Ethernet magnetic device winding closed loop control method

Abstract

This invention discloses a closed-loop control method for winding Ethernet magnetic devices, belonging to the field of closed-loop control technology for winding Ethernet magnetic devices. This method acquires input quantities such as spindle encoding, wire position, wire tension, guide posture, pin positioning, clamping state, and skeleton reference. After time alignment, outlier removal, window slicing, and normalization, a window quantity is formed. A target corridor is generated, jointly defined by leakage inductance, distributed capacitance, channel symmetry, and margin constraints. A state vector is constructed, and a low-energy correction response is collected at the end of the preset window to correct the mapping relationship. Subsequent winding control commands and quality labels are output, which helps improve the electrical performance convergence stability during the winding process of Ethernet magnetic devices.

Description

Technical Field

[0001] This invention relates to the field of closed-loop control technology for Ethernet magnetic device winding, specifically to online sensing, process judgment and control of Ethernet magnetic devices such as network transformers and integrated differential and common mode magnetic devices during the winding manufacturing process. Background Technology

[0002] In the manufacturing process of 10 / 100 / 1000Base-T network transformers and related Ethernet magnetic devices, the winding stage typically requires simultaneous consideration of the number of turns, wire placement, wire tension, lead-out path, and channel pairing consistency. The leakage inductance, inter-winding distributed capacitance, channel symmetry, and high-frequency transmission consistency of these devices often accumulate and change gradually with variations in the conductor's placement within the bobbin window, slot transposition, and interlayer compaction processes. Therefore, the winding process itself is a crucial factor affecting the final electrical performance.

[0003] Several representative existing technologies have emerged to address this scenario. Firstly, patent US4746075A discloses a precision coil winding control scheme, which controls the lateral movement and winding action of the wire through a preset geometric contour, making the coil landing point closer to the ideal arrangement. This approach indicates that existing technologies typically use geometric trajectory and landing accuracy as the core of control, but their analysis focuses on geometric consistency, making it difficult to directly determine the subsequent impact of geometric deviation on the leakage inductance and distributed capacitance of Ethernet magnetic devices. Secondly, patents CN112110297B address constant tension wire release and winding stability control, and CN114910009A addresses real-time visual monitoring of winding turn distance, demonstrating that existing technologies also improve process stability through tension closed-loop or visual monitoring. However, these approaches primarily focus on mechanical process quantities or abnormal appearance arrangements, especially when cross-slot transposition, pin lead-out, and multi-channel pairing coexist, making it difficult to determine whether the final electrical performance can still converge to the target range.

[0004] Besides process control approaches, existing technologies also focus on device structure design or post-production testing. Patent US6239557B1 reduces parasitic capacitance through winding topology and segmented arrangement; the IEEE paper "Modeling and Testing of Ethernet Transformers" explains that leakage inductance and cross-winding capacitance directly affect the high-frequency transmission behavior of Ethernet magnetic devices; and patent CN103234450A discloses an online monitoring scheme that uses high-frequency response to infer winding status. These published schemes and papers demonstrate that leakage inductance, distributed capacitance, and symmetry are clearly defined performance boundaries of concern for this type of device. However, existing technologies still largely remain at the level of structural pre-design, service monitoring, or post-production testing, making it difficult to establish a stable correlation between process quantity changes and final electrical performance risks before the winding process is complete, and also making it difficult to determine whether current deviations can be corrected in subsequent steps.

[0005] Therefore, in the process of manufacturing Ethernet magnetic devices, how to combine multi-source process data in continuous winding scenarios to determine the convergence risk of electrical performance related to the current winding state and the final leakage inductance, distributed capacitance, and channel symmetry, and further distinguish whether the current deviation can still be corrected through subsequent steps, has become a technical problem that needs to be solved in this field. Summary of the Invention

[0006] To address the problem that existing Ethernet magnetic device winding processes mainly focus on controlling turn counting, tension, and wire placement, making it difficult to determine whether the final leakage inductance, distributed capacitance, channel symmetry, and coupling consistency can still converge to the target range before the winding is finished, thus easily exposing electrical performance mismatches in subsequent electrical testing processes, this invention provides a closed-loop control method for Ethernet magnetic device winding, which moves the determination and correction of final electrical performance to the winding manufacturing process.

[0007] This invention establishes a closed-loop control chain based on the geometric and electromagnetic states that gradually develop during the winding process of Ethernet magnetic devices. The winding equipment continuously collects spindle encoding values, wire placement values, wire tension values, guide posture values, pin positioning values, clamping status values, and skeleton reference values, and simultaneously obtains correction values ​​for online correction. This method performs time alignment, outlier removal, window slicing, and normalization processing on these process quantities to form window quantities corresponding to the current step, ensuring that subsequent judgments are based on a unified window-level data reference.

[0008] After obtaining the window size, the system generates a target corridor by combining the product formula, the current process step, and the completed winding structure. The target corridor maps leakage inductance, distributed capacitance, channel symmetry, coupling consistency, lead-out path, and margin requirements to the convergent range of the current window, and is used to determine whether the current window still has the conditions to converge to the target electrical performance range under the action of the remaining process steps. Around this target corridor, the system further constructs a state vector, organizing the effective number of turns in the current layer, the coil centroid position, the layer margin margin, the cross-slot phase, the compaction degree, the torsion trend of the lead-out section, and the predicted quantities related to subsequent electrical performance into the same judgment object, thereby linking the geometric arrangement changes with the electrical performance convergence trend.

[0009] To reduce the cumulative deviation between geometric observations and the actual electromagnetic state, the system injects low-energy, narrow pulses into the formed winding using an auxiliary coupling coil at the end of the preset window and acquires the corresponding transient response. The transient response is used to correct the mapping relationship between the geometric state and electromagnetic parameters, forming a corrected state vector. Based on this, the system re-solves the control strategy and synchronously outputs the control results to the spindle controller, wire laying controller, tension controller, wire release brake, and transposition mechanism, ensuring that the spindle acceleration, wire laying step distance, tension reference value, guide attitude, and wire passing sequence are coordinated and adjusted around the same target corridor.

