Intelligent monitoring control system for winding of high-frequency transformer

Abstract

The invention belongs to the technical field of transformer manufacturing, and particularly relates to a high-frequency transformer winding intelligent monitoring control system, which is characterized in that an acoustic emission energy data set is constructed by binding high-frequency stress wave signals and winding point space coordinates, a pressure gradient vector of each winding point is determined through two-dimensional interpolation processing and gradient field calculation, and the pressure gradient vector of each winding point is calculated; based on a pressure gradient vector change trend, identifying a pressure non-uniform point and a non-uniform evolution type thereof, matching a corresponding feedforward fine tuning action from a preset strategy library in combination with a relative phase of the pressure non-uniform point in a winding path, and performing active suppression before microdefects are solidified, and an intelligent decision-making mechanism deeply associated with the defect physical type and the process context is provided, acoustic emission signals of subsequent winding points are continuously monitored after the fine tuning action is executed, and the system can continuously accumulate process knowledge through effect evaluation and adaptive optimization of a strategy library, so that the winding control precision is favorably improved.

Description

Technical Field

[0001] This invention belongs to the field of transformer manufacturing technology, specifically a high-frequency transformer winding intelligent monitoring and control system. Background Technology

[0002] In the manufacturing of high-frequency transformers, the winding process is the core factor determining their electrical performance and long-term reliability. During the winding process, the conductors and the frame, as well as the turns, are compacted to form a tight structure, and the uniformity of the internal micro-pressure distribution is crucial. If micro-defects such as inter-turn pressure imbalance or localized insulation stress concentration occur, they will be amplified by frequent power fluctuations and thermal cycling after the transformer is put into operation under dynamic conditions such as smart grids. This can lead to localized overheating, accelerated insulation aging, or even premature winding failure, severely restricting the service life of the equipment and the reliability of the power grid.

[0003] Currently, there are two main technical solutions for quality control during the winding process: First, process monitoring based on traditional electromechanical sensors, such as using tension sensors to monitor the average tension of the wires or using photoelectric encoders to ensure accurate turn counts. Second, the introduction of offline or post-processing inspection technologies, such as indirectly assessing the overall quality after winding through withstand voltage tests, inductance measurements, or using industrial vision to check the neatness of the surface wiring.

[0004] However, existing technologies still have the following limitations: 1. Existing controls are mostly constant tension control based on set values ​​or feedback correction based on visual deviations. This mode can only make delayed corrections after macro defects appear. It cannot predict and intervene in the early stage of the formation of micro pressure unevenness trend. It is a post-event remedy and cannot achieve preventive manufacturing.

[0005] 2. Existing control strategies are singular and rigid, failing to consider the specific physical evolution of uneven pressure or the specific winding process stage in which it occurs, such as in straight sections or end corners. This lack of process specificity leads to limited suppression effects and may even introduce new disturbances, thus restricting further improvement in process accuracy. Summary of the Invention

[0006] To overcome the shortcomings in the background art, embodiments of the present invention provide a high-frequency transformer winding intelligent monitoring and control system, which can effectively solve the problems involved in the background art.

[0007] The objective of this invention can be achieved through the following technical solution: a high-frequency transformer winding intelligent monitoring and control system, comprising: a signal acquisition module, a gradient calculation module, an identification and decision module, and an execution optimization module.

[0008] The signal acquisition module is connected to the gradient calculation module, the gradient calculation module is connected to the identification decision module, the identification decision module is connected to the execution optimization module, and the execution optimization module optimizes the mapping relationship in the preset strategy library and then feeds it back to the identification decision module.

[0009] The signal acquisition module collects high-frequency stress wave signals generated during winding in real time through an acoustic emission sensor array arranged on the winding skeleton. It uses the time difference positioning method to calculate the spatial coordinates of the winding point corresponding to each acoustic emission event and constructs an acoustic emission energy dataset according to the order of the winding path.

[0010] The gradient calculation module performs two-dimensional interpolation on the acoustic emission energy dataset and determines the pressure gradient vector at each winding point through gradient field calculation.

[0011] The identification and decision module identifies pressure non-uniformity points and their non-uniformity evolution types based on the pressure gradient vector change trend. Combining the relative phase of the pressure non-uniformity points in the winding path, it matches corresponding feedforward fine-tuning actions from the preset strategy library. The feedforward fine-tuning actions include fine-tuning of the cable guide pitch angle or fine-tuning of the tension setting value.

[0012] The optimization module continues to monitor the acoustic emission signals of subsequent winding points after the fine-tuning action is performed. It calculates the pressure gradient of the newly generated winding segment to evaluate the adjustment effect and uses the effect data to optimize the mapping relationship in the preset strategy library.

[0013] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) This invention constructs an acoustic emission energy dataset by binding high-frequency stress wave signals with the spatial coordinates of the winding point. The pressure gradient vector field is obtained by interpolation and gradient calculation. The pressure gradient vector change trend is determined by the pressure non-uniformity forming a continuous spatial trend. This triggers feedforward fine-tuning for the subsequent winding process, thereby actively suppressing micro-defects before they solidify, and realizing closed-loop control based on process state prediction.

