A multi-channel electrical parameter parallel test system for power adapters

Abstract

The present application relates to power supply test and automation detection technical field, specifically for a kind of multi-channel electric parameter parallel test system for power adapter, including parallel test loop, multi-channel signal acquisition module, state evaluation and quantization module, intelligent scheduling decision module and dynamic decoupling execution module;By obtaining the real-time electric parameter signal of each test channel and extracting to form multidimensional electrical state sequence, construct mutual impedance matrix between channel, generate electrical crosstalk entropy based on matrix perturbation model nonlinear coupling analysis, and predict system failure risk value, then determine target scheduling strategy according to electrical crosstalk entropy and preset safety threshold comparison, reconstruct timing and load adjustment to parallel test loop and test load;The present application realizes stable operation under high-density parallel test environment.

Description

Technical Field

[0001] This invention relates to the field of power supply testing and automated testing technology, specifically to a multi-channel parallel testing system for electrical parameters of power adapters. Background Technology

[0002] Currently, production line testing of power adapters is usually implemented based on multi-channel parallel methods. During high-density concurrent testing, adjacent test channels are easily affected by electromagnetic coupling, heat accumulation, and external power grid harmonic disturbances, resulting in crosstalk between the electrical parameters of each channel. Existing testing mechanisms usually only focus on the output results of a single channel, making it difficult to identify the instability trend of inter-channel coupling in a timely manner and make targeted adjustments, which reduces the stability of parallel testing and the reliability of test results. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a multi-channel parallel testing system for electrical parameters of power adapters. Specifically, the technical solution of this invention includes:

[0004] The parallel test circuit includes multiple test channels, which are used to connect to multiple power adapters and adjustable test loads respectively.

[0005] A multi-channel signal acquisition module is connected to a parallel test circuit to acquire real-time electrical parameter signals of the corresponding power adapter in each test channel through the parallel test circuit, and to extract features from the real-time electrical parameter signals to obtain a multi-dimensional electrical state sequence.

[0006] The state assessment and quantization module is connected to the multi-channel signal acquisition module. It is used to receive multi-dimensional electrical state sequences, construct an inter-channel mutual impedance matrix representing the equivalent impedance and coupling strength based on the multi-dimensional electrical state sequences, and perform nonlinear coupling analysis on the inter-channel mutual impedance matrix based on the matrix perturbation model to generate electrical crosstalk entropy, and predict the system failure risk value based on the electrical crosstalk entropy.

[0007] The intelligent scheduling decision module is communicatively connected to the state assessment and quantification module. It is used to receive electrical crosstalk entropy and system failure risk value, and compare the electrical crosstalk entropy with a preset safety threshold based on background crosstalk calibration to determine the target scheduling strategy.

[0008] The dynamic decoupling execution module is used to reconstruct the timing and adjust the load of parallel test loops and adjustable test loads based on the target scheduling strategy, and feed back the adjusted loop status to the multi-channel signal acquisition module.

[0009] Preferably, the multi-channel signal acquisition module includes a wideband sampling unit, a harmonic separation unit, and a state reconstruction unit; the wideband sampling unit is used to acquire real-time electrical parameter signals of each test channel, including voltage signals, current signals, and ripple signals; the harmonic separation unit is used to perform frequency domain decomposition on the voltage signals, current signals, and ripple signals to separate the fundamental component and harmonic components with frequencies below a preset frequency threshold; the state reconstruction unit is used to fuse the fundamental component and harmonic components with frequencies below a preset frequency threshold to generate a multi-dimensional electrical state sequence.

[0010] Preferably, the state assessment and quantification module includes an eigenvalue perturbation analysis unit and an entropy calculation unit; the eigenvalue perturbation analysis unit is used to calculate the eigenvalue drift of the inter-channel mutual impedance matrix; the entropy calculation unit is used to normalize the eigenvalue drift to obtain the drift probability distribution, and calculate the information entropy of the drift probability distribution as the electrical crosstalk entropy, wherein the electrical crosstalk entropy characterizes the degree of electromagnetic noise coupling in the parallel test circuit.

[0011] Preferably, the state assessment and quantification module further includes a risk mapping unit, which is used to input electrical crosstalk entropy into a preset risk assessment model to output a system failure risk value; the preset risk assessment model is a trained deep neural network, which is configured to map the input electrical crosstalk entropy into a failure probability value normalized to a preset numerical range, and output the failure probability value as the system failure risk value.

[0012] Preferably, the intelligent scheduling decision module includes a strategy generation unit, which generates a first scheduling strategy as the target scheduling strategy when the electrical crosstalk entropy is lower than a preset safety threshold, wherein the first scheduling strategy is a synchronous test instruction that maintains the current maximum concurrency rate; and generates a second scheduling strategy as the target scheduling strategy when the electrical crosstalk entropy is equal to or higher than the preset safety threshold, wherein the second scheduling strategy is an asynchronous degradation test instruction that reduces the system failure risk value.

[0013] Preferably, the asynchronous degradation test instruction includes at least one of the following: pseudo-random timing misalignment instruction, channel intermittent heat dissipation instruction, and load dynamic rate limiting instruction.

[0014] Preferably, the dynamic decoupling execution module includes a timing control unit and a dynamic electronic load unit; the timing control unit is used to adjust the trigger phase of each test channel according to the target scheduling strategy to achieve electrical isolation reconstruction between channels; the dynamic electronic load unit is used to adjust the impedance change rate of the test load according to the target scheduling strategy.

[0015] Preferably, the system further includes an adaptive compensation module, which is used to receive a multi-dimensional electrical state sequence and a target scheduling strategy, generate a dynamic impedance compensation signal based on the target scheduling strategy, and inject the dynamic impedance compensation signal into a parallel test circuit to suppress low-frequency resonance.

[0016] Preferably, the multi-channel signal acquisition module further includes an anomaly marking unit, used to identify abnormal channels with a signal-to-noise ratio lower than a preset signal-to-noise ratio threshold and containing oscillation characteristics of a target resonance preset frequency band calibrated by frequency sweep during the extraction of multi-dimensional electrical state sequences, and to mark the resonance time point of the abnormal channel, as well as to cache the data of the abnormal channel.

[0017] Preferably, a first synchronous channel and a second asynchronous channel are provided between the state assessment and quantification module and the intelligent scheduling decision module; the first synchronous channel is connected by a hard wire and is used to transmit electrical crosstalk entropy and system failure risk value; the second asynchronous channel adopts a bus communication protocol and is used to transmit historical test logs and model update parameters.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] 1. This invention addresses the problem that existing mechanisms struggle to identify instability trends in inter-channel coupling. This system generates electrical crosstalk entropy by constructing a mutual impedance matrix and performing nonlinear perturbation analysis, and predicts the system failure risk value. Based on this, the timing sequence is dynamically reconstructed and the load is adjusted. This mechanism upgrades static monitoring to active closed-loop control of inter-channel coupling states, improving the stable operation capability of high-concurrency testing.

[0020] 2. This invention performs frequency domain decomposition on electrical parameter signals, separates and fuses the fundamental component and low-frequency harmonic components to generate a multidimensional electrical state sequence; this effectively avoids the masking of low-frequency resonance anomalies by high-frequency switching noise, making the evaluation matrix more focused on the real coupling disturbances strongly correlated with system cascading instability, and greatly improving the accuracy of early crosstalk risk extraction;

[0021] 3. This invention compresses complex time-varying local coupling noise into a uniformly quantized electrical crosstalk entropy by calculating the eigenvalue drift of the inter-channel mutual impedance matrix and obtaining the information entropy of its probability distribution. This mechanism can accurately identify whether the coupling mode inside the test array has undergone hierarchical diffusion, effectively remove individual abnormal interference, and realize feedforward quantification of instability trends. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application and the prior art, the accompanying drawings used in the description of the embodiments and the prior art will be briefly introduced below:

[0023] Figure 1This is a schematic diagram of a module for a multi-channel parallel testing system for electrical parameters of a power adapter, provided in an embodiment of this application. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0025] A multi-channel parallel testing system for electrical parameters of power adapters, the system comprising:

[0026] The parallel test circuit includes multiple test channels, which are used to connect to multiple power adapters and adjustable test loads respectively.

