Intelligent transformer core stacking method based on leakage flux minimization optimization

Abstract

The application discloses a transformer core intelligent stacking method based on magnetic flux leakage pole minimization optimization, and belongs to the technical field of transformer manufacturing. The method comprises the following steps: in response to a silicon steel sheet to be placed entering a stacking station, generating a background excitation magnetic field based on an excitation magnetic source in the stacking station; monitoring the spatial magnetic flux leakage distribution introduced by the silicon steel sheet in real time through a sensor array; based on the spatial magnetic flux leakage distribution, iteratively calculating an optimal placement pose through an optimization algorithm, the optimal placement pose corresponding to a minimum point of the spatial magnetic flux leakage distribution; and driving a mechanical arm to place the silicon steel sheet at the optimal placement pose. The application constructs a closed-loop feedback with actual magnetic physical quantities as the target, replaces the traditional open-loop control based on geometric alignment, and directly guides the assembly with the final electromagnetic performance, thereby significantly reducing the magnetic resistance and iron loss of the finished product core.

Description

Technical Field

[0001] This application relates to the field of transformer manufacturing technology, and in particular to a method for intelligent stacking of transformer cores based on minimizing leakage flux. Background Technology

[0002] In transformer manufacturing, the quality of core stacking directly determines its core electromagnetic performance, such as no-load loss (iron loss) and excitation current. Traditional automated stacking technologies generally follow a technical practice that prioritizes achieving precise geometric alignment between silicon steel sheets. However, this paradigm faces a fundamental technical contradiction: the pursuit of ultimate geometric precision does not equate to the optimization of final electromagnetic performance. Due to unavoidable physical defects such as stamping burrs, internal material stress, and tolerance accumulation in silicon steel sheets, even with sub-micron-level geometric alignment, high magnetic reluctance regions caused by physical air gaps and stress concentration points will still form in the local magnetic circuit, resulting in significant leakage flux and energy loss. Existing technologies are trapped in a dilemma where continuous improvement in positioning accuracy leads to performance bottlenecks. Summary of the Invention

[0003] Firstly, this application provides a method for intelligent stacking of transformer cores based on minimizing leakage flux, aiming to solve the technical problem in the prior art that the electromagnetic performance of the core cannot be fundamentally optimized due to excessive reliance on geometric alignment.

[0004] This application provides a method for intelligent stacking of transformer cores based on minimizing leakage flux, comprising: responding to a silicon steel sheet to be placed entering a stacking station, generating a background excitation magnetic field based on an excitation magnetic source within the stacking station; monitoring a spatial leakage flux distribution introduced by the silicon steel sheet in real time through a sensor array; iteratively calculating an optimal placement posture based on the spatial leakage flux distribution using an optimization algorithm, the optimal placement posture corresponding to a minimum point of the spatial leakage flux distribution; and driving a robotic arm to place the silicon steel sheet in the optimal placement posture.

[0005] Optionally, the step of iteratively calculating an optimal placement pose using an optimization algorithm includes: in each iteration of the optimization algorithm, obtaining a current pose of the silicon steel sheet; based on the current pose, obtaining a gradient estimate of the spatial leakage flux distribution at the current pose by applying a perturbation displacement to the silicon steel sheet and performing at least two measurements by the sensor array; and updating the current pose based on the gradient estimate to generate a new current pose.

[0006] Optionally, the step of obtaining a gradient estimate of the spatial leakage flux distribution at the current pose by applying a perturbation displacement to the silicon steel sheet and performing at least two measurements by the sensor array is implemented based on a synchronous perturbation stochastic approximation algorithm, wherein the perturbation displacement is a random perturbation vector applied synchronously on all pose degrees of freedom.

[0007] Optionally, the synchronous perturbation random approximation algorithm employs a gain sequence that varies with the number of iterations, the gain sequence including a step-size gain sequence and a perturbation gain sequence.