[0010] Building upon this foundation, the system further utilizes subsequent compensation capacity to evaluate the recoverability of the current deviation within the remaining steps, and accordingly makes tiered decisions between online compensation, partial rework, and termination of processing. This allows for early differentiation between correctable and non-convergent deviations during the winding process, and synchronizes risk information to the assembly station interface and electrical testing station interface via quality tags. For multi-winding, multi-channel devices, the system also allows preceding windings to release constraint information to subsequent windings, and when both differential-mode and common-mode functional windings are involved, it integrates both types of performance constraints into the same control strategy solution process.

[0011] To achieve the above objectives, the present invention adopts the following technical solution.

[0012] The winding closed-loop control method acquires input and correction quantities during equipment operation. Input quantities at least cover spindle encoding values, wire placement values, wire tension values, guide attitude values, pin positioning values, fixture clamping status values, and skeleton reference values. Correction quantities reflect the online correction results at the end of the current window. This method first aligns the execution time of various sampling results according to the spindle cycle time and step boundaries, then combines outlier removal, window slicing, and normalization to form window quantities. The window boundaries are determined based on single-layer completion points, preset array completion points, cross-pin swapping points, and insulation insertion points, ensuring that window division aligns with changes in the winding structure.

[0013] The system generates a target corridor based on the product formula, the current process step, the completed winding structure, and window quantities. The target corridor characterizes the state range within which the current window can still converge to the target electrical performance range under subsequent process steps, and is jointly constrained by leakage inductance constraints, distributed capacitance constraints, channel symmetry constraints, coupling consistency constraints, lead-out path constraints, and edge distance constraints. Around the target corridor, the system constructs a state vector based on the window quantities. The state vector at least incorporates the current layer's effective number of turns, coil centroid position, layer edge distance margin, slot-crossing phase, compaction degree, lead-out section torsion trend, predicted leakage inductance deviation, predicted distributed capacitance gradient, channel symmetry error, and subsequent compensation capacity, characterizing the current window's geometric state, electromagnetic trend, and remaining correction space.

[0014] When the system reaches the preset window end position, it injects a low-energy narrow pulse into the formed winding through an auxiliary coupling coil and acquires the corresponding transient response. Based on the transient response, the system corrects the mapping relationship between the geometric state and electromagnetic parameters to obtain a corrected state vector. Subsequently, the system combines the corrected state vector and the target corridor to solve for the control strategy and outputs control commands to the spindle controller, wire laying controller, tension controller, wire release brake, and transposition mechanism. These control commands at least affect the subsequent spindle acceleration, wire laying step distance, tension reference value, guide attitude, and wire passing sequence, ensuring continuous adjustment of the subsequent winding process around the target corridor.

[0015] The system also performs tiered decision-making on the current deviation based on the subsequent compensation capacity. The subsequent compensation capacity is jointly determined by the remaining logarithm, the remaining number of layers, the allowed number of cross-line steps, the current coil centroid deviation, the current predicted leakage inductance deviation, the current predicted distributed capacitance gradient, and the current channel symmetry error. It is used to measure the likelihood that the current deviation will return to the target corridor within the remaining steps. When the subsequent compensation capacity meets the online compensation conditions, the system outputs a compensation path instruction; when the subsequent compensation capacity meets the local rework conditions, the system outputs a rework path instruction; when the subsequent compensation capacity no longer meets the convergence conditions, the system outputs a termination processing instruction and a quality label. The quality label at least includes the current window risk level, the convergence status of the remaining steps, and the downstream electrical testing concerns, and is output to the assembly station interface and the electrical testing station interface.

[0016] In a further embodiment, the target corridor is mapped to a combination of window-level geometric constraints and window-level electromagnetic constraints according to the current construction step. The window-level geometric constraints at least cover the centroid of the landing point, layer-to-layer distance, cross-slot phase, line-crossing sequence, and torsional trend of the lead-out section, while the window-level electromagnetic constraints at least cover leakage inductance, distributed capacitance, channel symmetry, and coupling consistency. The system updates the control reference for the next window based on whether the current window is still within the convergence range corresponding to the combined constraints.

[0017] In a further embodiment, time alignment, outlier removal, window slicing, and normalization are performed according to a unified process step time scale. Tension sampling, cable routing sampling, guide posture sampling, and pin positioning sampling are first uniformly aligned according to the spindle coding cycle, then window slicing is performed based on the window boundaries, and normalization is completed based on the skeleton reference size and the current process step reference. This allows process quantities with different sampling cycles and dimensions to be stably incorporated into the same window quantity caliber.

[0018] In a further embodiment, the transient response obtained from low-energy narrow-pulse correction is used to continuously update the mapping between the geometric state and electromagnetic parameters. The control strategy then redistributes the adjustments for spindle control, wire laying control, tension control, braking control, and transposition control based on the updated corrected state vector. Compensation path commands primarily affect the spindle acceleration, wire laying step distance, tension reference value, guide attitude, and wire passing sequence adjustments in subsequent windows. Rework path commands primarily affect the stabilization window rollback, release of the limited-length conductor, and determination of the rewinding start point. Quality tags record the current window risk level, remaining step convergence status, downstream electrical testing concerns, and the corresponding batch identifier.

[0019] In a further embodiment, when the target device is a multi-winding, multi-channel magnetic device, the preceding winding releases constraint information to the subsequent winding after completing the current window. This constraint information at least covers the occupied window area, coupling baseline, symmetry compensation interval, and lead-out path limitations. The subsequent winding then combines this constraint information with its own target corridor to solve for the control strategy. When the target device simultaneously contains differential-mode and common-mode functional windings, the target corridor further incorporates differential-mode and common-mode performance constraints, and maps these two types of constraints to the same control strategy through a unified state vector.