[0027] A multi-channel signal acquisition module is connected to a parallel test circuit to acquire real-time electrical parameter signals of the corresponding power adapter in each test channel through the parallel test circuit, and to extract features from the real-time electrical parameter signals to obtain a multi-dimensional electrical state sequence.

[0028] The state assessment and quantization module is connected to the multi-channel signal acquisition module. It is used to receive multi-dimensional electrical state sequences, construct an inter-channel mutual impedance matrix representing the equivalent impedance and coupling strength based on the multi-dimensional electrical state sequences, and perform nonlinear coupling analysis on the inter-channel mutual impedance matrix based on the matrix perturbation model to generate electrical crosstalk entropy, and predict the system failure risk value based on the electrical crosstalk entropy.

[0029] The intelligent scheduling decision module is communicatively connected to the state assessment and quantification module. It is used to receive electrical crosstalk entropy and system failure risk value, and compare the electrical crosstalk entropy with a preset safety threshold based on background crosstalk calibration to determine the target scheduling strategy.

[0030] The dynamic decoupling execution module is used to reconstruct the timing and adjust the load of parallel test loops and adjustable test loads based on the target scheduling strategy, and feed back the adjusted loop status to the multi-channel signal acquisition module.

[0031] This embodiment provides a multi-channel parallel testing mechanism for electrical parameters in high-density production lines of power adapters, such as... Figure 1 As shown; specifically, the system is deployed in the aging test station of electronic devices to simultaneously connect multiple gallium nitride fast charging power adapters on the same high-density test board and continuously perform test tasks such as full load, transition load, ripple and transient response under high concurrency.

[0032] The entire system operates around a link that senses the channel coupling status in real time, quantifies crosstalk risks, dynamically adjusts the test rhythm, and provides feedback closed-loop correction, so as to maintain the stable operation of the test array under conditions of limited physical space, significant power grid harmonic pollution, and natural electromagnetic and thermal coupling between adjacent channels.

[0033] Specifically, the parallel test circuit may include channels 1 to N, each channel is connected to a power adapter under test and a test load; the multi-channel signal acquisition module acquires the voltage, current and ripple signals of each channel in each sampling period, and combines multiple electrical parameters at the same time to form a frame state of that channel;

[0034] For ease of understanding, assume there are 4 channels, and the simplified data obtained at a certain sampling time are as follows: Channel 1 is 20.1 V, 2.95 A, and 0.08 V ripple; Channel 2 is 19.9 V, 3.02 A, and 0.10 V ripple; Channel 3 is 20.0 V, 2.97 A, and 0.16 V ripple; Channel 4 is 19.8 V, 3.01 A, and 0.09 V ripple.

[0035] The system can further splice the data from three consecutive sampling points of each channel into a state segment. For example, if the ripple values ​​of channel 3 at three times are 0.12, 0.16, and 0.21 volts, it indicates that the disturbance has a continuous upward trend. After the multi-channel signal acquisition module extracts features from these state segments, it forms a multi-dimensional electrical state sequence.

[0036] The state assessment and quantization module establishes an inter-channel mutual impedance matrix based on this sequence. To ensure the understandability of the explanation, a simplified 4×4 matrix can be used for illustration, where the diagonal lines represent the equivalent impedance of each channel and the off-diagonal lines represent the coupling strength. For example, if the first row of the matrix is ​​1.00, 0.12, 0.05, and 0.03, it means that the normalized impedance of channel 1 to itself is 1.00, and the coupling strengths to channels 2, 3, and 4 are 0.12, 0.05, and 0.03, respectively. If the first row of the matrix changes to 1.00, 0.19, 0.06, and 0.04 in the next cycle, it can be intuitively seen that the coupling between channel 1 and channel 2 is enhanced.

[0037] The system compares the mutual impedance matrix between two consecutive sampling periods based on the matrix perturbation model, calculates the nonlinear variation amplitude of the off-diagonal elements of the matrix as the specific execution logic for nonlinear coupling analysis, extracts the coupling change caused by spatial electromagnetic field or thermal accumulation, and then generates electrical crosstalk entropy.

[0038] Specifically, the matrix perturbation model quantifies the nonlinear variation amplitude of the coupling relationship between channels by solving the difference matrix of the mutual impedance matrices of two consecutive sampling periods and extracting the eigenvalue spectral norm of this difference matrix. The specific explanation of mapping the nonlinear coupling to matrix perturbation form is as follows: Let... Time and The inter-channel mutual impedance matrices at time t are respectively and Matrix perturbation parameters The physical definition of this perturbation parameter is the nonlinear transient increment of the inter-channel coupling impedance in the test array caused by spatial electromagnetic radiation or local thermal accumulation; through The spectral norm characterizes the intensity of the increment, demonstrating the effectiveness of the model in quantifying time-varying coupled disturbances; the higher the electrical crosstalk entropy, the more disordered the coupling state on the entire test board, and the greater the risk of subsequent misjudgment.

[0039] After obtaining the electrical crosstalk entropy, the system further outputs a system failure risk value. This risk value can be understood as a measure of the probability that the test array will enter a state of continuous misjudgment under the current concurrency conditions. For example, normalizing the risk value to between 0 and 1, when the risk value is 0.25, it means that high-concurrency testing can continue to be maintained; when the risk value rises to 0.74, it means that the array has approached the boundary of continuous misjudgment and protective scheduling needs to be initiated.

[0040] After receiving the electrical crosstalk entropy and system failure risk value, the intelligent scheduling decision module compares them with the preset safety threshold. If the current crosstalk entropy is lower than the threshold, the original synchronous test rhythm is maintained. If the current crosstalk entropy reaches or exceeds the threshold, the target scheduling strategy is selected to adjust the trigger time, load change rate and the order of some test tasks for each channel.

[0041] The preset safety threshold is determined as follows: under the no-load and single-channel full-load calibration test of the test system, the background electrical crosstalk entropy is collected as the reference value, and the tolerance coefficient is calculated in combination with the withstand voltage limit and heat capacity limit of the test hardware. The product of the reference value and the tolerance coefficient is used as the preset safety threshold.

[0042] The dynamic decoupling execution module performs timing reconstruction and load adjustment accordingly. For example, it changes the high-current load action that originally started simultaneously to staggered peak triggering, or reduces the load rise slope of some channels from 0.8 amps per millisecond to 0.4 amps per millisecond, thereby weakening the superposition effect of instantaneous resonance. The adjusted loop state is fed back to the acquisition side to form a closed loop.

[0043] As a fault-tolerance mechanism, if individual channel data is missing in a certain sampling period, such as a voltage value gap caused by a momentary disconnection of channel 2, the preset safe state of the previous stable period of that channel can be used for mutual impedance estimation in this period, and a data failure mark is configured for that channel; if data is missing for more than two consecutive periods, the dispatching side will temporarily isolate it from the high-concurrency array to avoid incomplete data interfering with the accuracy of the global assessment; if the crosstalk entropy of all channels increases rapidly at the same time, but a single channel does not show obvious abnormalities, it is determined that it is more likely to come from external power grid disturbances, and the system will prioritize global load reduction and phase reconstruction, rather than mistakenly attributing the fault to a certain adapter itself;

[0044] In the mass production test of 480 fast charging adapters in the same batch, a certain aging board was divided into 48 parallel channels and operated at a concurrency rate of over 98% for a long time; low-frequency harmonics were introduced after the factory stamping equipment started and stopped, causing the ripple of channels 12 to 18 to rise synchronously.

