Automatic calibration and assembly system and method for transformer winding

Abstract

The invention discloses an automatic calibration assembly system and method for a transformer winding. The method comprises the steps that a multi-mode sensing and fusion module synchronously collects geometric, tension, insulation and electromagnetic data of the winding and fuses the geometric, tension, insulation and electromagnetic data into a multi-physical field fusion sensing data set; the digital twin bidirectional interaction and pre-judgment module receives the data set to drive a multi-field coupling virtual model to synchronize in real time, and performs forward simulation based on a synchronous model to generate a pre-judgment calibration instruction; a multi-field coupling cooperative calibration module, the edge data processing unit is used for locally preprocessing real-time data; the cooperative control algorithm unit is used for receiving the preprocessed data and a pre-judgment instruction and outputting the preprocessed data and the pre-judgment instruction through a fusion graph neural network and an electromagnetic coupling quantification model to generate a cooperative calibration control instruction set; the instruction set is executed to control the winding machine in the system; and the execution control unit is used for carrying out cooperative action on the tension adjusting mechanism and the insulating part conveying device. According to the invention, high-precision, pre-judgment and multi-physical-field cooperative control of the winding assembly process is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of transformer assembly and intelligent control, and particularly relates to a transformer winding automatic calibration assembly system and method. BACKGROUND

[0002] With the continuous development of the power system towards high voltage level, large capacity and intelligentization, the multi-winding transformer applied to the scenes of power transmission and distribution, new energy grid connection and the like puts forward higher requirements on the winding assembly quality. The geometric precision, wire tension consistency, insulation gap stability and electromagnetic coupling state between the multiple windings in the winding assembly process directly affect the operation reliability, short-circuit resistance and long-term service life of the transformer.

[0003] Under the conditions of high voltage level and multi-winding nested structure, the winding assembly process presents significant multi-physical field coupling characteristics: on the one hand, mechanical factors such as wire tension change and structural deviation will cause dynamic changes in winding spatial distribution and insulation gap; on the other hand, the electromagnetic interaction between windings will in turn affect the wire stress state and local structural stability. In addition, the process rhythm in the assembly process is constantly accelerated, so that small deviations are rapidly accumulated in a short time, further amplifying the adverse effects of multi-field coupling on assembly precision.

[0004] However, the existing transformer winding assembly and calibration method is mainly based on the monitoring and adjustment of a single or a small number of physical quantities, and it is difficult to uniformly perceive and comprehensively evaluate the geometric, mechanical, electromagnetic and other multi-dimensional states existing simultaneously in the assembly process. Under this condition, the assembly system can only make passive corrections after the deviation has been formed, and it is difficult to identify the potential precision instability trend in time, especially in the high-speed winding or multi-winding collaborative assembly scene, which is more likely to cause problems such as local stress concentration, uneven insulation gap or electromagnetic performance degradation.

[0005] At the same time, with the deepening application of digital technology in the field of equipment manufacturing, the demand for virtual modeling and simulation of the assembly process is increasing. However, in actual application, there is often a lack of effective real-time interaction mechanism between the virtual model and the physical assembly process, and the virtual analysis results are difficult to directly participate in the assembly control decision, resulting in that the assembly process still mainly relies on experience parameters and static process settings, and cannot be dynamically optimized according to real-time state changes.

[0006] The above factors make it difficult for the existing assembly method to balance assembly efficiency and assembly quality when facing the winding assembly demand of high precision, multiple constraints and multi-field coupling, thereby limiting the further improvement of the overall performance of high voltage level and multi-winding transformer. Therefore, an automatic assembly system capable of facing the multi-physical field coupling characteristics and comprehensively perceiving, dynamically analyzing and collaboratively calibrating the winding assembly process is urgently needed to meet the assembly requirements of high precision, high consistency and high reliability for the new generation of transformers. SUMMARY

[0007] In view of the deficiencies of the prior art, the purpose of the present application is to provide a transformer winding automatic calibration assembly system and method for realizing the coordinated perception and dynamic calibration of winding geometry, tension, insulation and electromagnetic state during the assembly process, thereby improving the assembly precision, stability and overall electromagnetic performance of multiple windings.

[0008] To achieve the above-mentioned purpose, the present application provides the following technical solution: a transformer winding automatic calibration assembly system, comprising:

[0009] A multi-modal perception and fusion module is used to synchronously collect and fuse the geometric data, tension data, insulation data and electromagnetic data of the winding, and output a multi-physical field fusion perception dataset;

[0010] A digital twin bidirectional interaction and prediction module is connected to the multi-modal perception and fusion module, used to receive the multi-physical field fusion perception dataset to drive the internal multi-field coupled virtual model for state synchronization in real time, and generate predictive calibration instructions based on the synchronized multi-field coupled virtual model;

[0011] A multi-field coupled collaborative calibration module is connected to the multi-modal perception and fusion module and the digital twin bidirectional interaction and prediction module, comprising:

[0012] An edge data processing unit is used to locally pre-process the collected geometric data, tension data, insulation data and electromagnetic data to obtain pre-processed feature data;

[0013] A collaborative control algorithm unit is connected to the edge data processing unit, used to receive the pre-processed feature data and the predictive calibration instructions, output cross-field domain correlation calibration parameters through a fusion graph neural network model, and output electromagnetic compensation calibration parameters through an electromagnetic coupling quantization model, and generate a collaborative calibration control instruction set in combination with the predictive calibration instructions;

[0014] An execution control unit is connected to the collaborative control algorithm unit, used to control the winding machine, tension adjusting mechanism and insulation piece conveying device in the system to perform collaborative action according to the collaborative calibration control instruction set, so as to synchronously adjust the geometric positioning, tension distribution, insulation gap and electromagnetic coupling state of the winding.

[0015] Further, the digital twin bidirectional interaction and prediction module comprises:

[0016] A model construction unit is used to construct the multi-field coupled virtual model that fuses geometry, mechanics, thermodynamics and electromagnetic field based on the design parameters of the transformer and the core processing error data;

[0017] A bidirectional interaction unit, connected to the model building unit, includes:

[0018] The real-time synchronization subunit is used to inject the multi-physics field fusion sensing dataset into the multi-field coupled virtual model at a delay lower than a preset delay threshold, thereby driving the model state to synchronize with the physical entity.

[0019] The predictive subunit, connected to the real-time synchronization subunit, is used to perform multi-field coupling simulation ahead of the physical assembly process based on the synchronized multi-field coupling virtual model. When any one of the stress concentration, insulation gap electric field strength or electromagnetic coupling strength in the simulation result exceeds a preset threshold, the predictive calibration command containing compensation parameters is generated.

[0020] Furthermore, the digital twin bidirectional interaction and prediction module also includes:

[0021] The model iterative optimization unit connects the bidirectional interaction unit and the model construction unit. It is used to feed back the actual assembly calibration effect data to the multi-field coupled virtual model after a single or batch assembly is completed, and optimize the simulation parameters of the model through reinforcement learning algorithm so that the prediction accuracy of the predictive subunit is continuously improved in the iteration.

[0022] Furthermore, the digital twin bidirectional interaction and prediction module is connected to a redundant fault-tolerant module, which includes:

[0023] The sensing redundancy unit is used to enable the backup sensing unit for seamless switching when any sensing subunit of the multimodal sensing and fusion module fails.

[0024] The model reconstruction unit, connected to the sensing redundancy unit, is used to instruct the multi-field coupled virtual model to reconstruct the missing physical field parameters based on historical assembly data and real-time data of the associated field when sensing data is continuously missing, so as to support the cooperative control algorithm unit in generating uninterrupted calibration instructions.