[0020] The beneficial effects of this invention are: By uniformly collecting input values ​​such as spindle encoding values, wire laying position values, wire feeding tension values, guide posture values, pin positioning values, fixture clamping status values, and skeleton reference values, and performing time alignment, outlier removal, window slicing, and normalization on these input values, a window quantity corresponding to the current step can be formed. Existing technologies typically control based on the count value, tension value, or a single appearance quantity, making it difficult to stably incorporate data from different cycles and dimensions into the same processing caliber, which easily leads to process judgment being affected by local disturbances. Based on the above processing chain, subsequent state quantity construction and control strategy solution can be established on the same window-level data reference, which is beneficial to improving the stability of winding process judgment and providing a more reliable data foundation for subsequent closed-loop control.

[0021] By generating a target corridor based on the product formula, current process step, completed winding structure, and window quantity, and using leakage inductance constraints, distributed capacitance constraints, channel symmetry constraints, coupling consistency constraints, lead-out path constraints, and edge distance constraints to determine whether the current window is still within the target electrical performance convergence range, the final electrical performance boundary can be moved forward to the winding process control stage. Existing technologies focus more on adjusting geometric arrangement accuracy or tension stability; even if the process quantities appear normal, it is difficult to directly indicate whether the final leakage inductance and distributed capacitance can still converge. Utilizing this target corridor to continuously judge the current window allows the controlled object to be directly aligned with the electrical performance direction of interest to the Ethernet magnetic device, which is beneficial for improving the electrical performance consistency between different process steps and reducing the risk of mismatch being exposed only during subsequent electrical testing.

[0022] By constructing a state vector based on window quantities, which includes the current effective number of turns, coil centroid position, layer edge distance margin, slot phase, compaction degree, lead-out section torsion trend, predicted leakage inductance deviation, predicted distributed capacitance gradient, channel symmetry error, and subsequent compensation capacity, geometric, process, and electrical performance-related state quantities can be incorporated into a single judgment object. Existing technologies, when detecting deviations, often only provide a rough judgment of whether to stop or continue based on a single abnormal quantity, making it difficult to quantify the recoverability of the current deviation in the remaining steps. With the help of this state vector and its subsequent compensation capacity, the current deviation can be processed and controlled more finely, improving the rationality of anomaly handling and enhancing the closed-loop control's adaptability to different deviation sources.

[0023] By injecting low-energy, narrow pulses into the formed winding using an auxiliary coupling coil at the end of a preset window and acquiring the transient response, the mapping relationship between the geometric state and electromagnetic parameters can be corrected based on this transient response. This yields a corrected state vector that more closely reflects the actual conditions of the current batch and the current process step. Existing technologies, which infer electromagnetic states solely from geometric landing points, tension, and visual arrangement, are easily affected by skeleton tolerances, wire outer diameter drift, and fixture zero-position deviations, leading to accumulated mapping errors. Using this corrected response to re-feed back the mapping relationship allows the feedback chain to reflect the actual winding state more promptly, thereby improving the accuracy of control strategy updates and enhancing the reliability of the process closed loop.

[0024] By solving the control strategy based on the modified state vector and the target corridor, and then outputting control commands to the spindle controller, cable laying controller, tension controller, cable release brake, and transposition mechanism, spindle acceleration, cable laying step distance, tension reference value, guide attitude, and cable passing sequence can be incorporated into the same control chain. Existing technologies typically adjust individual actuators separately, which can easily introduce new layout deviations or path deviations during local correction, resulting in a lack of coordination between multiple control variables. Linked control based on a unified control strategy enables different actuators to coordinate their actions around the same state variable and the same target corridor, improving the consistency of cable routing and enhancing control connections between consecutive windows.

[0025] By implementing hierarchical decision-making based on the subsequent compensation capacity for the current deviation, outputting compensation path instructions when online compensation conditions are met, rework path instructions when local rework conditions are met, and termination of processing instructions and quality tags when convergence conditions are not met, the anomaly handling can be expanded from binary stop judgment to hierarchical path selection. Existing technologies, after a deviation occurs, often directly stop the machine, scrap the entire part, or rely on manual experience for backtracking. This makes it difficult to determine which deviations still have correction margins and easily leads to unstable rework paths. With this hierarchical decision chain and quality tag write-back, downstream assembly station interfaces and electrical testing station interfaces can obtain risk information in advance, which is beneficial for improving the traceability of anomaly handling and providing more favorable collaborative conditions for subsequent processes.

[0026] By releasing constraint information such as occupied window areas, coupling baselines, symmetry compensation intervals, and lead-out path limitations from preceding windings to subsequent windings in multi-winding, multi-channel device scenarios, and by using a unified state vector to map both types of performance constraints to the same control strategy when both differential-mode and common-mode functional windings are included, cross-winding disturbances can be incorporated into the same closed-loop framework. Even if existing technologies can handle single-winding steps or single-type performance indicators separately, they struggle to simultaneously consider the window occupancy relationship between preceding and subsequent steps, as well as the interplay between differential-mode and common-mode constraints, in multi-channel scenarios. Utilizing the aforementioned constraint release and unified mapping processing, subsequent windings can maintain a calculable control baseline under existing occupancy conditions, which is beneficial for improving the coordination of the winding process in multi-channel devices and enhancing the control effect of channel pairing consistency. Attached Figure Description

[0027] Figure 1 The process of this invention Figure 1 ; Figure 2 The process of this invention Figure 2 Figure 3 The process of this invention Figure 3 . Detailed Implementation

[0028] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0029] Example 1 like Figure 1 As shown, this embodiment takes a set of differential windings and their paired return windings on a single Ethernet network transformer skeleton as the object, and discloses a winding closed-loop control method driven by a target electrical performance corridor. The equipment inputs include spindle encoding values, cable placement values, cable tension values, guide attitude values, pin positioning values, clamping state values, and skeleton reference values. The spindle encoding value is in the dimension of the number of encoding pulses, the cable placement value is in millimeters, the cable tension value is in Newtons, the guide attitude value is in angles, and the skeleton reference value is in millimeters. The pin positioning value is given by the pin visual positioning result or pin detection record, and the clamping state value is given by the clamping state sampling channel. The above inputs are collected once per spindle servo control cycle, and the sampling time point is fixed at a preset encoding point before the spindle encoding cycle reaches the current window boundary. The window boundary is jointly determined by the single-layer completion point, the preset turn array completion point, the cross-pin transposition point, and the insulation insertion point. The product formula also specifies the current winding topology, target leakage inductance range, target distributed capacitance range, channel symmetry constraints, coupling consistency constraints, lead-out path constraints, and edge distance constraints. Outputs are divided into two categories: control outputs and decision outputs. Control outputs are routed to the spindle controller, cable management controller, tension controller, cable release brake, and transposition mechanism. Decision outputs are routed to the local compensation decision chain and downstream abnormal current distribution chain.