[0045] After the acquisition module extracted the multidimensional state sequences of these channels, the quantization module found that the relevant off-diagonal elements in the mutual impedance matrix collectively increased from 0.08 to 0.11 to 0.17 to 0.23, and the corresponding electrical crosstalk entropy increased from 0.31 to 0.67, and the system failure risk value increased to 0.79. Therefore, the scheduling module no longer maintained the fully synchronous test, but instead commanded the dynamic decoupling execution module to introduce peak load shifting and load dynamic rate limitation for the channels in this area, so that the risk value dropped back to below 0.42 in the following 5 cycles.

[0046] The purpose of this step is to not only measure the electrical parameters of each channel in a high-density parallel testing environment, but also to simultaneously evaluate the coupling instability trend between channels caused by space compression and external harmonics, so as to achieve proactive maintenance of the high reliability of the test system, rather than just passively waiting for misjudgment results to occur.

[0047] Furthermore, the multi-channel signal acquisition module includes:

[0048] The wideband sampling unit is used to acquire real-time electrical parameter signals from each test channel, including voltage signals, current signals, and ripple signals.

[0049] The harmonic separation unit is used to perform frequency domain decomposition on voltage signals, current signals and ripple signals to separate the fundamental component and harmonic components with frequencies below a preset frequency threshold.

[0050] The state reconstruction unit is used to fuse the fundamental component and harmonic components with frequencies below a preset frequency threshold to generate a multidimensional electrical state sequence.

[0051] This embodiment provides a wideband sampling and state reconstruction mechanism for scenarios with concentrated low-frequency harmonic interference. Specifically, in the aforementioned high-density parallel testing station, relying solely on the average instantaneous voltage or current can easily mask latent low-frequency coupling anomalies. For example, multiple channels may still meet factory specifications in terms of DC output values, but their low-frequency ripple components may have already oscillated synchronously. If these components are not separated, the subsequent mutual impedance matrix will be diluted by the appearance of normal average values, making it impossible to identify the systemic instability trend in a timely manner. Therefore, this embodiment further performs frequency domain decomposition and reconstruction on the real-time electrical parameter signals.

[0052] Specifically, the wideband sampling unit can simultaneously acquire voltage, current, and ripple signals in each channel. For ease of deduction, it is assumed that a certain channel obtains a ripple sequence of 0.06, 0.09, 0.13, 0.11, 0.08, 0.05, 0.07, and 0.12 at eight consecutive sampling points, and its overall fluctuation amplitude is within the preset tolerance range. However, after frequency domain decomposition, it can be found that it contains a stable low-frequency fluctuation component and a higher-frequency switching noise component.

[0053] The harmonic separation unit splits the original signal into a fundamental component and harmonic components with frequencies below a preset frequency threshold. This frequency threshold can be set according to the coupling characteristics of the test board, for example, 500 Hz, to prioritize the extraction of low-frequency disturbances that are more related to the resonance between channels. The coupling characteristics of the test board are specifically obtained by performing frequency sweep calibration on the high-density test board in advance, and the lowest resonant frequency point that causes a nonlinear change in the amplitude of the mutual impedance between adjacent test channels is selected as the preset frequency threshold.

[0054] The state reconstruction unit does not simply retain the entire spectrum, but rather merges the fundamental component with harmonic components below a preset frequency threshold to form a multidimensional electrical state sequence that better reflects the coupled state of the system.

[0055] A microscopic example can illustrate this: If a channel has a voltage fundamental amplitude of 20.0, a low-frequency harmonic amplitude of 0.25, a current fundamental amplitude of 3.0, a low-frequency harmonic amplitude of 0.18, a ripple fundamental amplitude of 0.09, and a low-frequency harmonic amplitude of 0.07 within a certain sampling window, then the reconstructed multidimensional electrical state sequence can be represented as 20.0, 0.25, 3.0, 0.18, 0.09, 0.07. Compared to directly using the original abbreviation of 20.0 volts, 3.0 amps, and 0.09 ripple volts, this sequence more clearly distinguishes between steady-state output and the source of coupled disturbances.

[0056] The reason for this design is that if the previous scheme directly inputs the original waveform after full-frequency domain mixing into the back-end evaluation, the high-frequency switching noise may be dominant in terms of energy, masking the low-frequency anomalies that are more sensitive to instability, making it difficult for the system to respond to crosstalk risks in a timely manner. This embodiment increases harmonic separation and state reconstruction, so that the subsequent mutual impedance matrix is ​​more focused on the signal part that is more strongly associated with cascading instability.

[0057] As a fault-tolerance mechanism, if the low-frequency harmonic amplitude is close to zero after frequency domain decomposition within a certain sampling window, the state reconstruction unit still retains the corresponding position, but fills in zero or near-zero minimum values ​​to maintain the consistency of the state vector dimension at each time step; if a channel is subjected to a transient impact that causes an abnormal spike in spectral energy, the system can first perform amplitude clipping before entering the reconstruction step to avoid individual outliers misjudging the entire window as severe resonance; if broadband sampling introduces a large amount of random noise due to poor probe contact, it can be required that the channel must show similar low-frequency characteristics in two consecutive windows before it is considered a valid anomaly.

[0058] On the same 48-channel aging board, the DC output of channels 15 and 16 remained around 20 volts for a long time, which would be considered normal by traditional mean value judgment. However, after wideband sampling, it was found that the low-frequency current harmonic components of channel 15 were 0.03, 0.07, 0.11, 0.15 and 0.18 in five windows, while channel 16 showed a synchronous increase. After state reconstruction, the low-frequency characteristics of these two channels were significantly prominent in the multidimensional electrical state sequence, and the subsequent module identified that a resonant coupling chain was forming between them.

[0059] The purpose of this step is to expand the original electrical parameters from a static perspective of whether they meet the standards to a dynamic perspective of the coexistence of steady-state components and low-frequency coupling components, thereby enabling more sensitive extraction of early crosstalk symptoms.

[0060] Furthermore, the state assessment and quantification module includes: an eigenvalue perturbation analysis unit, used to calculate the eigenvalue drift of the inter-channel mutual impedance matrix;

[0061] The entropy calculation unit is used to normalize the eigenvalue drift to obtain the drift probability distribution, and calculate the information entropy of the drift probability distribution as the electrical crosstalk entropy, where the electrical crosstalk entropy characterizes the degree of electromagnetic noise coupling in the parallel test circuit.

[0062] This embodiment provides a mechanism for characterizing the electromagnetic coupling degree of a parallel test board through eigenvalue perturbation and entropy quantization. Specifically, although the previous-level scheme has formed a multi-dimensional electrical state sequence that is more suitable for analysis, if only the fluctuation amplitude of a single channel is used to judge the risk, it is easy to make it difficult to accurately distinguish between individual anomalies and system coupling diffusion. For example, the quality fluctuation of an adapter itself may cause its ripple to increase, but it does not necessarily mean that the entire array is approaching the cascading instability boundary. Therefore, this embodiment further starts from the overall structural change of the mutual impedance matrix and measures the change of the channel coupling pattern through the eigenvalue drift.

[0063] Specifically, the eigenvalue perturbation analysis unit receives the mutual impedance matrix at two or more consecutive time points and calculates the changes in the matrix eigenvalues. For illustration, assume that at time T1, the main eigenvalues ​​of the matrix corresponding to a certain 4-channel region are 1.02, 0.31, 0.18, and 0.07; and at time T2, they become 1.08, 0.42, 0.21, and 0.09. Then the drift amounts can be obtained as 0.06, 0.11, 0.03, and 0.02, respectively. Here, the drift amount of the second eigenvalue is greater than the preset drift threshold, which usually means that a stronger group coupling is formed between some channels, rather than a single-point discrete fluctuation.

[0064] The entropy calculation unit normalizes these drift values. Specifically, the normalization rule is as follows: the absolute value of the drift of a single feature value is divided by the sum of the absolute values ​​of the drifts of all feature values ​​at the current time, thereby obtaining the probability distribution of the drift value representing the energy proportion of each coupling mode. For example, if the sum of the above four drift values ​​is 0.22, then the probability distributions are approximately 0.27, 0.50, 0.14, and 0.09. The information entropy is then calculated based on this distribution.