[0025] Furthermore, the multi-field coupled virtual model constructed by the model building unit includes:

[0026] A coupled co-simulation engine, and geometric sub-models, mechanical sub-models, thermal sub-models and electromagnetic sub-models respectively connected to the coupled co-simulation engine;

[0027] The coupled co-simulation engine is configured as follows:

[0028] During the advanced simulation process, based on the real-time synchronized physical field data, the following coupled calculation process is dynamically invoked and executed:

[0029] Step a, the winding electromagnetic force distribution data generated by the electromagnetic sub-model calculation is input into the mechanical sub-model in real time as its load boundary condition;

[0030] Step b, the conductor deformation and stress data generated by the mechanical sub-model calculation are fed back to the geometric sub-model in real time to update the winding topography, and are synchronized to the electromagnetic sub-model to correct the geometric boundary of its electromagnetic field calculation;

[0031] Step c, the winding local temperature rise data generated by the thermal sub-model calculation is input into the mechanical sub-model in real time for dynamic correction of the material elastic modulus parameter of the conductor;

[0032] The coupling co-simulation engine further comprehensively evaluates a comprehensive coupling deviation index of the winding assembly based on the results of the iterative calculation of steps a to c, and the predictive calibration instruction is generated by the predictive sub-unit based on whether the comprehensive coupling deviation index exceeds a preset threshold.

[0033] Further, the co-control algorithm unit comprises:

[0034] A feature graph construction sub-unit is configured to construct winding geometric deviation, tension fluctuation, insulation gap, electromagnetic intensity, temperature and environmental parameters in the preprocessed feature data into a relationship feature graph of nodes and edges;

[0035] A multi-objective optimization training sub-unit is connected to the feature graph construction sub-unit and is configured to train the graph neural network model based on a large number of historical assembly samples, with the joint optimization objectives of minimizing concentricity error, maximizing tension uniformity, ensuring insulation gap compliance rate and improving electromagnetic coupling balance.

[0036] The trained graph neural network model is configured to process the relationship feature graph and output a set of cross-field domain calibration suggestion parameters associated with each other, including conductor tension adjustment amount, winding machine speed correction value, insulation laying compensation position and winding phase difference correction value.

[0037] Further, the co-control algorithm unit further comprises:

[0038] A multi-port network parameterization sub-unit is configured to represent a multi-winding system as a network with a specific topology based on multi-port network theory, and extract a scattering parameter matrix representing electromagnetic interference;

[0039] An interference quantization compensation sub-unit is connected to the multi-port network parameterization sub-unit and is configured to calculate quantifiable electromagnetic compensation calibration parameters for offsetting electromagnetic coupling interference according to the amplitude or phase of key elements in the scattering parameter matrix and combining a compensation coefficient dynamically determined by the number of turns of the winding.

[0040] Furthermore, the system is configured to generate the collaborative calibration control instruction set via the following architecture:

[0041] A hierarchical decision-making framework, including:

[0042] The edge data processing unit is configured to perform real-time filtering and feature extraction on the real-time data, and generate a local fast calibration command with first priority when any feature data is detected to exceed its local response threshold.

[0043] The collaborative control algorithm unit is configured to execute:

[0044] Global fusion decision: Receive and synchronously process the preprocessed feature data, the local fast calibration command, and the predictive calibration command from each of the edge data processing units;

[0045] Conflict resolution and integration: When the local rapid calibration command and the predictive calibration command conflict in terms of adjustment target or amplitude, the coupling effect simulation of the conflicting command is performed based on the graph neural network model and the electromagnetic coupling quantization model, and a comprehensive global optimization calibration command is generated based on the principle of optimal overall assembly quality index.

[0046] Instruction set encapsulation: The global optimization calibration instruction, the conflict-free local fast calibration instruction, and the predictive calibration instruction are time- and logic-arranged and encapsulated to generate the final collaborative calibration control instruction set.

[0047] Furthermore, the execution control unit includes:

[0048] The instruction parsing and allocation subunit is used to receive and parse the collaborative calibration control instruction set, and decompose it into independent action instructions corresponding to the winding machine, the tension adjustment mechanism and the insulation component conveying device, respectively.

[0049] A multi-mechanism collaborative timing control subunit is connected to the instruction parsing and allocation subunit. It is used to plan the precise execution timing and logical sequence for each of the independent action instructions, and drive each of the execution mechanisms to start and stop within a preset time window, so as to achieve synchronization and connection of actions.

[0050] The closed-loop fine-tuning execution subunit is connected to the multi-mechanism collaborative timing control subunit. During the execution of the mechanism's actions, it dynamically fine-tunes the action amplitude or speed of each mechanism based on the real-time changes of the corresponding parameters in the multi-physics field fusion sensing dataset, until the target state threshold set by the collaborative calibration control instruction set is reached.

[0051] An automated calibration and assembly method for transformer windings, applied to the automated calibration and assembly system for transformer windings as described above, includes:

[0052] Step S1: Synchronously collect and fuse the geometric data, tension data, insulation data and electromagnetic data of the winding to output a multi-physics field fusion sensing dataset;

[0053] Step S2: Receive the multi-physics field fusion sensing dataset to drive the internal multi-field coupled virtual model to synchronize its state in real time, and perform forward simulation based on the synchronized multi-field coupled virtual model to generate predictive calibration instructions.

[0054] Step S3: Perform local preprocessing on the collected geometric data, tension data, insulation data, and electromagnetic data to obtain preprocessed feature data;

[0055] Step S4: Receive the preprocessed feature data and the predictive calibration command, and generate a collaborative calibration control command set by fusing the cross-field correlation calibration parameters output by the graph neural network model and the electromagnetic compensation calibration parameters output by the electromagnetic coupling quantization model, and combining the predictive calibration command.

[0056] Step S5: According to the coordinated calibration control instruction set, control the winding machine, tension adjustment mechanism and insulation component conveying device to perform coordinated actions to synchronously adjust the geometric positioning, tension distribution, insulation gap and electromagnetic coupling state of the winding.

[0057] The beneficial effects of this invention are:

[0058] Compared with the prior art, the present invention has at least the following beneficial effects:

[0059] 1. Overall improvement in assembly and calibration accuracy: By coordinating and comprehensively calibrating the geometric, mechanical, insulation and electromagnetic states during the winding assembly process, the synchronous optimization of multiple physical field parameters is achieved, effectively suppressing the chain instability problem caused by single parameter deviation, thereby significantly improving the winding concentricity control accuracy, conductor tension consistency and insulation gap stability, and reducing the adverse effects of electromagnetic coupling between multiple windings.

[0060] 2. Significantly enhanced real-time and forward-looking calibration response: By introducing a virtual-real state synchronization and calibration prediction mechanism during the assembly process, the system can perform calibration and adjustment before deviations actually form, significantly reducing the cumulative accuracy error caused by execution delays and dynamic operating condition changes. It is particularly suitable for high-speed winding and multi-winding collaborative assembly scenarios.

[0061] 4. Improved transformer operational reliability and structural stability: By identifying and suppressing potential defect risks caused by multi-physics coupling in advance during the assembly stage, the generation of problems such as stress concentration inside the winding, insulation performance degradation and electromagnetic performance deviation are reduced, thereby improving the winding's short-circuit withstand capability, heat dissipation performance and long-term operational stability, and reducing the probability of transformer failure during service.

[0062] 4. Strong system adaptability and scalability: The system architecture of this invention can adapt to the assembly requirements under different winding numbers, different voltage levels and different process parameters. Through parameter configuration and model adjustment, it can adapt to the assembly tasks of transformer windings of various specifications, and has good versatility and engineering promotion value.

