy, extracted by FFT);

[0216] Tool wear characteristic vector (such as flank wear width VB, tool geometric parameter change);

[0217] Environmental disturbance factors (such as workshop temperature fluctuations and grid voltage fluctuations).

[0218] Fuzzy rule design: Define fuzzy subsets (such as severe vibration, severe wear, and large disturbance) through expert experience, and establish fuzzy mapping rules between input variables and stability levels (such as stable, warning, and dangerous).

[0219] Digital twin dynamic prediction:

[0220] The real-time collected force-heat-wear data is input into the cloud-based digital twin model, and the process parameter offset trend (such as the cutting force rising slope and temperature accumulation trend) of the next 5-10 processing cycles is predicted through finite element simulation to correct the fuzzy evaluation results.

[0221] Step S6.4: Three-level threshold setting:

[0222] Level 1 warning (0.5≤index<0.7): triggers parameter fine-tuning, automatically adjusting minor parameters such as Δv (±5%) and α_f (±3%) without pausing processing;

[0223] Level 2 warning (0.3≤index<0.5): Start machining path optimization, such as adjusting the tool entry angle, increasing the air cutting transition path, and reducing the cutting load;

[0224] Level 3 warning (index < 0.3): Execute emergency retraction, the tool quickly retracts from the workpiece, processing is suspended and an alarm is issued, waiting for manual intervention.

[0225] Step S6.5: Data transmission and storage: Multimodal raw data (such as strain gauge array signals, infrared thermal imaging videos), parameter adjustment records (such as Δv = +8%, α_f = -5%), and processing effect evaluation (such as roughness measured value Ra = 0.6μm) are synchronously stored in the process knowledge base and classified by labels such as workpiece number, material type, and processing date.

[0226] Model correction algorithm:

[0227] Finite element model: Compare simulated predicted stress with measured stress and automatically adjust material constitutive parameters (such as elastic modulus and thermal expansion coefficient);

[0228] Optimization algorithm weight: According to the change of stability index before and after parameter adjustment, the objective function weight of NSGA-II is adopted.

[0229] Step S6.6: Real-time energy consumption monitoring:

[0230] The active power (kW) of the spindle motor and feed motor is collected in real time through power sensors, combined with the cumulative energy consumption (kWh) during processing time, and compared with the equipment energy consumption model (such as the theoretical power-load curve) to identify energy waste links (such as idling loss and overload cutting).

[0231] Multi-objective optimization embedding: A new energy efficiency objective function is added to the NSGA-II algorithm in step S5: the ratio of actual energy consumption to the theoretical minimum energy consumption.

[0232] Dynamic energy consumption balance: When energy consumption is detected to exceed 120% of the theoretical value, the cutting speed or feed rate is automatically reduced (sacrificing efficiency to prioritize energy efficiency); when energy consumption is lower than 80% of the theoretical value, the parameters are appropriately increased to balance efficiency and energy consumption, ensuring that the energy consumption reduction rate per unit workpiece is ≥15%.

[0233] like Figure 2 As shown, an embodiment of the present invention also provides a magnetic core intelligent cutting parameter adaptive optimization system based on multimodal perception, including:

[0234] An acquisition module is used to integrate a multimodal sensor module on the magnetic core cutting equipment. The module includes a force sensor, a visual sensor, and a temperature sensor, which are respectively used to collect real-time cutting force dynamic signals, cutting trajectory image sequences, and tool temperature time series data. The visual sensor captures the surface texture features and three-dimensional contour data of the magnetic core, and combines the results of the magnetic core material type identification, the associated parameters in the historical process database, and the preset process constraints to generate an initial cutting parameter set.

[0235] The fusion module is used to perform the first trial cutting based on the initial cutting parameter set and simultaneously start the multimodal data fusion acquisition to obtain the cutting force frequency domain feature vector, the tool temperature change rate curve and the cutting surface defect image characteristics to obtain multimodal data. The collected multimodal data is subjected to feature-level fusion processing, including time-frequency joint analysis of the cutting force signal, calculation of the fractal dimension of the cutting surface image and derivation of the tool thermal deformation compensation coefficient, to generate a multidimensional process state representation matrix.

[0236] A processing module is used to input the multidimensional process state characterization matrix into a pre-trained cutting parameter optimization model, set a first monitoring point and a second monitoring point in the tool-material contact area, respectively obtain the cutting force distribution difference characteristics and tool temperature gradient change data in real time, combine the historical optimal parameter mapping relationship in the process knowledge base, and compare and analyze the dynamic characteristics of the two monitoring points to correct the tool-material interaction coefficient in the model, and output an adaptive parameter group including a speed compensation amount, a feed correction factor and a tool angle optimization value; adjust the cutting equipment operation state in real time according to the adaptive parameter group, and establish a closed-loop feedback mechanism during the continuous processing process, so as to realize iterative optimization of cutting parameters and process stability control through a multimodal data dynamic monitoring window.

