Diagnostic methods and related devices for high-voltage switch coil current waveforms on mobile test platforms

Abstract

This application belongs to the field of power diagnostic technology. Addressing the difficulty in distinguishing between actual mechanical faults and waveform anomalies caused by measurement noise and loop parameter changes in high-voltage switch coil current detection on mobile test platforms, and the technical problem that traditional diagnostic methods rely on a large number of fault samples, this application provides a method and related device for diagnosing high-voltage switch coil current waveforms on mobile test platforms. It predicts coil current and electromagnetic-mechanical state predictions through a neural network outputting health benchmark physical information, calculates the test data fitting loss and test physical residuals respectively, and outputs diagnostic results based on dual threshold comparison. This method can distinguish between actual equipment faults and non-fault anomalies such as measurement noise and loop parameter changes from two dimensions: waveform deviation and physical consistency, without relying on a large number of fault samples, thus improving the reliability of on-site diagnosis on mobile test platforms.

Description

Technical Field

[0001] This application belongs to the field of power diagnostic technology, specifically relating to a method and related device for diagnosing the current waveform of a high-voltage switch coil on a mobile test platform. Background Technology

[0002] High-voltage switches are control and protection devices in power systems. The current waveform of the opening and closing coils reflects the movement of the iron core, the action of the contacts, and the state of the mechanism, serving as an important basis for diagnosing equipment faults. During on-site testing using mobile electrical testing platforms, factors such as strong electromagnetic interference, unstable wiring and grounding conditions can cause waveform amplitude drift, glitches, or timing misalignments, which can easily be confused with actual equipment fault waveforms.

[0003] Existing technologies mainly include diagnostic methods based on artificial features such as current peak value, pulse width, and slope combined with shallow classifiers, and end-to-end deep learning methods that input the original waveform into models such as CNN (Convolutional Neural Network) and TCN (Temporal Convolutional Network). Among them, the former relies on expert experience and has limited feature dimensions, resulting in weak generalization ability; the latter lacks physical interpretability, relies on a large number of labeled fault samples, and is difficult to effectively distinguish between real faults and non-fault waveform anomalies when fault samples are scarce or when wiring, gain, and grounding conditions change in the mobile field. Summary of the Invention

[0004] This application addresses the technical problem that it is difficult to distinguish between actual equipment faults and waveform anomalies caused by measurement noise and changes in circuit parameters in the detection of high-voltage switch coil current in mobile test platforms, and that traditional diagnostic methods rely on a large number of fault samples. It provides a method and related device for diagnosing the waveform of high-voltage switch coil current in mobile test platforms.

[0005] To achieve the above objectives, this application adopts the following technical solution: Firstly, this application proposes a method for diagnosing the current waveform of a high-voltage switch coil on a mobile test platform, including: Acquire test timing data during the opening and closing process of the high-voltage switch under test, and preprocess the test timing data; wherein, the test timing data includes time information, coil current information and coil voltage information; The preprocessed time information is input into the health benchmark physical information neural network to obtain the corresponding network prediction coil current and electromagnetic-mechanical state prediction quantities. The test data fitting loss is calculated based on the network predicted coil current and the coil current information, and the test physical residual is calculated based on the coil voltage information, the network predicted coil current and the electromagnetic-mechanical state prediction. The test data fitting loss is compared with a preset data fitting threshold, and the test physical residual is compared with a preset physical residual threshold. Based on the comparison results, the state diagnosis result of the high-voltage switch under test is output.

[0006] Furthermore, the step of outputting the state diagnosis result of the high-voltage switch under test based on the comparison result includes: When the test data fitting loss is less than the preset data fitting threshold and the test physical residual is less than the preset physical residual threshold, the state diagnosis result of the high voltage switch under test is healthy. When the test data fitting loss is less than the preset data fitting threshold and the test physical residual is greater than or equal to the preset physical residual threshold, the state diagnosis result of the high voltage switch under test is equipment failure. When the test data fitting loss is greater than or equal to the preset data fitting threshold, and the test physical residual is less than the preset physical residual threshold, the state diagnosis result of the high voltage switch under test is measurement noise or circuit parameter change. When the test data fitting loss is greater than or equal to the preset data fitting threshold, and the test physical residual is greater than or equal to the preset physical residual threshold, the state diagnosis result of the high voltage switch under test is a serious fault or a mixed factor.

[0007] Furthermore, when the state diagnosis result of the high-voltage switch under test is a device fault, the weight parameters of the health benchmark physical information neural network are fixed, and the physical parameters used to calculate the test physical residual are fine-tuned with the goal of minimizing the test physical residual, so as to obtain the optimal estimated value of the physical parameters. The test physical residuals include electromagnetic equation residuals and mechanical motion equation residuals. The electromagnetic equation residuals are calculated based on coil voltage information, network-predicted coil current, electromagnetic-mechanical state predictions, and coil resistance. The mechanical motion equation residuals are calculated based on network-predicted coil current, electromagnetic-mechanical state predictions, damping coefficients, and spring stiffness.

[0008] Furthermore, the electromagnetic-mechanical state prediction quantities include core displacement, core velocity, coil inductance, and inductance rate of change; The test physical residual is calculated based on the coil voltage information, the network-predicted coil current, and the electromagnetic-mechanical state prediction, including: The network predicts the coil current and automatically differentiates it based on the time information to obtain the coil current change rate. The electromagnetic equation residuals are calculated based on the coil voltage information, the network-predicted coil current, the coil current change rate, the coil inductance, the inductance change rate, and the coil resistance. The core displacement is automatically differentiated based on the time information to obtain the core acceleration. The residual of the mechanical motion equation is calculated based on the network predicted coil current, the core displacement, the core velocity, the core acceleration, the rate of change of coil inductance relative to the core displacement, the damping coefficient, and the spring stiffness. The test physical residual is obtained based on the residuals of the electromagnetic equation and the residuals of the mechanical motion equation.