[0063] 5. High level of automation and intelligence: This invention can realize automatic and coordinated control of winding assembly and calibration process, reduce the impact of manual intervention on assembly quality, and continuously optimize control strategy during assembly, thereby reducing dependence on operator experience and improving assembly consistency and production efficiency. Attached Figure Description

[0064] Figure 1 This is a schematic diagram of the structure of the automated calibration and assembly system for transformer windings in this invention; Figure 2 This is a flowchart of the steps in the automated calibration and assembly method for transformer windings in this invention.

[0065] Figure labeling: 1. Multimodal perception and fusion module; 2. Digital twin bidirectional interaction and prediction module; 21. Model building unit; 22. Bidirectional interaction unit; 221. Real-time synchronization subunit; 222. Predictive subunit; 23. Model iterative optimization unit; 3. Multi-field coupling collaborative calibration module; 31. Edge data processing unit; 32. Collaborative control algorithm unit; 321. Feature map construction subunit; 322. Multi-objective optimization training subunit; 323. Multi-port network parameterization subunit; 324. Interference quantization compensation subunit; 33. Execution control unit; 331. Instruction parsing and allocation subunit; 332. Multi-mechanism collaborative timing control subunit; 333. Closed-loop fine-tuning execution subunit; 4. Winding machine; 5. Tension adjustment mechanism; 6. Insulating component conveying device; 7. Redundancy fault-tolerant module; 71. Sensing redundancy unit; 72. Model reconstruction unit. Detailed Implementation

[0066] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Identical components are denoted by the same reference numerals. It should be noted that the terms "front," "rear," "left," "right," "upper," and "lower" used in the following description refer to directions in the accompanying drawings, and the terms "bottom surface," "top surface," "inner," and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0067] Example 1, refer to Figure 1 This is the first embodiment of the present invention, which provides an automated calibration and assembly system for transformer windings.

[0068] I. The overall system structure includes:

[0069] The multimodal sensing and fusion module 1 is used to synchronously collect and fuse the geometric data, tension data, insulation data and electromagnetic data of the winding, and output a multi-physics field fusion sensing dataset.

[0070] The digital twin bidirectional interaction and prediction module 2 is connected to the multimodal perception and fusion module 1. It is used to receive the multi-physics field fusion perception dataset to drive the internal multi-field coupled virtual model to synchronize its state in real time, and to perform forward simulation based on the synchronized multi-field coupled virtual model to generate predictive calibration instructions.

[0071] The multi-field coupled collaborative calibration module 3, connected to the multimodal sensing and fusion module 1 and the digital twin bidirectional interaction and prediction module 2, includes:

[0072] Edge data processing unit 31 is used to perform local preprocessing on the collected geometric data, tension data, insulation data and electromagnetic data to obtain preprocessed feature data;

[0073] The collaborative control algorithm unit 32 is connected to the edge data processing unit 31. It is used to receive preprocessed feature data and predictive calibration instructions, and generate a collaborative calibration control instruction set by fusing the cross-field correlation calibration parameters output by the graph neural network model and the electromagnetic compensation calibration parameters output by the electromagnetic coupling quantization model, and combining the predictive calibration instructions.

[0074] The execution control unit 33 is connected to the collaborative control algorithm unit 32. It is used to control the winding machine 4, tension adjustment mechanism 5 and insulation component conveying device 6 in the system to perform coordinated actions according to the collaborative calibration control instruction set, so as to synchronously adjust the geometric positioning, tension distribution, insulation gap and electromagnetic coupling state of the winding.

[0075] All modules operate collaboratively under a unified control architecture, forming a closed-loop automated assembly system covering "pre-assembly prediction - in-assembly calibration - post-assembly feedback - model iterative optimization".

[0076] This system is suitable for single-winding or multi-winding transformers, and is especially suitable for high-voltage multi-winding transformers with voltage levels of 110kV to 500kV and winding numbers of 2 to 6.

[0077] II. Multimodal Perception and Fusion Module 1:

[0078] The hardware configuration and deployment method for synchronously acquiring geometric data, tension data, insulation data and electromagnetic data during the winding assembly process are as follows.

[0079] (a) Geometric data acquisition hardware

[0080] The system deploys a set of three-dimensional vision positioning devices above and to the side of the winding machine's four frames, including: LiDAR: Velodyne VLP-16; Industrial camera: Basler acA2500-14uc;

[0081] This combination is used to construct a three-dimensional spatial model of the winding and the iron core. It can acquire the three-dimensional coordinate information of the iron core slot, the starting end of the winding and the adjacent winding in real time. Its positioning accuracy reaches ±0.5μm and the sampling frequency is 100Hz. It is used to solve the problems of iron core processing error (≤5μm) and geometric offset during the winding process.

[0082] (ii) Tension data acquisition hardware

[0083] A flexible tension sensing device is embedded at the conductor contact point of the conductor laying mechanism. Specifically, a fiber optic grating tension sensor (FBG-T200) is used. This sensor has a measurement range of 0–500 N, a measurement accuracy of ±1 N, and a sampling frequency of 2 kHz. It is used to monitor the changes in conductor tension in real time during the winding process. The system is set to a safe tension range of 50 N–300 N to prevent the conductor from being plastically stretched due to tension exceeding 300 N, or the winding from becoming loose due to tension falling below 50 N.

[0084] (III) Hardware for Insulation and Thermal Data Acquisition

[0085] Infrared thermal imaging and insulation detection devices are arranged on the four sides of the winding machine, aligned with the winding area, including: infrared thermal imager: FLIR A655sc; ultrasonic thickness gauge: Panametrics-NDT 38DL PLUS.

[0086] Infrared thermal imagers are used to monitor the temperature rise of the winding surface and local areas. The measurement range is -40℃ to 150℃. An early warning is triggered when the local temperature reaches or exceeds 80℃.

[0087] Ultrasonic thickness gauges are used to measure the interlayer insulation gap in real time. The measurement range is 0.1 to 10 mm, and the measurement accuracy is ±0.01 mm. They are used to determine whether the insulation laying meets the requirement of breakdown voltage ≥3kV / mm.

[0088] (iv) Electromagnetic data acquisition hardware

[0089] A dual-ring H-field electromagnetic coupling sensing device is deployed on the inner wall of the winding and in the gap between adjacent windings. Specifically, the dual-ring H-field probe is H-Probe HP-08 (2 to 4 sets). The measurement range of this probe is 0 to 100 mT, and the sampling frequency is 500 Hz. It is used to collect the electromagnetic coupling strength, electric field distribution uniformity and resonance characteristic parameters between multiple windings in real time.

[0090] (v) Data fusion

[0091] The data collected by the various sensors are aligned with a unified timestamp and then fused in the multimodal sensing and fusion module 1 to form a multi-physics fusion sensing dataset containing geometric, mechanical, insulation and electromagnetic features, which is then output to subsequent modules.

[0092] The technical effect of the multimodal sensing and fusion module 1 is that it realizes the synchronous and comprehensive sensing of the state of multiple physical fields during the winding assembly process, avoiding the calibration blind spot caused by insufficient sensing dimensions in traditional systems.

[0093] III. Digital Twin Two-Way Interaction and Prediction Module 2:

[0094] (a) Computing and Communication Hardware Configuration

[0095] The digital twin two-way interaction and prediction module 2 operates within an industrial control cabinet, and its hardware foundation includes:

[0096] Central controller: Siemens S7-1500 industrial controller;

[0097] Edge computing gateway: Advantech EIS-D720;

[0098] 5G Industrial Communication Module: Huawei ME909s-821;

[0099] This combination enables high-speed bidirectional interaction between multimodal sensing data and virtual models, with an overall data transmission latency of no more than 1ms.