[0030] At the beginning of the working process in this embodiment, time alignment, outlier removal, window slicing, and baseline normalization are performed on the input quantities, and the window quantities are generated according to formula (1): in, Indicates the first The original spindle encoding sequence within each window, and These represent the starting and ending coding boundaries of the window, respectively, and are derived from the current step boundary record; This represents the original cable position sequence. This indicates the zero position of the ribbon cable corresponding to the current window. This indicates the effective width of the skeleton corresponding to the window, which is derived from the skeleton reference value and the step reference record; Represents the original tension sequence. This indicates the reference tension for each work step, derived from the current product formulation. Indicates the guidance attitude sequence. , and These represent the center value, upper boundary value, and lower boundary value of the guide attitude, respectively, and are derived from the guide mechanism calibration record. Indicates the pin positioning sequence. Indicates the center position of the target pin. The reference pitch between adjacent pins is derived from pin detection records and skeleton drawings; This indicates the normalized result of the clamping state, derived from the mapping of the clamping state value. Formula (1) compresses input quantities of different dimensions and different cycles into the same window size, and the output goes to the subsequent state vector construction stage. The window length is determined by... The only certainty is that the update method is to recalculate in real time according to the boundary of the next window after each window ends.

[0031] After obtaining the window size, the system constructs a state vector and simultaneously calculates the subsequent compensation capacity. The state size is determined according to formula (2): in, This indicates the number of valid turns that have been formed in the current window. Indicates the position of the coil's center of gravity. Indicates the layer edge spacing allowance. Indicates the cross-slot phase. Indicates the degree of compaction. This indicates that the preceding paragraph reverses the trend. This indicates the predicted leakage inductance deviation. This represents the predicted distributed capacitance gradient. This represents the channel symmetry error; the above quantities are calculated jointly from the window quantity, the skeleton reference value, and the completed winding structure, and are used as the first... A description of the status of each window. , and These represent the remaining number of turns, the remaining number of layers, and the allowed number of crossings, respectively, derived from the current work progress record; This indicates the current deviation of the coil's center of gravity, originating from... The difference between the center of gravity and the target position. to These are non-negative weighting coefficients, and their range of values ​​is limited to [value range missing]. and satisfy as well as The compensation capacity calibration record in the current product formula is used; its update time is fixed when the product is switched or after the batch is approved, and it is not changed in real time within the window. Formula (2) converges the geometric state, electrical performance trend and remaining correction space into the same state object, and the output destination is the target corridor determination and subsequent control strategy solution link.

[0032] The system then generates the target corridor based on the state vector and product formula, and calculates the corridor deviation of the current window according to formula (3): in, , , , and These represent the allowable deviations for leakage inductance, distributed capacitance, symmetry error, edge distance bandwidth, and lead-out path bandwidth, respectively, derived from the target corridor constraint record in the current product formulation. Indicates the target center of the window-level distributed capacitance. This indicates the symmetrical target center of the window-level channel, derived from a verified process window; to The weighting coefficient for corridor deviation is limited to a range of values. And sum as The data originates from the target corridor configuration record and is updated after product switching or process window review. Formula (3) compresses multidimensional constraints into a single convergence criterion, and the output destination is the online correction update and control command solution stage; when When it is determined that the current window is still within the convergence range of the target electrical performance, It is determined that at least one type of constraint overflow exists.

[0033] At the end of each preset window, the auxiliary coupling coil injects a low-energy narrow pulse into the formed winding and acquires the transient response. Then, the state correction and control command solution are completed according to formula (4): in, This represents the time difference between the rising edge and the peak value of the transient response. This represents the equivalent impedance offset. These represent the coupling differential offset, and all three originate from the transient response sampling after the current window ends. , , and To correct the gain vector components, the values ​​are derived from historical calibration results and the current step number, and the range of values ​​is limited to non-negative real numbers, and remains fixed within the same batch; The state correction vector is of length 10, and its dimension is the same as that of the state vector. Consistent. This represents the recipe baseline control quantity for the current process step, with a dimension of 5, corresponding to spindle acceleration, threading step distance, tension reference value, guide posture correction amount, and threading sequence correction amount; and For scalar gain, For dimension The control mapping coefficient matrix is ​​initially derived from the reference control template corresponding to the current product topology. The update method is to replace the entire system according to the new mapping version only after the batch review is passed, and not to drift online within a single window. Formula (4) uses the online correction results to correct the state, and then maps the target corridor deviation and compensation capacity synchronously to the control instructions of the next window. The output destinations correspond to the spindle controller, the cable controller, the tension controller, the guide attitude actuator and the transfer mechanism, respectively.

[0034] When an anomaly occurs during operation, the system calculates the anomaly score according to formula (5) and decides whether to continue online compensation: in, This represents the tension peak, which is obtained from the maximum deviation of the tension sequence within the window relative to the reference tension. The position jump is represented by the absolute maximum value of the position difference between adjacent sampling points; This indicates the spindle encoding skip count, which is obtained by counting the encoding skip gaps within the window. to This is the weighting coefficient for abnormal ratings, and its value range is limited to [value range missing]. And sum as This information originates from the anomaly criteria records in the current product formulation. The maximum anomaly score threshold allowed for online compensation. The minimum compensation capacity threshold required for online compensation is derived from the abnormal threshold records that have passed process review, and is updated after product switchover or batch review. Formula (5) compresses multi-source anomalies into a unified criterion and provides a Boolean decision on whether to allow continued online compensation, with the output destination being the local compensation decision chain; where This indicates that the current window will continue to perform online compensation. This indicates that the current window has stopped the online compensation path of this embodiment and will proceed to the subsequent rework or termination decision chain.