[0009] Furthermore, when the status diagnosis result of the high-voltage switch under test is a device fault, it also includes: The percentage deviation of physical parameters is calculated based on the optimal estimated value of the physical parameters and the corresponding nominal health value, and the equipment failure type is determined based on the percentage deviation of physical parameters.

[0010] Furthermore, the equipment failure type is determined based on the percentage deviation of the physical parameters, including: When the optimal estimated value of the damping coefficient is greater than the corresponding nominal health value and the percentage deviation of the physical parameters is greater than 20%, the equipment failure type is determined to be core jamming or poor lubrication. When the optimal estimated value of spring stiffness is less than the corresponding nominal health value and the absolute value of the percentage deviation of physical parameters is greater than 15%, the equipment failure type is determined to be either opening spring fatigue or closing spring fatigue. When the optimal estimated value of the coil resistance is less than the corresponding nominal health value and the absolute value of the percentage deviation of the physical parameter is greater than 10%, the equipment fault type is determined to be an inter-turn short circuit in the coil.

[0011] Furthermore, the health benchmark physical information neural network includes an input layer, four fully connected hidden layers, and an output layer connected in sequence; Each fully connected hidden layer comprises 128 neurons; the fully connected hidden layer uses the hyperbolic tangent function as the activation function; the output layer uses linear activation.

[0012] Secondly, this application proposes a high-voltage switch coil current waveform diagnostic system for a mobile test platform, comprising: The preprocessing module is used to acquire test timing data during the opening and closing process of the high-voltage switch under test, and to preprocess the test timing data; wherein, the test timing data includes time information, coil current information and coil voltage information; The prediction module is used to input the preprocessed time information into the health benchmark physical information neural network to obtain the corresponding network prediction coil current and electromagnetic-mechanical state prediction quantities. The calculation module is used to calculate the test data fitting loss based on the network predicted coil current and the coil current information, and to calculate the test physical residual based on the coil voltage information, the network predicted coil current and the electromagnetic-mechanical state prediction. The diagnostic module is used to compare the test data fitting loss with a preset data fitting threshold, and compare the test physical residual with a preset physical residual threshold, and output the state diagnostic result of the high-voltage switch under test based on the comparison result.

[0013] Thirdly, this application proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described mobile test platform high-voltage switch coil current waveform diagnosis method.

[0014] Fourthly, this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described mobile test platform high-voltage switch coil current waveform diagnosis method.

[0015] Compared with the prior art, this application has the following beneficial effects: This application proposes a diagnostic method for high-voltage switch coil current waveforms on a mobile testing platform. By acquiring time information, coil current information, and coil voltage information during the opening and closing process of the high-voltage switch under test, and preprocessing the aforementioned test timing data, the method ensures that the test data obtained by the mobile testing platform under different field wiring conditions, different electromagnetic interference environments, and different acquisition conditions maintain good consistency in time reference, amplitude scale, and noise level, thereby reducing the impact of field operating condition fluctuations on the diagnostic results. Furthermore, this application inputs the preprocessed time information into a health benchmark physical information neural network to obtain the corresponding network-predicted coil current and electromagnetic-mechanical state predictions. This allows the diagnostic process to use the electromagnetic-mechanical coupling relationship under a healthy state as a benchmark, without relying on a large number of fault-labeled samples, making it suitable for mobile testing scenarios where high-voltage switch fault samples are scarce. Based on this, this application calculates the test data fitting loss according to the network-predicted coil current and coil current information, and calculates the test physical residual according to the coil voltage information, the network-predicted coil current, and the electromagnetic-mechanical state prediction. The test data fitting loss reflects the degree of deviation of the measured current waveform from the healthy reference waveform, while the test physical residual reflects the degree of consistency between the measured data and the electromagnetic-mechanical coupling law of the high-voltage switch opening and closing process. By comparing the test data fitting loss with a preset data fitting threshold and the test physical residual with a preset physical residual threshold, the application outputs the state diagnosis result. This application can judge the high-voltage switch under test from two dimensions: waveform deviation and physical consistency. This helps to distinguish between real equipment faults and non-fault waveform anomalies caused by measurement noise, changes in wiring status, or changes in circuit parameters, reducing the risk of false alarms and missed judgments, and improving the reliability and field applicability of the high-voltage switch coil current waveform diagnosis on the mobile test platform.

[0016] This application also proposes a high-voltage switch coil current waveform diagnostic system for a mobile test platform, an electronic device, and a computer-readable storage medium, which possess the advantages of the aforementioned high-voltage switch coil current waveform diagnostic method for a mobile test platform. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a method for diagnosing the current waveform of a high-voltage switch coil in a mobile test platform according to this application. Figure 2This is another flowchart illustrating the diagnostic method for the high-voltage switch coil current waveform of the mobile test platform in this application. Figure 3 This is a schematic diagram of a structure of a health benchmark physical information neural network in an embodiment of this application; Figure 4 This is a fault-noise decoupling determination logic diagram in the embodiments of this application; Figure 5 This is a schematic diagram of a high-voltage switch coil current waveform diagnostic system for the mobile test platform of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0020] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0021] This application proposes a method and related device for diagnosing the current waveform of a high-voltage switch coil on a mobile test platform. The following is a detailed description of this application in conjunction with embodiments and accompanying drawings.