[0100] (II) Virtual Model-Driven and Predictive Mechanism

[0101] The digital twin bidirectional interaction and prediction module 2 is connected to the multimodal perception and fusion module 1 to receive multi-physics field fusion perception datasets and drive the internal multi-field coupled virtual model to perform real-time state synchronization.

[0102] The digital twin bidirectional interaction and prediction module 2 is connected to the multimodal perception and fusion module 1 to receive multi-physics field fusion perception datasets and drive the internal multi-field coupled virtual model to perform real-time state synchronization.

[0103] The multi-field coupling virtual model establishes a one-to-one geometric model based on the actual transformer structure and introduces mechanical, electromagnetic and thermal simulation modules. The input parameters include the number of winding turns, wire diameter, insulation thickness, winding spacing and electromagnetic coupling baseline value under no-load conditions (≤20mT).

[0104] Real-Virtual Synchronization: The multi-physics fusion sensing dataset collected by the multi-modal sensing and fusion module 1 is transmitted to the virtual model in real time through the industrial communication network. The data transmission delay is no more than 1ms, enabling the virtual model to dynamically reflect the physical assembly state.

[0105] Virtual-Real Prediction: The virtual model performs forward simulation based on the synchronized state, predicting potential problems 5-10ms before the assembly action, including but not limited to: stress concentration trend (error threshold ≤3%); risk of insufficient insulation gap (threshold <3kV / mm); abnormal electromagnetic coupling strength (threshold >80mT).

[0106] When the above risks are predicted, the module generates corresponding predictive calibration instructions and outputs them to the multi-field coupling collaborative calibration module 3.

[0107] Model Iterative Optimization: The actual calibration results after assembly are fed back to the virtual model. The model parameters are adjusted through reinforcement learning algorithms so that the deviation between the virtual model and the physical entity gradually converges to ≤2%.

[0108] The technical effect of the digital twin bidirectional interaction and prediction module 2 is to transform calibration from "post-compensation" to "pre-intervention", significantly reducing error accumulation under high-speed assembly conditions.

[0109] IV. Multi-field Coupled Collaborative Calibration Module 3:

[0110] (a) Edge Data Processing Unit 31 Hardware

[0111] FPGA edge computing nodes are deployed in the following devices: winding machine 4, tension adjustment mechanism 5, insulating component conveying device 6, and electromagnetic sensing unit.

[0112] The FPGA chip is a Xilinx Kintex-7 K7K325T, used for local preprocessing of high-frequency data with a processing latency of ≤0.5ms.

[0113] (II) Cooperative Control Algorithm Unit 32 and Model Explanation

[0114] The collaborative control algorithm unit 32 runs in the central controller and integrates: preprocessed feature data output by the edge data processing unit 31 and predictive calibration instructions output by the digital twin module.

[0115] Among them, the electromagnetic coupling quantization model is based on scattering parameters. The electromagnetic coupling interference between multiple windings is quantified, among which, Indicates the amplitude of the incident electromagnetic wave. This represents the amplitude of the reflected electromagnetic wave. Using this model, electromagnetic coupling effects, which are inherently difficult to control directly, can be transformed into calculable compensation quantities. The compensation amount is configured as follows: The compensation coefficient A value of 0.8 to 1.2 is used to guide electromagnetic compensation calibration. This compensation amount is used to guide the adjustment of winding phase and tension, thereby reducing undesirable electromagnetic coupling between multiple windings.

[0116] The graph neural network takes multi-physical field features such as geometric offset, tension fluctuation, insulation gap, electromagnetic coupling strength, and temperature rise as input. By learning the correlation between different physical fields, it outputs the conductor tension adjustment, winding machine speed correction value, insulation component laying compensation position, and winding phase correction value.

[0117] (iii) Execution control unit 33 hardware

[0118] The execution control unit 33 controls the coordinated operation of the following devices: winding machine 4 (transformer-specific winding equipment); tension adjustment mechanism 5 (electric actuator: Thomson Electrak HD); and insulation component conveying device 6 (servo motor: Panasonic MSMF042L1U2M). The execution delay is controlled within 2ms.

[0119] (iv) Collaborative control process: The system operates according to the following steps: Multimodal perception and fusion module 1 collects real-time data; edge data processing unit 31 completes feature extraction; collaborative control algorithm unit 32 fuses predictive calibration instructions to generate control instruction set; execution control unit 33 drives the actions of each actuator with an execution delay ≤2ms; the adjustment result is perceived and fed back again to form closed-loop calibration.

[0120] The multi-field coupling collaborative calibration module 3 achieves synchronous dynamic calibration of winding geometric positioning, tension distribution, insulation gap and electromagnetic coupling status.

[0121] Overall technical effects of Example 1:

[0122] Through the above hardware configuration and collaborative working mechanism, this embodiment realizes real-time perception, predictive calibration and collaborative control of multi-physical field states during the assembly of high-voltage multi-winding transformer windings, significantly improving assembly accuracy, response speed and operational reliability.

[0123] Example 2 is the second embodiment of the present invention. Based on Example 1, this embodiment further defines and refines the digital twin bidirectional interaction and prediction module 2 and its internal multi-field coupling virtual model. It focuses on explaining the working principle, model construction method and technical effects of the module in the process of automated winding calibration and assembly.

[0124] This embodiment is also applicable to winding assembly scenarios of high-voltage, multi-winding transformers, and is especially suitable for complex working conditions where there are significant geometric-mechanical-thermal-electromagnetic multi-field coupling effects during the assembly process.

[0125] Working principle of Example 2:

[0126] Digital twin two-way interaction and prediction module 2 includes:

[0127] Model building unit 21 is used to build a multi-field coupled virtual model that integrates geometry, mechanics, thermodynamics and electromagnetic fields based on the transformer's design parameters and core processing error data;

[0128] The bidirectional interaction unit 22 and the connection model construction unit 21 include:

[0129] The real-time synchronization subunit 221 is used to inject the multi-physics field fusion sensing dataset into the multi-field coupled virtual model with a delay of less than a preset delay threshold (1ms), thereby driving the model state to synchronize with the physical entity.

[0130] The predictive subunit 222 is connected to the real-time synchronization subunit 221. It is used to perform multi-field coupling simulation that is ahead of the physical assembly process based on the synchronized multi-field coupling virtual model. When any one of the stress concentration, insulation gap electric field strength or electromagnetic coupling strength in the simulation result exceeds the preset threshold, a predictive calibration command containing compensation parameters is generated.

[0131] Digital twin two-way interaction and prediction module 2 also includes:

[0132] The model iteration optimization unit 23 connects the bidirectional interaction unit 22 and the model building unit 21. It is used to feed back the actual assembly calibration effect data to the multi-field coupled virtual model after a single or batch assembly is completed, and optimize the simulation parameters of the model through reinforcement learning algorithm so that the prediction accuracy of the prediction subunit 222 is continuously improved in the iteration.

[0133] The multi-field coupled virtual model constructed by model building unit 21 includes:

[0134] A coupled co-simulation engine, and a geometric sub-model, a mechanical sub-model, a thermal sub-model, and an electromagnetic sub-model respectively connected to the coupled co-simulation engine;

[0135] Among them, the coupled co-simulation engine is configured to:

[0136] During the advanced simulation process, according to the real-time synchronized physical field data, dynamically call and execute the following coupled calculation processes:

[0137] Step a, input the winding electromagnetic force distribution data calculated by the electromagnetic sub-model into the mechanical sub-model in real time as its load boundary condition;

[0138] Step b, feedback the wire deformation and stress data calculated by the mechanical sub-model to the geometric sub-model in real time to update the winding morphology, and synchronize it to the electromagnetic sub-model to correct the geometric boundary of its electromagnetic field calculation;

[0139] Step c, input the local temperature rise data of the winding calculated by the thermal sub-model into the mechanical sub-model in real time to dynamically correct the material elastic modulus parameter of the wire;

[0140] The above steps are executed in an iterative manner until the preset simulation convergence condition is reached.