[0035] In this embodiment, the final leakage inductance, distributed capacitance, and channel symmetry of the winding are not suddenly formed during the final electrical test, but are gradually accumulated in the center of gravity of each window, the margin of the edge, the phase across the slot, the degree of compaction, and the torsional trend of the lead-out section. By using formula (1) to unify input quantities of different dimensions and different beats into window quantities, the system can stably compare state changes between different windows; by using formula (2) to unify geometric state, electrical performance trend and remaining correction space into state vector, the system can determine whether the current state still has the ability to return to the target range before the winding is finished; by using formula (3) to compress the target corridor into a single deviation, the system can use the same criterion to coordinate leakage inductance, distributed capacitance, channel symmetry and edge distance constraints; by using formula (4) to introduce online correction into state correction and control command solution, the system can reduce the mapping error caused by skeleton tolerance, wire outer diameter drift and fixture zero position deviation; by using formula (5) to combine abnormal scoring and compensation capacity into online compensation gating, the system can distinguish between deviations that still have recovery space and deviations that have lost convergence conditions, thereby reducing miscompensation and invalid continuous winding.

[0036] When this embodiment enters an abnormal flow, the priority is executed in the order of main axis coding time loss, tension spike, position jump, and corridor overflow; if multiple types of abnormalities are triggered simultaneously within the same window, only the highest priority type is retained as the primary abnormality, and the online compensation actions of lower priority abnormalities are frozen. The system will then maintain continuous winding and continue to send data. The corresponding spindle, cable routing, tension, guide posture, and cable passing sequence control commands; if If the system stops the online compensation output for this window, it will write back the current window number and the anomaly score. Compensation capacity and corridor deviation The result is then passed on to subsequent rework or termination decision chains. The recovery condition is limited to two consecutive subsequent windows simultaneously meeting the condition. , And the input sequence returns to the valid sampling range, where To recover the threshold, records originating from the same abnormal threshold and satisfying the following conditions are required. Therefore, this embodiment discloses a complete main solution from input acquisition, state construction, target corridor determination, online correction, control output to anomaly quantification and diversion.

[0037] Example 2 like Figure 1 and Figure 2As shown in Example 1, this example still focuses on the winding process of Ethernet magnetic devices, but is geared towards multi-winding, multi-channel magnetic devices. It emphasizes disclosing the method for releasing constraint information from the preceding winding to the following winding, the continuous updating of the mapping relationship based on low-energy online correction results, and the unified solution method for differential-mode performance constraints and common-mode performance constraints within the same control chain. The input quantities in Example 1 include the spindle encoding value, cable position value, cable tension value, guide attitude value, pin positioning value, fixture clamping state value, skeleton reference value, and window quantity. Correcting the state vector and corridor deviation Based on this, add the occupied window area for the first winding release. Coupled baseline Symmetrical compensation interval Lead-out path restrictions Differential target quantity Common mode target quantity Multi-channel pairing relationship table and mirrored process step constraint table. Variable sources are as follows: , , and Constraint release record written when the first winding window was closed. and The multi-channel performance constraint records are derived from the current product formulation, while the multi-channel pairing relationship table and mirrored process step constraint table are derived from the product topology definition. The outputs of the above inputs are respectively directed to the correction feature extraction stage, the mapping matrix update stage, the joint constraint fusion stage, and the multi-channel control solution stage.

[0038] In the operation of this embodiment, the window quantity of the current subsequent winding is first obtained with reference to Embodiment 1. and corrected state vector Then, feature extraction is performed on the transient response acquired by the auxiliary coupling coil at the end of the preset window, and a correction feature vector is formed according to formula (6): The meanings of the variables are as follows, and their sources are as follows: This indicates the time difference offset from the rising edge to the peak value of the corrected narrow pulse. This represents the equivalent impedance offset. This represents the coupling differential offset. These represent the differential mode response residuals of the same paired channel within the current window; all four are derived from transient sampling records after the window ends. The dimension is The sampling time is fixed within the preset attenuation interval after the narrow pulse ends, and the sampling frequency is higher than the spindle control cycle and consistent with the clock of the calibration sampling channel. Formula (6) compresses the transient response into a unified feature quantity that can be directly called in subsequent mapping corrections, and the output destination is the mapping matrix update stage.

[0039] After obtaining the corrected eigenvector, the system performs an incremental update on the mapping matrix from geometric state to electromagnetic parameters and generates the mapping version corresponding to the current window according to formula (7): The meanings of the variables are as follows, and their sources are as follows: This represents the mapping matrix of the previous stable version. This indicates that the current corrected feature vector is used to represent the current corrected feature vector. With the modified state vector The candidate mapping matrix obtained by joint calculation, This indicates the mapping update gain. , and All dimensions are The six output components correspond to the predicted leakage inductance deviation correction, the predicted distributed capacitance gradient correction, the channel symmetry error correction, the landing point centroid correction, the lead-out section torsion trend correction, and the cross-slot phase correction, respectively. The four input components correspond to the four correction features in formula (6). The initialization method is to call the baseline mapping template corresponding to the current product topology, and the initial value comes from the mapping template version field in the product recipe; The range of values ​​is limited to The data originates from the gain configuration record corresponding to the correction residual level and the multi-channel consistency level; the update method is to perform an incremental update only when the mapping freeze criterion is not triggered in the current window, and to allow the overall replacement of the benchmark template after product switching, batch review, or manual rollback. Formula (7) enables the online correction results to continuously correct the mapping relationship between geometric and electromagnetic quantities, and the output destination is the joint constraint solution stage and the subsequent window control solution stage.

