Method and system for on-line detection of transformer production quality

Abstract

The application discloses a transformer production quality online detection method and system, and the method comprises the following steps: clamping a transformer to be detected and reading a unique identifier, measuring the primary terminal contact resistance and the secondary terminal contact resistance, and re-measuring if the contact resistance is out of limit; applying a bipolar high-slope pulse between the primary and the secondary, synchronously sampling the voltage, the response current and the reference noise, and aligning the sampling time; setting two levels of gate windows with the pulse edges, coherently superimposing the response current in the inner window, and calculating the offset amount according to the reference noise to obtain the net response in the outer window; calculating the cross-correlation peak value, the gated energy and the frequency band energy ratio from the net response, and generating the first index in combination with the batch reference. The application realizes the synchronous online identification of insulation hidden troubles and connection hidden troubles under the production line condition, and forms a stable judgment and a full-flow tracing closed loop.

Description

Technical Field

[0001] This application relates to the field of production testing technology, and in particular to online testing methods and systems for transformer production quality. Background Technology

[0002] As power electronics such as OBCs and DC / DC converters in new energy vehicles develop towards higher frequencies and higher power densities, small high-frequency transformers, which are subjected to high dv / dt pulses, electrothermal coupling, and vibration thermal cycling in automotive-grade scenarios, should be subject to rapid, full-inspection, and traceable online quality testing on the production line.

[0003] Existing methods mostly rely on visual inspection, random sampling and cutting, and final inspection of turns ratio, DC resistance, low-frequency LCR, withstand voltage, weak field, low frequency and short-time stress, which are out of touch with the actual working conditions, resulting in the failure to detect hidden failures.

[0004] Firstly, existing withstand voltage and insulation resistance are mostly determined in a short time, which is difficult to reflect the partial discharge margin of micro-voids and micro-moisture inside potting and impregnation. The reasons for this are insufficient degassing, dead corner filling, curing shrinkage and environmental humidity adsorption, and production line noise makes early partial discharge characteristics easy to be submerged. The adverse consequences are that the product is qualified at the factory, but partial discharge is accelerated under pulse voltage and thermal cycling, insulation deteriorates rapidly and suddenly breaks down.

[0005] Secondly, existing connection quality relies heavily on appearance and DC resistance, making it difficult to identify abnormal AC losses caused by broken strands, incomplete paint removal, or poor soldering at the Litz wire ends. This is because defects are not sensitive to DC indicators but significantly increase the equivalent AC resistance and cause localized heating at high frequencies. The adverse consequences are that hot spots lead to dendration and lifespan dispersion, resulting in poor consistency and early failure. Summary of the Invention

[0006] To address the above problems, embodiments of the present invention provide an online method for detecting the production quality of transformers, the method comprising: Mount the transformer under test and read its unique identifier. Measure the contact resistance of the primary terminal and the contact resistance of the secondary terminal. If the contact resistance exceeds the limit, repeat the measurement. A bipolar high-slope pulse is applied between the primary and secondary sides to synchronously sample the voltage, response current, reference noise, and align the sampling time. Two levels of gating windows are set with pulse edge. The response current is coherently superimposed in the inner window, and the net response is obtained by calculating the cancellation amount based on the reference noise in the outer window. The first indicator is generated by calculating the cross-correlation peak value and the ratio of gated energy to bandwidth energy from the net response and combining them with the batch benchmark. Apply constant current sweep frequency to the winding terminals, sample the terminal voltage and terminal current, and calculate the equivalent AC resistance curve and phase curve; The segmented slope and resistance ratio are calculated from the equivalent AC resistance curve, and the jump count is calculated from the phase curve. Combined with the batch reference, a second index is generated. The first and second thresholds are updated using qualified samples from the same batch. When the first indicator is greater than or equal to the first threshold, the output is unqualified and marked as an insulation hazard. When the second indicator is greater than or equal to the second threshold, the output is unqualified and marked as a connection hazard. When both indicators are less than the first and second thresholds, the output is qualified. The unique identifier, sampling data, indicators and judgment results are stored and written back to the process database.

[0007] Furthermore, the reference noise is acquired by a reference pickup that is electrically isolated from the transformer under test; no-load calibration is performed before applying a bipolar high-slope pulse, which includes acquiring the background noise waveform, establishing amplitude correction coefficients, and establishing phase correction coefficients; before performing the cancellation process in the outer window, amplitude correction and phase correction are performed on the reference noise, and then the cancellation amount is calculated based on the correlation between the reference noise and the response current in the outer window, and the cancellation amount is subtracted from the response current to obtain the net response.

[0008] Furthermore, a two-level gating system with an outer window and an inner window is adopted. The outer window covers the preset duration after the pulse edge, while the inner window is located inside the outer window and avoids the edge saturation region. In the inner window, the response current of each pulse is subjected to DC bias removal, amplitude normalization, and edge alignment, and then coherent superposition is performed according to the pulse sequence number. After coherent superposition, the zero-crossing drift is calculated and compared with the upper limit of drift. If the zero-crossing drift exceeds the limit, the system returns to the clamping positioning and unique identifier reading stage for retesting and writes a clamping instability tag.

[0009] Furthermore, the bipolar high-slope pulse is output in pulse groups, with each pulse group outputting a positive pulse and a negative pulse in sequence, and a rest interval is set between pulse groups; reference noise is collected and the noise mean square value is calculated within the rest interval; if the noise mean square value exceeds the noise upper limit, the test is paused and the shield grounding check procedure is executed; after the shield grounding check procedure is completed, the two-level gating and net response generation are continued, and the pause mark and the noise mean square value are written into the traceability record.

[0010] Furthermore, benchmarks for the same batch are established according to tooling number, and the most recent qualified samples with a fixed window number are used for rolling updates. During rolling updates, the center value and dispersion of the first and second indicators are calculated, and the first and second thresholds are generated from the center value and dispersion. When the change of the threshold relative to the previous window exceeds the drift limit, a drift alarm is written and the calibration process is triggered. The calibration process includes sampling channel time alignment verification, constant current excitation amplitude verification, and reference pickup no-load calibration.

[0011] Furthermore, the constant current frequency sweep uses a discrete frequency sequence, with the frequency points output in ascending order and a dwell time maintained at each frequency point; at each frequency point, multi-cycle terminal voltage waveforms and terminal current waveforms are acquired, and synchronous detection is performed on the acquired waveforms to obtain the fundamental voltage component and the fundamental current component. Based on the fundamental amplitude and fundamental phase, the complex impedance is calculated and the equivalent AC resistance curve and phase curve are generated. The frequency sequence and dwell time are written into the traceability record for cross-workstation consistency verification.