[0065] The specific structured logic for calculating information entropy is as follows: Calculate the natural logarithm of each obtained drift probability distribution value; multiply the calculated natural logarithm by the corresponding probability distribution value and take the negative number; sum the products of all feature patterns to output the final electrical crosstalk entropy; its mathematical definition and calculation basis are as follows: Let the... The feature value drift corresponding to each feature pattern is: Then the probability distribution of the drift amount Electrical crosstalk entropy The calculation formula is: The fact that the proportion of the second component has increased significantly indicates that energy is more concentrated in a certain coupling mode.

[0066] When multiple components rise simultaneously and their distribution tends to spread, it reflects that the coupling disturbance is spreading among more modes, and the quantified electrical crosstalk entropy will also increase accordingly; in other words, the electrical crosstalk entropy is not determined by whether a certain characteristic value is large, but by comprehensively assessing the degree of disorder and diffusion of coupling changes in the structure.

[0067] The reason for adopting this mechanism is that if the threshold judgment is made by simply using the changes in the original matrix elements of the previous level, it is easily affected by the local coupling of individual channels to neighboring channels, and it is difficult to identify whether the coupling mode inside the entire array has changed hierarchically. By introducing eigenvalue perturbation analysis, a large number of local changes can be compressed into a few representative modes, and then uniformly quantified by entropy calculation, which is more suitable for scheduling decisions.

[0068] As a fault-tolerance mechanism, if the calculation result of the mutual impedance matrix becomes ill-conditioned at a certain moment, for example, if individual elements are abnormally increased due to sensor jitter, causing the matrix value to be unstable, the system can first smooth or prune the matrix and then calculate the eigenvalues; if all drift values ​​are close to zero, it indicates that the current coupling state is basically stable, and the electrical crosstalk entropy can be directly assigned a low value.

[0069] If the total drift during normalization is zero, there is no need to continue entropy calculation. Instead, the low-coupling state is directly output to avoid division by zero. If a prominent single transition occurs in the eigenvalue but only lasts for one sampling window, it is not immediately raised to a high-risk level. Instead, it waits for the next window to be reviewed to reduce the instantaneous false alarm rate.

[0070] Eight adjacent channels were selected from the 48-channel aging board to form a monitoring sub-region. During the first 10 minutes of normal operation, the fluctuation range of the main eigenvalues ​​of the mutual impedance matrix in this sub-region was within the preset safety range, and the corresponding electrical crosstalk entropy remained between 0.20 and 0.28. After the heavy equipment was restarted and stopped again, the fourth and fifth eigenvalues ​​showed an upward trend for three consecutive windows, and the drift distribution gradually changed from a concentrated type to a diffuse type, with the entropy value rising to 0.63. At this time, although the DC output of most channels was still within tolerance, the system had already identified the diffusion of the coupling mode, providing a basis for subsequent scheduling in advance.

[0071] The purpose of this mechanism is to compress complex, multi-point, and time-varying coupled noise changes into a comparable, thresholdable, and continuously traceable electrical crosstalk entropy index, thereby achieving feedforward quantification of the instability trend of the test array.

[0072] Furthermore, the state assessment and quantification module also includes a risk mapping unit, which is used to input electrical crosstalk entropy into a preset risk assessment model to output a system failure risk value. The preset risk assessment model is a trained deep neural network, which is configured to map the input electrical crosstalk entropy into a failure probability value normalized to a preset numerical range, and output the failure probability value as the system failure risk value.

[0073] This embodiment provides a mechanism to further map electrical crosstalk entropy to a system failure risk value. Specifically, the aforementioned entropy value can indicate the degree of coupling disorder of the current array, but in industrial settings, what the scheduling system really needs is a risk measure of whether continuing to maintain the current concurrency will cause continuous misjudgments in a short period of time. In other words, it is not enough to just know whether the entropy value is high or low; it is also necessary to know the failure probability corresponding to the entropy value under the current equipment type, load conditions, and environmental harmonic background. Based on this, this embodiment introduces a risk mapping unit.

[0074] Specifically, the risk mapping unit can use a pre-trained deep neural network as a risk assessment model; the training samples come from historical test logs, including multiple sets of electrical crosstalk entropy—corresponding to whether subsequent misjudgment clusters or protection malfunctions occur;

[0075] Specifically, the training supervision logic of the preset risk assessment model is as follows: Electrical crosstalk entropy data collected in historical tests is used as input features; the ratio of the actual number of consecutive failures within a preset time window after the occurrence of this entropy value in the historical data to the total number of concurrent tests is normalized to a preset numerical range and used as a label representing the failure probability value for supervised learning; the specific failure mode corresponds to the increase in test result misjudgment rate caused by excessive inter-channel crosstalk; the quantitative relationship supported by statistical data is: failure risk prediction label. ,in This refers to the number of consecutive misjudgments in historical statistics within a preset time window. To establish a statistical quantitative benchmark that maps crosstalk entropy to the true false positive rate, the model can nonlinearly map any current electrical crosstalk entropy to the failure probability range of 0 to 1 and output the failure probability value as the system failure risk value.

[0076] For ease of explanation, the following training sample examples can be simplified: When the entropy value is 0.18, no anomalies occur in the subsequent 10 test cycles; when the entropy value is 0.42, there is a small probability of individual channels being falsely triggered; when the entropy value is 0.68, the number of samples with consecutive misjudgments increases significantly in the subsequent 3 cycles; when the entropy value is 0.81, most samples will trigger array degradation; for example, the output is 0.27 when the input is 0.42, 0.73 when the input is 0.68, and 0.91 when the input is 0.81.

[0077] This mapping is not a simple linear relationship. The reason is that the same entropy value may correspond to different levels of danger under different workstations, different power range adapters, and different environmental harmonic conditions. For example, under low concurrency and low load modes, an entropy value of 0.55 fails to trigger the preset system failure judgment threshold. However, when the concurrency rate is above 98% and most channels are simultaneously experiencing large current transients, an entropy value of 0.55 may be close to the instability boundary. The risk assessment model can incorporate different operating condition labels during training to make the output closer to the actual risks on site.

[0078] In terms of specific network structure configuration, deep neural networks can adopt a multi-layer perceptron architecture that includes an input layer, at least one fully connected hidden layer, and an output layer. Since the implementation example limits the mapping of the input electrical crosstalk entropy to a failure probability value, when only a single feature is input, the network can combine operating conditions by calling a specific set of network weight parameters pre-trained and saved for different operating condition labels.

[0079] The hidden layer uses a nonlinear activation function to amplify the features of a specific entropy value mutation range; the output layer is processed by a sigmoid activation function to ensure that the output system failure risk value is strictly limited to between 0 and 1; this nonlinear network mapping, compared with a simple linear scaling transformation, can better fit the nonlinear mutation physical mechanism of instability caused by electrical coupling from gradual accumulation to threshold mutation.

[0080] The reason for introducing this feature on top of the previous entropy quantization is that simply setting a fixed threshold for the entropy value often only allows for coarse-grained switching and makes it difficult to distinguish between three different intensities of handling: those that can be observed, those that require local degradation, and those that require immediate global degradation of concurrency. This embodiment uses risk mapping to convert the abstract entropy value into a unified risk probability, thereby making subsequent scheduling more hierarchical.

[0081] In the anomaly handling mechanism, if the training model's output becomes unstable due to input exceeding the historical range, the input entropy value can be pruned to the training coverage range. For example, if the training samples mainly cover 0 to 0.9, when an entropy value of 0.95 is detected, a conservative mapping based on 0.9 can be performed, and this can be recorded as an out-of-bounds event in the log. If the model output is abnormally missing or the calculation times out, the system adopts a degradation scheme, directly performing conservative scheduling based on the entropy value and static threshold to avoid interrupting the field protection due to the model's temporary unavailability. If the historical samples do not cover a certain new adapter model adequately, the conservative bias of the risk value can be temporarily increased, for example, by adding a safety margin of 0.05 to the model output.

