Open-type mutual inductor magnetic core cutting-molding cooperative intelligent control method and system

Abstract

The invention provides an open-type mutual inductor magnetic core cutting-molding cooperative intelligent control method and system, and relates to the technical field of intelligent manufacturing, and the method comprises the steps: collecting multi-dimensional feature data of a magnetic core raw material in real time through a multi-source sensing unit, including magnetic core surface topography point cloud data, magnetic conductivity, hardness parameters and internal defect distribution; based on the multi-dimensional feature data, a historical process library is matched through a transfer learning algorithm to generate a material compensation factor, and an initial cutting parameter final combination is calculated by using an improved particle swarm algorithm in combination with transformer specification parameters; cutting the magnetic core for the first time based on the final combination of the initial cutting parameters to obtain actual size data; based on actual size data, two points are selected on a cutting surface, space coordinates are determined through laser scanning data, and a reference straight line is generated. The machining precision, the production efficiency and the process reliability of the core part of the mutual inductor are improved through a full-process closed loop.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and in particular to a method and system for collaborative intelligent control of cutting and shaping of an open-type transformer core. Background Art

[0002] The performance of the open-type transformer core directly affects the monitoring, protection and metering accuracy of the power system, and its cutting and shaping process is key.

[0003] However, traditional magnetic core cutting processes lack the ability to adapt to raw materials of different batches and material properties. Therefore, it is difficult to adjust parameters in real time based on multi-dimensional characteristics such as the core surface morphology, magnetic permeability, hardness and internal defects, which can easily lead to poor cutting quality and high scrap rate. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an open-type transformer core cutting-shaping collaborative intelligent control method and system, which improves the processing accuracy, production efficiency and process reliability of the transformer core components through a full-process closed loop.

[0005] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0006] In a first aspect, a method for collaborative intelligent control of cutting and shaping of an open-type transformer core is provided, the method comprising:

[0007] Step S1: using a multi-source sensing unit to collect multi-dimensional feature data of the core raw material in real time, including core surface topography point cloud data, magnetic permeability, hardness parameters, and internal defect distribution;

[0008] Step S2: Based on the multi-dimensional feature data, the material compensation factor is generated by matching the historical process library through the transfer learning algorithm. Combined with the mutual inductor specification parameters, the improved particle swarm algorithm is used to calculate the final combination of the initial cutting parameters;

[0009] Step S3: performing a first cut on the magnetic core based on the final combination of the initial cutting parameters to obtain actual size data; based on the actual size data, selecting two points on the cutting surface, determining the spatial coordinates through the laser scanning data, and generating a reference line;

[0010] Step S4: Calculate the geometric deviation correction value based on the direction vector of the reference straight line and the angle between the design standard axis, generate a comprehensive correction factor based on the hardness fluctuation parameter of the core material, perform vector correction on the initial cutting parameters, and generate a corrected cutting parameter vector;

[0011] Step S5: combining the corrected cutting parameter vector with the core surface topography point cloud data to dynamically generate shaping parameters;

[0012] Step S6: Based on the shaping parameters and the deformation data collected by the distributed pressure sensor array, the mold deformation compensation is performed through the edge computing node to generate shaping compensation feedback data; the corrected cutting parameter vector is sent to the cutting machine tool in real time through the industrial Internet of Things gateway.

[0013] Furthermore, the multi-source sensing unit collects multi-dimensional characteristic data of the core raw materials in real time, including: core surface morphology point cloud data, magnetic permeability, hardness parameters and internal defect distribution, including:

[0014] The core surface is scanned at multiple angles using a laser 3D scanner to generate 3D point cloud data and extract the core surface curvature distribution characteristics.

[0015] Based on the curvature distribution characteristics of the 3D point cloud data, the detection path of the electromagnetic characteristic sensor is dynamically planned. The high curvature area of ​​the core surface is jointly tested for magnetic field strength and eddy current loss at a preset frequency to extract the dynamic change curves of magnetic permeability and hardness parameters.

[0016] Based on the 3D topography point cloud data and magnetic permeability test results, a hyperspectral imaging device is used to focus and scan the defective areas inside the core in the visible to near-infrared range. The spectral feature matching algorithm is used to identify the distribution coordinates and depth information of microcracks and pore defects.

[0017] The three-dimensional morphology point cloud data, electromagnetic parameter dynamic change curve and defect distribution depth information are aligned and fused in time and space to construct multi-dimensional feature data containing spatial position correlation relationships.

[0018] Furthermore, based on multi-dimensional feature data, a transfer learning algorithm is used to match the historical process library to generate material compensation factors. Combined with the mutual inductor specification parameters, an improved particle swarm algorithm is used to calculate the final combination of initial cutting parameters, including:

[0019] The multi-dimensional feature data is input into the pre-trained transfer learning model, and similarity matching is performed with the material characteristic parameters in the historical process database through the feature extraction layer to output the material adaptability parameter compensation factor;

[0020] Based on the material adaptability parameter compensation factor, the target transformer's winding turns, rated current, and accuracy grade specification parameters are weightedly modified to generate dynamic optimization constraints and construct a cutting parameter optimization objective function.

[0021] According to the dynamic optimization constraints, an improved particle swarm optimization algorithm is adopted to dynamically adjust the inertia weight and learning factor, and the final combination that meets the objective function is searched in parallel in the parameter space of cutting speed, laser power, and feed angle to generate the final combination of initial cutting parameters.

[0022] Furthermore, the magnetic core is first cut based on the final combination of the initial cutting parameters to obtain actual size data. Based on the actual size data, two points are selected on the cutting surface, and the spatial coordinates are determined by laser scanning data to generate a reference line, including:

[0023] According to the final combination of the initial cutting parameters, the laser cutting machine is controlled to perform the first cutting. After the cutting is completed, the width, flatness and angle deviation data of the actual cutting surface of the magnetic core are obtained through the linear array CCD sensor, and the angle deviation parameters are extracted;

[0024] Based on the angle deviation parameter, the selected position of the preset target point is dynamically adjusted within the geometric center symmetry area of ​​the cutting surface. The 3D spatial coordinates of the two target points are analyzed through the laser 3D scanning data and marked as the endpoints of the reference line.

[0025] According to the spatial coordinates of the endpoints of the reference line, the least squares method is used to fit and generate the reference line connecting the two target points.

[0026] Furthermore, the geometric deviation correction value is calculated based on the angle between the direction vector of the reference line and the design standard axis. The comprehensive correction factor is generated by combining the hardness fluctuation parameters of the core material. The initial cutting parameters are vectorized and corrected to generate a corrected cutting parameter vector, including:

[0027] Calculate the cosine value of the spatial angle between the reference straight line direction vector and the design standard axis, generate the geometric deviation correction value according to the preset geometric accuracy level mapping table, and extract the geometric accuracy level identifier of the current magnetic core;

[0028] Dynamically adjust the normalized weight of the core hardness fluctuation data based on the geometric accuracy grade identification to generate a hardness correction component that matches the geometric deviation;

[0029] The hardness correction component and the geometric deviation correction value are linearly superimposed to generate a comprehensive correction factor, and according to the sensitivity threshold of each parameter in the final combination of the initial cutting parameters, they are mapped to the cutting speed, laser power and feed angle in a preset proportion to generate a corrected cutting parameter vector.