[0040] The system then merges the constraint information released by the preceding winding with the current window corridor deviation into a unified constraint vector, and calculates it according to formula (8): in This indicates the compression constraint of the occupied window area on the current landing point's centroid. This represents the constraint amount exerted by the coupled baseline on the current symmetry error. This represents the remaining capacity of the symmetric compensation interval for the mirror offset. This indicates the constraint imposed by the lead-in path on the transposition order. Indicates the current corridor deviation. Additional pressure on multi-channel coordinated control. The sources of variables are as follows: to Depend on , , and Obtained through normalized mapping, The deviation was obtained from the corridor deviation record of Example 1. The dimension is The update method is to recalculate once after each window is closed, and keep the source of the constraint vector unchanged during mapping rollback. Formula (8) pushes the preceding winding constraints and the current window constraints into the same solution entry point, and the output destination is the multi-channel joint control loop.

[0041] When the target device contains both differential-mode and common-mode functional windings, the system further performs unified quantization on the dual-constraint pressure according to formula (9): The meanings of the variables are as follows, and their sources are as follows: This represents the deviation of differential performance, derived from the differential target value corresponding to the current window. The deviation between the predicted differential response and the actual response; This represents the deviation in common-mode performance, originating from the common-mode target value. The deviation from the predicted common-mode response; This indicates the mirror phase deviation, which originates from the mirror step constraint record of the paired channel; This indicates the coupling baseline deviation, which is derived from the difference between the coupling baseline record of the first winding release and the current channel center offset. to The weights are non-negative, and their range of values ​​is limited to 1. And sum as The threshold is derived from the dual-constraint priority record in the product formula, and is updated only after product switching or formula review. Formula (9) compresses the differential mode performance constraint and common mode performance constraint into the same pressure scale, and the output destination is multi-channel joint control solution and abnormal conflict judgment.

[0042] After obtaining the mapped version, joint constraint vector, and dual constraint score, the system solves for the multi-channel joint control command according to formula (10): in The meanings of the variables are as follows, and their sources are as follows: This represents the multi-channel baseline control vector given by the current product formulation, with dimensions of [dimension number missing]. The five output components correspond to the spindle acceleration correction, cable spacing correction, tension reference value correction, guide attitude correction, and mirror transposition sequence correction, respectively. The constraint vector gain matrix has dimensions of . ; This represents a column vector of double-constraint score gains, with dimensions of [missing information]. ; This represents the mapping correction gain matrix, with dimension 1. ; Indicates the mapping version Acting on the correction eigenvector The obtained six-dimensional state correction. , and The initialization method is to read the joint control template corresponding to the current product topology. The initial value comes from the joint control template version field. The update strategy is to only allow the template version that has been reviewed and approved to be used in the same batch. It does not drift spontaneously in a single window. If the template freeze or mapping rollback criterion is triggered, the previous stable version is maintained. Formula (10) maps the preceding constraints, double constraint pressure and online correction correction amount to the execution instructions of the next window. The output destination is the multi-channel spindle execution chain, the wiring execution chain, the tension execution chain and the next winding window solution chain.

[0043] The final electrical performance of the multi-winding multi-channel magnetic device in this embodiment is jointly determined by the structural conditions of the first winding and the structural conditions of the subsequent winding. If the preceding constraints are not explicitly introduced when solving the subsequent winding, the seemingly qualified landing point and attitude in the local window may still be manifested as channel mismatch in the finished product stage. By using formula (7) to make the mapping matrix continuously absorb the low-energy online correction results, the system can reduce the model distortion caused by skeleton tolerance, wire outer diameter drift and fixture zero position deviation; by using formula (8) and formula (9) to compress the preceding winding release constraints and differential mode and common mode dual constraints into a unified solution entry, the system can simultaneously evaluate "whether the current window can maintain its own convergence" and "whether the current window will destroy the overall channel consistency" in each window; by using formula (10) to map the three types of information into the same set of execution quantities, the system can complete multi-channel collaborative correction without breaking the control chain.

[0044] When anomalies, recovery, or conflict scenarios occur during the work process, this embodiment uses quantitative anomaly criteria instead of verbal classification. Anomaly criteria include two categories: one is multi-channel constraint conflict criteria, i.e., two consecutive windows satisfy... And simultaneously satisfy This indicates that the single-channel local convergence is acceptable, but the overall dual constraints have mismatched; the second is the mapping instability criterion, namely, three consecutive windows satisfying Furthermore, the corrected residuals increase in the same direction. The threshold sources are as follows: Records of double-constraint conflict limits from product formulations. From the target corridor configuration record, This information comes from the stability review record of the mapping version. Priorities are as follows: mapping instability has higher priority than multi-channel constraint conflicts, and multi-channel constraint conflicts have higher priority than simple corridor overflows. If two types of anomalies are triggered simultaneously in the same window, only the primary anomaly with the highest priority is retained, and the online compensation actions corresponding to the lower-priority anomaly are frozen. Recovery conditions are as follows: two consecutive subsequent windows must simultaneously meet the conditions. , Only after the current mapping template version has not been rolled back can the abnormal freeze be lifted and incremental updates resume. and All of them come from the same anomaly threshold record and respectively meet the following conditions: , If the recovery conditions are not met, this embodiment only rolls back the mapping version, without rolling back the structural constraints released by the preceding winding, and transfers the conflict level, freeze flag, and current template version to Embodiment 3 for rework or termination determination.

[0045] Example 3 like Figures 1 to 3 As shown in Examples 1 and 2, this example continues to use window size, target corridor, corrected state vector, dual-constraint scoring, and multi-channel conflict level as basic inputs, but focuses on disclosing anomaly hierarchical decision-making, local rework, processing termination, and quality label write-back chain. The inputs include the corridor deviation output from Example 1. Compensation capacity The draft control instructions and window risk levels also include the dual-constraint score output by Example 2. The mapping includes version status, conflict level, and freeze flag, as well as remaining turns, remaining layers, allowed cross-line count, and the index of the most recent stable window. The interface identifiers for the assembly station and the electrical testing station are specified. The variable sources are as follows: the aforementioned status variables are from the current batch operation record, window history record, and multi-channel control results; the remaining number of turns, remaining number of layers, and allowed number of line crossings are from the current work step progress record; and the interface identifiers for the assembly station and the electrical testing station are from the equipment configuration table. The output destinations of the above inputs are respectively the online compensation decision chain, the partial rework instruction generation chain, the processing termination decision chain, and the quality label write-back chain.