[0237] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A magnetic core intelligent cutting parameter adaptive optimization method based on multimodal perception, characterized in that: The method comprises: S1: A multimodal sensor module is integrated on the core cutting equipment. The module includes a force sensor, a visual sensor, and a temperature sensor, which are used to respectively collect cutting force dynamic signals, cutting trajectory image sequences, and tool temperature time series data in real time; S2: Capture the core surface texture features and 3D contour data through a visual sensor, combine the core material type identification results, associated parameters in the historical process database, and preset process constraints to generate an initial cutting parameter set; S3: Perform the first trial cutting based on the initial cutting parameter set, and simultaneously start multimodal data fusion acquisition to obtain the cutting force frequency domain feature vector, tool temperature change rate curve and cutting surface defect image features to obtain multimodal data; S4: Perform feature-level fusion processing on the collected multimodal data, including time-frequency joint analysis of cutting force signals, calculation of fractal dimension of cutting surface images, and derivation of tool thermal deformation compensation coefficients to generate a multidimensional process state representation matrix; S5: Inputting the multi-dimensional process state representation matrix into a pre-trained cutting parameter optimization model, setting a first monitoring point and a second monitoring point in the tool-material contact area, respectively obtaining cutting force distribution difference characteristics and tool temperature gradient change data in real time, combining the historical optimal parameter mapping relationship in the process knowledge base, and performing a comparative analysis of the dynamic characteristics of the two monitoring points to correct the tool-material interaction coefficient in the model, and outputting an adaptive parameter group including a speed compensation amount, a feed correction factor, and a tool angle optimization value; S6: Adjust the operating status of the cutting equipment in real time according to the adaptive parameter group, and establish a closed-loop feedback mechanism during the continuous processing process, and realize iterative optimization of cutting parameters and process stability control through the multimodal data dynamic monitoring window.

2. The method for adaptive optimization of magnetic core intelligent cutting parameters based on multimodal perception according to claim 1 is characterized in that: A multimodal sensor module is integrated into the core cutting equipment. The module includes a force sensor, a visual sensor, and a temperature sensor, which are used to collect real-time cutting force dynamic signals, cutting trajectory image sequences, and tool temperature time series data, including: S1.1: Configure the data bus of the multimodal sensor based on a time-sharing multiplexing strategy. Ensure that the conflict rate of concurrent acquisition of cutting force signals, image sequences, and temperature data is less than 0.5% through the data transmission timing of the force sensor, vision sensor, and temperature sensor. S1.2: Deploy an FPGA coprocessor at the sensor end, build a double-buffer queue structure to cache asynchronous data, and trigger a synchronization signal based on the tool displacement encoder to achieve microsecond-level timing alignment of the cutting force sampling time, image frame exposure time, and temperature acquisition point.

3. The method for adaptive optimization of magnetic core intelligent cutting parameters based on multimodal perception according to claim 2 is characterized in that: The core surface texture features and 3D contour data are captured by a visual sensor. Combined with the core material type identification results, associated parameters in the historical process database, and preset process constraints, an initial cutting parameter set is generated, including: S2.1: Based on the stereo image sequences collected by the vision sensor, the 3D point cloud of the magnetic core is reconstructed using adaptive structured light coding technology. An improved iterative closest point algorithm is used to match the preset process reference model to achieve a contour positioning accuracy of ±0.03mm. S2.2: Based on the contour positioning accuracy, the grayscale distribution matrix of the core surface is extracted. The texture quality evaluation index is constructed by combining the wavelet transform and the local binary pattern fusion feature to dynamically identify abnormal areas with grain orientation deviation exceeding 5°. S2.3: Based on the abnormal area, a lightweight material recognition model is deployed to receive the texture feature vector after edge preprocessing. The model then fuses the historical material map features through knowledge distillation technology to output the material type and the corresponding process parameter confidence level. S2.4: Based on the material identification results, the historical process database is called and a dynamic time warping algorithm is used to match the 3D contour characteristic curves of similar working conditions. Combined with the current tool wear compensation coefficient, the initial cutting speed, feed rate, and tool inclination angle parameters are generated. S2.5: Based on the initial cutting speed, feed rate, and tool inclination parameters, a cutting parameter feasibility verification model is established. The cutting stress distribution is predicted based on finite element simulation data. The parameter set is dynamically adjusted to meet the preset constraints and an initial parameter set with a safety margin greater than 15% is output.