[0030] Furthermore, the corrected cutting parameter vector is combined with the core surface topography point cloud data to dynamically generate shaping parameters, including:

[0031] The corrected cutting parameter vector and the core surface topography point cloud data are input into the shaping parameter prediction neural network. The geometric features of the cutting surface and the material distribution features are extracted through the convolution layer. The initial shaping parameters are output through the fully connected layer, and the pressure gradient distribution features in the initial parameters are extracted.

[0032] Based on the pressure gradient distribution characteristics, combined with the thermal expansion coefficient of the magnetic core and the material yield strength parameters, the temperature curve and holding time window of the initial molding parameters are compensated and adjusted through a dynamic interpolation algorithm to generate a molding mold pressure gradient, temperature curve and holding time window that match the corrected cutting parameters.

[0033] Furthermore, based on the shaping parameters and the deformation data collected by the distributed pressure sensor array, the edge computing node performs mold deformation compensation to generate shaping compensation feedback data; the corrected cutting parameter vector is sent to the cutting machine tool in real time through the industrial Internet of Things gateway, including:

[0034] At the molding station, the hydraulic actuator is controlled to apply step pressure according to the molding mold pressure gradient, and the deformation data of the contact surface between the mold and the core is collected in real time through a distributed pressure sensor array. The deformation data is processed by Kalman filtering using edge computing nodes to generate mold deformation compensation instructions and feed them back to the hydraulic control unit, which also outputs the molding compensation status indicator.

[0035] Based on the shaping compensation status identification, the corrected cutting parameter vector is converted and sent to the laser cutting machine tool through the OPCUA protocol, and the surface roughness and edge straightness are calculated based on the cutting surface image captured in real time by the linear array CCD; according to the correlation analysis between the shaping compensation status identification and the cutting surface quality data, a correction effect verification index including the synergistic effect of the cutting and shaping process is generated.

[0036] Furthermore, after step S6, step S7 is also included: based on the correction effect verification index and the shaping compensation feedback data, a multi-dimensional fusion analysis is performed with the performance parameters of the finished core product to generate a process deviation feature vector; if the threshold is exceeded, the digital twin simulation optimization correction logic is triggered, and the whole process data is stored in the ERP or MES platform through the blockchain to achieve quality traceability and parameter self-optimization closed loop, including:

[0037] Based on the correction effect verification indicators and the shaping compensation feedback data, a multi-dimensional weighted fusion is performed in combination with the hysteresis loop and loss parameters of the finished core to generate the process deviation feature vector and extract the key deviation dimension identifier;

[0038] Based on the key deviation dimension identifier, the digital twin simulation engine is triggered to load the current process parameters and equipment status data associated with the identifier, and the suppression effect of different correction strategies on the key deviation dimension is simulated in a virtual environment to generate a simulation result data set containing the strategy execution effect;

[0039] Based on the simulation result data set, the optimization contribution of each correction strategy to the key deviation dimension is evaluated through the reinforcement learning algorithm, and the strategy with the highest contribution is selected to update the dynamic process parameter optimization model; at the same time, the full-process process data packet is encrypted by timestamp and written into the blockchain trusted evidence chain, and associated and mapped with the work order data, key deviation dimension identification and simulation result data set of the ERP or MES platform to achieve quality traceability and parameter self-optimization closed loop.

[0040] In a second aspect, a collaborative intelligent control system for cutting and shaping the core of an open-type transformer is provided, comprising:

[0041] The acquisition module is used to collect multi-dimensional feature data of the core raw materials in real time through a multi-source sensing unit, including core surface topography point cloud data, magnetic permeability, hardness parameters and internal defect distribution;

[0042] The calculation module is used to generate material compensation factors based on multi-dimensional feature data and match the historical process library through a transfer learning algorithm. In combination with the mutual inductor specification parameters, an improved particle swarm algorithm is used to calculate the final combination of initial cutting parameters.

[0043] The reference line module is used to perform the first cut on the magnetic core based on the final combination of the initial cutting parameters to obtain the actual size data. Based on the actual size data, two points are selected on the cutting surface, and the spatial coordinates are determined by laser scanning data to generate the reference line.

[0044] The correction module is used to calculate the geometric deviation correction value based on the direction vector of the reference line and the angle between the design standard axis, generate a comprehensive correction factor based on the hardness fluctuation parameters of the core material, perform vector correction on the initial cutting parameters, and generate a corrected cutting parameter vector;

[0045] Dynamic generation module, used to combine the corrected cutting parameter vector with the core surface topography point cloud data to dynamically generate shaping parameters;

[0046] The compensation module is used to perform mold deformation compensation through the edge computing node based on the molding parameters and the deformation data collected by the distributed pressure sensor array, and generate molding compensation feedback data; the corrected cutting parameter vector is sent to the cutting machine tool in real time through the industrial Internet of Things.

[0047] According to a third aspect, a computing device includes:

[0048] one or more processors;

[0049] 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.

[0050] In a fourth aspect, a computer-readable storage medium stores a program, which implements the method when executed by a processor.

[0051] The above solution of the present invention includes at least the following beneficial effects:

[0052] Through full-process closed-loop control and multi-technology integration, the processing accuracy, efficiency and reliability of transformer cores are improved: laser scanning, electromagnetic detection, etc. are used to realize full feature perception of raw materials, and transfer learning and particle swarm algorithm are combined to optimize cutting parameters. Laser scanning is used to generate a reference straight line to dynamically correct geometric deviations. In the shaping process, parameters are predicted based on neural networks and combined with thermal expansion and yield strength compensation adjustments. Blockchain is used to store full-process data for quality traceability, and multi-dimensional data fusion and reinforcement learning are used to form a self-optimization closed loop of process parameters. At the same time, hyperspectral imaging and pressure sensors are used to realize defect prediction and risk suppression. Its "data-driven, technology-collaborative, full-cycle traceability" approach breaks through the limitations of traditional manufacturing and provides a replicable intelligent solution for the field of precision machining. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a flow chart of an open-type transformer core cutting-shaping collaborative intelligent control method provided by an embodiment of the present invention.

[0054] Figure 2 It is a schematic diagram of an open-type transformer core cutting-shaping collaborative intelligent control system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0055] 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 may 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.

[0056] like Figure 1 As shown, an embodiment of the present invention provides a method for collaborative intelligent control of cutting and shaping of an open-type transformer core, the method comprising the following steps:

[0057] Step S1: using a multi-source sensing unit to collect multi-dimensional feature data of the core raw material in real time, including: core surface topography point cloud data, magnetic permeability, hardness parameters and internal defect distribution;

[0058] Step S2: Based on the multi-dimensional feature data, the material compensation factor is generated by matching the historical process library through the transfer learning algorithm. Combined with the mutual inductor specification parameters, the improved particle swarm algorithm is used to calculate the final combination of the initial cutting parameters;

[0059] Step S3: performing a first cut on the magnetic core based on the final combination of the initial cutting parameters to obtain actual size data; based on the actual size data, selecting two points on the cutting surface, determining the spatial coordinates through the laser scanning data, and generating a reference line;

[0060] Step S4: Calculate the geometric deviation correction value based on the direction vector of the reference straight line and the angle between the design standard axis, generate a comprehensive correction factor based on the hardness fluctuation parameter of the core material, perform vector correction on the initial cutting parameters, and generate a corrected cutting parameter vector;

[0061] Step S5: combining the corrected cutting parameter vector with the core surface topography point cloud data to dynamically generate shaping parameters;

[0062] Step S6: Based on the shaping parameters and the deformation data collected by the distributed pressure sensor array, the mold deformation compensation is performed through the edge computing node to generate shaping compensation feedback data; the corrected cutting parameter vector is sent to the cutting machine tool in real time through the industrial Internet of Things gateway.