[0008] Optionally, after the silicon steel sheet enters the stacking station and before the optimal placement pose is iteratively calculated using the optimization algorithm, the method further includes: acquiring an initial planar offset of the silicon steel sheet using a vision sensor; and driving the robotic arm to perform compensating movement based on the initial planar offset to coarsely position the silicon steel sheet.

[0009] Optionally, at each moment of real-time monitoring of the spatial leakage magnetic flux distribution introduced by the silicon steel sheet, the method further includes: acquiring a current magnetic field map data output by the sensor array; calculating a set of statistical characteristic values ​​of the current magnetic field map data; and comparing the statistical characteristic values ​​with a preset health status characteristic threshold range to perform a health status self-check on the sensor array.

[0010] Optionally, before the method is executed, a calibration step for establishing the health status characteristic threshold range is further included. The calibration step includes: driving the robotic arm to grasp a standard silicon steel sheet and move it to multiple sampling poses within the stacking station; acquiring a calibration magnetic field map data in each sampling pose and calculating its corresponding statistical feature value; and determining the health status characteristic threshold range based on the statistical distribution of the statistical feature values ​​acquired from all the sampling poses.

[0011] Secondly, this application provides an intelligent stacking system for transformer cores.

[0012] This application provides an intelligent stacking system for transformer cores, comprising: an excitation magnetic source configured to generate a background excitation magnetic field within a stacking station when a silicon steel sheet to be placed enters a stacking station; a sensor array configured to monitor in real time a spatial leakage flux distribution introduced by the silicon steel sheet; a processing device configured to iteratively calculate an optimal placement posture based on the spatial leakage flux distribution using an optimization algorithm, the optimal placement posture corresponding to a minimum point of the spatial leakage flux distribution; and a robotic arm configured to receive instructions from the processing device and place the silicon steel sheet in the optimal placement posture.

[0013] Optionally, the processing device is specifically configured to: in each iteration of the optimization algorithm, obtain a current pose of the silicon steel sheet; control the robotic arm to apply a perturbation displacement based on the current pose, and obtain at least two measurement results from the sensor array to calculate a gradient estimate of the spatial leakage flux distribution at the current pose; and update the current pose based on the gradient estimate to generate a new current pose.

[0014] Optionally, the processing device calculates the gradient estimate based on a synchronous perturbation stochastic approximation algorithm, wherein the perturbation displacement is a stochastic perturbation vector applied synchronously across all pose degrees of freedom.

[0015] The beneficial effects of this application are as follows:

[0016] The technical solution provided in this application reconstructs the core stacking process from an open-loop positioning task based on geometric proxies into a closed-loop optimization task based on real-time magnetic physical performance by establishing a closed-loop feedback system targeting actual magnetic physical quantities. This technological shift allows stacking decisions to be directly determined by the final electromagnetic performance, rather than the geometry. This method fundamentally bypasses physical friction points that traditional geometric alignment methods cannot address, such as burrs on silicon steel sheets, material internal stress, and cumulative stacking errors. By finding the optimal magnetic point rather than the geometric center, it directly ensures that the finished core has the lowest magnetic reluctance and iron loss. This is not only a quantitative improvement in technical indicators but also a performance-oriented intelligent assembly logic, providing more efficient quality assurance for the manufacture of high-performance transformers. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a method for intelligent stacking of transformer cores based on leakage flux minimization optimization, provided in an embodiment of this application.

[0019] Figure 2 This is a schematic diagram of the structure of an intelligent stacking system for transformer cores provided in an embodiment of this application.

[0020] Figure 3 This is a flowchart illustrating the efficient estimation steps of leakage flux gradient in one embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0023] This application provides a method and system for intelligent stacking of transformer cores based on leakage flux minimization optimization. In a specific engineering implementation, this method employs a stacking station equipped with an excitation magnetic source and a sensor array, coupled to a multi-axis robotic arm scheduled by an optimization algorithm node. This robotic arm is configured to perform non-contact sensing of the influence of a single silicon steel sheet on the local magnetic circuit reluctance under multi-degree-of-freedom microscopic pose perturbations. The spatial leakage flux distribution matrix output by the sensor array is mapped to a unique scalar objective function, based on which the processing device performs a closed-loop pose inversion calculation. The combination of this hardware architecture and data flow logic shifts the assembly reference from the geometric center of the structural components to the minimum singularity of the electromagnetic field distribution, thereby compensating for the local high reluctance regions induced by interlayer air gaps in silicon steel sheets, uneven distribution of stamping stress, and accumulated tolerances. Ultimately, this minimizes the total no-load iron loss of the manufactured core globally.