[0014] (2) This invention determines the type of non-uniform evolution by analyzing the morphology and gradient direction of pressure non-uniformity points, and combines this with the specific phase of the point in the winding path to match targeted fine-tuning action combinations from a preset strategy library. This provides an intelligent decision-making mechanism that is deeply associated with the physical type of the defect and the process context, ensuring the targeting and effectiveness of the control actions. At the same time, through effect evaluation and adaptive optimization of the strategy library, the system can continuously accumulate process knowledge, which helps to improve control accuracy. Attached Figure Description

[0015] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the module connection of the present invention.

[0017] Figure 2 This is a schematic diagram of the process for identifying pressure unevenness points in this invention.

[0018] Figure 3 This is a schematic diagram of the feedforward fine-tuning action matching process of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Reference Figure 1 As shown, the present invention provides an intelligent monitoring and control system for high-frequency transformer windings, comprising: a signal acquisition module, a gradient calculation module, an identification and decision-making module, and an execution optimization module.

[0021] The signal acquisition module is connected to the gradient calculation module, the gradient calculation module is connected to the identification decision module, the identification decision module is connected to the execution optimization module, and the execution optimization module optimizes the mapping relationship in the preset strategy library and then feeds it back to the identification decision module.

[0022] The signal acquisition module collects high-frequency stress wave signals generated during winding in real time through an acoustic emission sensor array arranged on the winding skeleton. It calculates the spatial coordinates of the winding point corresponding to each acoustic emission event using the time difference positioning method and constructs an acoustic emission energy dataset according to the order of the winding path.

[0023] To achieve real-time monitoring of stress distribution during the winding process and obtain precise spatial location data of the winding points, the method of calculating the spatial coordinates of the winding points corresponding to each acoustic emission event using the time difference positioning method includes the following steps: recording the timestamp of each acoustic emission event arriving at each acoustic emission sensor.

[0024] Based on the spatial coordinates of each sensor and the propagation speed of sound waves in the winding skeleton material, the three-dimensional spatial coordinates of the acoustic emission event are calculated using the polygonal positioning principle. The specific implementation method is as follows: First, any sensor is used as a reference sensor, and the time difference between the arrival of the same acoustic emission event at other sensors and the arrival at the reference sensor is recorded. Based on the propagation speed of the sound wave in the winding skeleton material, each time difference is converted into a corresponding distance difference.

[0025] Then, using the spatial coordinates of the acoustic emission event as unknowns, and based on the known coordinates of each sensor and the distance difference calculated above, a set of positioning equations is established. This set of equations represents the difference between the distance from the acoustic emission event to each sensor and the distance to the reference sensor, which is equal to the corresponding distance difference.

[0026] Finally, the three-dimensional spatial coordinates of the acoustic emission event are calculated by solving the set of localization equations using existing numerical iterative algorithms such as the least squares method.

[0027] Furthermore, since the results of polygonal positioning are easily affected by sensor placement, sound wave propagation errors, and environmental noise, it is necessary to fuse and correct the real-time motion coordinates of the winding machine at the moment of acoustic emission event with the three-dimensional spatial coordinates. Specifically, the real-time motion coordinates in the winding machine coordinate system are first transformed to the acoustic emission sensor array coordinate system through a pre-calibrated transformation matrix to obtain the mechanical measurement coordinates. Then, based on the mechanical measurement coordinates and the three-dimensional spatial coordinates obtained by the polygonal positioning principle, a linear weighted average is used to fuse the data to obtain the final result.

[0028] The weighting of the mechanical measurement coordinates and the three-dimensional spatial coordinates in the above linear weighted average is independently and dynamically set based on the historical positioning errors of acoustic emission positioning and mechanical positioning for each coordinate axis: For each coordinate axis direction, the historical error variance of acoustic emission positioning and mechanical positioning in that direction is calculated. The sum of the historical error variances of the two is used as the denominator, and the historical error variance of each is used as the numerator to calculate the proportion of the historical error variance of each. The difference between 1 and the proportion of the historical error variance is used as the weighting. According to this rule, the larger the historical error variance, the lower its weight in the fusion. Thus, the weighting of the mechanical measurement coordinates and the three-dimensional spatial coordinates in each coordinate axis direction is obtained.

[0029] To enable gradient analysis of stress distribution during winding, the gradient calculation module performs two-dimensional interpolation on the acoustic emission energy dataset and determines the pressure gradient vector at each winding point through gradient field calculation.

[0030] To construct a continuously differentiable energy distribution for gradient calculation, the acoustic emission energy dataset is subjected to two-dimensional interpolation, including: projecting the current winding layer onto a two-dimensional plane coordinate system composed of the axial and radial directions of the winding skeleton.

[0031] Map the energy values ​​of all acoustic emission events located in the current winding layer in the acoustic emission energy dataset to the corresponding positions in the two-dimensional plane coordinate system.