[0082] During night shift mass production, a batch of 100W fast charging adapters experienced higher electrical crosstalk entropy on the aging boards compared to the day shift batch due to parasitic parameters exceeding the preset distribution tolerance. At one point, the entropy value of the second region was 0.61. While this might still be within an acceptable range in absolute terms, the risk mapping model, combined with the high-power, high-concurrency, and low-frequency harmonic background enhancement labels of this batch, output a system failure risk value of 0.78, indicating that this value was approaching the boundary of continuous misjudgment. Based on this, the scheduling module intervened in advance instead of waiting until the electrical parameters officially exceeded the tolerance before taking action.

[0083] The purpose of this mechanism is to transform the degree of physical coupling into the probability of failure risk for production line decision-making, thereby achieving a functional leap from detecting anomalies to predicting the consequences of misjudgments.

[0084] Furthermore, the intelligent scheduling decision module includes a strategy generation unit, which is used to: generate a first scheduling strategy as the target scheduling strategy when the electrical crosstalk entropy is lower than a preset safety threshold, wherein the first scheduling strategy is a synchronous test instruction that maintains the current maximum concurrency rate; and generate a second scheduling strategy as the target scheduling strategy when the electrical crosstalk entropy is equal to or higher than the preset safety threshold, wherein the second scheduling strategy is an asynchronous degradation test instruction that reduces the system failure risk value.

[0085] This embodiment provides a dual-mode scheduling strategy generation mechanism for high system reliability. Specifically, in high-density production lines, the goal of the testing system is not to always maintain the highest throughput, but to get as close to the maximum throughput as possible while ensuring safety. When the electrical crosstalk entropy is low, it is reasonable to continue to pursue full concurrency. However, when the crosstalk entropy is close to the danger boundary, if all channels are still tested simultaneously at full load, it may cause the entire batch to be reworked due to misjudgment and spread, ultimately reducing the overall testing throughput efficiency. Therefore, this embodiment sets two types of scheduling strategies and uses electrical crosstalk entropy as the switching basis.

[0086] Specifically, the strategy generation unit first reads the current electrical crosstalk entropy and compares it with a preset safety threshold. Assuming the safety threshold is set to 0.55, when the detection value is 0.34, it indicates that the system coupling is under control, and the first scheduling strategy is generated, which is to maintain the current maximum concurrency synchronous test command. At this time, each channel can maintain a unified trigger phase and execute no-load, half-load, full-load and dynamic load transition synchronously according to the established test process. Conversely, when the detection value reaches 0.55 or higher, such as 0.62, the strategy generation unit no longer maintains synchronous full concurrency, but switches to generating the second scheduling strategy, which is to generate asynchronous degradation test command. Degradation is not a simple shutdown, but a protective adjustment of the test timing and load action with the goal of reducing the risk of system failure.

[0087] A simplified example illustrates this process: In the 100th test cycle, the entropy value is 0.49, and the system maintains synchronous mode; in the 101st cycle, the entropy value rises to 0.57, and the system immediately switches to asynchronous mode; in the 102nd cycle, although the entropy value drops to 0.53, if the risk value remains high, the asynchronous mode can continue to be maintained for several more cycles to avoid frequent switching back and forth; thus, the strategy generation unit can make its initial judgment based on the entropy value, or it can combine the risk value to control whether to maintain or exit the system.

[0088] The reason for setting up the above dual modes is that a single fixed strategy cannot take into account the testing efficiency and stability at different stages. If a conservative asynchronous strategy is always used, although the risk of system failure can be reduced, it will lead to a decrease in the duty cycle of parallel testing. If synchronous full concurrency is always used, mass misjudgment is very likely to occur under conditions of concentrated harmonic interference or heat island accumulation. This embodiment introduces a segmented strategy generation mechanism so that the system can make full use of concurrency capabilities when the risk is low and actively adjust to a more stable operating point when the risk is high.

[0089] In the anomaly handling mechanism, if the entropy value fluctuates around the threshold, for example, oscillating between 0.54 and 0.56 for five consecutive cycles, a hysteresis interval can be introduced; for example, asynchronous mode is only switched when the entropy value is higher than 0.57, and synchronous mode is only restored when it is lower than 0.52, in order to avoid strategy jitter; if the entropy value is low but the risk value is abnormally high, the risk value is prioritized for conservative handling to prevent the risk from being underestimated by a single entropy value under certain special operating conditions; if the transmission delay causes the latest entropy value not to be received in time in this cycle, the strategy of the previous cycle is temporarily used and the sampling interval of the next cycle is shortened.

[0090] After the 48-channel aging board ran continuously for 2 hours, the system maintained synchronous testing in the early stages, with an average concurrency rate of 98.6%. When the large air compressor in the night shift factory started, low-frequency disturbances from the power grid side were transmitted to the test board, and the electrical crosstalk entropy rapidly increased from 0.41 to 0.59. The strategy generation unit identified that it had exceeded the safety threshold and immediately stopped the next round of full-channel synchronous full-load boost, instead issuing an asynchronous degradation test command to the dynamic decoupling execution module. Within 5 minutes, although the test concurrency rate in this area decreased, the system failure risk value dropped from 0.76 to 0.46, avoiding continuous misjudgments.

[0091] The purpose of this mechanism is to separate the optimal concurrency rate from the optimal system security, enabling the system to proactively adjust the local testing pace and ensure global stability when the risk of instability approaches.

[0092] Furthermore, the asynchronous degradation test instructions include at least one of the following: pseudo-random timing misalignment instructions, channel intermittent heat dissipation instructions, and load dynamic rate limiting instructions.

[0093] This embodiment provides a refined execution mechanism for asynchronous degradation test instructions. Specifically, the previous solution has distinguished between synchronous and asynchronous modes. However, if the asynchronous mode only generally indicates reducing concurrency or slowing down the test, it is still too coarse in engineering and cannot accurately suppress different sources of coupling. For example, some anomalies mainly come from electromagnetic superposition caused by simultaneous triggering, some from continuous high heat accumulation, and some from reverse disturbances caused by excessively rapid load transitions. Based on this, this embodiment subdivides the asynchronous degradation test instructions into several combinable action types.

[0094] Specifically, the first type is pseudo-random timing misalignment instructions; this means that multiple channels are no longer allowed to perform high current load or ripple tests at completely identical times, but rather controlled misalignment is introduced within an allowable test window; for example, if all 16 channels were originally increasing from 1 amp to 5 amp starting at 0 millisecond, it can now be changed to channel 1 at 0 millisecond, channel 2 at 0.7 millisecond, channel 3 at 1.1 millisecond, channel 4 at 0.3 millisecond, and so on.

[0095] Here, pseudo-randomness is not disordered, but rather a reproducible misalignment table is generated based on a preset seed, causing abrupt changes in adjacent channels at different times to reduce the superposition peak. The preset seed is generated by a hash algorithm based on the channel number with the highest electrical crosstalk entropy in the current test array and the current timestamp of its corresponding multidimensional electrical state sequence, to ensure that the generated misalignment table can specifically break up the temporal correlation of the strongest coupled channel at the current moment.

[0096] The second type is the channel intermittent heat dissipation command; it is applicable to areas with obvious heat island effect; specifically, if the temperature change rate of four adjacent channels exceeds the preset temperature threshold in the last 20 cycles, the preset temperature threshold is extracted based on the upper limit of the heat capacity of the test channel hardware and the safe operation tolerance margin calibration, then two of the channels can be commanded to insert a short no-load or light-load interval after completing a full-load window; for example, every 30 seconds of full-load test, a 3-second light-load recovery is inserted; this can break the local thermal coupling chain and prevent the temperature rise from pushing up the electrical parameter drift in the opposite direction;

[0097] The third type is the load dynamic rate limiting instruction; its function is to control the impedance change rate or current rise slope of the test load; for example, it was originally allowed to jump from 0 amps to 5 amps every millisecond, but now it is limited to an increase of 1 amp per millisecond, and it takes 5 milliseconds to complete the same jump; this will sacrifice some transient test speed, but can significantly reduce the disturbance injection caused to adjacent channels when the jump exceeds the preset slope at the same time.