[0012] Furthermore, the jump count is obtained as follows: the phase curve is expanded and first-order differential is performed, the number of differential sign inversions is counted to obtain the jump count, and the phase swing amplitude is calculated at the same time; when the jump count is greater than the jump threshold and the phase swing amplitude is greater than the swing threshold, the connection hazard tag is subdivided into contact discontinuity tags, and the frequency range that caused the jump is written into the traceability record for rework location.

[0013] Furthermore, after the constant current frequency sweep is completed, a thermal excitation test is performed, including: applying constant current excitation at a fixed frequency point and maintaining the heating time, and collecting the transient temperature curve at the temperature measurement point; calculating the temperature rise slope and cooling slope based on the transient temperature curve, and adding the temperature rise slope and cooling slope as thermal response features to the second index generation process; when the second index triggers the connection hazard judgment, the thermal response features and the equivalent AC resistance of the corresponding frequency point are simultaneously written into the traceability record to distinguish between virtual contact and abnormal wire loss.

[0014] The online inspection method for transformer production quality also includes: keeping the clamping and sampling configuration unchanged, maintaining the primary and secondary terminals at zero level and completing the switching according to the same triggering cycle, synchronously collecting the no-trigger response current, generating an inherent transient template for subsequent subtraction and writing into the traceability record.

[0015] On the other hand, this application also provides an online inspection system for transformer manufacturing quality, the system comprising: The clamping identification module clamps the transformer under test and reads its unique identifier, then measures the contact resistance of the primary terminal and the contact resistance of the secondary terminal. If the contact resistance exceeds the limit, the measurement is repeated. A pulse sampling module applies a bipolar high-slope pulse between the primary and secondary sides to synchronously sample voltage, response current, reference noise, and align sampling times. The gated purification module has two levels of gated windows set with pulse edges. The response current is coherently superimposed in the inner window, and the net response is obtained by calculating the cancellation amount based on the reference noise in the outer window. The insulation index module is a first index generated by calculating the cross-correlation peak value and the ratio of gated energy to bandwidth energy from the net response and combining them with the batch benchmark. A frequency sweep resistance measurement module applies a constant current frequency sweep to the winding terminals, samples the terminal voltage and terminal current, and calculates the equivalent AC resistance curve and phase curve. The connection index module calculates the segmented slope and resistance ratio from the equivalent AC resistance curve, calculates the jump count from the phase curve, and generates a second index by combining the batch benchmark. The threshold determination module updates the first threshold and the second threshold with qualified samples from the same batch. When the first indicator is greater than or equal to the first threshold, it outputs "unqualified" and marks an insulation hazard. When the second indicator is greater than or equal to the second threshold, it outputs "unqualified" and marks a connection hazard. When both indicators are less than the first threshold and the second threshold, it outputs "qualified". The unique identifier, sampling data, indicators and determination results are stored and written back to the process database.

[0016] The technical effects and advantages of the online transformer production quality inspection system provided by this invention are as follows: This invention enables simultaneous online identification of insulation and connection defects under production line conditions, forming a stable judgment and full-process traceability closed loop. Through a clamping contact resistance verification and retesting mechanism, this invention preemptively eliminates measurement deviations introduced by unstable clamping contacts, ensuring that subsequent sampling and judgment are based on consistent electrical connection conditions. It employs bipolar high-slope pulse synchronous sampling and introduces static interval noise mean square value monitoring and shielded grounding inspection processes, allowing testing to run under noise control and reducing misjudgments and missed detections caused by environmental interference. Through a combination of external window reference noise correlation cancellation and internal window coherent superposition processing, it enhances weak anomalous components synchronized with the excitation, suppresses random noise and asynchronous interference, and improves the distinguishability of the net response. Two-stage gating avoids edge saturation regions, and zero-crossing drift constraints identify timing anomalies caused by clamping, reducing false anomalies caused by equipment transients, trigger jitter, and other factors from entering the index calculation. Attached Figure Description

[0017] Figure 1 This is a flowchart of the online quality inspection method for transformer production in Example 1; Figure 2 This is a flowchart of the online transformer production quality inspection method in Example 2; Figure 3 This is a schematic diagram of the connection of the online transformer production quality inspection system in Example 3. Detailed Implementation

[0018] 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.

[0019] Example 1: Please see Figure 1 As shown, embodiments of the present invention provide an online inspection method for transformer manufacturing quality, the method comprising: S1: Clamp the transformer under test and read the unique identifier. Measure the contact resistance of the primary terminal and the contact resistance of the secondary terminal. If the contact resistance exceeds the limit, repeat the measurement. S2: Apply a bipolar high-slope pulse between the primary and secondary sides to synchronously sample the voltage, response current, reference noise, and align the sampling time; S3: Set two levels of gated windows with pulse edge, coherently superimpose the response current in the inner window, and calculate the cancellation amount based on the reference noise in the outer window to obtain the net response; S4: The first indicator is generated by calculating the cross-correlation peak value, the ratio of gated energy to bandwidth energy from the net response and combining it with the batch benchmark. S5: Apply constant current sweep frequency to the winding terminals, sample the terminal voltage and terminal current, and calculate the equivalent AC resistance curve and phase curve; S6: Calculate the segmented slope and resistance ratio from the equivalent AC resistance curve, calculate the jump count from the phase curve, and generate a second index by combining the batch reference. S7: Update the first threshold and the second threshold with qualified samples from the same batch. If the first indicator is greater than or equal to the first threshold, output "unqualified" and mark it as an insulation hazard. If the second indicator is greater than or equal to the second threshold, output "unqualified" and mark it as a connection hazard. If both indicators are less than the first threshold and the second threshold, output "qualified". Store the unique identifier, sampling data, indicators and judgment results and write them back to the process database.

[0020] In this embodiment, in order to stably obtain the effective component in the response current under the electromagnetic interference conditions of the production line, a reference noise channel is introduced in step S2, and no-load calibration is performed before step S2. The reference noise refers to the broadband interference generated by the equipment around the test station, the crosstalk coupling inside the tooling, and the noise component formed by the cable routing induction. Its characteristic is that it has no direct correspondence with the insulation state of the transformer under test, but it will be superimposed on the response current and cause judgment fluctuations. The reference pickup is a pickup structure installed in the test tooling. The reference pickup is electrically isolated from the transformer under test. The reference pickup is connected to the sampling channel through a shielded cable. The cable shielding layer is fixed at the tooling end according to the predetermined grounding method, thereby ensuring that the reference pickup only "sees" the noise of the work station and the coupling noise of the tooling, without introducing an additional electrical connection path.