[0032] By setting the axial and radial grid spacing, the two-dimensional plane is divided into regular rectangular grids, and the location points are mapped to regular grid nodes.

[0033] The event energy value is interpolated on the regular grid node using a bilinear interpolation method. The specific execution process is as follows: taking the current grid node as the interpolation target point, the known energy values ​​of the grid cell in which it is located and the four corner points of the cell are determined.

[0034] A linear interpolation is performed along the axial direction on the energy values ​​at the two diagonal points to obtain two intermediate values. Then, a linear interpolation is performed along the radial direction on these two intermediate values.

[0035] The result of the second interpolation is used as the energy estimate at that grid node. The calculation is completed by traversing all grid nodes, thereby generating a continuous energy density distribution covering the entire current winding layer region.

[0036] The step of determining the pressure gradient vector at each winding point through gradient field calculation includes: Based on the continuous energy density distribution, for each regular grid node, the central difference method is used to calculate its first directional partial derivative along the axial direction and the second directional partial derivative along the radial direction in the two-dimensional plane coordinate system.

[0037] Based on the first and second directional partial derivatives, a pressure gradient vector is constructed at each regular grid node. The direction angle of the pressure gradient vector is determined by the ratio of the second and first directional partial derivative values ​​and by inverse trigonometric function calculation, indicating the direction of the fastest decrease in pressure energy level. The magnitude of the pressure gradient vector is obtained by vector synthesis calculation of the magnitudes of the first and second directional partial derivatives, and is used to characterize the drastic degree of pressure change.

[0038] In this embodiment of the invention, an acoustic emission energy dataset is constructed by binding high-frequency stress wave signals with the spatial coordinates of the winding point. The pressure gradient vector field is obtained through interpolation and gradient calculation. By judging the trend of pressure gradient vector change, the non-uniformity of pressure is immediately identified when it forms a continuous spatial trend, and feedforward fine-tuning for subsequent winding processes is triggered, thereby actively suppressing micro-defects before they solidify, and realizing closed-loop control based on process state prediction.

[0039] The identification and decision module identifies pressure non-uniformity points and their non-uniformity evolution types based on the pressure gradient vector change trend. Combining the relative phase of the pressure non-uniformity points in the winding path, it matches corresponding feedforward fine-tuning actions from the preset strategy library. The feedforward fine-tuning actions include fine-tuning of the cable guide pitch angle or fine-tuning of the tension setting value.

[0040] To locate regions of abrupt stress changes within the energy density distribution, it is necessary to filter and aggregate all regular mesh nodes to form significantly non-uniform regions, referring to... Figure 2 As shown, the process of identifying pressure non-uniformity points in the identification decision module includes: traversing the pressure gradient vectors of all regular grid nodes and selecting nodes with a magnitude greater than a preset gradient magnitude threshold as candidate nodes.

[0041] Using the candidate nodes as seeds and the growth condition that the gradient magnitudes of adjacent nodes all exceed a preset gradient magnitude threshold, region growth is performed based on spatial adjacency relationships to aggregate and form several connected candidate regions.

[0042] For each candidate region, the consistency index of the pressure gradient vector direction of all nodes within it is obtained by calculating the average of the pairwise dot products of each vector in the region. If the consistency index is greater than the preset direction consistency threshold, the candidate region is determined to be a valid pressure non-uniform region.

[0043] The preset gradient modulus threshold and the preset direction consistency threshold are dynamically determined by statistical analysis of the distribution characteristics of historical qualified samples. Taking the preset gradient modulus threshold as a specific example: under stable process conditions, multiple qualified winding samples are wound, and the pressure gradient field data generated by each sample during the winding process are obtained.

[0044] Based on the pressure gradient field data of all qualified winding samples, the statistical distribution characteristics of the pressure gradient magnitude at each regular grid node are calculated, and the statistical distribution characteristics include the mean and standard deviation.

[0045] For each regular grid node, the mean of its corresponding pressure gradient magnitude is added to a certain number of standard deviations to serve as the preset gradient magnitude threshold for that node. Alternatively, the percentile of the statistical characteristics of all nodes, such as the 95th percentile, can be taken as a uniform preset gradient magnitude threshold.

[0046] For each effective pressure non-uniform region, the spatial coordinates of all regular grid nodes in the region are arithmetically averaged in a two-dimensional plane coordinate system. That is, the average value of the axial coordinates of all regular grid nodes is calculated as the axial coordinate of the representative point, and the average value of the radial coordinates of all regular grid nodes is calculated as the radial coordinate of the representative point. The average coordinate point obtained is defined as the geometric center of the region and serves as the representative point in space.

[0047] The pressure gradient vector of each regular grid node in the region is decomposed into axial and radial components. The axial and radial component values ​​of all regular grid nodes are summed to obtain the total axial and radial components of the vector sum. This total axial and radial components are then used to synthesize the region's directional representative vector. This vector sum reflects the overall trend of stress variation within the region.