[0024] In order to enable those skilled in the art to reproduce the above-described technical solutions of this application without ambiguity, specific embodiments of this application are described below in detail with reference to the accompanying drawings.

[0025] Example 1: Method Example

[0026] Reference Figure 1 The intelligent stacking method for transformer cores based on leakage flux minimization optimization provided in one embodiment of this application is suitable for deployment in the main control IPC (Industrial Personal Computer) of an automated transformer production line. This IPC performs high-frequency handshakes and data exchanges with the FPGA acquisition board of the robotic arm servo driver and sensor array via a fieldbus protocol (such as EtherCAT).

[0027] In a specific system operating cycle, the method includes the following instruction sequence:

[0028] S100: In response to the silicon steel sheet to be placed entering a stacking station, a background excitation magnetic field is generated based on an excitation magnetic source within the stacking station.

[0029] The stacking station entity configured to perform this step integrates an integrated platform containing a physical bearing surface and an electromagnetic interaction interface. The excitation magnetic source is configured as a coil array embedded in or surrounding the physical bearing surface. In response to a material placement interruption signal issued by the system photoelectric switch or the robotic arm controller, a signal generation module configured as a constant current source triggers the drive logic for the excitation magnetic source. The signal generation module outputs a sinusoidal alternating current through a digital-to-analog converter (DAC) channel, thereby inducing an alternating background excitation magnetic field with temporal periodicity and spatial continuity in the three-dimensional space above the physical bearing surface. When a silicon steel sheet with ferromagnetic properties is gripped by the robotic arm and moved into the three-dimensional space, due to its much higher permeability than air, the silicon steel sheet induces the convergence and distortion of magnetic field lines. The establishment of this magnetic field distortion state provides a steady-state physical field input for subsequent sensor array measurements characterizing the magnetic circuit impedance.

[0030] Exemplarily, the excitation magnetic source comprises a pair of parallel, coaxially arranged Helmholtz coils, which are physically fixed to the bottom of the stacking platform base. Upon receiving a start flag, a signal generation module located within the control cabinet applies a sinusoidal alternating current with a peak value of 1 ampere (A) and a frequency of 50 Hz to the coil pair. The Ampere torque generated by the coils is directly above the stacking platform. Axis coordinate range Within a three-dimensional workspace of millimeters (mm), a background excitation magnetic field with a constant vertical magnetic induction intensity of approximately 10 Gauss (Gs) is established. An E-shaped silicon steel sheet, held by an electromagnetic chuck at the end of a six-axis robotic arm, is moved into this background excitation magnetic field in response to trajectory planning commands. A magnetization vector is induced within the E-shaped silicon steel sheet, causing a sharp change in local magnetic flux density at its sharp corners and interlayer micro-gaps. The magnetic field lines overflowing into free space constitute a leakage magnetic flux field that can be captured by the array.

[0031] S200: Real-time monitoring of the spatial leakage flux distribution introduced by the silicon steel sheet via a sensor array.