[0022] like Figure 1 The diagram shown is a flowchart illustrating a method for diagnosing the coil current waveform of a high-voltage switch on a mobile testing platform according to this application. In this embodiment, the method for diagnosing the coil current waveform of the high-voltage switch under test is used to identify the state of the coil current waveform during the opening or closing process of the high-voltage switch. It may include: S101, acquire the test timing data during the opening and closing process of the high-voltage switch under test, and preprocess the test timing data.

[0023] The high-voltage switch under test can be installed at the testing station of a mobile electrical testing platform. After the mobile electrical testing platform is connected to the high-voltage switch operating circuit on-site, it synchronously collects test timing data during the opening and closing process. The test timing data includes time information, coil current information, and coil voltage information. The time information is used to characterize the timing position of the sampling point during the opening and closing process. The coil current information is used to reflect the influence of the core attraction after the coil is energized, the change in motion damping, and the operation process on the current waveform. The coil voltage information is used to reflect the electrical excitation applied to the two ends of the coil by the test circuit.

[0024] Preprocessing of test timing data can include noise suppression, amplitude scale unification, timing reference alignment, and effective time window truncation for time, coil current, and coil voltage information. This ensures that data collected from different test batches and under different field wiring conditions have a comparable timing basis. This preprocessing reduces the impact of electromagnetic interference, wiring fluctuations, and sensor gain differences on the mobile test platform on subsequent diagnostic judgments.

[0025] S102, input the preprocessed time information into the health benchmark physical information neural network to obtain the corresponding network prediction coil current and electromagnetic-mechanical state prediction quantities.

[0026] In practical applications, this health benchmark physical information neural network is pre-trained based on the opening and closing timing data of high-voltage switches under healthy conditions. The input to the health benchmark physical information neural network is time information, and the output includes the network-predicted coil current and the electromagnetic-mechanical state prediction quantity corresponding to this time information. The network-predicted coil current is used to characterize the prediction result of the coil current changing with time under healthy benchmark conditions, and the electromagnetic-mechanical state prediction quantity is used to characterize the coupling state between the electromagnetic process of the coil and the mechanical motion of the operating mechanism. Since the health benchmark physical information neural network uses the healthy state as a benchmark, when the data to be tested enters the diagnostic stage, the output of the health benchmark physical information neural network can serve as a reference for the healthy electromagnetic-mechanical coupling relationship.

[0027] S103, calculate the test data fitting loss based on the network predicted coil current and the coil current information, and calculate the test physical residual based on the coil voltage information, the network predicted coil current and the electromagnetic-mechanical state prediction.

[0028] Specifically, the test data fitting loss reflects the degree of deviation between the waveform of the coil current under test and the network-predicted coil current obtained from the physical information network of the health baseline. When the waveform of the coil current under test is affected by noise, wiring conditions, changes in test circuit parameters, or changes in the actual equipment conditions, the test data fitting loss will change with the difference in the coil current waveform. The test physical residual is used to characterize the degree of physical consistency between the test data and the electromagnetic process of the high-voltage switch coil and the mechanical process of the operating mechanism. Unlike simple differences in coil current waveform, the test physical residual can reflect whether the abnormality of the coil current waveform has disrupted the electromagnetic-mechanical coupling relationship that the high-voltage switch opening and closing process should satisfy.

[0029] S104, compare the test data fitting loss with the preset data fitting threshold, and compare the test physical residual with the preset physical residual threshold, and output the state diagnosis result of the high voltage switch under test based on the comparison result.

[0030] In practical applications, if both the test data fitting loss and the test physical residual are within the allowable range of the corresponding thresholds, a diagnostic result indicating that the high-voltage switch under test is in normal condition can be output. If either or both of these indicators exceed the corresponding thresholds, a corresponding state diagnostic result will be output based on the combined deviation of the two indicators. By jointly judging the data fitting loss and the test physical residual, this application can diagnose abnormal coil current waveforms without relying on a large number of fault samples, and can distinguish between waveform abnormalities caused by abnormal real equipment conditions and non-fault abnormalities caused by measurement noise, changes in wiring conditions, or changes in circuit parameters in a mobile test platform on-site testing scenario.

[0031] The present application will be further described in detail below through other embodiments.

[0032] In some embodiments of this application, the high-voltage switch coil current waveform diagnostic method for mobile test platforms is applied to rapid on-site testing of foldable mobile electrical test platforms. The high-voltage switch opening and closing coil current waveform contains mechanical characteristics such as core movement, operating mechanism action, and buffering processes. Faults such as core jamming, spring fatigue, and inter-turn short circuits in the coil can all cause characteristic distortions in the current waveform. In mobile environments, there are also factors such as strong electromagnetic interference, unstable wiring and grounding conditions, and sensor gain differences, which can cause current waveform amplitude drift, glitches, or time axis misalignment. To avoid misjudging non-fault distortions as equipment faults, this embodiment uses health status data to train a health benchmark physical information neural network, and simultaneously utilizes data fitting loss and test physical residuals during the diagnostic phase to decouple faults from noise identification.

[0033] like Figure 2The diagram shown illustrates another flowchart of the high-voltage switch coil current waveform diagnostic method for the mobile test platform proposed in this application. The flowchart sequentially includes data acquisition and preprocessing, design of a neural network architecture for health baseline physical information, network training, fault diagnosis and decoupling, and fault root cause localization. This process establishes a coupling benchmark between the electromagnetic process of the high-voltage switch coil and the mechanical movement process of the iron core using the healthy coil current waveform, eliminating the need for pre-collection of a large number of tagged fault samples during field testing.