[0021] No-load calibration is performed once after each tooling change, shift change, and workstation maintenance, and resumed when the noise environment changes significantly. During no-load calibration, the tooling maintains the same clamping state as in formal testing, the excitation remains off, and the sampling link maintains the same range and bandwidth settings as in formal testing. The sampled waveform is recorded as the noise floor waveform, which is used to characterize the noise amplitude level and phase response of the reference pickup channel in the current workstation environment. Because the reference pickup channel and the response current channel differ in sensor structure, wiring length, front-end filtering, and sampling clock alignment, the same noise source exhibits "inconsistent amplitude and phase" in the two channels. Therefore, no-load calibration requires the establishment of two types of correction quantities: Amplitude correction coefficient: used to convert the amplitude scale of the reference noise to a scale consistent with the response current noise component; in practice, the energy index of the background noise waveform within a specified time window is used as a benchmark to obtain the amplitude ratio between the reference channel and the response channel, and this ratio is solidified as the amplitude correction coefficient. The amplitude correction coefficient is associated with the tooling number and stored.

[0022] Phase correction coefficient: used to compensate for the equivalent time delay and phase shift between the two channels; in practice, the background noise waveform and the noise segment of the response current channel are time-delayed to obtain the equivalent alignment amount, and the phase correction coefficient is established accordingly, so that the reference noise is aligned with the noise component in the response current on the same time reference after correction.

[0023] After entering step S2, a bipolar high-slope pulse is applied to the primary and secondary sides and the applied voltage, response current, and reference noise are sampled simultaneously, while the sampling time alignment is completed. After entering step S3, a gated window is established with the pulse edge. The outer window covers the response attenuation and interference duration interval after the edge, which is used to calculate the noise cancellation amount. The inner window is used for subsequent feature calculation. In order to avoid mistaking the real response as noise cancellation, the interval in the outer window that avoids the inner window is used as the calculation segment when calculating the cancellation amount.

[0024] The specific implementation is as follows: First, amplitude and phase corrections are performed on the reference noise to obtain a corrected reference noise with the same scale and phase as the noise component of the response current. Then, the correlation between the corrected reference noise and the response current is calculated in the outer window. The correlation is used to measure the similarity between the two waveforms in this time segment, reflecting "how much noise component in the response current can be explained by the reference noise". Based on this, the cancellation amount is calculated: the cancellation amount is the waveform component formed by scaling the corrected reference noise. The scaling factor is obtained by minimizing "the residual signal energy after subtracting the cancellation amount from the response current". In other words, the cancellation amount is adaptively calculated based on the correlation in the outer window segment, so that the cancellation mainly targets the interference components that are from the same source as the reference noise. Finally, the cancellation amount is subtracted from the response current to obtain the net response. The net response is used as the input signal for calculating the first index in the subsequent step S4, and is associated with the unique identifier and tooling number recorded in step S1 in the same traceability record to ensure that the calibration coefficients of the same tooling can be correctly called in subsequent sample testing.

[0025] For example: There is periodic switching power supply interference near a certain workstation. When directly using the response current to calculate the characteristics, the noise energy in the outer window fluctuates with the start and stop of the equipment, causing the first index of the same batch of samples to become more discrete. After adopting the above-mentioned no-load calibration and cancellation process, the reference noise is aligned with the response current noise component after amplitude and phase correction. The scaling factor is obtained in the outer window segment and the cancellation amount is generated. The net response obtained after deduction shows that the residual noise in the outer window is significantly converged, while the response details synchronized with the excitation in the inner window are preserved. This keeps the first index stable within the batch and avoids misjudging the noise fluctuation of the workstation as an insulation abnormality.

[0026] In this embodiment, step S3 is used to stably extract the response current synchronized with the pulse excitation when there is production line noise and clamping fluctuation, and to screen out the timing drift caused by clamping instability; step S3 sets two levels of gating control, outer window and inner window, based on the response current waveform and pulse triggering time obtained in step S2.

[0027] The outer window extracts a segment of response current data starting from the pulse trigger moment. The duration of the outer window is set by the test formula to fix the data segment corresponding to each pulse and provide a unified time reference for the positioning of the inner window. The inner window is located inside the outer window, avoiding the edge saturation region. The edge saturation region refers to the nonlinear region near the pulse rising edge caused by amplifier saturation, sampling recovery, and probe overshoot. The waveform in this region does not reflect the true response. The start and end positions of the inner window are fixed in the test formula so that a stable response segment is extracted from the outer window according to the same rule for each measurement.

[0028] For each pulse's inner window data, DC bias removal, amplitude normalization, and edge alignment are performed sequentially. During DC bias removal, the average value of a reference interval without the main response is calculated within the outer window, and this bias is subtracted from the inner window sampling point to obtain the debiased inner window waveform. During amplitude normalization, the normalization coefficient is calculated based on the amplitude scale of the inner window waveform, and the inner window waveform is scaled to ensure that different pulses are under the same dimension. During edge alignment, an inner window waveform is selected as an alignment reference, the time shift of each pulse's inner window waveform relative to the reference is calculated, and the waveform is shifted according to this time shift to make the main response structure coincide on the time axis.

[0029] Then, coherent superposition is performed according to the pulse sequence number to obtain the coherent superposition waveform. Coherent superposition makes the response components synchronized with the excitation accumulate in phase, and random noise is weakened in the superposition, thereby improving the stability of subsequent feature calculation.

[0030] After coherent superposition, the zero-crossing drift is calculated and compared with the upper limit of the drift. The zero-crossing point is the zero-crossing moment of the coherent superposition waveform within the predetermined search interval. The zero-crossing time is obtained by interpolation of adjacent sampling points. For the aligned inner window waveform of each single pulse, the corresponding zero-crossing time is located using the same search interval. The difference between the zero-crossing time and the coherent superposition zero-crossing time is calculated, and the maximum absolute value of the difference is taken as the zero-crossing drift. The zero-crossing drift reflects the inconsistency of residual timing between pulses, which comes from clamping instability factors such as terminal clamping micro-motion and clamp springback.