[0032] A sensor array configured to perform physical signal digitization is fixed directly beneath the insulating panel of the stacked platform. Each sensing unit in the array is used to synchronously sense the magnetic flux density component perpendicular to the platform. Driven by an FPGA hardware timer, the analog-to-digital converter front-end synchronously latches the output levels of all sensing units and serializes and encapsulates them into a digital frame format, which is then transmitted to the main processor's direct memory access (DMA) buffer via a high-speed backplane bus. The main processor's interrupt service routine responds to the DMA transfer completion flag by storing the original magnetic field map matrix of the current frame. Extracted to the cache. Based on this original matrix, the processing unit invokes the SIMD (Single Instruction Multiple Data) instruction set to execute a background magnetic field matrix residing in memory. The corresponding elements are subtracted. The resulting difference magnetic field matrix is ​​calculated by subtracting these elements. , is defined as the net spatial leakage flux distribution introduced by the current pose of the silicon steel sheet.

[0033] For example, the sensor array is configured as a The Hall element grid topology, with a spatial physical resolution (sensor center-to-center distance) of 2 mm, enables an effective measurement aperture of 32 mm. 32mm. Each Hall element pin is coupled to a cutoff frequency of A Hz anti-aliasing low-pass filter and a filter with A bit-deep analog-to-digital converter (ADC). Responding to a hardware trigger pulse with a 1-millisecond period, all 256 ADC channels of the array synchronously complete sampling and push the data word to the FPGA. The FPGA sends a data packet carrying 256 floating-point values ​​to the main processing unit via the UDP / IP protocol stack. The main processing unit parses the packet in kernel space and reconstructs the data. floating-point matrix Subsequently, the Arithmetic Logic Unit (ALU) loads from a contiguous block of memory addresses the data pre-calibrated in a workpiece-free state. Background magnetic field matrix Perform matrix difference The generated Differential magnetic field matrix The median coordinate is scalar elements Its physical meaning is the increment of the vertical magnetic induction intensity at the two-dimensional coordinate point caused by the silicon steel sheet, and the unit is millitalas (mT).

[0034] S300: Based on the spatial leakage flux distribution, an optimal placement pose is iteratively calculated using an optimization algorithm. The optimal placement pose corresponds to a minimum point of the spatial leakage flux distribution.

[0035] The optimization decision unit, equipped with a floating-point arithmetic core, reads the differential magnetic field matrix output by the S200. This unit first performs a norm-1 aggregation operation, mapping the high-dimensional matrix data into a single scalar value, which is then used as the objective function to evaluate the current system state, namely the integral value of the total leakage flux. Based on the scalar objective function, the optimization decision unit loads a numerical optimizer instance residing in non-volatile memory. This optimizer instance is configured to calculate the gradient vector that guides the descent of the objective function within a six-dimensional physical pose space spanned by the three translation axes (x, y, z) and three Euler angles (rx, ry, rz) of the robotic arm's Cartesian coordinate system. Based on the obtained gradient vector, the optimizer instance calculates a new pose command sequence according to its internal update equations and sends it to the robotic arm servo controller via the fieldbus to drive the silicon steel sheet to produce a small spatial displacement. This sampling-computation-displacement loop logic is configured to execute repeatedly until the norm of the calculated gradient vector decays below a preset tolerance threshold, thus determining that the system has converged to a local minimum of the objective function. The six-dimensional coordinate vector corresponding to this minimum point is then marked in memory and output as the optimal placement pose.

[0036] For example, the optimization decision unit is configured to execute a Synchronous Perturbation Stochastic Approximation (SPSA) optimization algorithm based on a normalized spatial mapping. This optimizer instance enforces all iterative logic constraints within a dimensionless mathematical space. Its underlying dataflow engine is configured to execute the following sub-steps:

[0037] Spatial normalization initialization: The processing module instantiates a diagonal scaling matrix in memory. .in It is preset to 100 millimeters (mm). It is preset to 0.1 radians (rad). The processing module reads the current initial physical pose vector from the robotic arm's register. and call the linear algebra library to execute The left multiplication operation generates a dimensionless initial normalized pose vector. .