[0063] In an embodiment of the present invention, multi-source sensing technologies such as laser scanning, electromagnetic detection, and hyperspectral imaging are used to achieve micron-level precision detection of magnetic cores and identify quality risks in advance; transfer learning and particle swarm algorithm are used to intelligently optimize cutting parameters to improve matching efficiency and reduce material waste; parameters are dynamically corrected based on cutting deviation and hardness fluctuation to improve geometric accuracy; matching shaping parameters are generated through neural networks, industrial Internet of Things gateways achieve millisecond-level coordinated response of equipment, and visual servoing and pressure sensors achieve real-time closed loop of detection-execution-verification to reduce manual intervention; the full-process closed loop improves the processing accuracy, production efficiency and process reliability of the core components of the transformer.

[0064] In a preferred embodiment of the present invention, the above step S1: real-time acquisition of multi-dimensional feature data of the magnetic core raw material by a multi-source sensing unit, including: magnetic core surface morphology point cloud data, magnetic permeability, hardness parameters and internal defect distribution, may include:

[0065] Step S11, scanning the surface of the magnetic core at multiple angles using a laser 3D scanner to generate 3D topography point cloud data and extract the curvature distribution characteristics of the magnetic core surface;

[0066] Step S12: dynamically planning the detection path of the electromagnetic characteristic sensor based on the curvature distribution characteristics of the three-dimensional topography point cloud data, performing a joint detection of magnetic field strength and eddy current loss on the high curvature area of ​​the core surface at a preset frequency, and extracting the dynamic change curve of magnetic permeability and hardness parameters;

[0067] Step S13: Based on the three-dimensional topography point cloud data and the magnetic permeability test results, a hyperspectral imaging device is used to focus and scan the defective area inside the magnetic core in the visible to near-infrared band, and a spectral feature matching algorithm is used to identify the distribution coordinates and depth information of microcracks and pore defects;

[0068] In step S14, the three-dimensional morphology point cloud data, the electromagnetic parameter dynamic change curve and the defect distribution depth information are temporally and spatially aligned and fused to construct multi-dimensional feature data containing spatial position correlation relationships.

[0069] In an embodiment of the present invention, the geometric features of the core surface are restored at the micron level, the positioning error in high curvature areas is reduced, and spatial prior information is provided for detection path planning to avoid blind scanning; the detection path is dynamically planned to improve detection efficiency, and eddy current and magnetic field are jointly detected + dynamic point arrangement is used to improve parameter accuracy; focused scanning is combined with spectral feature matching to improve defect recognition rate, accurately detect micro defects, and integrate spatial and spectral data to achieve full-dimensional defect positioning; spatiotemporal alignment eliminates multi-source data misalignment and delay, and multi-dimensional feature vectors provide rich input for the algorithm, thereby improving the generalization ability of the parameter optimization model.

[0070] In an embodiment of the present invention, the specific steps include:

[0071] Step S11, using a laser 3D scanner to emit a laser beam from at least three orthogonal viewpoints such as the top and side of the magnetic core to obtain the three-dimensional coordinates (X, Y, Z) of the surface reflection point to form dense point cloud data; using statistical filtering or radius filtering algorithm to eliminate outliers (such as dust, equipment noise interference points); selecting k nearest neighbor points (such as k = 20) for each point, and calculating the Gaussian curvature and mean curvature through local polynomial surface fitting (such as quadratic surface), and identifying high curvature areas (such as edges and corners); according to the preset grid division (such as dividing the magnetic core surface into 10×10 areas), the curvature mean, maximum value and distribution variance in each grid are counted to generate a curvature distribution heat map.

[0072] Step S12: According to the curvature heat map generated in S11, the surface of the magnetic core is divided into a "high curvature area" (curvature > threshold C) and a "low curvature area"; in the high curvature area, detection points are dynamically added at intervals of 5 mm, and in the low curvature area, points are evenly distributed at intervals of 20 mm to form a spiral or serpentine detection path; the electromagnetic characteristic sensor moves along the planned path and synchronously collects magnetic field strength (H) and eddy current loss (P) data at a frequency of 20 Hz; for each detection point, three repeated measurement values ​​are recorded and the average is taken.

[0073] The magnetic permeability (μ) is estimated by combining the relationship between the magnetic field intensity H and the induced electromotive force with Ampere's loop law. The hardness (HB) is converted using an empirical mapping table of eddy current loss P and material hardness (such as a P-HB curve established through historical data). The parameter fluctuation curve along the detection path is generated with the spatial coordinates of the detection point as the horizontal axis and the magnetic permeability and hardness values ​​as the vertical axis.

[0074] In step S13, the defect area is located by combining point cloud data (such as surface depressions and protrusions) and areas of abnormal magnetic permeability (such as μ below the standard value ±10%) to delineate the potential defect region (ROI). The hyperspectral imaging device scans the ROI line by line, collecting reflectance spectrum data in the 400-1000nm band with a resolution of 5nm / band. Each pixel generates a spectral vector containing 120 bands. The detected spectrum is matched with the "sharp absorption peak @ 700nm" feature in the crack spectrum library, and the cosine similarity is calculated (threshold > 0.8 is used to determine it is a crack). Based on the spectral "low reflectivity platform @ 800-900nm" feature, it is compared with the pore sample library and the matching degree is calculated using the spectral angle mapping (SAM) algorithm. The defect depth is estimated based on the penetration depth of light of different wavelengths (such as short wavelength < 200μm, long wavelength > 500μm) combined with the imaging depth of field.

[0075] Step S14: The coordinates of the electromagnetic detection point (X e , Y e , Z e ) and hyperspectral pixel coordinates (X h , Y h , Z h ) is converted to (X l , Y l , Z l ), eliminate the installation errors between sensors through rigid body transformation (translation + rotation matrix); sort the data streams of different sensors by timestamp, and use linear interpolation to fill the data gaps caused by asynchronous acquisition (such as the time difference between laser scanning and electromagnetic detection is less than 50ms); associate the point cloud data (curvature), electromagnetic parameters (μ, HB), and spectral characteristics (defect type) according to spatial position to form a "coordinate-geometry-material-defect" quadruple; for each coordinate point, construct a multidimensional feature vector: [X, Y, Z, curvature, μ, HB, defect probability].

[0076] In a preferred embodiment of the present invention, the above step S2: based on the multi-dimensional feature data, generating a material compensation factor by matching the historical process library through a transfer learning algorithm, and calculating the final combination of the initial cutting parameters using an improved particle swarm algorithm in combination with the mutual inductor specification parameters, may include:

[0077] Step S21: Input the multi-dimensional feature data into the pre-trained transfer learning model, perform similarity matching with the material characteristic parameters in the historical process database through the feature extraction layer, and output the material adaptability parameter compensation factor;

[0078] Step S22: Based on the material adaptability parameter compensation factor, weighted correction is performed on the target transformer's winding turns, rated current, and accuracy grade specification parameters, dynamic optimization constraints are generated, and a cutting parameter optimization objective function is constructed;

[0079] In step S23, based on the dynamic optimization constraints, an improved particle swarm optimization algorithm is used to dynamically adjust the inertia weight and learning factor to search in parallel for the final combination that satisfies the objective function in the parameter space of cutting speed, laser power, and feed angle to generate the final combination of initial cutting parameters.