[0031] The methods for calculating drift amount used to determine clamping instability include: ; in, ; ; ; ; This also includes: ; ; ; ; In the formula, For zero-point drift, the unit is time. This amount is used to compare with the drift limit. If the limit is exceeded, the return step S1 is triggered to retest and write the clamping instability label. For pulse sequence number index; The number of pulses involved in coherent superposition in this test is determined by the test formula and kept consistent to ensure repeatability and traceability. The sampling period is equal to the sampling interval time, which converts the location of discrete sampling points into actual time. The discrete sampling point number; For the first Aligned normalized waveform of each pulse The index at which the sign is flipped within the inner window search interval satisfies the condition. and The sign is opposite, and it is used for the zero-crossing point of the envelope; Coherent superposition waveform The index at which the sign flips within the same inner window search interval is defined in the same way as... Consistent; For the first The original sampling sequence of the response current corresponding to each pulse comes from the response current sampling channel in step S2; For edge alignment, the first The discrete time shift applied by each pulse, which is obtained through a relevant criterion, is used to eliminate the equivalent time delay difference introduced by trigger jitter and clamping. The range of time-shifted sets allowed for search is limited by the test recipe to avoid unreasonable large shifts; To align with the reference waveform, the reference inner window waveform from the same test is usually taken in engineering implementation as the normalized inner window waveform of the first qualified pulse, or the iterative update result of coherent superposition, to ensure the stability of the alignment reference. This is a DC bias value used to compensate for the baseline offset introduced by the zero-point drift of the response current channel, low-frequency bias, and the tooling ground potential difference. This is the set of reference interval indices within the outer window used to estimate the DC bias. These intervals avoid the main response segment, making... It represents the "baseline" rather than the "response"; For set The number of sampling points used to form the mean; As a scale for amplitude normalization, the root mean square within the inner window is used as the scale to ensure that the amplitudes of different pulses can be compared and superimposed under the same dimension. The inner window index set corresponds to the "stable response segment after avoiding the edge saturation region". The start and end rules of the inner window are fixed in the test formula to ensure consistency. For set The number of sampling points used to form the root mean square. The weights are taken as the inverse of the noise variance to reduce the impact of noisy pulses on the superposition result. For the first The root mean square of noise for each pulse within the noise evaluation interval is used to quantify the background disturbance level of the pulse. The index set used to estimate noise intensity is located within the outer window and avoids the main response segment, reflecting the production line background noise rather than the device response; Noise assessment range The mean value within the range is used to remove the DC component when calculating the noise variance; The traversal index is used in summation and related calculations to represent the index of a sampling point in the set.

[0032] For example: In a certain test, the sampling rate was set to 10 MHz, the number of pulse groups was set to 16, the outer window length was set to 8 microseconds, the inner window length was set to 3 microseconds, and the starting point of the inner window avoided the saturation recovery region of 0.5 microseconds after the pulse edge. After debiasing, normalization and correlation alignment of the inner window waveform of each pulse, a coherent superimposed waveform was obtained. The zero-crossing point deviation of each single pulse was calculated based on the zero-crossing point of the coherent superimposed waveform. The maximum deviation was 0.12 microseconds, which was less than the drift limit of 0.2 microseconds, so the test proceeded to the next step S4. If the maximum deviation reached 0.35 microseconds, the test was repeated in step S1 and the unstable mounting label was recorded.

[0033] In this embodiment, step S2 uses bipolar high-slope pulses and outputs them in pulse group mode to obtain response data under positive and negative stress conditions under the same clamping state. At the same time, it provides a stable triggering reference for the gating processing in the subsequent step S3. A pulse group refers to the sequential output of positive and negative pulses in the same group, which share the same trigger sequence and sampling cycle. The interval between the positive and negative pulses is controlled by the waveform schedule of the excitation source to ensure that the sampling channel completes recovery between the two polarity pulses and maintains the same sampling alignment rules.

[0034] A set interval is set between adjacent pulse groups. The purpose of the set interval is not to apply electrical stress, but to provide the sampling system with an observation window that only includes environmental interference, so that the statistics of the reference noise are not mixed in by the pulse edge response. During the set interval, the excitation source keeps the output at zero level, and the transformer under test still maintains the clamping and wiring state determined in step S1. The reference pickup continuously collects the reference noise waveform. The reference pickup is a pickup channel that is electrically isolated from the transformer under test. Its installation position is fixed inside the fixture, and the grounding and shielding structure is solidified during assembly. It mainly reflects the electromagnetic interference introduced by switching power supplies, drivers and welding equipment in the production line environment, and reflects the coupling mode of the interference in the current fixture.

[0035] Within the settling interval, the noise mean square value is calculated for the reference noise. The noise mean square value is calculated based on the discrete sampling sequence within the settling interval. During the calculation, the sequence is first subjected to baseline removal to avoid zero-point drift of the sampling channel raising the statistic. Then, the baseline-removed sequence is squared point by point and averaged to obtain the mean square value used to characterize the noise intensity. The noise mean square value is a value with the same dimensions as the sampling channel and is used to quickly determine whether the current electromagnetic environment meets the test conditions on the production line. The noise upper limit is written into the test formula and fixed with the tooling number, serving as a consistent judgment boundary across shifts for the same machine.

[0036] When the noise mean square value exceeds the noise upper limit, the control process sets a pause mark for this test and pauses the process to step S3. The pause is not a simple wait, but a transition to the shielding grounding check process. This process is executed in a fixed order and the results are recorded to avoid difficulties in reproduction due to differences in human operation. Specifically, it includes: confirming the continuity of the tooling protective grounding, confirming the connection status between the excitation source shell grounding and the acquisition system grounding terminal, confirming the shielding layer termination position and termination tightness, confirming the connection status between the reference pickup shielding shell and the tooling ground, and confirming the continuity of the sampling cable shielding layer and the locking status of the connector. After completing the above checks, an unloaded sampling is performed to verify that the amplitude distribution of the reference noise waveform has returned to the allowable range of the test formula. After the shielding grounding check process is completed, the pause mark is removed and step S3 continues to be executed, so that subsequent gating, coherent superposition and other operations are based on noise-controlled sampling, thereby reducing the frequency of misjudgment and retesting caused by sudden noise increases.

[0037] To ensure a closed-loop traceability system, when the noise mean square value exceeds the noise limit and triggers a pause, the pause marker and the corresponding noise mean square value are written into the traceability record and bound to the unique identifier read in step S1. The trigger time and workstation number are also written into the traceability record so that subsequent quality analysis can distinguish between "abnormal device response" and "test conditions not being met due to excessive environmental noise" and provide a location basis for machine maintenance.

[0038] For example: Under normal circumstances, the reference noise mean square value of a production line within the static interval is stable below a given threshold, and the test process continuously enters step S3; when the start-up and shutdown of nearby welding equipment causes increased interference, the noise mean square value rises briefly and exceeds the noise upper limit. The system sets a pause mark and performs a shield grounding check. After completing the cable shield termination reset and grounding tightening, it samples again under no-load conditions. The noise mean square value falls back to the allowable range, the test process is unsuspended and continues to execute step S3. At the same time, the pause mark and the noise mean square value are written into the traceability record.

[0039] In this embodiment, step S7 establishes the same batch benchmark for the first indicator and the second indicator respectively, and updates it with the most recent qualified samples in a fixed window. The qualified samples are limited to: the qualified judgment output by step S7, and the unstable clamping label and the pause mark are not written in the traceability record. The purpose of this limitation is to ensure that the data entering the benchmark sample library comes from effective measurement under controlled test conditions, and to avoid writing false anomalies introduced by clamping fluctuations and environmental noise exceeding limits into the threshold generation link.

[0040] During the rolling update, the central statistic and discrete statistic are calculated for the first indicator sample sequence and the second indicator sample sequence within the window, respectively. This embodiment selects the statistic that is not sensitive to outliers because production line interference is sudden: the same tooling may experience an electromagnetic pulse, fixture contact micro-motion, cable sway, or other sporadic events in a short period of time. These events will form isolated outliers in the indicator sequence. If the mean and variance are used, outliers will significantly pull the mean and amplify the variance, which will cause the threshold to jump due to non-process reasons within an update cycle, allowing defects that should be intercepted to pass or misjudging a large number of normal products.