[0038] Iterative optimization (entering the stage) (next iteration)

[0039] Bidirectional Normalized Perturbation Generation and Physical Mapping: A Pseudo-Random Number Generator Engine Produces a Number Based on Uniform Distribution 3D perturbation vector Its elements can only take values. The calculation core reads the preset dimensionless perturbation gain variable. And by using dimensionless vector addition and subtraction commands, a positive perturbation vector is generated. With negative perturbation vector Subsequently, the conversion module executes... and The inverse normalization operation generates robotic arm drive coordinates with physical dimensions. These two sets of coordinates are sent out time-divisionally via the EtherCAT bus to drive the robotic arm movement and simultaneously trigger the S200 module to perform two scalar integral measurements of leakage flux at corresponding poses, obtaining the measurement values ​​respectively. and .

[0040] The processing module receives the two sets of measures mentioned above and executes the equations. The value of is to be determined. Since the molecule has the dimension of leakage magnetic flux... The denominator is a dimensionless pure number. The gradient estimation vector output by this operation Granted and maintained legal dimension.

[0041] Processing module loads step gain coefficient and execute the state update equation. The gain coefficient in accordance with During generation, hyperparameters Forced to be configured to have dimensions (That is, the reciprocal of the leakage flux). Thus, the product term... The operation reduces the result to a dimensionless vector of pure numbers, making the new state variable... It still maintains its strictly dimensionless property.

[0042] Interruption of adjudication logic calculation Euclidean norm In response to the norm falling to a threshold (set as...), Below, the iteration loop is interrupted. The optimization decision unit will then... Locked into the optimal normalized pose Finally, perform the inverse normalization product. This will force the dimensionless data in the result buffer to be converted into a target floating-point array containing physical dimensions of length and angle.

[0043] S400: Drive a robotic arm to place the silicon steel sheet in the optimal placement position.

[0044] A communication daemon configured with a real-time kernel operating system polls the above-mentioned... A buffer for physical pose. In response to a data ready event, the process encapsulates this array containing six degrees of freedom coordinates into a PDO (Process Data Object) frame conforming to the Industrial Ethernet standard and pushes it to the robotic arm servo controller node in the next communication cycle of the bus. The inverse kinematics resolver configured on the servo controller maps the received target Cartesian coordinate system to the desired rotor angle sequence of the six joint motors. Six independent PID closed-loop control circuits operate in parallel, adjusting the PWM (Pulse Width Modulation) duty cycle to drive the servo motors to rotate. This drive mechanism enables the robotic arm's end flange and its attached electromagnetic chuck to carry silicon steel sheets, approximating the interpolation trajectory. Coordinate point. During the approximation process, the robotic arm performs linear interpolation descent along the direction of gravity (Z-axis) until the torque sensor detects a step signal of the contact reaction force generated by the physical contact between the silicon steel sheet and the base. In response to this step signal, the I / O control module sends a power-off trip command to the relay of the electromagnetic chuck, thereby eliminating residual magnetism and releasing the silicon steel sheet.

[0045] For example, the optimal placement pose array output by the S300 module is represented by a memory copy as follows: The floating-point data block is written to the controller's designated register address via the EtherCAT bus. The servo mechanism drives the flange center point to align with the absolute spatial coordinates according to the interpolation algorithm. And adjust the orientation to Euler angles. This pose is generated based on the nominal geometric centroid through closed-loop feedback. Translation offset and Tilting bias. The physical mechanism of this bias is to counteract the air gap magnetic resistance caused by the burrs on the laminates. Subsequently, the robotic arm controls the Z-axis servo to descend to... A millimeter-high plane. After the electromagnetic chuck coil is de-energized, the workpiece is unloaded by gravity, and the single silicon steel sheet completes its assembly cycle. The entire timing chain is constrained to operate within a strict clock cycle, ensuring industrial production efficiency.

[0046] It should be noted that in the SPSA optimization algorithm based on normalized space, to ensure the stability of algorithm convergence, all hyperparameters involved in the gain sequence are configured with concrete numerical values. For example, the theoretical constants controlling the decay rate are written into memory variables as follows: (dimensionless) (Dimensionless). The stability constant controlling the initial step size damping is set to... (Dimensionless). The amplitude constant controlling the disturbance radius is calibrated based on the noise level of the Hall sensor's background. (Dimensionless). The fundamental coefficient controlling the initial learning rate is explicitly assigned a value. Its data structure in memory is strongly typed and associated with units. (i.e., the reciprocal of millitalas) By assigning deterministic constant values ​​and binding them to dimensions, the compiler can ensure the stability of the computation cycle.