[0080] In an embodiment of the present invention, by reusing historical process data through a transfer learning model, material characteristics can be matched in seconds, parameter debugging time can be reduced, and the error between the compensation factor generated based on multi-dimensional feature matching and the actual material characteristics is less than 5%, thereby improving the initial adaptability of the cutting parameters; at the same time, the specification parameter weights are adaptively adjusted according to the material characteristics to reduce the cutting error of the high-precision mutual inductor, and a multi-objective balance is achieved by clarifying the priorities of cutting accuracy, efficiency, and tool life; the improved particle swarm optimization algorithm improves search efficiency and shortens parameter optimization time by dynamically adjusting the inertia weight and learning factor, and expands the coverage of the final solution by dynamically switching between global and local searches, thereby improving the comprehensive performance of the initial parameter combination.

[0081] In an embodiment of the present invention, the specific steps include:

[0082] In step S21, the multidimensional feature vector (including core curvature, permeability, hardness, defects, etc.) is input into a pre-trained transfer learning model (such as ResNet or a fully connected neural network). The model's convolutional or fully connected layers extract material features (such as hardness gradient and permeability distribution pattern) and compare them dimensionally with thousands of sets of material sample features (such as silicon steel and Permalloy) stored in the historical process database. Using cosine similarity or Euclidean distance algorithms, the matching degree between the current core features and the historical samples is calculated, and the top five similar material samples are screened. Based on the historical process parameter deviations (such as cutting speed adjustment) of similar samples, the material adaptability parameter compensation factors (such as hardness compensation coefficient K1 and permeability compensation coefficient K2) are generated by weighted average according to the matching degree.

[0083] Step S22: extract parameters such as the number of winding turns (N), rated current (I), and precision level (P) of the target transformer. For example, the cutting size error requirement for a high-precision transformer (P=0.1 level) is <10 μm.

[0084] Weighted correction:

[0085] If the compensation factor indicates that the material hardness is high (K1>1.2), a higher weight (e.g., weight coefficient × 1.5) is assigned to the rated current parameter I to increase the laser power compensation;

[0086] If the magnetic permeability is low (K2 < 0.8), the weight of the number of winding turns N is reduced (such as weight coefficient × 0.8) to reduce the impact of eddy current loss on cutting.

[0087] Constraint generation:

[0088] Physical constraints: laser power ≤ maximum power of the device, feed angle ∈ [-45°, 45°];

[0089] Process constraints: The cutting speed must match the heat dissipation characteristics of the material (e.g., speed for hard materials ≤ 50 mm / s).

[0090] Objective function construction:

[0091] Main goal: Minimize cutting size error (accounting for 60%);

[0092] Secondary goals: maximize cutting efficiency (accounting for 30%) and minimize tool wear (accounting for 10%), forming a multi-objective optimization function.

[0093] In step S23, 50 particles (each particle represents a set of parameter combinations) are randomly generated in the parameter space of cutting speed (20-100 mm / s), laser power (50-300 W), and feed angle (-30°-30°); dynamic parameter adjustment: in the early iteration (the first 30 generations), the inertia weight ω decreases linearly from 0.9 to 0.4, the learning factor C1 decreases from 2.5 to 2.0, and C2 increases from 2.0 to 2.5 to enhance the global search capability; in the late iteration (the last 20 generations), ω is fixed at 0.4, C1=2.0, C2=2.0, and local fine search is focused.

[0094] For each particle, the objective function is substituted to calculate the comprehensive fitness value (the smaller the error and the higher the efficiency, the higher the fitness), and the individual final solution (pbest) and the global final solution (gbest) are updated. After 50 generations of iteration, the particle with the highest fitness is selected as the final combination of initial cutting parameters (such as speed 60 mm / s, power 200 W, angle -15°).

[0095] In a preferred embodiment of the present invention, the above step S3: performing a first cutting of the magnetic core based on the final combination of initial cutting parameters to obtain actual size data; selecting two points on the cutting surface based on the actual size data, determining spatial coordinates through laser scanning data, and generating a reference line may include:

[0096] Step S31, controlling the laser cutting machine to perform the first cutting according to the final combination of the initial cutting parameters, and obtaining the width, flatness and angle deviation data of the actual cutting surface of the magnetic core through the linear array CCD sensor after the cutting is completed, and extracting the angle deviation parameter;

[0097] Step S32: Based on the angle deviation parameter, the selected positions of the preset target points are dynamically adjusted within the geometric center symmetric area of ​​the cutting surface. The three-dimensional spatial coordinates of the two target points are analyzed using the laser three-dimensional scanning data and marked as the endpoints of the reference line.

[0098] Step S33 , generating a reference straight line connecting the two target points by using the least squares fitting method according to the spatial coordinates of the endpoints of the reference straight line.

[0099] In an embodiment of the present invention, sub-millimeter level real-time detection of the cutting surface size is achieved through a linear array CCD sensor, with a width measurement error of <0.05mm and an angular deviation detection accuracy of ±0.1°. In addition, a single scan can synchronously collect multi-dimensional data such as width, flatness, and angular deviation, providing a comprehensive basis for subsequent corrections. At the same time, the preset target point selection position of the cutting surface is dynamically adjusted based on the angular deviation, focusing on the symmetrical area of ​​the geometric center to avoid the baseline offset caused by edge anomalies. The reference point positioning error is <0.2mm, ensuring that the reference straight line reflects the overall geometric trend. The reference straight line is fitted by the least squares method, and noise interference is suppressed to make the fitting error <0.03mm. The angular deviation of the cutting surface and the design axis can be accurately calculated based on the direction vector of the fitted straight line.

[0100] In an embodiment of the present invention, the specific steps include:

[0101] In step S31, according to the initial cutting parameters (such as speed 60 mm / s and laser power 200 W), the laser cutting machine is controlled to complete the first cutting along the preset trajectory of the magnetic core; after the cutting is completed, the linear array CCD sensor moves horizontally along the cutting surface to scan, collect the grayscale image data of the cutting surface, and identify the cutting boundary through image threshold segmentation.

[0102] Calculation of size parameters:

[0103] Width: Measure the pixel distance between the two ends of the cutting surface and convert it into the actual physical width based on the sensor calibration factor (e.g. 1 pixel = 10 μm).

[0104] Flatness: Perform Fourier transform on the cut surface image, extract the high-frequency noise component (representing surface undulations), and calculate the root mean square error (RMS) as the flatness indicator;

[0105] Angle deviation: Detect the angle between the edge of the cutting surface and the axis of the core, compare it with the design angle (such as 90°), and calculate the deviation value (unit: degree).

[0106] Step S32, angle deviation judgment: If the detected angle deviation is greater than 0.5°, the preset target point position is dynamically adjusted in the geometric center area of ​​the cutting surface (e.g., within the range of ±10mm from the center) to avoid the edge area with excessive deviation; if the deviation is ≤0.5°, the symmetrical points on both sides of the center point are selected according to the original plan;

[0107] Laser 3D scanning: The laser 3D scanner transmits a laser beam to the target point, receives the time difference of the reflected light and calculates the 3D coordinates (X, Y, Z) of the target point. The scan is repeated 3 times and the average value is taken to reduce noise.