[0048] The spatial location representative point is marked as the pressure non-uniformity point, and its corresponding direction representative vector and the range information of the area to which it belongs are recorded.

[0049] To distinguish different types of stress distribution defects to support precise control, it is necessary to classify the evolution types of the identified pressure non-uniform regions. Therefore, the process of identifying the non-uniform evolution type in the identification decision module includes: extracting the contour boundary of the pressure non-uniform region and calculating the ratio of its axial span to radial span as the aspect ratio feature.

[0050] The directional angle range from 0° to 360° is divided into several directional intervals at equal intervals according to a preset angular span. All pressure gradient vectors in the pressure non-uniform region are traversed, and each vector is assigned to its corresponding directional interval according to its directional angle. The number of vectors contained in each directional interval is counted, and this number is divided by the total number of vectors in the pressure non-uniform region to obtain the normalized frequency of each directional interval. Thus, a histogram of the directional distribution of pressure gradient vectors in the pressure non-uniform region is drawn.

[0051] Based on the aforementioned directional distribution histogram, the existence of a dominant direction or a ring-shaped rotation pattern is identified. The specific identification process is as follows: Dominant direction: If the histogram of this direction contains one or two consecutive peak intervals with normalized frequencies significantly higher than other intervals, for example, the normalized frequency of the peak interval can be set to be greater than twice the average frequency of all other intervals, or greater than 1.5 times the frequency of the second-highest peak interval, and the cumulative frequency of the peak interval accounts for more than 60% of the total vector, then the region is determined to have a dominant direction. It should be noted that the specific values ​​used in the above criteria for determining the dominant direction are conventional examples, and those skilled in the art can adjust them according to actual monitoring sensitivity requirements.

[0052] The specific angle value of the dominant direction can be obtained by calculating the average direction angle of all vectors within the peak interval. The positive axial direction is set as the 0° reference direction, and the positive radial direction as the 90° reference direction. The maximum historical error variance of acoustic emission positioning in axial and radial measurements is selected, and an angle tolerance threshold is defined as an integer multiple thereof. .

[0053] If the specific angle value of the dominant direction falls within or If the range is within a certain range, then the dominant direction is determined to be concentrated in the axial direction.

[0054] If the specific angle value of the dominant direction falls within or Within the interval, the dominant direction is determined to be radial.

[0055] If the specific angle value of the dominant direction does not belong to any of the above intervals, it is determined to be another direction.

[0056] Circular Rotation Mode: The information entropy of the actual directional distribution relative to the ideal directional distribution is calculated from the histogram of directional distribution. The ideal directional distribution refers to a distribution where the normalized frequencies of each interval are equal. This information entropy is then compared to the nominal information entropy of an ideal uniform distribution within the same interval division, and normalization is performed to obtain the information entropy ratio of the actual directional distribution. If this information entropy ratio is greater than a preset uniformity judgment value, it indicates that the directional distribution is nearly uniform and there is no single dominant peak. Furthermore, this uniformity judgment value can be exemplified as 0.85, based on the fact that the entropy ratio of an ideal uniform distribution is 1. If the entropy ratio of the actual distribution reaches or exceeds 0.85, from an information theory perspective, it can be considered that most of the uncertainty has been preserved, and the distribution is close to a uniform state, sufficient to determine that there is no dominant direction. This value is set based on experience, and those skilled in the art can adjust it within a range of, for example, 0.80 to 0.95, depending on the stringency of the directional consistency requirements.

[0057] Based on this, the normalized frequency sequence of each interval of the directional distribution histogram is regarded as a periodic signal, and a discrete Fourier transform is performed on it to analyze its frequency domain characteristics. If the transformed spectrum shows discrete peaks with energy higher than the noise floor at the fundamental frequency corresponding to one or more complete circular periods, this mathematical characteristic indicates that the original directional distribution exhibits a multi-periodic undulation pattern similar to a sine wave, thus determining that there is a ring-shaped rotation mode in this region.

[0058] If the aspect ratio of the shape is greater than the baseline standard and the gradient direction is concentrated in the axial direction, it is determined to be axial accumulation type. Axial accumulation type represents the distribution pattern of stress anomaly accumulating and extending along the skeleton axis during the winding process. It usually corresponds to the linear stress concentration zone in a single layer winding caused by uneven wire arrangement or instantaneous tension change.

[0059] If the aspect ratio of the shape is close to the baseline standard and the gradient direction is concentrated in the radial direction, it is determined to be radial diffusion type. Radial diffusion type represents the distribution pattern of stress anomaly spreading in the radial direction of the winding. It usually corresponds to the ring stress distribution caused by uneven interlayer pressure, coil stacking or local deformation of insulation material.

[0060] It should be noted that the aforementioned benchmark specifically refers to the value 1. This is based on the fact that in a two-dimensional plane composed of axial and radial axes, a value of 1 represents a region whose axial and radial spans are equal, meaning its shape is square or circular. Geometrically, this state is characterized as an isotropic benchmark without a clearly defined direction of extension. Therefore, a value greater than 1 is used as the mathematical condition for judging whether a region has a significant axial extension.