[0047] Example 2: System Example

[0048] Reference Figure 2 This application provides an embodiment of a transformer core intelligent stacking system 800, which is configured as a physical hardware carrier to implement the data flow and control command flow of the aforementioned method embodiments. In a specific physical deployment topology, the system 800 includes an excitation magnetic source 810, a sensor array 820, a processing device 830, and a robotic arm 840, all coupled in communication.

[0049] The excitation magnetic source 810 includes a coil winding and its corresponding power amplifier driver board. In the system topology, the control pins of the power amplifier driver board are hardwired to the I / O expansion card of the processing device 830 via analog output (AO) terminal blocks.

[0050] The sensor array 820 includes a Hall element array and an FPGA acquisition chip configured on a printed circuit board (PCB). The FPGA chip is configured to encapsulate a TCP / IP protocol stack, and its Ethernet physical layer transceiver (PHY) is directly connected to the dedicated gigabit network interface card (NIC) of the processing device 830 via a Category 6 twisted pair cable to ensure that the data link is not interrupted by other broadcast storms.

[0051] The processing device 830 is configured as a system main control rack, its motherboard equipped with a central processing unit (CPU) with out-of-order execution and AVX instruction set extension capabilities, and a non-volatile solid-state drive. The multi-threaded background daemons residing within its operating system are divided into:

[0052] Multi-source data fusion unit: It is configured to bind to a specific socket port, repeatedly call the recv() function to read UDP packets sent by the sensor array, and deserialize them into the process's heap memory space.

[0053] Health status diagnosis unit: It is configured to read the above heap memory data, perform variance and mean statistical calculations, and trigger a software exception throwing mechanism in response to out-of-limit results.

[0054] The optimization algorithm execution unit is configured to read the memory state and perform iterative floating-point operations according to the normalized space SPSA formula detailed in the aforementioned embodiments to calculate the optimal physical coordinates.

[0055] Motion command generation unit: configured to read a coordinate floating-point array from the output pointer of the algorithm unit, format it according to the CIA402 driver specifications, and push it to the EtherCAT master stack for transmission to the servo driver.

[0056] The robotic arm 840 is configured as an execution-level slave station. The EtherCAT slave controller in its servo control cabinet parses the above message and injects three-phase AC power into the stator winding of the servo motor by the motor driver, thereby converting the digital command into torque output.

[0057] In an optional hardware expansion embodiment, the system also includes a vision sensor 850 driven by the GigE Vision protocol. The sensor's mounting bracket is vertically aligned with the stacking table, and its gigabit Ethernet port is connected to a second network interface card (NIC) of the processing unit 830. In response to a trigger level prior to the start of the fine positioning process, the sensor pushes a two-dimensional grayscale image pixel matrix to the processing unit 830 for image processing operators to extract the centroid coordinate offset pixel values ​​of the silicon steel sheet. The robotic arm performs a coarse alignment displacement based on vision servoing according to these pixel offset values, thereby pulling the workpiece into an effective convergence basin optimized by the magnetic field gradient, greatly suppressing the probability of algorithm divergence caused by coarse loading.

[0058] In summary, through the strict data flow coupling between the aforementioned hardware modules and software threads, this application embodiment constructs a closed-loop automated equipment system configured to directly minimize the physical parameters of magnetoresistive reluctance.

[0059] Although the embodiments of this application have been described above in conjunction with the accompanying drawings, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and these all fall within the scope of protection of this application.

[0060] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit described above can be implemented in hardware.