[0041] In practice, the central statistic is the median of the window sample, and the discrete statistic is the median of the absolute deviation around the center. Similarly, the second indicator yields the corresponding central and discrete statistics, which are then used to generate the first and second thresholds. In this embodiment, the threshold is written as "the central statistic plus the proportional term of the discrete statistics". The proportional coefficient is fixed in the test formula and bound to the tooling number, so that the same tooling uses the same threshold generation rule in different shifts. After the threshold is generated, the central statistic, discrete statistics, threshold, and corresponding window number are written into the traceability record to ensure that the source of the threshold can be reproduced during auditing.

[0042] Threshold drift monitoring is performed on a window update cycle. After each new threshold is generated, the change in threshold relative to the previous window threshold is calculated and compared with the drift limit. The calculation criteria for the change and the drift limit are fixed in the test formula. The drift limit is used to define "slow changes caused by process fluctuations" and "abrupt changes caused by changes in machine status". When the change exceeds the drift limit, the trace record is written to the drift alarm and the calibration process is triggered. The calibration process sequentially performs sampling channel time alignment verification, constant current excitation amplitude verification, and reference pickup no-load calibration. After calibration is completed, rolling updates resume and the calibration completion mark is written to the trace record, forming a closed-loop link from drift alarm to calibration action.

[0043] For example: with 50 windows, 49 of them have the first indicator concentrated around 1.00, while 1 window is disturbed when the welding machine is started and stopped during the shift and reads 10.00. If the mean is used, the mean will be significantly raised, and the threshold will move up accordingly. When the median is used as the central statistic, the central statistic still falls around 1.00, and the discrete statistic will not be significantly amplified by a single outlier. The threshold remains stable. This stability makes the threshold more reflective of the true state of the tooling and batch, rather than being led by an occasional disturbance.

[0044] In this embodiment, step S5 is used to perform frequency domain scanning measurement on the end connection state of the transformer winding under test to form an equivalent AC resistance curve and a phase curve. These curves are used as input data for step S6 to generate the second index. Step S5 adopts a constant current frequency sweep method. The constant current index is adjusted by the closed loop of the excitation source during the frequency sweep process to keep the fundamental amplitude of the winding end current consistent at each frequency point, thereby reducing the amplitude drift caused by the change of winding impedance with frequency and making the impedance estimation between different frequency points comparable.

[0045] The constant current frequency sweep uses a discrete frequency point sequence, which is a set of frequency points predefined by the test formula. The frequency points are output sequentially in ascending order. Each frequency point has a dwell time, which is used to ensure that the excitation source enters a steady state and that the acquisition window covers multiple cycles. After the dwell time is reached, the system switches to the next frequency point until the entire frequency point sequence is completed. The frequency point sequence and dwell time are bound to the tooling number as part of the measurement formula and are written into a traceability record in step S7 for subsequent cross-station consistency verification, avoiding incomparability of curves due to differences in frequency point configuration.

[0046] At each frequency point, multi-cycle terminal voltage waveforms and terminal current waveforms are acquired. The terminal voltage waveform is the voltage sample at both ends of the winding terminals, and the terminal current waveform is the sample of the current injected into the winding terminals. Both are acquired synchronously from the same trigger reference to ensure that the phase information is available. Multi-cycle means that the acquisition window covers several complete cycles of the excitation signal at that frequency point, which is used to suppress random noise and reduce phase estimation jitter in subsequent synchronous detection.

[0047] Synchronous detection is performed on the acquired waveform to obtain the fundamental voltage and current components. Synchronous detection refers to extracting the orthogonal components with the same frequency as the current excitation frequency from the terminal voltage and terminal current waveforms, thereby obtaining the fundamental amplitude and fundamental phase. In engineering implementation, the excitation source output simultaneously provides reference sine and cosine sequences. The acquisition system multiplies and accumulates the terminal voltage waveform with the reference sequence and averages them within the acquisition window to obtain the in-phase and orthogonal components of the voltage. The terminal current waveform is processed in the same way to obtain the in-phase and orthogonal components of the current. The fundamental amplitude and fundamental phase relative to the reference are calculated from the in-phase and orthogonal components. In this way, non-in-phase components are suppressed by averaging, which is suitable for impedance measurement in the background of production line noise.

[0048] Based on the fundamental voltage amplitude and phase, and the fundamental current amplitude and phase, the complex impedance at that frequency point is calculated. The complex impedance is obtained by dividing the fundamental voltage by the fundamental current. Its amplitude reflects the magnitude of the impedance, and its phase angle reflects the phase difference between the voltage and the current. The complex impedances at each frequency point are arranged in the frequency point sequence to obtain the curve data of the complex impedance changing with frequency. The equivalent AC resistance curve is generated from the real part of the complex impedance, and the phase curve is generated from the phase angle of the complex impedance. The equivalent AC resistance curve represents the curve of the equivalent dissipation component of the winding end changing with frequency in the frequency sweep range, and the phase curve represents the curve of the phase relationship between voltage and current changing with frequency. Both curves are indexed by the frequency point sequence on the horizontal axis, which serves as the basis for the subsequent step S6 to extract the segment slope, resistance ratio, and jump count.

[0049] To ensure consistency across workstations, step S5 writes the frequency sequence and dwell time into the traceability record and associates it with the unique identifier read in step S1. When verifying consistency across workstations, the frequency sequence and dwell time generated by the same unique identifier at different workstations are read. If the two are inconsistent, the equivalent AC resistance curve and phase curve are marked as incomparable data and a retest or manual review is triggered to avoid using curves obtained from different formulations directly in the same judgment threshold system.

[0050] For example: The frequency point sequence is set as a number of discrete points from low frequency to high frequency, and the dwell time of each point covers multiple cycles. The fundamental voltage and fundamental current of a transformer under test are obtained by synchronous detection at each frequency point. Then, the complex impedance is calculated and the equivalent AC resistance curve and phase curve are formed. If the frequency point sequence of the same unique identifier at another station is missing some high-frequency points, the frequency point sequence is found to be inconsistent after tracing and checking. The data of that station is marked and a retest is triggered to ensure that the calculation of the second index in step S6 is based on consistent frequency domain sampling conditions.