4. The method for adaptive optimization of magnetic core intelligent cutting parameters based on multimodal perception according to claim 3 is characterized in that: Perform the first trial cutting based on the initial cutting parameter set, and simultaneously start multimodal data fusion acquisition to obtain the cutting force frequency domain feature vector, tool temperature change rate curve, and cut surface defect image features, obtaining multimodal data, including: S3.1: Based on the execution sequence of the initial cutting parameter set, a synchronous trigger pulse signal is generated through the tool spindle encoder to jointly control the force sensor sampling window, the visual sensor exposure time, and the temperature sensor acquisition cycle to ensure the spatiotemporal consistency of multimodal data; S3.2: Based on the spatiotemporal consistency of the data, a parallel computing channel was constructed to perform fast Fourier transform and short-time energy analysis on the cutting force signal, extracting the energy proportion of the 0.5-1.5kHz characteristic frequency band. A sliding window difference method was used on the tool temperature time series data to calculate the temperature change rate gradient. An improved U-Net segmentation network was applied to the cut surface image to detect the edge chipping defect area and crack extension direction in real time. S3.3: By detecting the chipping defect area and crack extension direction, a dynamic mapping relationship between multimodal data and process parameters is established. Based on the deviation value between the cutting force frequency domain characteristics and the initial parameter set, the similarity between the tool temperature change rate and the preset thermal stability curve, and the matching degree between the cutting surface defect distribution and the three-dimensional contour prediction model, a process state evaluation matrix is ​​generated to obtain multimodal data.

5. The method for adaptive optimization of magnetic core intelligent cutting parameters based on multimodal perception according to claim 4 is characterized in that: The collected multimodal data is subjected to feature-level fusion processing, including time-frequency joint analysis of cutting force signals, calculation of fractal dimension of cutting surface images, and derivation of tool thermal deformation compensation coefficients, to generate a multidimensional process state representation matrix, including: S4.1: Perform a modified Hilbert-Huang transform on the cutting force signal, extract 6-8 intrinsic mode functions through adaptive noise-assisted empirical mode decomposition, and construct a time-frequency energy density spectrum based on the instantaneous frequency calculation to quantify the energy entropy distribution characteristics in the 0.5-2 kHz frequency band; S4.2: Based on the energy entropy distribution characteristics, an improved multi-scale box counting method is used to analyze the cutting surface image. The mesh density is adaptively adjusted based on the tool feed direction. The anisotropic fractal index is calculated by combining the grayscale gradient co-occurrence matrix, and the fractal dimension calculation threshold is dynamically set. S4.3: Based on the fractal dimension, the threshold is calculated and the tool temperature gradient data is integrated with the finite element heat conduction simulation results to establish a dynamic prediction model for the axial / radial thermal expansion coefficient. Through the regression analysis of the temperature change rate and cutting parameters, the tool thermal deformation compensation coefficient is derived. S4.4: Based on the tool thermal deformation compensation coefficient, a feature weighted fusion network based on the attention mechanism is designed to dynamically normalize the time-frequency energy entropy, fractal dimension, and thermal compensation coefficient to generate a 16-dimensional process state feature vector. S4.5: Map the eigenvectors to the process parameter space, and construct a three-dimensional process state representation matrix including the cutting stability index, surface quality score, and thermal-mechanical coupling coefficient through joint optimization of principal component analysis and kernel density estimation.

6. The method for adaptive optimization of magnetic core intelligent cutting parameters based on multimodal perception according to claim 5, characterized in that: The multi-dimensional process state representation matrix is ​​input into a pre-trained cutting parameter optimization model. A first monitoring point and a second monitoring point are set in the tool-material contact area to respectively obtain the cutting force distribution difference characteristics and tool temperature gradient change data in real time. Combined with the historical optimal parameter mapping relationship in the process knowledge base, the tool-material interaction coefficient in the model is corrected through dynamic feature comparison and analysis of the two monitoring points. The adaptive parameter group containing the speed compensation amount, feed correction factor and tool angle optimization value is output, including: S5.1: Set the first and second monitoring points at the leading and trailing edges of the tool-material contact area, respectively. Use a high-density strain gauge array to capture the spatial distribution of cutting forces in real time. Combined with an infrared thermal imager, simultaneously obtain the temperature gradient change rate at the two monitoring points. S5.2: Based on the temperature gradient change rate and the multidimensional process state representation matrix, an improved grey correlation analysis method is used to calculate the difference in cutting force distribution between two monitoring points, a dynamic correlation coefficient matrix of the temperature gradient change rate is constructed, and the weight coefficient of the comparative analysis is optimized using a genetic algorithm; S5.3: By comparing and analyzing the weight coefficients, a dynamic feedback layer is embedded in the pre-trained cutting parameter optimization model. The tool-material interaction coefficient is updated online based on the quantified results of the feature differences between the two monitoring points and the historical optimal parameter mapping relationship in the process knowledge base. S5.4: Based on the interaction coefficient, design a multi-objective optimizer based on the NSGA-II algorithm. With cutting efficiency, surface quality, and tool life as optimization objectives, and combined with the modified interaction coefficient, iteratively solve the Pareto optimal solution set containing speed compensation, feed correction factor, and tool angle optimization value. S5.5: Based on the Pareto optimal solution set, verify the feasibility of the parameter group through digital twin system simulation, dynamically adjust the parameter weight distribution based on real-time working condition data, and output an adaptive parameter group that meets the process constraints.