[0141] The coupled co-simulation engine further comprehensively evaluates the comprehensive coupling deviation index of the winding assembly based on the results of the iterative calculations in steps a to c, and the predictive subunit 222 specifically generates a predictive calibration instruction based on whether the comprehensive coupling deviation index exceeds a preset threshold.

[0142] In this embodiment, the model construction unit 21 constructs a multi-field coupling virtual model based on the following input data:

[0143] Transformer design parameters: including winding turns, wire diameter, insulation thickness, winding spacing, and rated voltage level;

[0144] Core processing error data: including slot position offset and lamination flatness error, and the error range does not exceed ±5μm;

[0145] Initial process parameters: including set value of winding tension, winding speed, ambient temperature and humidity;

[0146] No-load electromagnetic baseline data: electromagnetic coupling intensity under the no-load condition of multiple windings, set not higher than 20mT.

[0147] The multi-field coupling virtual model includes: a geometric sub-model, a mechanical sub-model, a thermal sub-model, an electromagnetic sub-model, and a coupled co-simulation engine. Each sub-model does not operate independently, but performs multi-field linkage calculations under the unified scheduling of the coupled co-simulation engine.

[0148] Design of the comprehensive coupling deviation index function for multi-field coupled virtual models:

[0149] To comprehensively evaluate the combined impact of geometric, mechanical, thermal, and electromagnetic multi-field coupling on winding assembly accuracy, this embodiment two designs a comprehensive coupling deviation index function, which serves as the criterion for the predictive subunit 222 to generate predictive calibration commands. The comprehensive coupling deviation index function is defined as follows:

[0150] ;

[0151] Where Ω is the comprehensive deviation value output by the multi-field coupling virtual model within a unit assembly time window, i.e., the comprehensive coupling deviation index; Λ is the physical length of the winding along the axial direction, i.e., the effective axial length of the winding; Γ(α) is the mechanical stress weighting factor in the form of a Gamma function, and α is the stress concentration index;

[0152] ζ(β) is the Riemann Zeta function form of the complexity factor of multi-winding electromagnetic coupling, β is the order of electromagnetic coupling; Θ is the upper limit of the time integration of the ahead simulation, i.e. the simulation time look-ahead window;

[0153] Let ν be the radial distribution function of the electromagnetic field in Bessel function form, where ν is the Bessel order. Let r be the spatial frequency of the electromagnetic field, and r be the radial coordinate.

[0154] The thermal diffusion response function is in the form of an error function, where σ is the temperature rise diffusion coefficient and τ is the time variable;

[0155] This is the nonlinear response function of the insulation gap. ε is the insulation sensitivity coefficient, and ε is the insulation gap deviation.

[0156] This is the geometric deformation attenuation coefficient, i.e., the geometric damping factor. The normalized deformation variables output by the mechanical sub-model, i.e., the wire deformation variables;

[0157] This is the length of the integration interval for the elliptic function, i.e., the normalized interval length;

[0158] Let be a Jacobi elliptic function used to characterize the periodic perturbation features of multi-field coupling, where ρ is the coupling modulation frequency. The modulus of the ellipse. For integration variables;

[0159] ω is the normalization function for multi-physics fusion sensing data, and ω is the fusion sensing data vector.

[0160] The theoretical range of the comprehensive coupling deviation index Ω is (0, +∞). When Ω < 1, it indicates that the current multi-field coupling state is in the safe assembly range. When 1 ≤ Ω < Ωth, it indicates that there is a potential deviation trend but has not yet exceeded the safety threshold. When Ω ≥ Ωth, it indicates that the comprehensive coupling deviation exceeds the preset threshold Ωth and triggers the generation of a predictive calibration command. Ωth is set to 1.5 to 2.2 according to different voltage levels and winding numbers.

[0161] This invention introduces a comprehensive coupling deviation index function into the multi-field coupling virtual model to uniformly quantify multi-field coupling effects such as geometric offset, conductor stress, insulation gap variation, and multi-winding electromagnetic coupling.

[0162] By mapping the calculation results from geometric, mechanical, thermal, and electromagnetic sub-models into a single comprehensive deviation index value, the system can reflect the overall coupling state during the assembly process with a comparable and thresholdable index, thereby avoiding the limitations of relying solely on single physical quantities such as geometric errors, tension fluctuations, or electromagnetic strength for judgment.

[0163] When the overall coupling deviation index exceeds a preset threshold, the digital twin bidirectional interaction and prediction module 2 can identify potential risks before physical assembly deviations actually occur, and generate predictive calibration instructions containing compensation parameters, thus enabling proactive intervention in multi-field coupling instability trends. This effectively solves the technical problems of difficult-to-identify multi-field interference, delayed calibration actions, and easy error accumulation in traditional assembly systems, improving the overall accuracy, stability, and reliability of the multi-winding transformer assembly process.

[0164] Working principle of predictive subunit 222:

[0165] The real-time synchronization subunit 221 injects the multi-physics fusion sensing dataset into the multi-field coupled virtual model with a delay of less than 1ms, so that the model state is synchronized with the physical assembly state.

[0166] The predictive subunit 222 performs simulation calculations based on a multi-field coupling virtual model that are 5-10ms ahead of the physical assembly process. When the comprehensive coupling deviation index Ω exceeds the preset threshold Ωth, or when any of the following conditions are met: stress concentration exceeds 3%, electric field strength corresponding to insulation gap exceeds 3kV / mm, or electromagnetic coupling strength exceeds 80mT, a predictive calibration command is generated that includes tension compensation, winding speed correction, insulation laying compensation, and phase correction.

[0167] Technical effect: By using a unified quantitative assessment of comprehensive coupling deviation indicators, the risk of multi-field coupling can be identified in advance, avoiding misjudgment or missed judgment caused by a single physical quantity threshold judgment.

[0168] The working principle of model iterative optimization unit 23:

[0169] After a single or batch assembly is completed, the model iteration optimization unit 23 feeds back the actual assembly calibration effect data to the multi-field coupled virtual model, and adaptively adjusts the parameters α, β, σ, κ, μ, ρ, etc. through reinforcement learning algorithm, so that the error between the comprehensive coupling deviation index Ω and the actual assembly result gradually converges to ≤2%.

[0170] Technical effect: As the number of assembly batches increases, the prediction accuracy of the predictive subunit 222 for multi-field coupling risk continues to improve, further reducing the probability of error accumulation.

[0171] Overall technical effects of Example 2:

[0172] Compared with Example 1, Example 2 introduces a comprehensive coupling deviation index function with multi-function coupling characteristics, enabling the multi-field coupling virtual model to not only reflect the physical state but also to perform in-depth quantitative evaluation of multi-field interaction relationships, thereby achieving higher precision and more robust predictive calibration control in complex multi-winding assembly scenarios.

[0173] Example 3 is the third embodiment of the present invention. Based on the previous embodiments, Example 3 further elaborates on the working mode of the collaborative control algorithm unit 32, the hierarchical decision architecture and the execution control unit 33 to address the problems of parallel changes in multi-field data, calibration command conflicts and the difficulty of collaborative control of multiple actuators. It is applicable to high-speed winding and multi-winding synchronous assembly scenarios.

[0174] II. Structure and working principle of cooperative control algorithm unit 32:

[0175] The collaborative control algorithm unit 32 includes: a feature map construction subunit 321, a multi-objective optimization training subunit 322, a multi-port network parameterization subunit 323, and an interference quantization compensation subunit 324.