[0051] In this embodiment, after obtaining the phase curve output in step S5, step S6 further subdivides and distinguishes the contact state of the end connection, which is used to further distinguish the connection hazard label into the contact discontinuity label, and outputs the frequency range that can be used for rework positioning. The phase curve refers to the sequence data formed by the phase angle of the complex impedance in the order of the frequency points on the discrete frequency point sequence of constant current sweep frequency; each point corresponds to the phase difference between the voltage fundamental wave and the current fundamental wave at a frequency point; the phase curve not only reflects the inductive and capacitive components of the winding equivalent network, but is also sensitive to small changes in the end contact state, especially when there is intermittent conduction, micro-arcling or discontinuity of the contact point, the equivalent network may have abrupt changes between adjacent frequency points.

[0052] To avoid interference from phase jumps in numerical representation, step S6 first performs phase unrolling on the phase curve. Phase unrolling refers to the continuous processing of the phase curve, eliminating the back jumps caused by the limitation of the phase value range, so that the phase changes with the frequency point presents a continuous trajectory. In engineering implementation, phase unrolling is processed point by point in ascending order of frequency points: the phase difference between two adjacent points is compared, and when the phase difference crosses the boundary, an integer equivalent value is added or subtracted to the subsequent phase point to restore the phase continuity. The phase unrolled data is used as the input for subsequent differential and statistical analysis to ensure that the jump comes from the change in the state of the measured object, rather than the back jump caused by the numerical representation method.

[0053] After phase expansion, the phase curve is subjected to first-order difference. First-order difference refers to calculating the difference between the phase values ​​of adjacent frequency points to obtain a difference sequence. The sign of the difference sequence reflects the direction of local phase change as the frequency increases, and the difference amplitude reflects the severity of the local change. Under normal connection conditions, the phase curve usually changes slowly in the same direction as the frequency point, and the number of sign reversals in the difference sequence is small. When there is a discontinuity in the end contact, the phase curve may experience local swings or even reverse changes, and the difference sequence will show frequent positive and negative switching, which is manifested as an increase in the number of sign reversals.

[0054] Jump counts are obtained by counting the number of sign reversals in the differential sequence. Jump counts are defined as the number of times the signs of adjacent differential terms in the differential sequence change. During the statistics, differential terms close to zero are processed according to the zero interval rules specified in the test formula to avoid false flips caused by small noises near zero. Jump counts are used to characterize the local instability of the phase curve and are an important basis for subdividing the contact state.

[0055] Simultaneously, the phase swing amplitude is calculated. The phase swing amplitude is used to quantify the reverse offset of the phase curve within a certain frequency range, which is different from slow unidirectional drift. In implementation, candidate intervals corresponding to sign reversal are first identified in the phase curve. Then, the local peak and local valley values ​​of the phase are calculated in each candidate interval, and the difference between the two is used as the swing amplitude of that interval. The maximum value of the swing amplitude of the candidate interval is used as the phase swing amplitude. This quantity reflects whether the phase curve has obvious reversal. Combined with the jump count, it can suppress false flips caused only by noise.

[0056] Step S6 compares the jump count with the jump threshold and the phase swing amplitude with the swing threshold. When the jump count is greater than the jump threshold and the phase swing amplitude is greater than the swing threshold, the connection hazard label is subdivided into a contact discontinuity label. The jump threshold and swing threshold are written into the test formula and maintained in association with the same batch benchmark in step S7 to ensure that the judgment boundary under different tooling and different batches has a consistent management method. The reason for using dual conditions is that using only the jump count is easy to misjudge multiple small flips caused by high noise as contact discontinuity, while using only the swing amplitude may miss multiple small swings caused by frequent intermittent conduction. When both conditions are met at the same time, it can better correspond to the discontinuous characteristics of the end contact state.

[0057] To achieve rework location, step S6 also writes the frequency range that generates the jump into the traceability record. The frequency range refers to the range from the frequency point where the sign inversion first occurs to the frequency point where the sign inversion ends in the discrete frequency sequence, or the range determined by the start and end frequencies of the candidate range. The traceability record saves the correspondence between the frequency range and the unique identifier, tooling number, second indicator and connection hazard label, so that the rework station can trace back the corresponding connection status change segment according to the frequency range and narrow down the investigation scope by combining the end process information.

[0058] For example: After phase expansion, the phase curve of a transformer under test shows multiple local swings in the mid-to-high frequency band. The first-order differential sequence shows multiple sign inversions in this segment, and the jump count is 7. At the same time, the maximum swing amplitude is a certain significant value. When both exceed the jump threshold and swing threshold in the test formula, the connection hazard label is subdivided into contact discontinuity label, and the corresponding frequency range is written into the traceability record for key inspection of the contact continuity of end crimping and welding during rework.

[0059] In this embodiment, after the constant current frequency sweep is completed in step S5 and the equivalent AC resistance curve and phase curve are obtained, a thermal excitation test is performed to introduce independent observations related to local heating. This allows the electrical characteristics and thermal response characteristics to be referenced simultaneously when the second index is generated in step S6, thereby improving the ability to distinguish micro-defects in the end connection. The thermal excitation test is completed in the same clamping state, using the unique identifier and wiring relationship established in step S1, to avoid changes in the thermal path and contact state caused by re-clamping.

[0060] The thermal excitation test selects a fixed frequency point and applies constant current excitation. The fixed frequency point refers to a frequency point selected from the discrete frequency point sequence in step S5. This frequency point is solidified in the test formula and bound to the tooling number to ensure that the thermal excitation conditions are consistent for different work stations and different batches. The purpose of selecting this frequency point is to generate stable AC loss at the conductor and end connection under a controllable frequency. Constant current excitation means that the excitation source is adjusted through a closed loop to keep the fundamental amplitude of the winding end current stable during the heating stage, so that the temperature change is mainly determined by the equivalent AC resistance and contact state, rather than by current fluctuations. The heating time is also defined by the test formula. After the heating time is reached, the excitation is stopped and the natural cooling stage begins.

[0061] Temperature transient curves are collected at temperature measurement points during the heating and cooling phases. Temperature measurement points refer to pre-determined temperature acquisition locations in the tooling, and their relative positions to the transformer under test are fixed to improve batch comparability. Temperature measurement points can be located on the outer surface near the end connection area or on the outer surface near the winding end. The temperature measurement method can be a contact temperature sensor or a non-contact infrared temperature measurement. The temperature measurement method is kept consistent on the same tooling and written into the test formula. The temperature transient curve refers to the temperature change sequence over time from the start of heating to the end of cooling. The curve is associated with a unique identifier and stored for traceability and verification.

[0062] The temperature rise slope and cooling slope are calculated based on the transient temperature curve. The temperature rise slope characterizes the rate of temperature increase per unit time during the heating phase. In practice, a linear interval avoiding the initial transient is selected during the heating phase, and a linear fit is performed on temperature and time. The slope of the fitted line is taken as the temperature rise slope. This approach can reduce the impact of sensor response hysteresis and initial contact thermal resistance changes on the results. The cooling slope characterizes the rate of cooling after excitation stops. In practice, a stable interval is selected during the cooling phase, and a linear fit is performed on temperature and time. The absolute value of the fitted slope is taken as the cooling slope. The temperature rise slope and cooling slope together constitute the thermal response characteristics. The temperature rise slope is more sensitive to local losses and hot spot formation, while the cooling slope is more sensitive to changes in thermal path and contact thermal resistance. The combination of the two can reduce the risk of a single characteristic being affected by installation conditions.