[0034] First, a foldable mobile electrical test platform is used to synchronously collect coil current signals and coil voltage signals during the opening or closing process of the high-voltage switch. The coil current signal collected at each moment is denoted as The voltage signal across the coil is denoted as As an example, sampling frequency The duration of a single continuous signal acquisition is 50ms, which is then converted into a discrete time series after analog-to-digital conversion. As the original time series data, This represents the total number of sampling points for the current waveform of a single coil. For the first Time information for each sampling point For the first Coil current information at each sampling point For the first The coil voltage information at each sampling point. For high-voltage switches of the same model and in a healthy state, at least 50 sets of valid coil current waveform data are collected to construct a healthy state training dataset.

[0035] After acquisition, the raw time-series data is preprocessed. Specifically, a fourth-order zero-phase Butterworth low-pass filter can be used to reduce noise in the raw acquired signal, with a cutoff frequency of [missing information]. To reduce the impact of high-frequency electromagnetic noise and power frequency interference on waveform recognition, the filtered coil current signal is then linearly normalized, mapping the amplitude to the [0, 1] interval, ensuring a uniform amplitude scale for coil current waveforms from different acquisition batches or under different sensor gain conditions. Furthermore, the moment when the rising edge of the coil current signal first reaches 5% of its peak value is used as the timing reference. A fixed-length time window containing 2000 sampling points is extracted from this timing reference point, thereby aligning different waveforms on the time axis. Preprocessed health status training dataset. It can be recorded as:

[0036] In the formula, This refers to the sequence number of the healthy coil current waveform. The total number of current waveforms in the healthy coil.

[0037] like Figure 3 The diagram shown is a schematic representation of a physical information neural network for health benchmarks. In some embodiments of this application, a multi-layer fully connected physical information neural network is constructed. As a health benchmark physical information neural network, a multilayer fully connected physical information neural network uses continuous time coordinates. As input, the output layer is configured with 5 output parameters, namely the network prediction coil current. Core displacement Core speed Coil inductance and inductance change rate Among them, the core speed With core displacement Between satisfy Inductance change rate With coil inductance Between satisfy By simultaneously outputting current, inductance, and mechanical motion, a multilayer fully connected physical information neural network can express the coupling relationship between the electromagnetic dynamics of the coil and the mechanical motion of the iron core in a time-domain model.

[0038] In terms of network structure, the health benchmark physical information neural network can include an input layer, four fully connected hidden layers, and an output layer connected sequentially. Each fully connected hidden layer contains 128 neurons, and the activation function of the fully connected hidden layer is the hyperbolic tangent function tanh. The output layer uses a linear activation method. The learnable parameters of the health benchmark physical information neural network are denoted as follows: The health baseline physical information neural network structure in this application can balance the fitting capability of rapidly changing timing signals during the opening and closing process with the computational requirements of on-site inference on the mobile test platform.

[0039] In one embodiment of this application, the output physical quantity of the health benchmark physical information neural network is differentiated based on an automatic differentiation algorithm to obtain the rate of change of the network prediction coil current. Core speed Iron core acceleration and the rate of change of coil inductance The rate of change of coil inductance relative to the displacement of the iron core. The coil current can be predicted based on the output of the neural network of the health baseline physical information. L and core displacement Numerical differences can be obtained by following the correspondence of time sampling points, or, under the condition of satisfying computational stability, by automatically differentiating the chain relationship. Through the above differentiation process, the output of the health benchmark physical information neural network is not only used to fit the coil current waveform, but also to construct the residuals of the electromagnetic equations and the mechanical motion equations.

[0040] During the training phase, the total loss function Loss due to data fitting Electromagnetic equation residual loss and residual loss of mechanical motion equations Composition, residual loss of electromagnetic equations and residual loss of mechanical motion equations Together they constitute the test physical residual loss, expressed as:

[0041] In the formula, , and These are the weight coefficients for the corresponding items. As an example, , , .

[0042] The data fitting loss, supervised by the measured current, is used to constrain the degree of matching between the network prediction coil current output by the health baseline physical information neural network and the on-site acquired health status signal. It is expressed as:

[0043] In the formula, For health benchmark physical information neural networks The network predicts the coil current at all times. for The measured coil current is collected at any time.

[0044] The residual loss of the electromagnetic equations embeds the electromagnetic dynamics of the coil into the training process of a health benchmark physical information neural network. The electromagnetic dynamic equations of the coil circuit can be expressed as:

[0045] In the formula, This is the power supply voltage. for Current at any moment For inductance, Let be the coil resistance. Substituting the physical quantities output by the health baseline physical information neural network into the above electromagnetic dynamic equations, we can obtain the electromagnetic equation residuals. :

[0046] The corresponding residual loss of the electromagnetic equation is:

[0047] in, To be In Replace with .

[0048] The residual loss of the mechanical motion equations embeds the mechanical motion laws of the iron core into the training process of the health benchmark physical information neural network. The mechanical motion equations of the iron core system can be expressed as:

[0049] In the formula, For equivalent quality, For displacement, The damping coefficient is... For spring stiffness, The current is Displacement is Electromagnetic attraction at that time. Using a solenoid analytical model, it can be represented as follows:

[0050] Substituting the core displacement, core velocity, core acceleration, network-predicted coil current, and inductance-displacement relationship output by the health baseline physical information neural network into the mechanical motion equations, the residuals of the mechanical motion equations can be obtained. :

[0051] The corresponding residual loss of the mechanical motion equation is:

[0052] in, To be In Replace with .