[0108] Coordinate marking: Mark the two target points as P1 (X1, Y1, Z1) and P2 (X2, Y2, Z2) as the endpoints of the reference line.

[0109] Step S33, coordinate preprocessing: filtering the coordinate data of P1 and P2 to remove outliers (such as Z coordinate mutation points) caused by laser reflection interference;

[0110] Least squares fitting: Assuming the equation of the reference line is Ax+By+Cz+D=0, the line parameters (A, B, C, D) are solved by minimizing the sum of the squares of the vertical distances from P1 and P2 to the line to obtain the fitted reference line.

[0111] Direction vector calculation: Calculate the direction vector V = (X2-X1, Y2-Y1, Z2-Z1) of the straight line from the coordinates of P1 and P2 for subsequent geometric deviation calculation.

[0112] In a preferred embodiment of the present invention, the above step S4: calculating the geometric deviation correction value based on the angle between the direction vector of the reference straight line and the design standard axis, generating a comprehensive correction factor based on the hardness fluctuation parameter of the core material, performing vectorized correction on the initial cutting parameters, and generating a corrected cutting parameter vector, may include:

[0113] Step S41, calculating the cosine value of the spatial angle between the reference straight line direction vector and the design standard axis, generating a geometric deviation correction value according to a preset geometric accuracy level mapping table, and extracting the geometric accuracy level identifier of the current magnetic core;

[0114] Step S42, dynamically adjusting the normalized weight of the core hardness fluctuation data based on the geometric accuracy grade identifier to generate a hardness correction component that matches the geometric deviation;

[0115] In step S43, the hardness correction component and the geometric deviation correction value are linearly superimposed to generate a comprehensive correction factor, and according to the sensitivity threshold of each parameter in the final combination of the initial cutting parameters, they are mapped to the cutting speed, laser power and feed angle according to a preset ratio to generate a corrected cutting parameter vector.

[0116] In an embodiment of the present invention, the geometric deviation is converted into a calculable correction value through the cosine value of the vector angle, and the preset geometric accuracy level mapping table is combined to ensure that the correction amplitude meets the design requirements, with an error matching degree of >98%. There is no need to manually switch parameters, and the production flexibility is improved by 50%. For magnetic cores with large hardness fluctuations, a larger correction weight for hardness fluctuations is given in high-precision scenarios, so that the cutting parameters are more adaptable to material changes, and the parameter adaptation error is reduced by 30%. At the same time, multi-factor collaborative correction is performed in combination with geometric deviation and material hardness to avoid insufficient compensation caused by single-factor correction. The correction amount is dynamically allocated according to the sensitivity of each parameter to achieve multi-parameter collaborative optimization, avoid process imbalance caused by excessive adjustment of a single parameter, and directly generate parameter vectors through vectorized correction to shorten correction time and improve production rhythm.

[0117] In an embodiment of the present invention, the specific steps include:

[0118] Step S41, obtain the direction vector of the reference straight line (generated by step S33) and the direction vector of the design standard axis, calculate the cosine value of the angle between the two by vector dot product, and judge the degree of deviation between the cutting surface and the design axis; according to the design requirements of the mutual inductor, determine the geometric accuracy level of the current magnetic core (such as level 0.1, level 0.5), match the preset "accuracy level-correction value" mapping table (for example: level 0.1 corresponds to the correction threshold of ±0.01mm, level 1 corresponds to ±0.1mm), and generate a geometric deviation correction value (ΔG); record the geometric accuracy level identifier of the current magnetic core (such as "G0.1" represents level 0.1 accuracy).

[0119] Step S42, normalization of hardness data: obtain the core hardness fluctuation data detected in step S1 (such as HB = 180 ± 20), and map it to the [-1, 1] interval through normalization processing (subtracting the mean and dividing by the standard deviation); according to the geometric accuracy level identifier extracted in S41, dynamically adjust the normalization weight coefficient (such as 0.1 level accuracy gives the hardness fluctuation weight 0.7, and 1 level accuracy weight 0.3). In high-precision scenarios, the hardness has a greater impact on the cutting deviation, so the weight is higher.

[0120] The weighted normalized hardness data is multiplied by the geometric deviation correction value ΔG to obtain the hardness correction component (ΔH), that is, ΔH = weight × normalized hardness × ΔG.

[0121] Step S43, the geometric deviation correction value ΔG and the hardness correction component ΔH are linearly superimposed to generate a comprehensive correction factor (ΔC = ΔG + ΔH), which reflects the joint influence of geometric error and material properties; the sensitivity thresholds of cutting speed, laser power, and feed angle to deviation are preset (for example, every 10mm / s change in speed affects the error by 0.05mm, and every 50W change in power affects the error by 0.03mm), and the correction ratio of each parameter is determined (for example, speed accounts for 40%, power accounts for 50%, and angle accounts for 10%); according to the comprehensive correction factor, the correction ratio of each parameter is determined. Combine the correction factor ΔC and the sensitivity of each parameter, and distribute the correction amount proportionally to the cutting speed (ΔV), laser power (ΔP), and feed angle (Δθ) to generate the corrected parameter vector: Cutting speed: V'=V0+ΔC×40%×sensitivity V; Laser power: P'=P0+ΔC×50%×sensitivity P; Feed angle: θ'=θ0+ΔC×10%×sensitivity θ; Among them, V0 is the cutting speed before correction, P0 is the laser power before correction, and θ0 is the feed angle before correction.

[0122] In a preferred embodiment of the present invention, the above step S5: combining the corrected cutting parameter vector with the core surface topography point cloud data to dynamically generate shaping parameters may include:

[0123] Step S51: The corrected cutting parameter vector and the core surface topography point cloud data are input into a shaping parameter prediction neural network. The geometric features of the cutting surface and the material distribution features are extracted through the convolution layer. The initial shaping parameters are output through the fully connected layer, and the pressure gradient distribution features in the initial parameters are extracted.

[0124] In step S52, based on the pressure gradient distribution characteristics, combined with the thermal expansion coefficient of the magnetic core and the material yield strength parameters, the temperature curve and the holding time window of the initial molding parameters are compensated and adjusted through a dynamic interpolation algorithm to generate a molding mold pressure gradient, temperature curve and holding time window that match the corrected cutting parameters.

[0125] In an embodiment of the present invention, by combining cutting parameters and core surface topography point cloud data, the molding parameters can be predicted more accurately, reducing the impact of errors introduced by the cutting process on the final molding effect; the dynamic interpolation algorithm can compensate and adjust the initial parameters according to the actual situation of the core, so that the molding process can better adapt to the characteristic differences of different cores; through precise adjustment of the temperature curve and the holding time window, the deformation and stress distribution of the core during the molding process can be effectively controlled, thereby improving the consistency and reliability of the product; the use of neural network prediction and dynamic compensation algorithms reduces the trial and error process in traditional process development.

[0126] In an embodiment of the present invention, the specific steps include:

[0127] In step S51, the corrected cutting parameter vector is normalized and each parameter is scaled to the interval [0, 1] to eliminate the dimension effect; the core surface topography point cloud data is downsampled to reduce data complexity while retaining key geometric features; and then coordinate system 1 is performed to map the point cloud data and the cutting parameter vector into the same coordinate system.