[0176] (a) Feature map construction subunit 321

[0177] The feature map construction subunit 321 is used to construct a relational feature map for graph neural network model processing from the preprocessed feature data output by the edge data processing unit 31.

[0178] Preprocessed feature data includes at least: winding geometric offset (μm); conductor tension fluctuation amplitude (N);

[0179] Minimum insulation gap (mm); electromagnetic intensity (mT); local temperature rise (°C); ambient temperature and humidity parameters.

[0180] In the feature graph: each winding node, insulation layer node, and electromagnetic coupling node is abstracted as a graph node;

[0181] Geometric adjacency, mechanical coupling, thermal conduction, and electromagnetic coupling relationships are abstracted as graph edges; the weights of the graph edges are dynamically updated with real-time sensing data.

[0182] Technical effect: By expressing the relationship between multiple physics fields through graph structure, the originally scattered multidimensional data has a structural basis that can be jointly optimized.

[0183] (II) Multi-objective optimization training subunit 322

[0184] The multi-objective optimization training subunit 322 trains the graph neural network model offline and online based on a large number of historical assembly samples. Its joint optimization objectives include: minimizing concentricity error (target threshold ≤ ±0.5μm); maximizing tension uniformity (fluctuation ≤ ±3%); maintaining the insulation gap compliance rate at 100% (breakdown voltage ≥3kV / mm); and improving electromagnetic coupling balance (≥90%).

[0185] The trained graph neural network model is configured to process relational feature maps and output a set of interrelated cross-field calibration recommendation parameters, including at least: conductor tension adjustment ΔT; winding machine speed correction Δn; insulation laying compensation position ΔP; and winding phase difference correction Δφ.

[0186] Technical effect: Avoids the "one gains while the other loses" problem caused by independent adjustment of a single physical quantity, and achieves synergistic optimization of multiple field parameters.

[0187] (III) Multi-port network parameterization subunit 323 and interference quantization compensation subunit 324

[0188] The multi-port network parameterization subunit 323 is based on multi-port network theory, which equates the multi-winding system to a network model with a specific topology and extracts the scattering parameter matrix S characterizing electromagnetic interference.

[0189] Among them, the amplitude and phase of the key scattering parameter elements are used to characterize the electromagnetic coupling strength between different windings.

[0190] The interference quantization compensation subunit 324 calculates the quantifiable electromagnetic compensation calibration parameter ΔE used to counteract electromagnetic coupling interference based on the key elements in the scattering parameter matrix and the compensation coefficient k (range 0.8 to 1.2) dynamically determined by the number of winding turns. The calculation formula is consistent with that in Example 1 and will not be repeated here.

[0191] Technical effect: It transforms the electromagnetic coupling effect, which was originally difficult to directly participate in control decision-making, into an executable compensation quantity, thereby improving the electromagnetic consistency of multi-winding assemblies.

[0192] III. Generation Mechanism of Hierarchical Decision-Making Architecture:

[0193] The system adopts a hierarchical decision-making architecture to generate a collaborative calibration control instruction set.

[0194] (a) Local rapid decision-making of edge data processing unit 31

[0195] The edge data processing unit 31 performs high-frequency filtering and feature extraction on real-time data and sets local response thresholds, such as: tension mutation threshold ≥ ±10N; local temperature rise rate ≥ 5℃ / s; instantaneous electromagnetic intensity exceedance ≥ 80mT.

[0196] When any feature is detected to exceed the local response threshold, the edge data processing unit 31 immediately generates a local fast calibration instruction with first priority to suppress sudden risks.

[0197] (ii) Global fusion decision-making of collaborative control algorithm unit 32

[0198] The collaborative control algorithm unit 32 receives and synchronously processes: preprocessed feature data output by each edge data processing unit 31; local fast calibration instructions; and predictive calibration instructions output by the digital twin bidirectional interaction and prediction module 2.

[0199] When there is a conflict between local rapid calibration instructions and predictive calibration instructions in terms of adjustment target or magnitude, the system uses a graph neural network model and an electromagnetic coupling quantization model to simulate and evaluate the coupling effect of different instruction combinations, and generates a global optimized calibration instruction based on the principle of optimal overall assembly quality index.

[0200] (III) Instruction Set Encapsulation

[0201] The system performs unified timing and logic orchestration on conflict-free local rapid calibration commands, predictive calibration commands, and global optimization calibration commands, and encapsulates them to generate the final collaborative calibration control command set.

[0202] Technical effect: By using a hierarchical decision-making and conflict resolution mechanism, it balances rapid response and global optimization, avoiding assembly instability caused by the superposition of multiple instructions.

[0203] IV. Cooperative execution and closed-loop fine-tuning of the execution control unit 33:

[0204] The execution control unit 33 includes: instruction parsing and allocation subunit 331, multi-mechanism collaborative timing control subunit 332, and closed-loop fine-tuning execution subunit 333.

[0205] (I) Instruction parsing and allocation

[0206] The instruction parsing and allocation subunit 331 parses the collaborative calibration control instruction set into independent action instructions corresponding to the winding machine 4, the tension adjustment mechanism 5, and the insulation component conveying device 6, respectively.

[0207] (ii) Multi-agency collaborative timing control

[0208] The multi-mechanism collaborative timing control subunit 332 plans the precise execution timing and logical sequence for each independent action instruction, so that each actuator can start and stop synchronously within the preset time window, and the execution delay is controlled within 2ms.

[0209] (III) Closed-loop fine-tuning execution

[0210] During the execution of the mechanism's actions, the closed-loop fine-tuning execution subunit 333 dynamically fine-tunes the action amplitude or speed based on the real-time changes of the corresponding parameters in the multi-physics fusion sensing data set, until the relevant parameters reach the following target state thresholds: concentricity error ≤ ±0.5μm; tension fluctuation ≤ ±3%; insulation gap ≥ 3kV / mm; electromagnetic coupling strength ≤ 80mT.

[0211] Technical effect: Through real-time fine-tuning during the execution phase, the impact of execution errors and environmental disturbances on assembly quality is further eliminated.

[0212] Overall technical effects of Example 3:

[0213] Through the above structure and working principle, this embodiment three realizes cross-field collaborative calibration and multi-mechanism synchronous control driven by multi-field data. It effectively solves the problems of difficult resolution of multiple command conflicts, difficulty in balancing response speed and global optimization, and difficulty in suppressing execution stage errors in traditional winding assembly systems, and significantly improves the accuracy, stability and reliability of the automated assembly process of multi-winding transformers.

[0214] Example 4 is the fourth embodiment of the present invention. Based on the previous embodiments, this example 4 addresses the problems mentioned in the background art, such as the high dependence of the assembly system on the reliability of the sensors and the easy occurrence of assembly interruption or loss of accuracy due to local failures. It further expands the digital twin bidirectional interaction and prediction module 2 by introducing a redundant fault-tolerant module 7, so that the system can still maintain multi-field coupling calibration capability and assembly continuity even when some sensor units fail or data is continuously missing.

[0215] II. System Structure of Redundant Fault-Tolerant Module 7:

[0216] The redundancy fault-tolerant module 7 is connected to the digital twin bidirectional interaction and prediction module 2. The whole includes a sensing redundancy unit 71 and a model reconstruction unit 72, and forms a collaborative working relationship with the multimodal perception and fusion module 1 and the multi-field coupled virtual model.