[0053] By simultaneously constructing electromagnetic equation residuals and mechanical motion equation residuals, the health benchmark physical information neural network is jointly constrained by coil current waveform data and the physical laws of electromagnetic-mechanical coupling during the training phase. During the diagnostic phase, the aforementioned test physical residuals can be reused as state judgment indicators, allowing abnormal coil current waveforms to be further differentiated based on whether the electromagnetic-mechanical coupling relationship has been disrupted, rather than solely on waveform similarity.

[0054] During the training of the health benchmark physical information neural network, only the preprocessed health status training dataset is used. The Adam optimizer can be used to update network parameters, with an initial learning rate set to... First-order moment estimation of exponential decay rate Second-order moment estimation of exponential decay rate The batch size is set to 32. As a training method, each training round randomly selects a waveform from the health state training dataset, and then randomly selects 200 time sampling points from that waveform to participate in the iteration. These time sampling points are then input into the health baseline physical information neural network for forward propagation, yielding... , , , , The corresponding derivatives are calculated, and the loss value is determined based on the total loss function. The learnable parameters are then updated using the backpropagation algorithm. The total number of training iterations can be set to 500. After training, the network forms a high-voltage switch electromagnetic-mechanical coupling mapping relationship in a healthy state, and the physical quantities output at any sampling point at any time are constrained by electromagnetic equations and mechanical motion equations.

[0055] During the on-site diagnostic phase, for the high-voltage switch under test, test timing data are obtained according to the aforementioned data acquisition and preprocessing methods. The test current can be denoted as... The test voltage is recorded as The preprocessed test time sampling points are input into the trained health benchmark physical information neural network to obtain the network's predicted coil current, core displacement, core velocity, coil inductance, and rate of change of inductance for each test time point. Based on the test current... With predicted coil current Calculate the fitting loss of the test data According to the test voltage Predicting current Calculation and testing of electromagnetic equation residual loss, including coil inductance, rate of change of inductance, coil resistance, core motion state, damping coefficient, and spring stiffness. and test mechanical motion equation residual loss Overall test physical residuals It can be represented as:

[0056] Optionally, if it is necessary to maintain consistency with the loss weights during the training phase, the residual loss of the test electromagnetic equation and the residual loss of the test mechanical motion equation can be weighted to form the total test physical residual. However, a consistent residual construction method should be used in diagnostic threshold calibration and test judgment.

[0057] In some embodiments of this application, a health status validation set that was not involved in model training is used to set the diagnostic threshold. Specifically, the data fitting loss in the health status validation set is statistically analyzed. mean and standard deviation Statistical analysis of the total test physical residuals mean and standard deviation And based on the 3σ criterion, a preset data fitting threshold is set. and preset physical residual threshold :

[0058] like Figure 4 The diagram shown is a logic diagram for fault-noise decoupling determination. During the fault-noise decoupling determination process, based on the test samples... and and , The comparison results output the state diagnosis results. Specifically, when < and < When the condition of the high-voltage switch under test is healthy, the diagnostic result is as follows: < and ≥ When the status diagnosis result of the high-voltage switch under test is a device fault, this device fault can be further traced back to its physical parameters according to the actual mechanical fault in this embodiment; when ≥ and < When the condition diagnosis result of the high-voltage switch under test is measured noise or changes in circuit parameters, then... ≥ and ≥ When the condition diagnosis result of the high-voltage switch under test is a serious fault or a mixed factor, manual verification can be prompted in field applications. Through the above dual-indicator cross-judgment, the data fitting loss mainly reflects the morphological difference between the waveform of the current waveform of the coil under test and the healthy reference waveform, while the test physical residual mainly reflects whether the measured data has violated the electromagnetic-mechanical coupling constraint. Both are used together to distinguish between real equipment abnormalities and non-fault measurement abnormalities.

[0059] When the diagnosis indicates equipment failure and further investigation is needed to pinpoint the root cause, the learnable parameters of the trained health baseline physical information neural network are fixed. The network structure and weights are no longer updated. Then, the damping coefficient is selected. c Spring stiffness k 1 and coil resistance R As physical parameters to be optimized, to minimize the test physical residual To optimize the objective, parameters are fine-tuned to obtain optimal estimates of the physical parameters. Damping coefficient. c The damping term in the mechanical motion equation reflects the magnitude of the resistance to the movement of the iron core. Iron core jamming or poor lubrication will significantly increase the damping; spring stiffness. k1 represents the elastic term in the mechanical motion equation, reflecting the elastic properties of the opening or closing spring. The stiffness of the spring decreases due to fatigue after long-term use; coil resistance... R For electrical parameters in the electromagnetic dynamics equations, when an inter-turn short circuit occurs in the coil, the effective number of turns decreases, and the DC resistance drops. Optionally, the equivalent mass is an inherent parameter of the iron core and is usually not used as a field-adjustable health parameter during normal use; the coil inductance is a dynamic quantity that changes with the displacement of the iron core and has been output by the network and participates in coupling constraints; the power supply voltage is a measurement input quantity and is not used as a parameter for optimizing the health status of the equipment.

[0060] Before fine-tuning the parameters, the healthy nominal values ​​of the damping coefficient, spring stiffness, and coil resistance can be obtained, and these are denoted as follows: , and As an example, in the offline state of the high-voltage switch opening and closing mechanism, an initial displacement is applied to the iron core and then released. The free decaying vibration displacement signal is collected, and the damping coefficient's healthy nominal value is calculated after fitting an exponentially decaying envelope.

[0061] In the formula, For the damping ratio, Equivalent mass. Healthy nominal value of spring stiffness. The force-displacement curve can be obtained by performing a compression test on the spring using a spring tension-compression testing machine, recording the force-displacement curve, and taking the slope of the linear segment as the result. Coil resistance nominal value (healthy) The resistance across the coil can be measured using a four-wire digital micro-ohmmeter or a high-precision multimeter at room temperature (20℃±5℃). Repeat the measurement five times and take the average value. .