[0128] The downsampled point cloud data is convolved using a 3D convolution kernel to extract the geometric features of the cutting surface, such as curvature and normal vector. At the same time, a 1D convolution is performed on the cutting parameter vector to extract material distribution characteristics, such as the parameter variation patterns of different material regions. The extracted features are reduced in dimensionality through a pooling layer to retain the main feature information and reduce the amount of computation.

[0129] The convolution and pooling features are input into the fully connected layer, and the initial molding parameters are generated through nonlinear transformation, including the mold pressure gradient, initial temperature curve and holding time window. The pressure gradient distribution characteristics, such as the maximum pressure position and pressure change rate, are extracted from the generated initial parameters to provide a basis for subsequent compensation adjustments.

[0130] Step S52: Determine the thermal expansion coefficient of each region of the magnetic core under different pressures based on the pressure gradient distribution characteristics; calculate the thermal expansion amount of each region of the magnetic core under the action of the mold pressure based on the thermal expansion coefficient; and superimpose a compensation amount related to the thermal expansion amount on the initial temperature curve through a dynamic interpolation algorithm to adjust the temperature curve to ensure the dimensional stability of each region of the magnetic core during the molding process.

[0131] Combining the pressure gradient distribution characteristics and material yield strength parameters, the risk of plastic deformation in various regions of the magnetic core under current pressure and temperature conditions is evaluated. For areas where plastic deformation may occur, a dynamic interpolation algorithm is used to extend the pressure holding time window to allow the material sufficient time to reach a stable state and avoid plastic deformation.

[0132] The results of thermal expansion compensation and yield strength adjustment are integrated to generate the final molding parameters, including the adjusted temperature curve and holding time window. The generated molding parameters are then physically verified for feasibility to ensure that the parameters are within the equipment capability and will not cause core quality issues.

[0133] In a preferred embodiment of the present invention, step S6: performing mold deformation compensation through an edge computing node based on the molding parameters and the deformation data collected by the distributed pressure sensor array to generate molding compensation feedback data; and sending the corrected cutting parameter vector to the cutting machine tool in real time through the industrial Internet of Things gateway may include:

[0134] Step S61: At the molding station, a hydraulic actuator is controlled to apply stepped pressure according to the molding mold pressure gradient, and deformation data of the mold and core contact surface are collected in real time through a distributed pressure sensor array. The deformation data is processed by Kalman filtering using an edge computing node to generate a mold deformation compensation instruction, which is fed back to the hydraulic control unit and a molding compensation status indicator is simultaneously output.

[0135] In step S62, based on the shaping compensation status mark, the corrected cutting parameter vector is converted and sent to the laser cutting machine through the OPCUA protocol, and the surface roughness and edge straightness are calculated based on the cutting surface image captured in real time by the linear array CCD; based on the correlation analysis between the shaping compensation status mark and the cutting surface quality data, a correction effect verification index including the synergistic effect of the cutting-shaping process is generated.

[0136] In an embodiment of the present invention, real-time processing and Kalman filtering of distributed pressure sensor data by edge computing nodes are used to achieve precise compensation for mold deformation, improve molding accuracy and product consistency; dynamically adjust cutting parameters based on the molding compensation status indicator to achieve coordinated control of the cutting-molding process, reduce error transmission between processes, and improve overall processing quality. Utilize the cutting surface quality data fed back in real time by the linear array CCD to form a complete closed-loop control from cutting to molding, timely discover and correct machining deviations, and reduce scrap rates; use the OPCUA protocol to achieve data interaction between heterogeneous devices, improve the openness and interoperability of the system, and facilitate digital management and intelligent decision-making of the production process; by analyzing the correlation between molding compensation and cutting quality, generate quantitative correction effect verification indicators, provide data support for the continuous optimization of process parameters, and accelerate the improvement of process maturity.

[0137] In an embodiment of the present invention, the specific steps include:

[0138] In step S61, the set pressure value is decomposed into multiple step pressure values ​​based on the mold pressure gradient. The hydraulic actuator is then controlled to apply the corresponding pressure at preset time intervals. At each pressure step, a distributed pressure sensor array collects real-time pressure and deformation data from each measuring point on the mold-core contact surface. The collected raw deformation data is time-synchronized to ensure the consistency of timestamps across all sensor data. A Kalman filter algorithm is then used to filter the deformation data, fusing historical data with current measurements to estimate the true mold deformation state while suppressing noise interference.

[0139] Based on the filtered deformation data, the deformation deviation of each area of ​​the mold is calculated and compared with the preset ideal deformation model; according to the deformation deviation, the mold deformation compensation instruction is generated through the edge computing node, and the pressure output of the hydraulic actuator is adjusted to make the mold deformation approach the ideal state; the plastic compensation status identification is generated, including the compensation direction, compensation amount and compensation effect evaluation index, which is used for the collaborative control of subsequent processes.

[0140] Step S62, according to the shaping compensation status identifier, determine the timing and priority of sending the corrected cutting parameter vector, convert the corrected cutting parameter vector into the OPCUA protocol format through the industrial Internet of Things gateway, and send it to the control system of the laser cutting machine tool in real time.

[0141] The linear array CCD is used to capture the cutting surface image in real time during the laser cutting process, and image preprocessing is performed, including denoising, enhancement and binarization. Based on the image processing results, the surface roughness parameters (such as Ra value) and edge straightness indicators (such as straightness error and edge burr height) of the cutting surface are calculated.

[0142] Analyze the correlation between the shaping compensation status identification and the cutting surface quality data, such as the correlation between the pressure compensation amount and the cutting surface roughness; based on the correlation analysis results, generate correction effect verification indicators including the synergistic effect of the cutting-shaping process, such as synergistic matching degree, quality stability index, etc.; the verification indicators are fed back to the production management system to provide a basis for subsequent process optimization.

[0143] In a preferred embodiment of the present invention, step S6 is followed by step S7: based on the correction effect verification index and the shaping compensation feedback data, a multi-dimensional fusion analysis is performed with the performance parameters of the finished magnetic core product to generate a process deviation feature vector; if the threshold is exceeded, the digital twin simulation optimization correction logic is triggered, and the whole process data is stored in the ERP or MES platform through the blockchain to achieve quality traceability and parameter self-optimization closed loop, which may include:

[0144] Step S71: Based on the correction effect verification index and the shaping compensation feedback data, a multi-dimensional weighted fusion is performed in combination with the hysteresis loop and loss parameters of the finished magnetic core to generate a process deviation feature vector and extract a key deviation dimension identifier;

[0145] Step S72: Based on the key deviation dimension identifier, trigger the digital twin simulation engine to load the current process parameters and equipment status data associated with the identifier, simulate the suppression effects of different correction strategies on the key deviation dimension in a virtual environment, and generate a simulation result data set containing the strategy execution effects;

[0146] In step S73, based on the simulation result data set, the optimization contribution of each correction strategy to the key deviation dimension is evaluated through the reinforcement learning algorithm, and the strategy with the highest contribution is selected to update the dynamic process parameter optimization model; at the same time, the full-process process data packet is encrypted by timestamp and written into the blockchain trusted evidence chain, and is associated and mapped with the work order data, key deviation dimension identifier and simulation result data set of the ERP or MES platform to achieve quality traceability and parameter self-optimization closed loop.