[0098] The reason for introducing these refined features is that if asynchronous degradation simply reduces the number of channels, it will significantly extend the overall testing cycle and make it difficult to specifically eliminate the root causes of instability. This embodiment breaks down the degradation action into three dimensions: timing, thermal, and load, so that the system can be used in combination according to the actual risk causes. For example, if the main problem is resonance superposition, pseudo-random timing misalignment is preferred; if the main problem is heat island accumulation, heat dissipation intervals are preferred; if the main problem is excessively fast load jump, the load dynamic rate is reduced preferred.

[0099] As a fault-tolerance mechanism, if the risk value does not decrease within 2 to 3 cycles after an action is executed, a second or third type of action can be superimposed; if three types of actions have been superimposed simultaneously but the risk continues to rise, a higher level of protection should be triggered, such as suspending high-power testing in a local area; if pseudo-random timing misalignment may affect certain items that must be sampled synchronously, a limited synchronization window should be reserved for such items, and misalignment should only be introduced during the pre-load stage; if a channel is close to the end of the delivery cycle and is not suitable for inserting a long heat dissipation interval, a shorter load slope limit can be used instead.

[0100] After a low-frequency resonance band was formed in channels 12 to 18, the system issued pseudo-random timing misalignment commands to these channels, causing their high-current load actions to be staggered within a 2-millisecond test window. After observing two cycles, although the electrical crosstalk entropy decreased, it was still higher than the threshold. Therefore, a load dynamic rate limit was added, changing the original 5-amp step to five 1-amp steps. Since the temperature rise of channels 15 and 16 was still high, the system added intermittent heat dissipation commands to these two channels. After the combination of these three actions, the risk value of the relevant area decreased significantly.

[0101] The purpose of this mechanism is to transform asynchronous degradation from a vague concept into decomposable, composable, and cause-oriented execution instructions, thereby achieving more effective decoupling at a lower cost.

[0102] Furthermore, the dynamic decoupling execution module includes: a timing control unit, used to adjust the trigger phase of each test channel according to the target scheduling strategy to achieve electrical isolation reconstruction between channels; and a dynamic electronic load unit, used to adjust the impedance change rate of the test load according to the target scheduling strategy.

[0103] This embodiment provides a dynamic decoupled execution mechanism that implements scheduling strategies at the test hardware layer. Specifically, the previous-level scheme has already specified the specific type of asynchronous degradation instructions, but without a corresponding hardware execution link, the scheduling result remains at the logic layer and cannot actually change the coupling state of the parallel test loop. Therefore, this embodiment uses a timing control unit and a dynamic electronic load unit to directly apply the target scheduling strategy to each test channel.

[0104] Specifically, the timing control unit is used to adjust the trigger phase of each channel; it can be understood as being responsible for deciding which channel starts executing a certain test action at what time; for example, originally channels 1 to 4 would trigger full-load tests simultaneously under a unified clock, but now they can be set to phases of 0 degrees, 40 degrees, 85 degrees, and 120 degrees respectively according to the target scheduling strategy, or more intuitively set to 0 milliseconds, 0.5 milliseconds, 1.2 milliseconds, and 1.8 milliseconds; through this phase staggering, the current pulses that were originally superimposed at the same instant are dispersed in time, thereby restoring the electrical isolation effect between channels;

[0105] The dynamic electronic load unit is responsible for changing the impedance change rate of the test load. To illustrate with a simplified example: if the original setting is to switch from a 10-ohm load to a 4-ohm load within 1 millisecond, the dynamic electronic load unit can decompose this process into three stages according to the scheduling command: 10 ohms to 8 ohms, 8 ohms to 6 ohms, and 6 ohms to 4 ohms, each stage lasting 1 millisecond, thus changing the rapid transition to a gradual one. If the target strategy requires reducing load surges, the maximum rate of change can also be limited, for example, the impedance change cannot exceed 2 ohms per millisecond. This can significantly reduce current spikes in the loop and induced disturbances in adjacent channels.

[0106] The necessity of setting up this execution layer is that if there are only software decisions without the two key execution methods of timing and load, the aforementioned assessment of crosstalk entropy and risk value will not be able to close the loop; especially in the parallel testing scenario of high power adapters, many couplings are not determined by the absolute current value, but are caused by simultaneous triggering and rapid changes. Therefore, the loop must be reconstructed from the phase dimension and the rate of change dimension respectively.

[0107] As a fault-tolerant mechanism, if certain test items require standardized transient response according to regulations or industry standards, such as requiring a load step within a fixed time, the system can adjust the phase only in the auxiliary phases before and after the test, and then restore the standard action at the moment of the core measurement. If the dynamic electronic load unit cannot achieve the target rate of change due to hardware limitations, its executable upper limit should be fed back, and the scheduling module should regenerate the secondary strategy. If the trigger phase of a certain channel is delayed and may exceed the test window, the system can shorten its non-critical steps and prioritize ensuring that the critical measurement action still falls within the effective window.

[0108] In a certain risk-enhancing event, the scheduling module decided to reconfigure the timing of channels 10 to 14. The timing control unit changed the full-load start points of these 5 channels from being completely coincident to being staggered by 0.4 milliseconds. At the same time, the dynamic electronic load unit extended the original 1-millisecond 5-amp load to be completed in three segments within 3 milliseconds. After execution, the acquisition module observed in the next test window that the low-frequency coupling components of these channels were synchronously weakened, and the key off-diagonal elements in the mutual impedance matrix decreased.

[0109] The purpose of this mechanism is to transform risk-reducing scheduling strategies into actual physical interventions on test loop timing and load boundaries, thereby achieving a closed-loop implementation from assessment to decoupling.

[0110] Furthermore, the system also includes an adaptive compensation module, which receives a multi-dimensional electrical state sequence and a target scheduling strategy, generates a dynamic impedance compensation signal based on the target scheduling strategy, and injects the dynamic impedance compensation signal into a parallel test circuit to suppress low-frequency resonance.

[0111] This embodiment provides an adaptive compensation mechanism for suppressing low-frequency resonance. Specifically, the aforementioned timing reconstruction and load regulation can reduce coupling risks at the global electrical level. However, in some extreme scenarios, simply adjusting the action sequence and rate of change is insufficient to suppress the low-frequency resonance that has already formed. For example, when external grid harmonics and parasitic parameters of adjacent adapters work together, some channels will experience low-frequency oscillations that last for several cycles. If the circuit is not compensated at the electrical level, the resonance may be repeatedly triggered. Therefore, this embodiment further adds an adaptive compensation module to the system.

[0112] Specifically, the adaptive compensation module receives a multi-dimensional electrical state sequence and a target scheduling strategy to comprehensively determine what kind of dynamic impedance compensation signal needs to be injected. Specifically, the compensation signal can be understood as an additional correction quantity that changes over time, used to fine-tune the equivalent impedance of the test circuit. For example, if the system determines that a certain adjacent channel group has a continuous resonance around 80 Hz, the compensation module can generate an impedance adjustment signal that is opposite to the resonance trend, so that the local circuit is no longer easily amplified at that frequency.

[0113] For ease of illustration, assuming the resonance intensity of a certain channel group is 0.18, 0.24, and 0.29 within three windows, the compensation module can inject three levels of enhanced compensation, such as normalized impedance correction amplitudes of 0.05, 0.09, and 0.12. In terms of hardware execution mechanism, the dynamic impedance compensation signal is injected into the parallel test circuit, which can be achieved by converting the compensation signal into a micro AC bias command for the electronic test load of that channel.