[0062] During parameter fine-tuning, the test time and test voltage information of the sample under test are used for residual calculation, and the damping coefficient is used... c Spring stiffness k 1 and coil resistance R The initial values ​​are respectively set to , and The Adam optimizer is used to iteratively optimize the above physical parameters, and the learning rate can be set to... The iteration count can be set to 100 steps. In each iteration, only the damping coefficient, spring stiffness, and coil resistance are updated; the learnable parameters of the health baseline physical information neural network are not changed. θ Substitute the updated parameters back into the residuals of the electromagnetic equations and the mechanical motion equations to calculate the new... The physical parameters to be optimized are then adjusted through backpropagation until the residual converges or the preset number of iterations is reached. The final optimal estimates of the damping coefficient, spring stiffness, and coil resistance are then output, denoted as follows: It can also be uniformly recorded as .

[0063] After obtaining the optimal estimate, calculate the percentage deviation of each physical parameter from its corresponding nominal health value:

[0064] in, for The corresponding initial value. The equipment fault type is determined based on the direction and magnitude of parameter deviation. Specifically, when... and When the equipment malfunction is determined to be either core jamming or poor lubrication; and When the fault type is determined to be either opening spring fatigue or closing spring fatigue, the equipment fault type is determined to be either opening spring fatigue or closing spring fatigue. and At that time, the equipment fault type was determined to be an inter-turn short circuit in the coil.

[0065] In one embodiment of this application, the aforementioned fault determination threshold can be determined jointly through health sample fluctuation statistics and fault simulation experiments. Specifically, 30 sets of untrained healthy waveforms can be used to calculate the natural fluctuation range of parameters under healthy conditions by freezing network weights and fine-tuning physical parameters, thereby obtaining the percentage deviation Δ of the damping coefficient under healthy conditions. c health ≤ 8%, spring stiffness deviation percentage Coil resistance deviation percentage Furthermore, mild core jamming, slight spring fatigue, and inter-turn short circuit faults were artificially induced under laboratory conditions. Mild core jamming could be simulated by adding damping plates, slight spring fatigue by partially unloading the preload, and inter-turn short circuits by shorting 2% to 5% of the turns. The measured deviation range of mild jamming can be represented as the percentage deviation of the damping coefficient from its corresponding healthy nominal value. The spring fatigue deviation range can be defined as the percentage deviation of the spring stiffness from its corresponding healthy nominal value. The deviation range of inter-turn short circuit can be defined as the percentage deviation of the coil resistance from its corresponding nominal health value. Considering the normal fluctuation boundaries of healthy samples, the detectability of minor faults, and the requirement to avoid false alarms, the damping coefficient change threshold can be set to 20%, the spring stiffness change threshold to 15%, and the coil resistance change threshold to 10%.

[0066] Using the above method, this application relies solely on health state waveform data during the training phase, avoiding the problem of obtaining a large number of fault labeling samples at the high-voltage switch site. During the diagnosis phase, cross-judgment is performed using data fitting loss and test physical residuals, which can distinguish real mechanical faults from false anomalies caused by measurement noise, wiring abnormalities, grounding abnormalities, or changes in circuit parameters. During the fault root cause location phase, by freezing the neural network of health baseline physical information and fine-tuning the damping coefficient, spring stiffness, and coil resistance in reverse, the output results not only include the state category but also provide the fault root cause corresponding to the physical mechanism, providing interpretable diagnostic basis for the on-site operation and maintenance of the mobile test platform.

[0067] In one application example of this application, a foldable mobile electrical testing platform is used to conduct tests on a 10kV high-voltage switch. The test data includes 80 sets of healthy state waveforms, 30 sets of typical fault waveforms (core jamming, spring fatigue, and inter-turn short circuit), and 40 sets of noise interference waveforms caused by on-site electromagnetic interference and wiring abnormalities. During the test, coil current signals and voltage signals across the coil are simultaneously acquired for each waveform set, and preprocessed according to the aforementioned filtering, amplitude normalization, and time alignment methods.

[0068] In this application example, health status waveforms are used to train and validate the health benchmark physical information neural network, while fault waveforms and noise interference waveforms are used to validate the fault identification capability and fault-noise decoupling capability of the diagnostic method. Comparison methods include Temporal Convolutional Network (TCN), Long Short-Term Memory Network (LSTM), Transformer temporal diagnostic model, Support Vector Machine (SVM), and traditional threshold determination methods. Performance metrics for different diagnostic methods are shown in Table 1 below.

[0069] Table 1. Comparison of Performance Indicators of Different Diagnostic Methods

[0070] As demonstrated in this application example, the method of this application achieves an accuracy rate of 98.3% in identifying three typical faults: core jamming, spring fatigue, and inter-turn short circuits in the coil. It also achieves a decoupling accuracy of 97.5% in decoupling real faults from non-fault-related anomalies caused by environmental noise and wiring abnormalities. Compared to diagnostic methods that rely solely on waveform characteristics or end-to-end timing models, the method of this application utilizes a health baseline network, electromagnetic-mechanical dual-physical residuals, and parameter deviation quantification analysis. This enables the diagnostic results to simultaneously possess waveform-level anomaly identification capabilities and physical mechanism-level explanatory capabilities, meeting the engineering requirements for rapid on-site testing on mobile testing platforms.