[0061] The term "approaching the baseline standard" specifically refers to the aspect ratio characteristics falling within a preset tolerance range centered on the baseline value 1. This tolerance range can be determined by fitting the region's contour and using statistical tests to determine whether its aspect ratio differs significantly from the value 1, such as a t-test. If the p-value of the test is greater than the set significance level, usually 0.05 as statistically empirically defined, then it can be determined as approaching the baseline.

[0062] If the gradient direction exhibits a ring-shaped rotation pattern, it is determined to be a local vortex type. This type indicates that the stress distribution presents a complex pattern of local rotation or vortex, which is usually caused by transient events such as instantaneous slippage, slight torsion, or dynamic interference with the guide device during the winding process.

[0063] If none of the above conditions are met, it is classified as an undefined "other type." The pressure gradient characteristics, spatial location, and contextual process parameters of this "other type" case are automatically stored in a case unit to be analyzed. When a certain number of such cases are accumulated, technical personnel or the upper-level system can be prompted to perform offline analysis to determine whether it is a new, definable type of non-uniform evolution, and to provide a data foundation for subsequent updates to the classification system and strategy entries of the pre-built strategy library.

[0064] To achieve targeted process compensation, identified problem points need to be mapped to specific equipment control commands. This process relies on a pre-built strategy library, which stores multiple strategy entries. Each entry is associated with at least one winding path phase interval, one non-uniform evolution type, and at least one fine-tuning action generation rule, as shown in the reference... Figure 3 As shown, the feedforward fine-tuning action matching process is as follows: The identified pressure non-uniformity points are mapped to the corresponding winding path phase intervals. Combined with the identified non-uniformity evolution type, a joint search is performed in the preset strategy library to lock in the matching strategy entries.

[0065] Based on the fine-tuning action generation rules associated with the matched strategy entry, and combined with the axial and radial component values ​​of the current pressure gradient vector, the specific adjustment command value is calculated.

[0066] The fine-tuning action generation rule specifies the quantitative relationship between the adjustment command value and the pressure gradient vector component value.

[0067] As a quantitative example of specific implementation, this quantitative relationship can be defined in the form of a linear mapping function. For the pitch angle adjustment of the cable guide, its function output is the angle compensation amount. This compensation amount is related to the magnitude and sign of the component of the pressure gradient vector in the direction perpendicular to the winding direction. The first proportional coefficient is determined by regression analysis of the axial gradient change and the optimal pitch angle compensation amount in historical data. The product of the first proportional coefficient and the component of the pressure gradient vector in the direction perpendicular to the winding direction is used as the pitch angle fine-tuning value of the cable guide.

[0068] For tension setting adjustment, the function output is the tension correction amount, which is related to the magnitude and sign of the component of the pressure gradient vector along the winding direction. Similarly, the second proportional coefficient is determined by regression analysis of the radial gradient change and the optimal tension compensation amount in historical data. The product of the second proportional coefficient and the component of the pressure gradient vector along the winding direction is used as the fine-tuning value of the tension setting.

[0069] In practical applications, the fine-tuning action generation rules can also be represented as piecewise linear functions, polynomials, or nonlinear mapping relationships based on machine learning models, with the specific form determined according to the actual response characteristics of the process system. The parameters of the first and second proportional coefficients or more complex models are trained and calibrated using historical process data and feedback on control effects, and stored in the corresponding strategy entries of the preset strategy library.

[0070] Furthermore, during system initialization, the policy entries in the pre-built policy library can establish an initial mapping relationship through one or more of the following methods: (a) Based on the winding process simulation model, corresponding optimization and fine-tuning actions are generated for different phases and preset typical defect types.

[0071] (b) Summarize the process adjustment experience rules of domain experts and quantify them into initial adjustment rules.

[0072] (c) Analyze a large amount of acoustic emission gradient data from historical qualified winding processes and the corresponding equipment parameter logs, and establish an initial phase-gradient mode-adjustment correlation through data mining. The construction of this initial mapping relationship can be achieved based on conventional techniques or existing data in this field, and will not be elaborated on here. After the system is put into operation, continuous online optimization will be performed by executing the optimization module.

[0073] When the matching result indicates that a coordinated fine-tuning of the cable guide pitch angle and tension setting value is required, the primary and secondary relationship and synthesis method between the two need to be determined. The generation of the adjustment command value follows the following: Based on the non-uniform evolution type and winding path phase, determine whether the current adjustment method should be primarily the cable guide pitch angle fine-tuning or primarily the tension setting value fine-tuning. The specific determination process is as follows: The winding path phase is pre-divided into multiple continuous intervals with clear mechanical characteristics based on the geometric model of the winding skeleton and the preset winding law. Typical intervals include the starting segment, straight segment, transition segment, curved segment, and ending segment.