[0063] The temperature rise slope and cooling slope are added as thermal response features to the second index generation process in step S6. In implementation, step S6 introduces thermal response features in addition to the original set of connection risk features based on equivalent AC resistance curves and phase curves. These features are then mapped to the same dimension as the second index according to the normalization rules specified in the test formula, and then summarized together with the original connection risk features to generate the second index. The central statistic and discrete statistic on which the normalization depends are provided by the same batch benchmark in step S7 to ensure the comparability of thermal response features between different tooling and different batches.

[0064] When the second indicator triggers the connection hazard judgment, the thermal response characteristics and the equivalent AC resistance at the corresponding frequency point are simultaneously written into the traceability record to distinguish between virtual contact and abnormal conductor loss. The equivalent AC resistance at the corresponding frequency point refers to the equivalent AC resistance value obtained at the fixed frequency point in step S5, which reflects the electrical dissipation component at the same frequency. The thermal response characteristics reflect the rate of temperature change under the same excitation conditions. In engineering, virtual contact often manifests as a coexistence of electrical characteristic fluctuations and local thermal resistance changes, with a more prominent abnormal temperature rise slope, and the cooling slope may exhibit different attenuation behavior than normal samples. Abnormal conductor loss is more often manifested as an overall higher equivalent AC resistance and a thermal response closer to a stable uniform heating mode. The traceability record saves both types of data simultaneously, enabling rework and process traceability to distinguish between modes such as "abnormal resistance but insensitive temperature rise" and "abnormal temperature rise accompanied by an unstable thermal path," without having to make a judgment based solely on a single electrical characteristic.

[0065] For example: The fixed frequency point is selected at a mid-to-high frequency point in the frequency sweep sequence. During the heating stage, the constant current is maintained for 5 seconds, and the sampling frequency of the temperature measurement point is 100 Hz. Sample A has a slightly higher equivalent AC resistance at this frequency point, a significantly increased temperature rise slope, and an abnormal cooling slope. After the second indicator triggers the connection defect judgment, the traceability record shows that the resistance and thermal response are abnormal at the same time. The rework prioritizes checking the end crimping and solder joint continuity. Sample B has a higher equivalent AC resistance at this frequency point, but the temperature rise slope and cooling slope are close to the baseline distribution of the same batch. The traceability record suggests that it is more likely to be due to wire loss. The rework focuses on checking the consistency of wire specifications and the quality of end paint removal.

[0066] Example 2: like Figure 2 As shown, this embodiment further improves the design based on Embodiment 1. The difference is that in actual operation of Embodiment 1, it was found that the reference noise collected within the static interval can reflect the strength of the background interference on the production line, but it is difficult to cover the synchronous transient component introduced by the excitation source switching network and the parasitic parameters of the tooling. This synchronous transient component appears in phase with the pulse trigger and is amplified in coherent superposition after entering step S3, causing the first indicator to rise abnormally in some shifts. It fails to stably distinguish between the early abnormal response of insulation and the inherent transient response of equipment without increasing the false judgment rate. Based on this, the online detection method for transformer production quality also includes: Add step S2a after step S2 and before step S3.

[0067] S2a: Under the same clamping state, keep the sampling channel and two-level gating configuration unchanged, control the excitation source to execute the no-trigger sequence according to the same triggering rhythm as in step S2, keep the primary and secondary terminals at zero level output and complete the same switching action as in step S2, synchronously acquire the no-trigger response current waveform and generate the inherent transient template; then in step S3, for the response current of each pulse, first complete the time alignment according to the pulse triggering time, then calculate the template scaling factor in the outer window and subtract the inherent transient template according to the scaling factor to obtain the response current after removing the inherent transient, then perform two-level gating of the outer window and inner window, DC bias removal, amplitude normalization, edge alignment and coherent superposition, and write the no-trigger response current waveform, inherent transient template, template scaling factor and subtraction mark of step S2a into the traceability record along with the unique identifier.

[0068] The above step S2a constructs the inherent transient template corresponding to the current tooling and the current switching state in each measurement by "empty trigger sampling consistent with the test cycle". This makes the estimation of the inherent transient independent of offline calibration and fixed filter settings, and separates the inherent transient of the device from the response of the object under test from the implementation link.

[0069] Example 3: like Figure 3 As shown, based on the same inventive concept as the online transformer production quality detection method in the foregoing embodiments, this application provides an online transformer production quality detection system. The system and method embodiments in this application are based on the same inventive concept. The system includes: Clamping identification module: The clamping identification module clamps the transformer under test and reads its unique identifier, measures the contact resistance of the primary terminal and the contact resistance of the secondary terminal, and remeasures if the contact resistance exceeds the limit; Pulse sampling module: The pulse sampling module applies a bipolar high-slope pulse between the primary and secondary sides to synchronously sample the voltage, response current, reference noise, and align the sampling time. Gated purification module: The gated purification module sets two levels of gated windows with pulse edge. The response current is coherently superimposed in the inner window, and the net response is obtained by calculating the cancellation amount based on the reference noise in the outer window. Insulation Index Module: The insulation index module calculates the cross-correlation peak value and the ratio of gated energy to bandwidth energy from the net response and generates the first index by combining it with the batch benchmark. Frequency sweep resistance measurement module: The frequency sweep resistance measurement module applies constant current to the winding terminals, samples the terminal voltage and terminal current, and calculates the equivalent AC resistance curve and phase curve; Connection Index Module: The connection index module calculates the segmented slope and resistance ratio from the equivalent AC resistance curve, calculates the jump count from the phase curve, and generates a second index by combining the batch benchmark. Threshold Judgment Module: The threshold judgment module updates the first threshold and the second threshold with qualified samples from the same batch. When the first indicator is greater than or equal to the first threshold, it outputs "unqualified" and marks an insulation hazard. When the second indicator is greater than or equal to the second threshold, it outputs "unqualified" and marks a connection hazard. When both indicators are less than the first threshold and the second threshold, it outputs "qualified". The unique identifier, sampling data, indicators and judgment results are stored and written back to the process database.

[0070] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0071] The above description is merely a preferred embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present application, based on the technical solution and concept of the present application, should be covered within the scope of protection of the present application.