[0114] That is, on the basis of the original DC load current command, an alternating small current command with the opposite phase and the same frequency as the detected low-frequency resonance is superimposed; through the above processing, while consuming DC power, the test load will exhibit equivalent dynamic damping impedance characteristics at the resonance frequency point, directly absorbing and neutralizing the electrical disturbance energy at that frequency point.

[0115] The reason this module is introduced after the aforementioned dynamic decoupling mechanism is that if timing and load control are not performed first and impedance compensation is performed directly, the compensation signal itself may be re-amplified by the synchronous action of multiple channels, resulting in unstable effect. In other words, the timing and load control in the preceding steps first reduces the global coupling of the system, and this module then specifically suppresses the residual oscillations at local frequency points. The combination of the two makes it easier to achieve stable operation of the test system.

[0116] In addition, the adaptive compensation module can select the compensation intensity in combination with the target scheduling strategy; for example, when the system is still in synchronous mode but crosstalk is slightly increased, only weak compensation is applied; when the system has entered asynchronous degrade mode and there are obvious low-frequency fluctuations in local areas, a stronger but controlled amount of compensation is applied; this can avoid new disturbances caused by overcompensation.

[0117] In the anomaly handling mechanism, if the crosstalk entropy of the relevant channel increases instead of decreasing within two consecutive cycles after the compensation signal is injected, it should be determined that the current compensation direction may be mismatched, and the system can automatically reduce the compensation amplitude or switch the compensation phase; if the source of resonance in a certain area is unclear, it is preferred to use a small step size to try successively rather than injecting an excessive compensation signal at once; if some channels are in the measurement window limited by regulations, and the compensation action may affect the metrological impartiality, compensation can be injected in the non-measurement window, and only timing and load control are retained in the measurement window;

[0118] In the aging board, in channels 20 to 24, although pseudo-random misalignment and load slope limiting have been implemented, low-frequency resonance near 80 Hz still persists. After reading the reconstructed state sequence of these channels, the adaptive compensation module identifies that the oscillation component at this frequency point is continuously increasing. Combined with the current asynchronous degradation strategy, it injects a set of gradually increasing dynamic impedance compensation signals into the parallel test circuit. After 4 cycles, the amplitude of low-frequency harmonics in this region decreases, and the risk value falls back to near the safe zone.

[0119] The purpose of this mechanism is to provide an active vibration suppression channel that acts directly on the equivalent impedance of the circuit, in addition to timing and load reconstruction, thereby achieving fine-grained suppression of low-frequency resonance.

[0120] Furthermore, the multi-channel signal acquisition module also includes: an anomaly marking unit, used to identify anomaly channels with a signal-to-noise ratio lower than a preset signal-to-noise ratio threshold and containing oscillation characteristics of a target resonance preset frequency band calibrated by frequency sweep during the extraction of multi-dimensional electrical state sequences, and to mark the resonance time points of the anomaly channels, as well as to cache the data of the anomaly channels.

[0121] This embodiment provides a marking and caching mechanism for tracking abnormal data. Specifically, the aforementioned scheme can quantify and schedule the global coupling state, but in field operation, a key problem still needs to be solved: when the signal quality of some channels suddenly deteriorates, if these abnormal windows are not clearly marked, subsequent modules may mix them into normal samples, which will affect matrix evaluation and make it difficult to trace the root cause afterward. Therefore, this embodiment adds an abnormal marking unit to identify suspicious channels and their resonance time points in advance during the state extraction process.

[0122] Specifically, the anomaly marking unit checks the signal-to-noise ratio (SNR) of each channel signal. To ensure computational efficiency and the coherence of contextual data, the SNR can be obtained by directly reusing the frequency domain decomposition results of the aforementioned harmonic separation unit. The sum of the energy of the extracted fundamental components is taken as the effective signal power, and the sum of the energy of the remaining broadband scattered background components after removing the fundamental and low-frequency harmonics is taken as the noise power. The ratio of the two is calculated to obtain the current SNR.

[0123] For ease of explanation, the preset signal-to-noise ratio (SNR) threshold is calculated based on the floor noise power distribution of the test environment under no-load conditions, and can be set to 12 dB. If the SNR of a certain channel in the current window is only 8 dB, and a significant oscillation peak is detected in the preset frequency band, such as between 50 Hz and 200 Hz, then the channel can be determined to be an abnormal channel. At this time, the system not only adds an abnormal mark to the channel, but also records its resonance time point, such as the interval between 2.4 ms and 3.1 ms in the 235th sampling window. At the same time, the original waveform segment, the decomposed low-frequency harmonic data, and the reconstructed state vector corresponding to the channel are cached together.

[0124] The reason for adopting this mechanism is that if the previous level scheme only outputs the final entropy value and risk value, although it is sufficient to drive scheduling, it is not conducive to tracking which channels and when the initial resonance characteristics appeared. It is very important to retain these abnormal fragments when tracing production line quality and retraining the model. On the one hand, it can prevent abnormal data from being included in global statistics indiscriminately. On the other hand, it can provide real samples for subsequent risk model and compensation strategy optimization.

[0125] Anomaly marking can also be linked with scheduling; for example, if a channel does not cause the global entropy value to exceed the threshold immediately, but it has been marked as having low signal-to-noise ratio and oscillation in a preset frequency band multiple times, the scheduling module can reduce its test dynamic rate in advance, or remove it from the critical synchronization group in the next round.

[0126] As a fault-tolerance mechanism, if the signal-to-noise ratio of a channel is lower than the threshold, but no oscillation in the preset frequency band is detected, it may just be a loose probe or an occasional sampling anomaly. In this case, it is only marked as general noise and is not immediately identified as a resonant channel. If a channel has oscillation in the preset frequency band, but the signal-to-noise ratio is still higher than the threshold, it indicates that the oscillation may be part of a real measurable signal. In this case, it can be entered into the observation queue, and it will be upgraded to an anomaly mark only after it repeats for more than two consecutive windows. If the buffer space is close to the limit, priority is given to retaining abnormal data with longer duration, more concentrated frequency band, and subsequent entropy increase, while short-lived and weakly correlated anomalies are saved in summary mode.

[0127] During a night shift test, the average output voltage of channel 17 remained close to the nominal value, but the anomaly marker unit found that its signal-to-noise ratio dropped from 14 dB to 9 dB, and a significant oscillation peak appeared near 120 Hz. The system therefore recorded the resonance time point of channel 17 in the window from 410 to 413 and cached the relevant data. Subsequent review revealed that channel 17 was the starting point that triggered the increased coupling of adjacent channels 18 and 19. This cached data was used to optimize the risk mapping model and compensation strategy parameters.

[0128] The purpose of this mechanism is to further refine the global risk assessment to the local anomaly location and evidence retention, thereby achieving a balance between runtime protection and post-event analysis.

[0129] Furthermore, the state assessment and quantification module and the intelligent scheduling decision module include a first synchronous channel and a second asynchronous channel. The first synchronous channel is hardwired and used to transmit electrical crosstalk entropy and system failure risk values. The second asynchronous channel uses a bus communication protocol to transmit historical test logs and model update parameters.

[0130] This embodiment provides a dual-channel communication mechanism. Specifically, in the aforementioned high-risk parallel testing scenario, the scheduling link needs to simultaneously meet the dual requirements of high-precision calculation and low-latency transmission. If real-time protection indicators, historical logs, and model parameter updates are all placed on the same communication channel, bus congestion or update tasks occupying bandwidth may cause the most critical entropy and risk values ​​to be delayed in reaching the scheduling side, thus missing the best intervention opportunity. Therefore, this embodiment sets up a first synchronous channel and a second asynchronous channel between the state assessment and quantification module and the intelligent scheduling decision module.