[0217] III. Structure and working principle of the sensing redundancy unit 71:

[0218] (a) Sensor redundancy configuration method

[0219] In the multimodal sensing and fusion module 1, key sensing subunits are configured with primary and backup redundancy, including at least: the primary sensor and backup sensor of the tension sensing subunit; the primary sensor and backup sensor of the electromagnetic coupling sensing subunit; and the primary sensor and backup sensor of the insulation gap detection sensing subunit.

[0220] The main and backup sensors collect data simultaneously under normal operating conditions and compare them through a consistency verification mechanism.

[0221] (II) Fault diagnosis and seamless switching mechanism

[0222] When the deviation between the data collected by the main sensor and the data collected by the backup sensor exceeds a preset consistency threshold, the system determines that the main sensor is abnormal, where the consistency threshold is set to ±3%.

[0223] Once the main sensor is determined to be abnormal, the sensing redundancy unit 71 completes the data switching from the main sensor to the backup sensor within a time window of no more than 10ms, and sends the sensing status update information to the digital twin bidirectional interaction and prediction module 2.

[0224] During the switching process, the system maintains the original data interface and data format unchanged, so that the multi-physics fusion sensing dataset output by the multimodal sensing and fusion module 1 remains continuous.

[0225] Technical benefits: The seamless switching mechanism avoids data interruption or calibration logic failure due to a single sensor malfunction, significantly improving the operational reliability of the assembly system.

[0226] IV. Structure and working principle of model reconstruction unit 72:

[0227] (a) Triggering conditions

[0228] The model reconstruction unit 72 is triggered when any of the following occurs: the main and backup sensors of a certain sensing subunit fail simultaneously; the continuous loss of sensing data exceeds 20ms; or the sensing data is not completely lost, but the noise level exceeds the preset confidence threshold and cannot be directly used for calibration.

[0229] (II) Reconstructing Data Sources and Reconstruction Logic

[0230] After being triggered, the model reconstruction unit 72 instructs the multi-field coupled virtual model to enter the parameter reconstruction mode and reconstructs the missing physical field parameters based on the following data sources: historical assembly data of nearly 100 sets of transformers of the same type; real-time data of other physical fields that are still available in the current assembly process; and the coupling relationship between the sub-models in the multi-field coupled virtual model.

[0231] For example, when electromagnetic intensity data is missing, the model reconstruction unit 72 performs reverse calculations on the input parameters of the electromagnetic sub-model based on the winding topography parameters output by the current geometric sub-model, the conductor stress state output by the mechanical sub-model, and historical electromagnetic coupling samples, thereby obtaining the reconstructed electromagnetic intensity estimate.

[0232] (III) Reconstruction accuracy and threshold control

[0233] The model reconstruction unit 72 sets accuracy constraints on the reconstruction results, requiring that the deviation between the reconstructed physical field parameters and the historical statistical distribution does not exceed ±5%. When the reconstruction accuracy meets the above conditions, the reconstruction parameters are injected into the multi-field coupled virtual model and, as part of the multi-physics fusion sensing dataset, continue to participate in the generation of predictive calibration instructions.

[0234] Technical effect: Even in the absence of sensor data, it can still provide continuous and reliable input to the collaborative control algorithm unit 32, avoiding the interruption of calibration control.

[0235] V. Working principle of the redundancy fault-tolerant module 7 and the digital twin module:

[0236] With the intervention of the redundancy fault-tolerant module 7, the workflow of the digital twin bidirectional interaction and prediction module 2 is as follows:

[0237] The multimodal sensing and fusion module 1 normally collects and outputs a multi-physics fusion sensing dataset; when a sensing anomaly is detected, the sensing redundancy unit 71 completes the master-slave switch; if the data is still missing, the model reconstruction unit 72 triggers and reconstructs the missing parameters; the reconstructed data is injected into the multi-field coupled virtual model in real time for state synchronization and advanced simulation; the digital twin bidirectional interaction and prediction module 2 generates predictive calibration instructions based on the complete or reconstructed data; the collaborative control algorithm unit 32 continuously generates a collaborative calibration control instruction set without interruption during the assembly process.

[0238] Overall technical effects of Example 4:

[0239] By introducing the redundant fault-tolerant module 7, this embodiment effectively solves the problems of high dependence on a single sensing path and easy system shutdown or accuracy loss caused by local faults in the traditional transformer winding assembly system, enabling the system to have the ability to continuously sense, predict and calibrate in complex industrial environments.

[0240] This implementation significantly improves the stability, robustness, and engineering applicability of the transformer winding automated assembly system, and is particularly suitable for continuous production scenarios involving high voltage levels and multi-winding assembly.

[0241] An automated calibration and assembly method for transformer windings, applied to the automated calibration and assembly system for transformer windings as described above, refers to... Figure 2 ,include:

[0242] Step S1: Synchronously collect and fuse the geometric data, tension data, insulation data and electromagnetic data of the winding to output a multi-physics field fusion sensing dataset;

[0243] Step S2: Receive the multi-physics field fusion sensing dataset to drive the internal multi-field coupled virtual model to synchronize its state in real time, and perform forward simulation based on the synchronized multi-field coupled virtual model to generate predictive calibration instructions.

[0244] Step S3: Perform local preprocessing on the collected geometric data, tension data, insulation data, and electromagnetic data to obtain preprocessed feature data;

[0245] Step S4: Receive the preprocessed feature data and the predictive calibration command, and generate a collaborative calibration control command set by fusing the cross-field correlation calibration parameters output by the graph neural network model and the electromagnetic compensation calibration parameters output by the electromagnetic coupling quantization model, and combining the predictive calibration command.

[0246] Step S5: According to the collaborative calibration control instruction set, control the winding machine 4, tension adjustment mechanism 5 and insulation component conveying device 6 to perform coordinated actions to synchronously adjust the geometric positioning, tension distribution, insulation gap and electromagnetic coupling state of the winding.

[0247] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. An automated calibration and assembly system for transformer windings, characterized in that, include: The multimodal sensing and fusion module (1) is used to synchronously collect the geometric data, tension data, insulation data and electromagnetic data of the winding and fuse them to output a multi-physics field fusion sensing dataset. The digital twin bidirectional interaction and prediction module (2) is connected to the multimodal perception and fusion module (1) and is used to receive the multi-physics field fusion perception dataset to drive the internal multi-field coupling virtual model to synchronize its state in real time, and to perform forward simulation based on the synchronized multi-field coupling virtual model to generate a predictive calibration instruction. A multi-field coupled collaborative calibration module (3), connected to the multimodal perception and fusion module (1) and the digital twin bidirectional interaction and prediction module (2), includes: The edge data processing unit (31) is used to perform local preprocessing on the collected geometric data, tension data, insulation data and electromagnetic data to obtain preprocessed feature data; The collaborative control algorithm unit (32) is connected to the edge data processing unit (31) and is used to receive the preprocessed feature data and the predictive calibration instruction. It generates a collaborative calibration control instruction set by fusing the cross-field correlation calibration parameters output by the graph neural network model and the electromagnetic compensation calibration parameters output by the electromagnetic coupling quantization model, and combining the predictive calibration instruction. The execution control unit (33) is connected to the cooperative control algorithm unit (32) and is used to control the winding machine (4), tension adjustment mechanism (5) and insulation component conveying device (6) in the system to perform cooperative actions according to the cooperative calibration control instruction set, so as to synchronously adjust the geometric positioning, tension distribution, insulation gap and electromagnetic coupling state of the winding.