[0147] In an embodiment of the present invention, through multi-dimensional data fusion and key deviation dimension identification, the root cause of process deviation affecting the performance of the magnetic core is accurately found, thereby improving the efficiency of problem diagnosis; digital twin technology is used to verify the correction strategy in a virtual environment, avoiding trial and error in actual production, reducing optimization costs and time; based on the reinforcement learning algorithm, the correction strategy is evaluated to achieve automatic optimization of process parameters and improve the accuracy of decision-making; blockchain technology is used to achieve credible evidence of process data throughout the entire process, ensuring that the data cannot be tampered with and meeting quality traceability and compliance requirements. A complete closed loop from production, testing to optimization is established to achieve continuous improvement of process parameters, improve product consistency and production efficiency; the digital twin simulation results, blockchain evidence data and business data of the ERP or MES platform are deeply associated to support enterprise-level production collaboration and decision-making optimization.

[0148] In an embodiment of the present invention, the specific steps include:

[0149] Step S71, normalize the correction effect verification index, plastic compensation feedback data and core finished product performance parameters (hysteresis loop, loss parameters) to eliminate dimensional differences; extract key features such as remanence, coercive force, saturation magnetic induction intensity, etc. from the hysteresis loop, and extract iron loss, copper loss and other features at different frequencies from the loss parameters.

[0150] Different weight coefficients are assigned based on the degree of influence of each data dimension on core performance. For example, parameters directly related to magnetic properties are given higher weights. A weighted summation approach is used to integrate the correction effect verification indicators, molding compensation feedback data, and core performance characteristics to generate the initial process deviation feature vector.

[0151] Calculate the degree of deviation between each dimension in the initial process deviation feature vector and the preset ideal value, identify the key deviation dimension that has the greatest impact on the core performance through principal component analysis or correlation analysis, and generate the corresponding key deviation dimension identifier.

[0152] Step S72, digital twin model loading and configuration: according to the key deviation dimension identifier, load the virtual model that matches the current process and equipment status from the digital twin model library; input the current process parameters (such as cutting parameters, shaping parameters) and equipment status data (such as temperature, pressure, power) into the digital twin model for initialization.

[0153] Correction strategy simulation and evaluation: Design multiple correction strategies for key deviation dimensions, such as adjusting laser power, changing holding time, and optimizing mold temperature curves. Execute each correction strategy separately in a digital twin environment, simulate the core machining process, and collect data on changes in key deviation dimensions under different strategies.

[0154] Simulation result data generation: Statistical analysis is performed on the simulation results of each correction strategy to calculate the improvement range and stability indicators of key deviation dimensions. A simulation result data set is constructed that includes changes in correction strategies, process parameters, equipment status, and key deviation dimensions to provide a basis for subsequent optimization.

[0155] Step S73, optimization contribution evaluation and strategy selection: using a reinforcement learning algorithm, with the degree of improvement in the key deviation dimension as the reward function, evaluate the optimization contribution of each correction strategy; select the correction strategy with the highest contribution and its corresponding process parameter combination, and update the dynamic process parameter optimization model.

[0156] Blockchain evidence storage and data association: The entire process data (including original parameters, corrected process data, and simulation results) is encrypted by timestamp and written into the blockchain trusted evidence chain; an association mapping relationship is established between the process data and the work order data, key deviation dimension identifiers, and simulation result data sets of the ERP or MES platform.

[0157] Quality traceability and closed-loop optimization: Develop a blockchain-based quality traceability interface to support querying full-process process data and quality information based on work order number or product batch; feed back optimized process parameters to the production system to achieve a parameter self-optimization closed loop from detection, analysis to optimization.

[0158] like Figure 2 As shown, an embodiment of the present invention further provides an open-type transformer core cutting-shaping collaborative intelligent control system, comprising:

[0159] The acquisition module is used to collect multi-dimensional feature data of the core raw materials in real time through a multi-source sensing unit, including core surface topography point cloud data, magnetic permeability, hardness parameters and internal defect distribution;

[0160] The calculation module is used to generate material compensation factors based on multi-dimensional feature data and match the historical process library through a transfer learning algorithm. In combination with the mutual inductor specification parameters, an improved particle swarm algorithm is used to calculate the final combination of initial cutting parameters.

[0161] The reference line module is used to perform the first cut on the magnetic core based on the final combination of the initial cutting parameters to obtain the actual size data. Based on the actual size data, two points are selected on the cutting surface, and the spatial coordinates are determined by laser scanning data to generate the reference line.

[0162] The correction module is used to calculate the geometric deviation correction value based on the direction vector of the reference line and the angle between the design standard axis, generate a comprehensive correction factor based on the hardness fluctuation parameters of the core material, perform vector correction on the initial cutting parameters, and generate a corrected cutting parameter vector;

[0163] Dynamic generation module, used to combine the corrected cutting parameter vector with the core surface topography point cloud data to dynamically generate shaping parameters;

[0164] The compensation module is used to perform mold deformation compensation through the edge computing node based on the molding parameters and the deformation data collected by the distributed pressure sensor array, and generate molding compensation feedback data; the corrected cutting parameter vector is sent to the cutting machine tool in real time through the industrial Internet of Things.

[0165] It should be noted that this system is a system corresponding to the above method, and all implementation methods in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0166] An embodiment of the present invention further provides a computing device comprising: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the above-described method. All implementations in the above-described method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0167] 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 collaborative intelligent control method for cutting and shaping the core of an open-type transformer, characterized in that: The method comprises: Step S1: using a multi-source sensing unit to collect multi-dimensional feature data of the core raw material in real time, including core surface topography point cloud data, magnetic permeability, hardness parameters, and internal defect distribution; Step S2: Based on the multi-dimensional feature data, the material compensation factor is generated by matching the historical process library through the transfer learning algorithm. Combined with the mutual inductor specification parameters, the improved particle swarm algorithm is used to calculate the final combination of the initial cutting parameters; Step S3: performing a first cut on the magnetic core based on the final combination of the initial cutting parameters to obtain actual size data; based on the actual size data, selecting two points on the cutting surface, determining the spatial coordinates through the laser scanning data, and generating a reference line; Step S4: Calculate the geometric deviation correction value based on the direction vector of the reference straight line and the angle between the design standard axis, generate a comprehensive correction factor based on the hardness fluctuation parameter of the core material, perform vector correction on the initial cutting parameters, and generate a corrected cutting parameter vector; Step S5: combining the corrected cutting parameter vector with the core surface topography point cloud data to dynamically generate shaping parameters; Step S6: Based on the shaping parameters and the deformation data collected by the distributed pressure sensor array, the mold deformation compensation is performed through the edge computing node to generate shaping compensation feedback data; the corrected cutting parameter vector is sent to the cutting machine tool in real time through the industrial Internet of Things gateway.

2. The method for collaborative intelligent control of split-type transformer core cutting and shaping according to claim 1, characterized in that: The multi-source sensing unit collects multi-dimensional characteristic data of the core raw materials in real time, including: core surface morphology point cloud data, magnetic permeability, hardness parameters and internal defect distribution, including: The core surface is scanned at multiple angles using a laser 3D scanner to generate 3D point cloud data and extract the core surface curvature distribution characteristics. Based on the curvature distribution characteristics of the 3D point cloud data, the detection path of the electromagnetic characteristic sensor is dynamically planned. The high curvature area of ​​the core surface is jointly tested for magnetic field strength and eddy current loss at a preset frequency to extract the dynamic change curves of magnetic permeability and hardness parameters. Based on the 3D topography point cloud data and magnetic permeability test results, a hyperspectral imaging device is used to focus and scan the defective areas inside the core in the visible to near-infrared range. The spectral feature matching algorithm is used to identify the distribution coordinates and depth information of microcracks and pore defects. The three-dimensional morphology point cloud data, electromagnetic parameter dynamic change curve and defect distribution depth information are aligned and fused in time and space to construct multi-dimensional feature data containing spatial position correlation relationships.