[0071] This application reduces reliance on fault samples and is adapted to the practical engineering needs of on-site testing on mobile testing platforms. It trains a health benchmark physical information neural network using time-series data of coil current and voltage under healthy conditions. During on-site diagnosis, the network-predicted coil current and electromagnetic-mechanical state predictions output by the health benchmark network are used as references to calculate the test data fitting loss and test physical residual. Therefore, this application eliminates the need for pre-collecting a large number of high-voltage switch fault label samples, enabling condition diagnosis of the high-voltage switch under test. This is suitable for application scenarios where the probability of mechanical faults in high-voltage switches is low and sufficient fault samples are difficult to obtain on-site. This application introduces the coupling relationship between the electromagnetic process of the high-voltage switch coil and the mechanical motion process of the operating mechanism into the health benchmark physical information neural network, and simultaneously calculates the test data fitting loss and test physical residual during the diagnosis phase. The test data fitting loss characterizes the deviation of the waveform from the health benchmark, while the test physical residual characterizes the deviation of the test data from the electromagnetic-mechanical physical laws. Through the joint judgment of these two indicators, this application can distinguish between real mechanical faults and false anomalies caused by on-site electromagnetic interference, wiring abnormalities, changes in grounding status, or changes in loop parameters, thereby reducing the risk of false alarms and ineffective maintenance and improving the reliability of diagnostic results. Furthermore, the health benchmark physical information neural network not only outputs the network's predicted coil current but also outputs electromagnetic-mechanical state predictions related to the high-voltage switch's operation, such as core displacement, core velocity, coil inductance, and rate of change of inductance. After determining that the high-voltage switch under test has a fault, the learnable parameters of the health benchmark physical information neural network can be fixed, and the damping coefficient, spring stiffness, and coil resistance can be fine-tuned with the goal of minimizing the test physical residual. By comparing the optimal estimated values ​​of the physical parameters with the corresponding nominal health values, the fault type can be determined based on the percentage deviation of the physical parameters, such as core jamming or poor lubrication, fatigue of the opening or closing spring, or short circuit between coil turns, thus making the diagnostic results correspond to the specific physical mechanism. This application can also adapt to complex working conditions in mobile testing scenarios. Mobile test platforms are easily affected by electromagnetic interference, unstable wiring conditions, changes in grounding conditions, differences in sensor gain, and time axis misalignment during on-site testing. This application reduces the differences in amplitude, noise, and timing reference of field-acquired data by preprocessing the test timing data. Simultaneously, it evaluates whether the test data meets the electromagnetic-mechanical coupling relationship during the opening and closing process of high-voltage switches through test physical residuals. The combined effect of preprocessing and test physical residual constraints enables this application to maintain relatively stable diagnostic capabilities under the variable field conditions of a mobile test platform, improving the applicability of rapid field testing of high-voltage switches.

[0072] like Figure 5 The diagram shown is a schematic of a high-voltage switch coil current waveform diagnostic system for a mobile test platform according to this application, which may include: The preprocessing module is used to acquire test timing data during the opening and closing process of the high-voltage switch under test, and to preprocess the test timing data; wherein, the test timing data includes time information, coil current information and coil voltage information; The prediction module is used to input the preprocessed time information into the health benchmark physical information neural network to obtain the corresponding network prediction coil current and electromagnetic-mechanical state prediction quantities. The calculation module is used to calculate the test data fitting loss based on the network predicted coil current and the coil current information, and to calculate the test physical residual based on the coil voltage information, the network predicted coil current and the electromagnetic-mechanical state prediction. The diagnostic module is used to compare the test data fitting loss with a preset data fitting threshold, and compare the test physical residual with a preset physical residual threshold, and output the state diagnostic result of the high-voltage switch under test based on the comparison result.

[0073] It should be noted that, in the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of each block is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple blocks may be combined or integrated into another device, or some features may be ignored or not executed. The modules described as separate components may or may not be physically separated. The components shown as modules may be one or more physical units, that is, they may be located in one place or distributed in multiple different places. Some or all of the modules can be selected to achieve the purpose of the solution in this embodiment according to actual needs.

[0074] Furthermore, in the various embodiments of the present invention, the modules can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The integrated unit described above can be implemented in hardware or as a software functional unit.

[0075] This application also provides an electronic device, which may include one or more processors, memory and communication interfaces.

[0076] The memory, communication interface, and processor are coupled together. For example, the memory, communication interface, and processor can be coupled together via a bus.

[0077] The communication interface is used for data transmission with other devices. The memory stores computer program code. This computer program code includes computer instructions, which, when executed by the processor, cause the electronic device to perform the steps of the aforementioned mobile test platform high-voltage switch coil current waveform diagnostic method.

[0078] The processor can be a processor or controller, such as a Central Processing Unit (CPU), a general-purpose processor, a Digital Signal Processor (DSP), an Application-Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with this disclosure. The processor can also be a combination that implements computational functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc. The processor can be used to support an electronic device in performing the method steps provided in the above embodiments.

[0079] The bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. These buses can be categorized as address buses, data buses, control buses, etc.

[0080] This application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the steps of the above-described mobile test platform high-voltage switch coil current waveform diagnosis method.

[0081] The computer-readable storage media involved in this application include random access memory (RAM), memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage media known in the art.

[0082] The above are merely preferred embodiments of this application and are not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for diagnosing the current waveform of a high-voltage switch coil in a mobile test platform, characterized in that, include: Acquire test timing data during the opening and closing process of the high-voltage switch under test, and preprocess the test timing data; wherein, the test timing data includes time information, coil current information and coil voltage information; The preprocessed time information is input into the health benchmark physical information neural network to obtain the corresponding network prediction coil current and electromagnetic-mechanical state prediction quantities. The test data fitting loss is calculated based on the network predicted coil current and the coil current information, and the test physical residual is calculated based on the coil voltage information, the network predicted coil current and the electromagnetic-mechanical state prediction. The test data fitting loss is compared with a preset data fitting threshold, and the test physical residual is compared with a preset physical residual threshold. Based on the comparison results, the state diagnosis result of the high-voltage switch under test is output.