[0131] Specifically, the first synchronization channel uses a hard-wired connection to carry strong real-time data such as electrical crosstalk entropy and system failure risk values. The hard-wired connection here can be understood as a fixed, low-latency, and highest-priority direct transmission path. For ease of explanation, the system can be set to require that the single transmission delay of this channel is no more than 1 millisecond, so that when the risk value of a certain window increases sharply, the dispatching side can receive and trigger the protection action almost immediately.

[0132] The second asynchronous channel uses a bus communication protocol to transmit historical test logs, anomaly cache data, and model update parameters that do not require immediate closed-loop control. For example, waveform segments cached by the anomaly marking unit, recent sample statistical results, and compensation strategy optimization parameters can all be uploaded in batches through this asynchronous channel. This does not affect real-time protection data and facilitates gradual optimization of the system model in the background.

[0133] The reason for introducing a dual-channel structure is that a complex closed loop has been formed between the modules at the upper level. If all information shares a single communication path, when night shift batch log uploads, model parameter updates, or fault backtracking occur simultaneously, it is very easy to crowd out the bandwidth of real-time risk data, causing the degradation strategy that should be executed immediately to be delayed, thereby weakening the protection effect of the entire solution. This embodiment ensures the priority operation of the risk closed loop by separating hard real-time and non-real-time traffic.

[0134] As a fault-tolerance mechanism, if the second asynchronous channel experiences a short-term blockage, it will only affect the transmission of logs and model parameters and should not block the real-time risk reporting of the first synchronous channel. If the first synchronous channel detects a link anomaly, the system will immediately enter a conservative mode, such as maintaining the current asynchronous degradation strategy or directly reducing the concurrency rate until the synchronous link is restored. If the timestamps of the same batch of information transmitted from the two channels are inconsistent, the risk data of the first synchronous channel shall prevail, and the data of the second asynchronous channel shall only be used for post-analysis and model correction and shall not overwrite the current protection decision.

[0135] During a sustained low-frequency harmonic disturbance, the background system simultaneously initiated a new round of model parameter updates and uploaded the abnormal waveform logs from the previous 30 minutes. Since these data are transmitted via the second asynchronous channel, even if bus occupancy increases, it will not affect the first synchronous channel's continuous transmission of the critical signal with the current electrical crosstalk entropy of 0.64 and risk value of 0.81 to the scheduling side. Based on this, the scheduling module can still trigger timing misalignment and load slope limitation within milliseconds, while the model update is completed in the background without interfering with real-time protection.

[0136] The purpose of this mechanism is to decouple the real-time protection link from the non-real-time optimization link, ensuring that the system prioritizes maintaining the response speed of the risk closed loop under extreme operating conditions, while retaining the ability to continuously learn and trace operational maintenance.

[0137] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A multi-channel parallel testing system for electrical parameters of a power adapter, characterized in that, The system includes: A parallel test circuit, comprising multiple test channels, which are respectively connected to multiple power adapters and adjustable test loads; A multi-channel signal acquisition module is connected to the parallel test circuit and is used to acquire the real-time electrical parameter signals of the corresponding power adapter in each test channel through the parallel test circuit, and to extract features from the real-time electrical parameter signals to obtain a multi-dimensional electrical state sequence. The state assessment and quantization module is connected to the multi-channel signal acquisition module. It is used to receive the multi-dimensional electrical state sequence, construct an inter-channel mutual impedance matrix representing the equivalent impedance and coupling strength based on the multi-dimensional electrical state sequence, perform nonlinear coupling analysis on the inter-channel mutual impedance matrix based on the matrix perturbation model to generate electrical crosstalk entropy, and predict the system failure risk value based on the electrical crosstalk entropy. The intelligent scheduling decision module is communicatively connected to the state assessment and quantification module. It is used to receive the electrical crosstalk entropy and the system failure risk value, and compare the electrical crosstalk entropy with a preset safety threshold based on background crosstalk calibration to determine the target scheduling strategy. The dynamic decoupling execution module is used to perform timing reconstruction and load adjustment on the parallel test loop and the adjustable test load based on the target scheduling strategy, and to feed back the adjusted loop status to the multi-channel signal acquisition module. The state assessment and quantification module includes: The eigenvalue perturbation analysis unit receives the mutual impedance matrix at two or more consecutive moments and calculates the difference between the eigenvalues ​​of the mutual impedance matrix at two consecutive moments as the eigenvalue drift of the inter-channel mutual impedance matrix. The entropy calculation unit is used to normalize the drift of the feature value to obtain the probability distribution of the drift, and to calculate the information entropy of the probability distribution of the drift as the electrical crosstalk entropy, wherein the electrical crosstalk entropy characterizes the degree of electromagnetic noise coupling in the parallel test circuit. The state assessment and quantification module also includes: A risk mapping unit is used to input the electrical crosstalk entropy into a preset risk assessment model to output the system failure risk value. The preset risk assessment model is a trained deep neural network, which is configured to map the input electrical crosstalk entropy into a failure probability value normalized to a preset numerical range, and output the failure probability value as the system failure risk value. The dynamic decoupling execution module includes: A timing control unit is used to adjust the trigger phase of each test channel according to the target scheduling strategy in order to achieve the reconstruction of electrical isolation between channels; A dynamic electronic load unit is used to adjust the impedance change rate of the test load according to the target scheduling strategy.

2. The multi-channel parallel testing system for electrical parameters of a power adapter according to claim 1, characterized in that, The multi-channel signal acquisition module includes: A wideband sampling unit is used to acquire real-time electrical parameter signals of each test channel, including voltage signals, current signals and ripple signals; The harmonic separation unit is used to perform frequency domain decomposition on the voltage signal, the current signal and the ripple signal to separate the fundamental component and the harmonic components with frequencies lower than a preset frequency threshold. The state reconstruction unit is used to fuse the fundamental component and the harmonic components with frequencies lower than a preset frequency threshold to generate the multidimensional electrical state sequence.

3. The multi-channel parallel testing system for electrical parameters of a power adapter according to claim 1, characterized in that, The intelligent scheduling decision module includes a strategy generation unit, which is used for: When the electrical crosstalk entropy is lower than the preset safety threshold, a first scheduling strategy is generated as the target scheduling strategy, wherein the first scheduling strategy is a synchronous test instruction that maintains the current maximum concurrency rate; When the electrical crosstalk entropy is equal to or higher than the preset safety threshold, a second scheduling strategy is generated as the target scheduling strategy, wherein the second scheduling strategy is an asynchronous degradation test instruction to reduce the system failure risk value.

4. A multi-channel parallel testing system for electrical parameters of a power adapter according to claim 3, characterized in that, The asynchronous degradation test command includes at least one of the following: pseudo-random timing misalignment command, channel intermittent heat dissipation command, and load dynamic rate limiting command.

5. A multi-channel parallel testing system for electrical parameters of a power adapter according to claim 1, characterized in that, The system also includes: An adaptive compensation module is used to receive the multidimensional electrical state sequence and the target scheduling strategy, generate a dynamic impedance compensation signal based on the target scheduling strategy, and inject the dynamic impedance compensation signal into the parallel test circuit to suppress low-frequency resonance.

6. A multi-channel parallel testing system for electrical parameters of a power adapter according to claim 1, characterized in that, The multi-channel signal acquisition module also includes: An anomaly marking unit is used to identify, during the extraction of the multidimensional electrical state sequence, anomaly channels with a signal-to-noise ratio lower than a preset signal-to-noise ratio threshold and containing oscillation characteristics of a target resonant preset frequency band calibrated by frequency sweep, and to mark the resonant time point of the anomaly channel and to cache the data of the anomaly channel.

7. A multi-channel parallel testing system for electrical parameters of a power adapter according to claim 1, characterized in that, The state assessment and quantification module and the intelligent scheduling decision module include a first synchronous channel and a second asynchronous channel, wherein the first synchronous channel is hard-wired and used to transmit the electrical crosstalk entropy and the system failure risk value; The second asynchronous channel uses a bus communication protocol to transmit historical test logs and model update parameters.

Images