2. The automated calibration and assembly system for transformer windings according to claim 1, characterized in that, The digital twin two-way interaction and prediction module (2) includes: The model building unit (21) is used to build a multi-field coupled virtual model that integrates geometry, mechanics, thermodynamics and electromagnetic fields based on the transformer design parameters and core processing error data; A bidirectional interaction unit (22), connected to the model building unit (21), includes: The real-time synchronization subunit (221) is used to inject the multi-physics field fusion sensing dataset into the multi-field coupled virtual model with a delay lower than a preset delay threshold, thereby driving the model state to synchronize with the physical entity. The predictive subunit (222) is connected to the real-time synchronization subunit (221) and is used to perform multi-field coupling simulation ahead of the physical assembly process based on the synchronized multi-field coupling virtual model. When any one of the stress concentration, insulation gap electric field strength or electromagnetic coupling strength in the simulation result exceeds a preset threshold, the predictive calibration instruction containing compensation parameters is generated.

3. The transformer winding automated calibration and assembly system according to claim 2, characterized in that, The digital twin two-way interaction and prediction module (2) also includes: The model iteration optimization unit (23) connects the bidirectional interaction unit (22) and the model construction unit (21) and is used to feed back the actual assembly calibration effect data to the multi-field coupled virtual model after a single or batch assembly is completed, and optimize the simulation parameters of the model through reinforcement learning algorithm.

4. The automated calibration and assembly system for transformer windings according to claim 1, characterized in that, The digital twin bidirectional interaction and prediction module (2) is connected to a redundant fault-tolerant module (7), which includes: The sensing redundancy unit (71) is used to enable the backup sensing unit for seamless switching when any sensing subunit of the multimodal sensing and fusion module (1) fails. The model reconstruction unit (72) is connected to the sensing redundancy unit (71) and is used to instruct the multi-field coupled virtual model to reconstruct the missing physical field parameters based on historical assembly data and real-time data of associated fields when sensing data is continuously missing, so as to support the cooperative control algorithm unit (32) in generating uninterrupted calibration instructions.

5. The transformer winding automated calibration and assembly system according to claim 2, characterized in that, The multi-field coupled virtual model constructed by the model building unit (21) includes: A coupled co-simulation engine, and geometric sub-models, mechanical sub-models, thermal sub-models and electromagnetic sub-models respectively connected to the coupled co-simulation engine; The coupled co-simulation engine is configured as follows: During the advanced simulation process, based on the real-time synchronized physical field data, the following coupled calculation process is dynamically invoked and executed: Step a: Input the winding electromagnetic force distribution data generated by the electromagnetic sub-model into the mechanical sub-model in real time as its load boundary conditions; Step b: The conductor deformation and stress data calculated by the mechanical sub-model are fed back to the geometric sub-model in real time to update the winding morphology, and synchronized to the electromagnetic sub-model to correct the geometric boundary of its electromagnetic field calculation. Step c: Input the local temperature rise data of the winding generated by the thermal sub-model into the mechanical sub-model in real time to dynamically correct the material elastic modulus parameter of the conductor. The coupled co-simulation engine further evaluates the comprehensive coupling deviation index of the winding assembly based on the results of iterative calculations from step a to step c. The predictive subunit (222) specifically generates the predictive calibration command based on whether the comprehensive coupling deviation index exceeds a preset threshold.

6. The automated calibration and assembly system for transformer windings according to claim 1, characterized in that, The cooperative control algorithm unit (32) includes: The feature graph construction subunit (321) is used to construct the relationship feature graph between nodes and edges from the winding geometric offset, tension fluctuation, insulation gap, electromagnetic strength, temperature and environmental parameters in the preprocessed feature data; The multi-objective optimization training subunit (322) is connected to the feature map construction subunit (321) and is used to train the graph neural network model based on a large number of historical assembly samples with the joint optimization objectives of minimizing concentricity error, maximizing tension uniformity, ensuring insulation gap compliance rate and improving electromagnetic coupling balance. The trained graph neural network model is configured to process the relational feature graph and output a set of interrelated cross-field calibration suggestion parameters, which include at least the conductor tension adjustment, winding machine speed correction value, insulation laying compensation position, and winding phase difference correction value.

7. The automated calibration and assembly system for transformer windings according to claim 1, characterized in that, The cooperative control algorithm unit (32) further includes: Multiport network parameterization subunit (323) is used to characterize a multi-winding system as a network with a specific topology based on multiport network theory and extract the scattering parameter matrix characterizing electromagnetic interference. Interference quantization compensation subunit (324), connected to the multi-port network parameterization subunit (323), is used to calculate quantifiable electromagnetic compensation calibration parameters for offsetting electromagnetic coupling interference based on the amplitude or phase of the key elements in the scattering parameter matrix and the compensation coefficient dynamically determined by the number of winding turns.

8. The automated calibration and assembly system for transformer windings according to claim 1, characterized in that, The system is configured to generate the collaborative calibration control instruction set via the following architecture: A hierarchical decision-making framework, including: The edge data processing unit (31) is configured to perform real-time filtering and feature extraction on the real-time data, and generate a local fast calibration instruction with a first priority when any feature data exceeds its local response threshold. The collaborative control algorithm unit (32) is configured to execute: Global fusion decision: Receive and synchronously process the preprocessed feature data, the local fast calibration command, and the predictive calibration command from each of the edge data processing units (31); Conflict resolution and integration: When the local rapid calibration command and the predictive calibration command conflict in terms of adjustment target or amplitude, the coupling effect simulation of the conflicting command is performed based on the graph neural network model and the electromagnetic coupling quantization model, and a comprehensive global optimization calibration command is generated based on the principle of optimal overall assembly quality index. Instruction set encapsulation: The global optimization calibration instruction, the conflict-free local fast calibration instruction, and the predictive calibration instruction are time- and logic-arranged and encapsulated to generate the final collaborative calibration control instruction set.

9. The automated calibration and assembly system for transformer windings according to claim 1, characterized in that, The execution control unit (33) includes: The instruction parsing and allocation subunit (331) is used to receive and parse the collaborative calibration control instruction set and decompose it into independent action instructions corresponding to the winding machine (4), the tension adjustment mechanism (5) and the insulation conveying device (6), respectively. A multi-mechanism collaborative timing control subunit (332) is connected to the instruction parsing and allocation subunit (331) and is used to plan the precise execution timing and logical sequence for each of the independent action instructions, and drive each of the execution mechanisms to start and stop within a preset time window; The closed-loop fine-tuning execution subunit (333) is connected to the multi-mechanism collaborative timing control subunit (332) and is used to dynamically fine-tune the action amplitude or speed of each mechanism according to the real-time changes of the corresponding parameters in the multi-physics field fusion sensing dataset during the execution of the mechanism action, until the target state threshold set by the collaborative calibration control instruction set is reached.

10. An automated calibration and assembly method for transformer windings, applied to the automated calibration and assembly system for transformer windings as described in any one of claims 1-9, characterized in that, include: Step S1: Synchronously collect and fuse the geometric data, tension data, insulation data and electromagnetic data of the winding to output a multi-physics field fusion sensing dataset; Step S2: Receive the multi-physics field fusion sensing dataset to drive the internal multi-field coupled virtual model to synchronize its state in real time, and perform forward simulation based on the synchronized multi-field coupled virtual model to generate predictive calibration instructions. Step S3: Perform local preprocessing on the collected geometric data, tension data, insulation data, and electromagnetic data to obtain preprocessed feature data; Step S4: Receive the preprocessed feature data and the predictive calibration command, and generate a collaborative calibration control command set by fusing the cross-field correlation calibration parameters output by the graph neural network model and the electromagnetic compensation calibration parameters output by the electromagnetic coupling quantization model, and combining the predictive calibration command. Step S5: According to the collaborative calibration control instruction set, control the winding machine (4), tension adjustment mechanism (5) and insulation component conveying device (6) to perform coordinated actions to synchronously adjust the geometric positioning, tension distribution, insulation gap and electromagnetic coupling state of the winding.

Images