[0074] If the uneven evolution type is axially cumulative, and the pressure unevenness point falls within the winding start segment or straight segment phase, then the current adjustment method should be primarily based on fine-tuning of the wire guide pitch angle. This is because during winding in the start and straight segments, the axial positioning accuracy of the wire is extremely sensitive to the projection angle of the wire guide, and the accumulation of axial stress is directly related to the pitch angle deviation. Performing pitch angle-dominant adjustments at this phase can most directly correct the axial landing point trajectory of the wire.

[0075] If the uneven evolution type is radial diffusion, and the pressure unevenness point falls within the transition section of the winding layer, then the primary adjustment method should be fine-tuning the tension setpoint. This is because such phases are usually accompanied by changes in coil density or inter-layer switching, and radial stress diffusion mainly stems from uneven inter-turn pressure distribution. Adjusting the tension can directly and effectively change the radial embedding force and circumferential tightness of the wire, making it the preferred direct means to optimize the pressure distribution in this phase.

[0076] If the non-uniform evolution type is local eddy current, and if the phase corresponds to the high-speed winding section, which usually corresponds to the straight section or the high-speed section of a specific process design, then tension fine-tuning should be the primary approach. Dynamic instability caused at high speed is the main cause of eddy current generation. Tension adjustment can quickly change the lateral stiffness and dynamic characteristics of the wire. If the phase corresponds to the curved section, then pitch angle fine-tuning should be the primary approach. When winding the curved section, the wire is prone to directional accumulation and torsion due to geometric constraints. Targeted correction of the pitch angle can effectively alleviate the local rotation mode caused by the path curvature.

[0077] Assign weights to the pitch angle fine-tuning and tension setpoint fine-tuning of the cable guide, where the weight of the dominant attribute object is higher than that of the auxiliary attribute object, and the sum of their weights is 1.

[0078] The final feedforward compensation action is obtained by multiplying the pitch angle adjustment value and the tension setting value of the cable guide by their corresponding weights.

[0079] Based on the aforementioned matching and fine-tuning actions, to ensure the safety and timeliness of control commands, the system also performs feasibility verification and conflict avoidance processing before executing the feedforward fine-tuning actions: obtaining the mechanical limit parameters of the current winding machine's wire guide and the response delay characteristics of the tension control system. The mechanical limit parameters include the maximum and minimum pitch angles, which define the permissible mechanical angle range. The response delay characteristics include the stabilization time and the delay time, with the sum of these two times used as the minimum action response period.

[0080] The matched pitch angle adjustment is superimposed on the current pitch angle to determine whether it exceeds the mechanically permissible angle range.

[0081] The fine-tuning amount of the tension setpoint is superimposed on the current tension value to determine whether it exceeds the upper or lower limit of the tension controller's output.

[0082] If any fine-tuning amount exceeds the feasible range, it will be scaled proportionally to the feasible range. The specific scaling process is as follows: For each fine-tuning amount that exceeds the feasible range boundary, calculate the proportion of its excess portion to the original fine-tuning instruction size. This proportion reflects the severity of the exceedance.

[0083] Among all the control quantities that need to be adjusted, the calculated maximum over-limit ratio is selected, and a uniform scaling factor is calculated and applied to both the original pitch angle fine-tuning amount and the tension setpoint fine-tuning amount. This results in a set of final fine-tuning instructions that are reduced by the same ratio. This scaling factor ensures that if the original pitch angle and tension fine-tuning instructions are reduced proportionally by this factor, all the expected values ​​calculated after scaling will fall exactly within or be within their safe boundaries.

[0084] The system synchronously checks whether the time interval between the current feedforward fine-tuning action and the action executed in the previous cycle is less than the minimum action response cycle. If so, the current instruction is delayed until the next available cycle, or it is superimposed and merged with the instruction to be executed in the next cycle.

[0085] After the fine-tuning action is performed, the execution optimization module continues to monitor the acoustic emission signals of subsequent winding points, calculates the pressure gradient of the newly generated winding segment to evaluate the adjustment effect, and uses the effect data to optimize the mapping relationship in the preset strategy library.

[0086] The above optimization process specifically includes: The state space is defined as the spatial location, gradient direction, and non-uniformity evolution type of the current non-uniform pressure point in the winding layer.

[0087] The motion space is defined as a weighted combination of the pitch angle fine adjustment of the cable guide, the tension setpoint fine adjustment, and their combined adjustments.

[0088] The reward function is defined as the percentage decrease in the magnitude of the pressure gradient vector of the newly generated winding segment downstream of the pressure non-uniformity point after the fine-tuning action is performed.

[0089] After each fine-tuning action is executed, a new acoustic emission signal is collected and a reward value is calculated. The state, action and reward value are combined to form an experience tuple and stored in the experience playback buffer. The state includes two states before and after the fine-tuning action is executed, which are marked as the original state and the updated state.

[0090] Periodically sample experience tuples from the experience replay buffer and use the policy gradient algorithm to update the parameters of the corresponding state-action pairs in the pre-set policy library: For each sampled experience tuple, obtain the probability density of the action space before and after the fine-tuning action is executed, and use the ratio of the probability density after execution to the probability density before execution as the importance sampling ratio.