2. The method for diagnosing the current waveform of a high-voltage switch coil on a mobile test platform according to claim 1, characterized in that, The step of outputting the status diagnosis result of the high-voltage switch under test based on the comparison result includes: When the test data fitting loss is less than the preset data fitting threshold and the test physical residual is less than the preset physical residual threshold, the state diagnosis result of the high voltage switch under test is healthy. When the test data fitting loss is less than the preset data fitting threshold and the test physical residual is greater than or equal to the preset physical residual threshold, the state diagnosis result of the high voltage switch under test is equipment failure. When the test data fitting loss is greater than or equal to the preset data fitting threshold, and the test physical residual is less than the preset physical residual threshold, the state diagnosis result of the high voltage switch under test is measurement noise or circuit parameter change. When the test data fitting loss is greater than or equal to the preset data fitting threshold, and the test physical residual is greater than or equal to the preset physical residual threshold, the state diagnosis result of the high voltage switch under test is a serious fault or a mixed factor.

3. The method for diagnosing the current waveform of a high-voltage switch coil on a mobile test platform according to claim 2, characterized in that, When the status diagnosis result of the high voltage switch under test is a device fault, the weight parameters of the health benchmark physical information neural network are fixed, and the physical parameters used to calculate the test physical residual are fine-tuned with the goal of minimizing the test physical residual, so as to obtain the optimal estimate of the physical parameters. The test physical residuals include electromagnetic equation residuals and mechanical motion equation residuals. The electromagnetic equation residuals are calculated based on coil voltage information, network-predicted coil current, electromagnetic-mechanical state predictions, and coil resistance. The mechanical motion equation residuals are calculated based on network-predicted coil current, electromagnetic-mechanical state predictions, damping coefficients, and spring stiffness.

4. The method for diagnosing the current waveform of a high-voltage switch coil on a mobile test platform according to claim 3, characterized in that: The electromagnetic-mechanical state prediction quantities include core displacement, core velocity, coil inductance, and inductance rate. The test physical residual is calculated based on the coil voltage information, the network-predicted coil current, and the electromagnetic-mechanical state prediction, including: The network predicts the coil current and automatically differentiates it based on the time information to obtain the coil current change rate. The electromagnetic equation residuals are calculated based on the coil voltage information, the network-predicted coil current, the coil current change rate, the coil inductance, the inductance change rate, and the coil resistance. The core displacement is automatically differentiated based on the time information to obtain the core acceleration. The residual of the mechanical motion equation is calculated based on the network predicted coil current, the core displacement, the core velocity, the core acceleration, the rate of change of coil inductance relative to the core displacement, the damping coefficient, and the spring stiffness. The test physical residual is obtained based on the residuals of the electromagnetic equation and the residuals of the mechanical motion equation.

5. The method for diagnosing the current waveform of a high-voltage switch coil on a mobile test platform according to claim 3, characterized in that, When the status diagnosis result of the high-voltage switch under test is a device fault, it also includes: The percentage deviation of physical parameters is calculated based on the optimal estimated value of the physical parameters and the corresponding nominal health value, and the equipment failure type is determined based on the percentage deviation of physical parameters.

6. The method for diagnosing the current waveform of a high-voltage switch coil on a mobile test platform according to claim 5, characterized in that, The equipment failure type is determined based on the percentage deviation of the physical parameters, including: When the optimal estimated value of the damping coefficient is greater than the corresponding nominal health value and the percentage deviation of the physical parameters is greater than 20%, the equipment failure type is determined to be core jamming or poor lubrication. When the optimal estimated value of spring stiffness is less than the corresponding nominal health value and the absolute value of the percentage deviation of physical parameters is greater than 15%, the equipment failure type is determined to be either opening spring fatigue or closing spring fatigue. When the optimal estimated value of the coil resistance is less than the corresponding nominal health value and the absolute value of the percentage deviation of the physical parameter is greater than 10%, the equipment fault type is determined to be an inter-turn short circuit in the coil.

7. The method for diagnosing the current waveform of a high-voltage switch coil on a mobile test platform according to claim 1, characterized in that, The health benchmark physical information neural network includes an input layer, four fully connected hidden layers, and an output layer connected in sequence. Each fully connected hidden layer comprises 128 neurons; the fully connected hidden layer uses the hyperbolic tangent function as the activation function; the output layer uses linear activation.

8. A diagnostic system for high-voltage switch coil current waveforms on a mobile test platform, characterized in that, include: The preprocessing module is used to acquire test timing data during the opening and closing process of the high-voltage switch under test, and to preprocess the test timing data; wherein, the test timing data includes time information, coil current information and coil voltage information; The prediction module is used to input the preprocessed time information into the health benchmark physical information neural network to obtain the corresponding network prediction coil current and electromagnetic-mechanical state prediction quantities. The calculation module is used to calculate the test data fitting loss based on the network predicted coil current and the coil current information, and to calculate the test physical residual based on the coil voltage information, the network predicted coil current and the electromagnetic-mechanical state prediction. The diagnostic module is used to compare the test data fitting loss with a preset data fitting threshold, and compare the test physical residual with a preset physical residual threshold, and output the state diagnostic result of the high-voltage switch under test based on the comparison result.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the mobile test platform high-voltage switch coil current waveform diagnosis method as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the mobile test platform high-voltage switch coil current waveform diagnosis method as described in any one of claims 1-7.

Images