7. The method for adaptive optimization of magnetic core intelligent cutting parameters based on multimodal perception according to claim 6, characterized in that: The cutting equipment operating status is adjusted in real time based on the adaptive parameter set, and a closed-loop feedback mechanism is established during the continuous processing process. The iterative optimization of cutting parameters and process stability control are achieved through the multi-modal data dynamic monitoring window, including: S6.1: Generates a device control instruction stream based on the adaptive parameter set, and synchronously sends the speed compensation, feed correction factor, and tool angle optimization value to the cutting device controller via the time-sensitive network protocol, achieving precise matching of parameter adjustment and processing timing; S6.2: By precisely matching parameter adjustments with machining timing, the length of the multimodal data acquisition window is adaptively adjusted according to the characteristics of the machining stage. A sliding window algorithm is used to extract the cutting force standard deviation, temperature fluctuation rate, and defect density gradient as process stability evaluation indicators. S6.3: Construct a fuzzy comprehensive evaluation model for process stability, integrating the real-time cutting vibration spectrum, tool wear characteristic vector, and environmental disturbance factors. Dynamically predict process deviation trends through a cloud-based digital twin system to generate a stability index. S6.4: Design a three-level fault-tolerant mechanism. When the stability index is detected to be lower than 0.7, parameter fine-tuning, machining path optimization, and emergency rollback are triggered in sequence to ensure seamless connection of continuous machining processes. S6.5: The multimodal data, parameter adjustment records, and processing effect evaluations generated during the closed-loop optimization process are synchronously transmitted back to the process knowledge base. By comparing the simulation predictions with the actual results, the finite element model boundary conditions and optimization algorithm weight coefficients are automatically corrected. S6.6: An energy efficiency evaluation module is embedded in the parameter iterative optimization process. Based on the real-time monitoring data of cutting power and the equipment energy consumption model, the processing efficiency and energy consumption are dynamically balanced to achieve the optimization goal of reducing the energy consumption per unit workpiece by ≥15%.

8. A magnetic core intelligent cutting parameter adaptive optimization system based on multimodal perception, the system implementing the method according to any one of claims 1 to 7, characterized in that: include: An acquisition module is used to integrate a multimodal sensor module on the magnetic core cutting equipment. The module includes a force sensor, a visual sensor, and a temperature sensor, which are respectively used to collect real-time cutting force dynamic signals, cutting trajectory image sequences, and tool temperature time series data. The visual sensor captures the surface texture features and three-dimensional contour data of the magnetic core, and combines the results of the magnetic core material type identification, the associated parameters in the historical process database, and the preset process constraints to generate an initial cutting parameter set. The fusion module is used to perform the first trial cutting based on the initial cutting parameter set and simultaneously start the multimodal data fusion acquisition to obtain the cutting force frequency domain feature vector, the tool temperature change rate curve and the cutting surface defect image characteristics to obtain multimodal data. The collected multimodal data is subjected to feature-level fusion processing, including time-frequency joint analysis of the cutting force signal, calculation of the fractal dimension of the cutting surface image and derivation of the tool thermal deformation compensation coefficient, to generate a multidimensional process state representation matrix. A processing module is used to input the multidimensional process state characterization matrix into a pre-trained cutting parameter optimization model, set a first monitoring point and a second monitoring point in the tool-material contact area, respectively obtain the cutting force distribution difference characteristics and tool temperature gradient change data in real time, combine the historical optimal parameter mapping relationship in the process knowledge base, and compare and analyze the dynamic characteristics of the two monitoring points to correct the tool-material interaction coefficient in the model, and output an adaptive parameter group including a speed compensation amount, a feed correction factor and a tool angle optimization value; adjust the cutting equipment operation state in real time according to the adaptive parameter group, and establish a closed-loop feedback mechanism during the continuous processing process, so as to realize iterative optimization of cutting parameters and process stability control through a multimodal data dynamic monitoring window.

9. A computing device, characterized in that include: one or more processors; A storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a program, which, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images