[0046] In the operation of this embodiment, the current deviation and disturbance are first uniformly quantized, and the anomaly score is calculated according to formula (11): The meanings of the variables are as follows, and their sources are as follows: This indicates the corridor deviation given in Example 1. This represents the peak tension within the window, which originates from the deviation between the current maximum tension sampled value within the window and the reference tension for the work step. This represents the abrupt change in the guidance attitude, which originates from the maximum absolute value of the adjacent differences in the guidance attitude sequence. This indicates the amount of mirror constraint conflict, derived from the multi-channel conflict output of Example 2. This represents the online correction residual, which originates from the residual between the current window correction response and the mapping prediction result. to The weights are non-negative, and their range of values ​​is limited to 1. And sum as The threshold is derived from the abnormal level configuration record in the product formula. The update method is to reload when switching products and only allow updates under administrator authorization within the same product batch. Formula (11) compresses multi-source disturbances into the same risk level, and the output destination is the subsequent three-stage adjudication chain.

[0047] The system then first determines whether online compensation is allowed to continue, and calculates the online compensation determination flag according to formula (12): The meanings of the variables are as follows, and their sources are as follows: This represents the compensation capacity calculated in Example 1. This represents the minimum compensation capacity threshold required for online compensation. This indicates the maximum anomaly score threshold allowed for online compensation. This represents the maximum allowed dual-constraint scoring threshold for online compensation; all three threshold types are derived from the stratified decision threshold records in the product formulation. If If the deviation is corrected online, it means that the current deviation is still correctable online. The system maintains the current winding continuity and continues to issue compensation path instructions using the control vectors generated in Examples 1 and 2. Formula (12) converges "whether to continue online compensation" into an executable Boolean criterion, and the output destination is the spindle controller, the cable laying controller, the tension controller, and the transposition mechanism.

[0048] When online compensation is no longer sufficient, the system further determines whether to enter a local rework path and calculates the rework judgment flag according to formula (13): The meanings of the variables are as follows, and their sources are as follows: and They jointly define the applicable compensation capacity range for partial repairs. This indicates the upper limit of abnormal scores before forced termination. Indicates the most recent stable window number that can be traced. The source is the most recent one in the current batch window history that simultaneously meets the condition. , and Window index, This information originates from records of abnormal recovery thresholds. If... If so, the system generates a rework path instruction, which includes at least the return window number, wire release length, rearrangement order, and rewinding start point. Formula (13) distinguishes the intermediate state of "reworkable but not suitable for continued online compensation" from other states, and the output destination is the local rework execution mechanism and the window history write-back chain.

[0049] When the rework conditions are not met, the system executes the termination of processing judgment according to formula (14): The meanings of the variables are as follows, and their sources are as follows: This represents the minimum convergence capacity threshold. This indicates the upper limit of the double-constraint mismatch. This indicates the threshold for mandatory termination of abnormal scoring; all three are derived from the termination criterion record after the process review is passed. If If the error occurs, the system will output a termination processing command, stopping the current window from continuing winding, flushing or converging the current execution buffer, freezing the subsequent automatic release status of the current batch, and recording the current window number and anomaly score. Compensation capacity Dual-constraint scoring The mapping version identifier and freeze flag are written to the termination event object. Formula (14) blocks the semi-finished products that have lost the convergence condition from continuing to consume equipment cycle time and downstream station resources, and the output destination is the termination processing module and the quality label generation module.

[0050] While generating the rework path or termination path, the system generates a quality label object according to formula (15): The meanings of the variables are as follows, and their sources are as follows: This indicates the current window's risk level, derived from anomaly scoring. Deviation from the corridor The joint classification results; This indicates the convergence state of the remaining steps, derived from the compensation capacity. Termination Judgment The joint judgment result; This indicates the downstream electrical testing concerns, derived from the current deviation's corresponding risk categories of leakage inductance, capacitance, or symmetry. This indicates the mapping version identifier, which originates from the mapping version status output in Example 2; This indicates the current batch identifier, which originates from the equipment production record. The dimension is The key field is the batch identifier and window number, the value field is the risk level, convergence status and electrical test concern item, the version field is the mapped version identifier, and the update time is the current window closing time; the update method is that the same window can only cover the label below the same version, and rollback is allowed after manual review and confirmation of misjudgment. After the termination processing label takes effect, it remains frozen. Formula (15) converts the current abnormal conclusion into quality control information that can be directly executed by the assembly station interface and the electrical test station interface, and the output destination is the assembly station interface, the electrical test station interface and the local batch freeze object.

[0051] In this embodiment, deviations occurring during the winding process are not equivalent to requiring a shutdown and scrapping, because different deviations have different recoverability under the constraints of the remaining number of turns, the remaining number of layers, and the allowed number of cross-lines. If the decision to keep or discard is based solely on a single tension deviation or a single position deviation, it is still possible to scrap deviations that can be corrected online in advance, or to continue sending deviations that have lost convergence conditions into subsequent steps. By incorporating compensation capacity, anomaly score, and dual-constraint score into the online compensation judgment using formula (12), the system can retain windows with remaining recovery space; by introducing the most recent stable window index and compensation capacity range into the rework judgment using formula (13), the system can converge traceable local mismatches into limited rework actions; by linking the termination boundary and the quality label write-back chain together using formulas (14) and (15), the system can move the current anomaly conclusion forward to the assembly station and electrical testing station, instead of leaving all risks to the final unified electrical testing stage.