[0091] The generalized advantage estimation is used to calculate the advantage function value of the state-action pair, and an alternative loss function is constructed by combining the importance sampling ratio.

[0092] Calculate the gradient of the substitution loss with respect to the policy network parameters, and update the parameters along the gradient direction using an adaptive optimizer. Essentially, this step adjusts the policy network to increase the probability of selecting actions that yield a higher advantage in the original state.

[0093] After the update is completed, the output distribution parameters of the original state corresponding to the trained policy network are extracted and written into the preset policy library, overwriting the original corresponding entries, thereby completing the iterative optimization of the state-action mapping relationship in the policy library.

[0094] It should also be noted that if the reward function value does not increase in multiple consecutive adjustments for the same type of uneven pressure region, a strategy reset is triggered, re-initializing a set of conservative adjustment parameters for that state-action pair. This set of conservative adjustment parameters is derived from statistical analysis of defect-free normal winding process data. For any initial state, its corresponding action parameters are initialized to the statistical average of the fine-tuning amounts used during historical normal winding in that state.

[0095] This invention analyzes the morphology and gradient direction of pressure non-uniformity points to determine their non-uniformity evolution type. Combined with the specific phase of this point in the winding path, it matches targeted fine-tuning action combinations from a pre-set strategy library. This provides an intelligent decision-making mechanism deeply correlated with defect physical types and process context, ensuring the targeting and effectiveness of control actions. Simultaneously, through effect evaluation and adaptive optimization of the strategy library, the system can continuously accumulate process knowledge, contributing to improved control accuracy.

[0096] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined by the present invention, they should all fall within the protection scope of the present invention.

Claims

1. A high-frequency transformer winding intelligent monitoring and control system, characterized in that, include: The signal acquisition module collects high-frequency stress wave signals generated during winding in real time through an acoustic emission sensor array arranged on the winding skeleton. It uses the time difference positioning method to calculate the spatial coordinates of the winding point corresponding to each acoustic emission event and constructs an acoustic emission energy dataset according to the order of the winding path. The gradient calculation module performs two-dimensional interpolation on the acoustic emission energy dataset and determines the pressure gradient vector at each winding point through gradient field calculation. The identification and decision module identifies pressure non-uniformity points and their non-uniformity evolution types based on the pressure gradient vector change trend. Combining the relative phase of the pressure non-uniformity points in the winding path, it matches corresponding feedforward fine-tuning actions from the preset strategy library. The feedforward fine-tuning actions include fine-tuning of the cable tray pitch angle or fine-tuning of the tension setting value. The optimization module continues to monitor the acoustic emission signals of subsequent winding points after the fine-tuning action is performed. It calculates the pressure gradient of the newly generated winding segment to evaluate the adjustment effect and uses the effect data to optimize the mapping relationship in the preset strategy library.

2. The intelligent monitoring and control system for high-frequency transformer windings according to claim 1, characterized in that, The calculation of the spatial coordinates of the winding point corresponding to each acoustic emission event using the time difference positioning method includes: For each acoustic emission event, record the timestamp of its arrival at each acoustic emission sensor; Based on the spatial coordinates of each sensor and the propagation speed of sound waves in the winding skeleton material, the three-dimensional spatial coordinates of the acoustic emission event are calculated using the principle of polygonal positioning. The real-time motion coordinates of the winding machine at the moment of the acoustic emission event are fused and corrected with the three-dimensional spatial coordinates to obtain the spatial coordinates of the winding point corresponding to the acoustic emission event after correction.

3. The intelligent monitoring and control system for high-frequency transformer windings according to claim 1, characterized in that, Two-dimensional interpolation processing is performed on the acoustic emission energy dataset, including: Project the current winding layer onto a two-dimensional plane coordinate system consisting of the axial and radial axes of the winding skeleton; Map the energy values ​​of all acoustic emission events located in the current winding layer in the acoustic emission energy dataset to the corresponding position points in the two-dimensional plane coordinate system; The location points are then processed into a grid to form regular grid nodes; Using bilinear interpolation, the event energy value is interpolated at the regular grid nodes to generate a continuous energy density distribution covering the entire current winding layer region.

4. The intelligent monitoring and control system for high-frequency transformer windings according to claim 3, characterized in that, The step of determining the pressure gradient vector at each winding point through gradient field calculation includes: Based on the continuous energy density distribution, for each regular grid node, its first directional partial derivative along the axial direction and its second directional partial derivative along the radial direction in the two-dimensional plane coordinate system are calculated respectively. Based on the first directional partial derivative and the second directional partial derivative, a pressure gradient vector at each regular grid node is constructed by combining them. The direction angle of the pressure gradient vector is determined by the ratio of the second directional partial derivative value to the first directional partial derivative value through inverse trigonometric function calculation, and its magnitude is obtained by vector synthesis calculation of the magnitudes of the first directional partial derivative and the second directional partial derivative.