Claims

1. An online inspection method for transformer manufacturing quality, characterized in that, The methods include: Mount the transformer under test and read its unique identifier. Measure the contact resistance of the primary terminal and the contact resistance of the secondary terminal. If the contact resistance exceeds the limit, repeat the measurement. A bipolar high-slope pulse is applied between the primary and secondary sides to synchronously sample the voltage, response current, reference noise, and align the sampling time. Two levels of gating windows are set with pulse edge. The response current is coherently superimposed in the inner window, and the net response is obtained by calculating the cancellation amount based on the reference noise in the outer window. The first indicator is generated by calculating the cross-correlation peak value and the ratio of gated energy to bandwidth energy from the net response and combining them with the batch benchmark. Apply constant current sweep frequency to the winding terminals, sample the terminal voltage and terminal current, and calculate the equivalent AC resistance curve and phase curve; The segmented slope and resistance ratio are calculated from the equivalent AC resistance curve, and the jump count is calculated from the phase curve. Combined with the batch reference, a second index is generated. The first and second thresholds are updated using qualified samples from the same batch. When the first indicator is greater than or equal to the first threshold, the output is unqualified and marked as an insulation hazard. When the second indicator is greater than or equal to the second threshold, the output is unqualified and marked as a connection hazard. When both indicators are less than the first and second thresholds, the output is qualified. The unique identifier, sampling data, indicators and judgment results are stored and written back to the process database.

2. The online inspection method for transformer production quality according to claim 1, characterized in that, The reference noise is acquired by a reference pickup that is electrically isolated from the transformer under test. No-load calibration is performed before applying a bipolar high-slope pulse. The no-load calibration includes acquiring the background noise waveform, establishing amplitude correction coefficients, and establishing phase correction coefficients. Before performing the cancellation process in the outer window, amplitude correction and phase correction are performed on the reference noise. Then, the cancellation amount is calculated based on the correlation between the reference noise and the response current in the outer window, and the cancellation amount is subtracted from the response current to obtain the net response.

3. The online inspection method for transformer production quality according to claim 1, characterized in that, The system employs a two-level gating system with an outer window and an inner window. The outer window covers the preset duration after the pulse edge, while the inner window is located inside the outer window and avoids the edge saturation region. Within the inner window, DC bias removal, amplitude normalization, and edge alignment are performed on the response current of each pulse, and then coherent superposition is performed according to the pulse sequence number. After coherent superposition, the zero-crossing drift is calculated and compared with the upper limit of drift. If the zero-crossing drift exceeds the limit, the system returns to the clamping positioning and unique identifier reading stage for retesting and writes the clamping instability tag.

4. The online inspection method for transformer production quality according to claim 1, characterized in that, The bipolar high-slope pulse is output in pulse groups. Each pulse group outputs a positive pulse and a negative pulse in sequence, and a rest interval is set between pulse groups. Reference noise is collected and the noise mean square value is calculated within the rest interval. If the noise mean square value exceeds the noise upper limit, the test is paused and the shield grounding check procedure is executed. After the shield grounding check procedure is completed, the two-level gating and net response generation are continued, and the pause mark and the noise mean square value are written into the traceability record.

5. The online inspection method for transformer production quality according to claim 1, characterized in that, Establish benchmarks for the same batch according to tooling number, and use the most recent qualified samples with a fixed window number for rolling updates; during rolling updates, calculate the center value and dispersion of the first index and the second index, and generate the first threshold and the second threshold from the center value and dispersion; when the change of the threshold relative to the previous window exceeds the drift limit, write a drift alarm and trigger the calibration process, which includes sampling channel time alignment verification, constant current excitation amplitude verification and reference pickup no-load calibration.

6. The online inspection method for transformer production quality according to claim 1, characterized in that, The constant current frequency sweep uses a discrete frequency sequence, with the frequency points output in ascending order and a dwell time maintained at each frequency point. At each frequency point, multi-cycle terminal voltage and terminal current waveforms are acquired, and synchronous detection is performed on the acquired waveforms to obtain the fundamental voltage and current components. Based on the fundamental amplitude and phase, the complex impedance is calculated, and the equivalent AC resistance curve and phase curve are generated. The frequency sequence and dwell time are written into the traceability record for cross-station consistency verification.

7. The online inspection method for transformer production quality according to claim 1, characterized in that, The jump count is obtained as follows: perform phase expansion and first-order difference on the phase curve, count the number of differential sign inversions to obtain the jump count, and calculate the phase swing amplitude at the same time; when the jump count is greater than the jump threshold and the phase swing amplitude is greater than the swing threshold, the connection hazard tag is subdivided into contact discontinuity tags, and the frequency range of the jump is written into the traceability record for rework location.

8. The online inspection method for transformer production quality according to claim 1, characterized in that, After the constant current frequency sweep is completed, a thermal excitation test is performed, including: applying constant current excitation at a fixed frequency point and maintaining the heating time, and collecting the transient temperature curve at the temperature measurement point; calculating the temperature rise slope and cooling slope based on the transient temperature curve, and adding the temperature rise slope and cooling slope as thermal response features into the second index generation process; when the second index triggers the connection hazard judgment, the thermal response features and the equivalent AC resistance of the corresponding frequency point are simultaneously written into the traceability record to distinguish between virtual contact and abnormal wire loss.

9. The online inspection method for transformer production quality according to claim 1, characterized in that, Also includes: Keeping the clamping and sampling configuration unchanged, the primary and secondary terminals are kept at zero level and the switching is completed according to the same triggering cycle. The no-trigger response current is collected synchronously to generate an inherent transient template for subsequent subtraction and writing into the traceability record.

10. An online quality inspection system for transformer production, characterized in that, The system includes: The clamping identification module clamps the transformer under test and reads its unique identifier, then measures the contact resistance of the primary terminal and the contact resistance of the secondary terminal. If the contact resistance exceeds the limit, the measurement is repeated. A pulse sampling module applies a bipolar high-slope pulse between the primary and secondary windings to synchronously sample voltage, response current, reference noise, and align sampling times. The gated purification module has two levels of gated windows set with pulse edges. The response current is coherently superimposed in the inner window, and the net response is obtained by calculating the cancellation amount based on the reference noise in the outer window. The insulation index module is a first index generated by calculating the cross-correlation peak value and the ratio of gated energy to bandwidth energy from the net response and combining them with the batch benchmark. A frequency sweep resistance measurement module applies a constant current frequency sweep to the winding terminals, samples the terminal voltage and terminal current, and calculates the equivalent AC resistance curve and phase curve. The connection index module calculates the segmented slope and resistance ratio from the equivalent AC resistance curve, calculates the jump count from the phase curve, and generates a second index by combining the batch benchmark. The threshold determination module updates the first threshold and the second threshold with qualified samples from the same batch. When the first indicator is greater than or equal to the first threshold, it outputs "unqualified" and marks an insulation hazard. When the second indicator is greater than or equal to the second threshold, it outputs "unqualified" and marks a connection hazard. When both indicators are less than the first threshold and the second threshold, it outputs "qualified". The unique identifier, sampling data, indicators and determination results are stored and written back to the process database.

Images