[0052] When the exception chain and recovery chain are actually triggered, this embodiment executes according to the quantization priority. The exception criteria are as follows: If If so, it will enter the online compensation path; if and If so, it will enter the partial rework path; if If the condition is met, the processing will terminate. The threshold sources are as follows: , , , , , , and All data comes from stratified decision threshold records that have passed process review under the same product formula. Priorities are as follows: termination of processing has higher priority than partial rework, and partial rework has higher priority than online compensation. If both the rework criterion and termination criterion are met simultaneously in the same window, only termination of processing is retained as the primary anomaly, and the rework action is frozen. Recovery conditions are as follows: after partial rework is completed, two consecutive subsequent windows simultaneously meet the conditions. , and Only then was the repair freeze lifted and continuous winding resumed, among which... and Records originating from the same recovery threshold and respectively satisfying , Once the processing is terminated, the freeze can only be lifted after manual review, batch rebuilding, and reloading of the mapped version. Thus, this embodiment discloses the complete action chain of online compensation, partial rework, processing termination, and quality label rewriting.

[0053] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be modified within the scope of the concept described herein by means of the above teachings or the technology or knowledge in related fields.

Claims

1. A winding closed loop control method for an Ethernet magnetic device, characterized by, include: The system acquires input and correction values, including spindle encoding values, wire position values, wire tension values, guide posture values, pin positioning values, clamping state values, and skeleton reference values. It performs time alignment, outlier removal, window slicing, and reference normalization on the input and correction values ​​to obtain window values. The window boundaries are determined based on the single-layer completion point, the preset turn array completion point, the cross-pin swapping point, and the insulation insertion point. A target corridor is generated based on the product formula, the current process step, the completed winding structure, and the window values. This target corridor is jointly defined by leakage inductance constraints, distributed capacitance constraints, channel symmetry constraints, coupling consistency constraints, lead-out path constraints, and edge distance constraints, and is used to characterize whether the current window remains within the target electrical performance convergence range in subsequent processes. A state vector is constructed based on the window values. This state vector includes at least the current layer's effective number of turns, coil centroid position, layer edge distance margin, cross-slot phase, compaction degree, lead-out section torsion trend, predicted leakage inductance deviation, predicted distributed capacitance gradient, channel symmetry error, and subsequent compensation capacity. At the end of the preset window, an auxiliary coupling line is used... A low-energy narrow pulse is injected into the formed winding, and the transient response is acquired. Based on the transient response, the mapping relationship between the geometric state and electromagnetic parameters is corrected to obtain a corrected state vector. Based on the corrected state vector and the target corridor, the control strategy is solved, and control commands are output to the spindle controller, wire laying controller, tension controller, wire release brake, and transposition mechanism. The control commands are used to adjust at least the spindle acceleration, wire laying step distance, tension reference value, guide attitude, and wire passing sequence corresponding to the subsequent turns. A hierarchical decision is performed on the current deviation based on the subsequent compensation capacity. The subsequent compensation capacity is jointly determined based on the remaining number of turns, the remaining number of layers, the allowed number of wire crossings, the current coil center of gravity deviation, the current predicted leakage inductance deviation, the current predicted distributed capacitance gradient, and the current channel symmetry error. When the subsequent compensation capacity meets the online compensation condition, a compensation path command is output. When the subsequent compensation capacity meets the local rework condition, a rework path command is output. When the subsequent compensation capacity does not meet the convergence condition, a termination processing command and a quality tag are output, and the quality tag is output to the assembly station interface and the electrical testing station interface.

2. The Ethernet magnetic device winding closed-loop control method according to claim 1, characterized in that, The target corridor is mapped to the landing point centroid constraint, edge distance constraint, slot phase constraint, line crossing sequence constraint, and lead-out segment torsion constraint corresponding to the current winding window according to the current work step. The control reference for the next window is determined by judging whether the current winding window is still in the target electrical performance convergence range.

3. The Ethernet magnetic device winding closed-loop control method according to claim 1, characterized in that, The time alignment, outlier removal, window slicing, and reference normalization include: performing unified time-scale alignment on tension sampling, cable laying position sampling, guide posture sampling, and pin positioning sampling according to the spindle encoding cycle; performing window slicing on the aligned sampling results based on the winding window boundary; and performing normalization on the slicing results based on the skeleton reference size and the current step reference.

4. The Ethernet magnetic device winding closed-loop control method according to claim 1, characterized in that, The subsequent compensation capacity is determined jointly based on the remaining number of turns, the remaining number of layers, the allowed number of cross-line steps, the current coil center of gravity deviation, the current predicted leakage inductance deviation, the current predicted distributed capacitance gradient, and the current channel symmetry error. It is used to characterize the degree of convergence of the current deviation back to the target corridor in subsequent steps.

5. The Ethernet magnetic device winding closed-loop control method according to claim 1, characterized in that, The correction response is formed by acquiring the transient response after injecting a low-energy narrow pulse excitation into the formed winding through an auxiliary coupling coil, and the mapping relationship is updated by comparing the deviation between the correction parameters obtained from the transient response and the predicted parameters.

6. The Ethernet magnetic device winding closed-loop control method according to claim 1, characterized in that, The compensation path command is used to adjust the spindle acceleration, wire laying step, tension reference value, guide attitude and wire passing sequence corresponding to the subsequent turns. The rework path command is used to control the equipment to return to the nearest stable window and perform release rearrangement and rewinding on the wire of the limited length.

7. The Ethernet magnetic device winding closed-loop control method according to claim 1, characterized in that, The quality label includes the current window risk level, the convergence status of the remaining steps, and the downstream electrical testing concerns, and is output to the assembly station interface and the electrical testing station interface.

8. The Ethernet magnetic device winding closed-loop control method according to claim 1, characterized in that, When the target device is a multi-winding, multi-channel magnetic device, the first winding releases constraint information to the subsequent winding after completing the current window. The constraint information includes the occupied window area, coupling baseline, symmetric compensation interval, and lead-out path restriction. The subsequent winding solves the control strategy based on this constraint information and in combination with its own target corridor.

9. The Ethernet magnetic device winding closed-loop control method according to claim 8, characterized in that, When the target device includes both differential-mode and common-mode functional windings, the target corridor further includes differential-mode performance constraints and common-mode performance constraints, and performs the mapping of the two types of constraints to the same control strategy through a unified state vector.

Images