5. The intelligent monitoring and control system for high-frequency transformer windings according to claim 4, characterized in that, The process of identifying pressure non-uniformity points in the identification decision module includes: Iterate through the pressure gradient vectors of all regular grid nodes and select nodes whose magnitude is greater than the preset gradient magnitude threshold as candidate nodes; Using the candidate nodes as seeds, region growth is performed based on spatial adjacency relationships to aggregate and form several connected candidate regions; For each candidate region, calculate the consistency index of the pressure gradient vector direction of all nodes within it. If the consistency index is greater than the preset direction consistency threshold, the candidate region is determined to be a valid pressure non-uniform region. For each effective pressure non-uniform region, the spatial representative point is determined by calculating the geometric center, and the vector sum of the pressure gradient vectors within the region is calculated to determine the direction representative vector. The spatial location representative point is marked as the pressure non-uniformity point, and its corresponding direction representative vector and the range information of the area to which it belongs are recorded.

6. The intelligent monitoring and control system for high-frequency transformer windings according to claim 5, characterized in that, The process of identifying uneven evolution types in the identification decision module includes: Extract the contour boundary of the pressure uneven region and calculate the ratio of its axial span to radial span as the aspect ratio feature. Analyze the histogram of the directional distribution of pressure gradient vectors in regions of uneven pressure to identify whether there is a dominant direction or a ring-shaped rotation pattern. If the aspect ratio of the shape is greater than the baseline standard and the gradient direction is concentrated in the axial direction, it is determined to be an axially cumulative type. If the aspect ratio of the shape is close to the baseline standard and the gradient direction is concentrated in the radial direction, it is determined to be a radial diffusion type; If the gradient direction exhibits a circular rotation pattern, it is determined to be a local vortex type; If none of the above conditions are met, then it is classified as "other".

7. The intelligent monitoring and control system for high-frequency transformer windings according to claim 1, characterized in that, The pre-set strategy library stores multiple strategy entries, each entry being associated with at least one winding path phase interval, one non-uniform evolution type, and at least one fine-tuning action generation rule. The feedforward fine-tuning action matching process is as follows: The identified pressure non-uniformity points are mapped to the corresponding winding path phase intervals. Combined with the identified non-uniformity evolution type, a joint search is performed in the preset strategy library to lock in the matching strategy entries. Based on the fine-tuning action generation rules associated with the matched strategy entry, and combined with the axial and radial component values ​​of the current pressure gradient vector, the specific adjustment command value is calculated. The fine-tuning action generation rule specifies the quantitative relationship between the adjustment command value and the pressure gradient vector component value.

8. The intelligent monitoring and control system for high-frequency transformer windings according to claim 7, characterized in that, When the feedforward compensation action is matched with the coordinated fine-tuning of the cable tray pitch angle and tension setting value, the generation of the adjustment command value follows the following: Based on the aforementioned non-uniform evolution type and winding path phase, it is determined that the current adjustment method should be either fine-tuning the pitch angle of the cable guide or fine-tuning the tension setting value. Assign weights to fine-tune the pitch angle and tension setpoint of the cable guide, with the dominant attribute object having a higher weight than the auxiliary attribute object; The final feedforward compensation action is obtained by multiplying the pitch angle adjustment value and the tension setting value of the cable guide by their corresponding weights.

9. The intelligent monitoring and control system for high-frequency transformer windings according to claim 1, characterized in that, Optimizing the mapping relationships in the pre-built strategy library using performance data includes: The state space is defined as the spatial location, gradient direction, and non-uniformity evolution type of the current non-uniform pressure point in the winding layer. The motion space is defined as a weighted combination of the cable guide pitch angle fine adjustment, the tension setpoint fine adjustment, and their combined adjustments. The reward function is defined as the percentage decrease in the magnitude of the pressure gradient vector of the newly generated winding segment downstream of the pressure non-uniformity point after the fine-tuning action is performed. After each fine-tuning action is performed, a new acoustic emission signal is collected and a reward value is calculated. The state, action and reward value are combined to form an experience tuple and stored in the experience replay buffer. Periodically sample experience tuples from the experience replay buffer and use the policy gradient algorithm to update the parameters of the corresponding state-action pairs in the pre-set policy library.

10. The intelligent monitoring and control system for high-frequency transformer windings according to claim 1, characterized in that, Before the feedforward fine-tuning action is executed, feasibility verification and conflict avoidance handling are also included: Obtain the mechanical limit parameters of the current winding machine's wire guide and the response delay characteristics of the tension control system; The matched cable tray pitch angle fine-tuning amount is superimposed with the current pitch angle to determine whether it exceeds the mechanical allowable angle range; The fine-tuning amount of the tension setpoint is added to the current tension value to determine whether it exceeds the upper or lower limit of the tension controller's output. If any adjustment exceeds the feasible range, it will be scaled down proportionally to the feasible range. The system synchronously checks whether the time interval between the current feedforward fine-tuning action and the action executed in the previous cycle is less than the minimum action response cycle. If so, the action instructions are delayed or merged.

Images