[0061] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for intelligent stacking of transformer cores based on minimizing leakage flux, characterized in that, include: In response to the silicon steel sheet to be placed entering a stacking station, a background excitation magnetic field is generated based on an excitation magnetic source within the stacking station. The spatial leakage flux distribution introduced by the silicon steel sheet is monitored in real time by a sensor array. Based on the spatial leakage flux distribution, an optimal placement pose is iteratively calculated using an optimization algorithm. This optimal placement pose corresponds to a minimum point in the spatial leakage flux distribution. A robotic arm is driven to place the silicon steel sheet in the optimal placement position.

2. The method according to claim 1, characterized in that, The step of iteratively calculating an optimal placement pose using an optimization algorithm includes: In each iteration of the optimization algorithm, a current pose of the silicon steel sheet is obtained; Based on the current pose, by applying a perturbation displacement to the silicon steel sheet and performing at least two measurements by the sensor array, a gradient estimate of the spatial leakage flux distribution at the current pose is obtained; and The current pose is updated based on the gradient estimate to generate a new current pose.

3. The method according to claim 2, characterized in that, The step of obtaining a gradient estimate of the spatial leakage flux distribution at the current pose by applying a perturbation displacement to the silicon steel sheet and performing at least two measurements by the sensor array is implemented based on a synchronous perturbation stochastic approximation algorithm, wherein the perturbation displacement is a random perturbation vector applied synchronously on all pose degrees of freedom.

4. The method according to claim 3, characterized in that, The synchronous perturbation random approximation algorithm uses a gain sequence that varies with the number of iterations. The gain sequence includes a step-size gain sequence and a perturbation gain sequence.

5. The method according to claim 1, characterized in that, After the silicon steel sheet enters the stacking station, and before the optimal placement pose is iteratively calculated using the optimization algorithm, the process further includes: An initial planar offset of the silicon steel sheet is obtained using a vision sensor; and The robotic arm is driven to perform a compensated movement based on the initial planar offset in order to coarsely position the silicon steel sheet.

6. The method according to claim 1, characterized in that, The method further includes real-time monitoring of the spatial leakage flux distribution introduced by the silicon steel sheet at every moment: Obtain a current magnetic field map data output by the sensor array; Calculate a set of statistical characteristic values ​​for the current magnetic field map data; and The statistical feature values ​​are compared with a preset health status feature threshold range to perform a health status self-check on the sensor array.

7. The method according to claim 6, characterized in that, Prior to the execution of the method, a calibration step is included to establish the threshold range of the health status characteristics. The calibration step includes: The robotic arm is driven to grasp a standard silicon steel sheet and move it to multiple sampling positions within the stacking station; For each of the sampling poses, a calibration magnetic field map is acquired, and its corresponding statistical characteristic value is calculated; and The health status feature threshold range is determined based on the statistical distribution of the statistical feature values ​​obtained from all the sampled poses.

8. A transformer core intelligent stacking system, characterized in that, include: An excitation magnetic source is configured to generate a background excitation magnetic field within a stacking station when the silicon steel sheet to be placed enters a stacking station. A sensor array configured for real-time monitoring of a spatial leakage flux distribution introduced by the silicon steel sheet; A processing device is configured to iteratively calculate an optimal placement pose based on the spatial leakage flux distribution using an optimization algorithm, wherein the optimal placement pose corresponds to a minimum point of the spatial leakage flux distribution. as well as A robotic arm is configured to receive instructions from the processing device and place the silicon steel sheet in the optimal placement position.

9. The system according to claim 8, characterized in that, The processing device is specifically configured for: In each iteration of the optimization algorithm, a current pose of the silicon steel sheet is obtained; The robotic arm is controlled to apply a perturbation displacement based on the current pose, and at least two measurement results are obtained from the sensor array to calculate a gradient estimate of the spatial leakage flux distribution at the current pose; as well as The current pose is updated based on the gradient estimate to generate a new current pose.

10. The system according to claim 9, characterized in that, The processing device calculates the gradient estimate based on a synchronous perturbation stochastic approximation algorithm, wherein the perturbation displacement is a random perturbation vector applied synchronously across all pose degrees of freedom.

Images