3. The method for collaborative intelligent control of split-type transformer core cutting and shaping according to claim 2, characterized in that: Based on multi-dimensional feature data, the material compensation factor is generated by matching the historical process library through the transfer learning algorithm. Combined with the mutual inductor specification parameters, the improved particle swarm algorithm is used to calculate the final combination of initial cutting parameters, including: The multi-dimensional feature data is input into the pre-trained transfer learning model, and similarity matching is performed with the material characteristic parameters in the historical process database through the feature extraction layer to output the material adaptability parameter compensation factor; Based on the material adaptability parameter compensation factor, the target transformer's winding turns, rated current, and accuracy grade specification parameters are weightedly modified to generate dynamic optimization constraints and construct a cutting parameter optimization objective function. According to the dynamic optimization constraints, an improved particle swarm optimization algorithm is adopted to dynamically adjust the inertia weight and learning factor, and the final combination that meets the objective function is searched in parallel in the parameter space of cutting speed, laser power, and feed angle to generate the final combination of initial cutting parameters.

4. The method for collaborative intelligent control of split-type transformer core cutting and shaping according to claim 3, characterized in that: Perform the first cutting on the magnetic core based on the final combination of the initial cutting parameters to obtain actual size data; Based on the actual size data, two points are selected on the cutting surface, and the spatial coordinates are determined by the laser scanning data to generate the reference line, including: According to the final combination of the initial cutting parameters, the laser cutting machine is controlled to perform the first cutting. After the cutting is completed, the width, flatness and angle deviation data of the actual cutting surface of the magnetic core are obtained through the linear array CCD sensor, and the angle deviation parameters are extracted; Based on the angle deviation parameter, the selected position of the preset target point is dynamically adjusted within the geometric center symmetry area of ​​the cutting surface. The 3D spatial coordinates of the two target points are analyzed through the laser 3D scanning data and marked as the endpoints of the reference line. According to the spatial coordinates of the endpoints of the reference line, the least squares method is used to fit and generate the reference line connecting the two target points.

5. The method for collaborative intelligent control of split-type transformer core cutting and shaping according to claim 4, characterized in that: The geometric deviation correction value is calculated based on the angle between the direction vector of the reference straight line and the design standard axis. The comprehensive correction factor is generated by combining the hardness fluctuation parameters of the core material. The initial cutting parameters are vectorized and corrected to generate the corrected cutting parameter vector, including: Calculate the cosine value of the spatial angle between the reference straight line direction vector and the design standard axis, generate the geometric deviation correction value according to the preset geometric accuracy level mapping table, and extract the geometric accuracy level identifier of the current magnetic core; Dynamically adjust the normalized weight of the core hardness fluctuation data based on the geometric accuracy grade identification to generate a hardness correction component that matches the geometric deviation; The hardness correction component and the geometric deviation correction value are linearly superimposed to generate a comprehensive correction factor, and according to the sensitivity threshold of each parameter in the final combination of the initial cutting parameters, they are mapped to the cutting speed, laser power and feed angle in a preset proportion to generate a corrected cutting parameter vector.

6. The method for collaborative intelligent control of split-type transformer core cutting and shaping according to claim 5, characterized in that: The corrected cutting parameter vector is combined with the core surface topography point cloud data to dynamically generate shaping parameters, including: The corrected cutting parameter vector and the core surface topography point cloud data are input into the shaping parameter prediction neural network. The geometric features of the cutting surface and the material distribution features are extracted through the convolution layer. The initial shaping parameters are output through the fully connected layer, and the pressure gradient distribution features in the initial parameters are extracted. Based on the pressure gradient distribution characteristics, combined with the thermal expansion coefficient of the magnetic core and the material yield strength parameters, the temperature curve and holding time window of the initial molding parameters are compensated and adjusted through a dynamic interpolation algorithm to generate a molding mold pressure gradient, temperature curve and holding time window that match the corrected cutting parameters.

7. The method for collaborative intelligent control of split-type transformer core cutting and shaping according to claim 6, characterized in that: Based on the shaping parameters and the deformation data collected by the distributed pressure sensor array, the edge computing node performs mold deformation compensation and generates shaping compensation feedback data. The corrected cutting parameter vector is sent to the cutting machine tool in real time through the Industrial Internet of Things, including: At the molding station, the hydraulic actuator is controlled to apply step pressure according to the molding mold pressure gradient, and the deformation data of the contact surface between the mold and the core is collected in real time through a distributed pressure sensor array. The deformation data is processed by Kalman filtering using edge computing nodes to generate mold deformation compensation instructions and feed them back to the hydraulic control unit, which also outputs the molding compensation status indicator. Based on the shaping compensation status identification, the corrected cutting parameter vector is converted and sent to the laser cutting machine tool through the OPCUA protocol, and the surface roughness and edge straightness are calculated based on the cutting surface image captured in real time by the linear array CCD; according to the correlation analysis between the shaping compensation status identification and the cutting surface quality data, a correction effect verification index including the synergistic effect of the cutting and shaping process is generated.

8. The method for collaborative intelligent control of split-type transformer core cutting and shaping according to claim 7, characterized in that: After step S6, the method further includes: Step S7: Based on the correction effect verification indicators and molding compensation feedback data, a multi-dimensional fusion analysis is performed with the performance parameters of the finished core product to generate a process deviation feature vector; if the threshold is exceeded, the digital twin simulation optimization correction logic is triggered, and the entire process data is stored in the ERP or MES platform through the blockchain to achieve quality traceability and parameter self-optimization closed loop.

9. A collaborative intelligent control system for cutting and shaping the core of an open-type mutual inductor, the system implementing the method according to any one of claims 1 to 8, characterized in that: include: The acquisition module is used to collect multi-dimensional feature data of the core raw materials in real time through a multi-source sensing unit, including core surface topography point cloud data, magnetic permeability, hardness parameters and internal defect distribution; The calculation module is used to generate material compensation factors based on multi-dimensional feature data and match the historical process library through a transfer learning algorithm. It also uses an improved particle swarm algorithm to calculate the final combination of initial cutting parameters in combination with the mutual inductor specification parameters. A reference linear module is used to perform the first cutting of the magnetic core based on the final combination of initial cutting parameters to obtain actual size data; Based on the actual size data, two points are selected on the cutting surface, and the spatial coordinates are determined by the laser scanning data to generate the reference line; The correction module is used to calculate the geometric deviation correction value based on the direction vector of the reference line and the angle between the design standard axis, generate a comprehensive correction factor based on the hardness fluctuation parameters of the core material, perform vector correction on the initial cutting parameters, and generate a corrected cutting parameter vector; Dynamic generation module, used to combine the corrected cutting parameter vector with the core surface topography point cloud data to dynamically generate shaping parameters; The compensation module is used to perform mold deformation compensation through edge computing nodes based on the molding parameters and the deformation data collected by the distributed pressure sensor array to generate molding compensation feedback data; The corrected cutting parameter vector is sent to the cutting machine tool in real time through the industrial Internet of Things.

10. 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 8.

Images