GIS isolation switch mechanical fault diagnosis method and system based on adaptive filtering

Abstract

The invention provides a GIS isolation switch mechanical fault diagnosis method and system based on adaptive filtering, and the method comprises the steps: S1, data obtaining and preprocessing: obtaining a power signal of a GIS isolation switch driving motor through a sensor, and carrying out the denoising, standardization processing and common frequency analysis; s2, performing multi-scale distribution modulation decoupling enhancement processing: performing sideband feature extraction and enhancement on the preprocessed signal by adopting a multi-scale adaptive modulation decomposition step, a nonlinear generalized inverse modulation compensation step, a dynamic feature adaptive spectrum enhancement step and an adaptive energy constrained variational spectrum reconstruction step; and S3, fault diagnosis: based on the enhanced sideband features, performing fault diagnosis in combination with a rule base method or a machine learning model. The method is applied to a corresponding system, a sideband enhancement algorithm of adaptive filtering is adopted, sideband features in motor power signals can be adaptively recognized and enhanced in dynamically changing signals, and therefore the reliability and accuracy of fault diagnosis are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical fault diagnosis of GIS disconnectors, and particularly to a method and system for mechanical fault diagnosis of GIS disconnectors based on adaptive filtering. Background Art

[0002] In modern power systems, as an important switching device, the operation state monitoring and fault diagnosis of GIS (Gas Insulated Switch) disconnectors are crucial for ensuring the safety and stability of the power grid. With the continuous development of technology, traditional fault diagnosis methods mainly rely on monitoring the working state of equipment through sensors, such as physical quantities like temperature and vibration, or analyzing electrical parameters such as current and voltage. However, with the increasing complexity of the system, these traditional methods face challenges in terms of accuracy, real-time performance, and comprehensiveness. In recent years, using motor power signals for mechanical fault diagnosis has become a new research direction. By analyzing the motor power signals, the working state and potential faults of the equipment can be effectively identified. In the process of fault diagnosis, the spectral analysis of motor power signals is of great significance, especially in the sideband feature part of the spectrum. Sidebands usually appear on both sides of the main frequency and reflect the modulation effects within the system, such as amplitude modulation (AM) or frequency modulation (FM). These sideband features often change significantly when the equipment fails, so they become an important basis for fault diagnosis. However, the actually collected motor power signals usually have problems such as noise interference and non-stationarity, which bring difficulties to the spectral analysis of the signals. Therefore, how to extract clear and accurate sideband features has become an urgent problem to be solved in the field of fault diagnosis.

[0003] Deficiencies of the Existing Technology The existing motor power signal fault diagnosis technologies mainly rely on traditional signal processing methods, such as Fast Fourier Transform (FFT), Wavelet Transform, Empirical Mode Decomposition (EMD), etc. These methods have achieved certain results in extracting spectral features and sideband analysis. However, these traditional methods have some obvious deficiencies when facing complex motor power signals, which are specifically reflected in the following aspects:

[0004] Insufficient spectral resolution: Although the Fast Fourier Transform (FFT) can effectively transform the signal from the time domain to the frequency domain, its frequency resolution is limited by the signal sampling rate and window length. For the recognition of weak changes in the sideband features within the frequency band of motor power signals, FFT often cannot provide sufficiently detailed spectral information, resulting in the fuzziness of sideband features and affecting the accurate diagnosis of faults.

[0005] Problems in the processing of noise and non-stationary signals: In practical applications, motor power signals often contain various noises and interferences. These noises may come from the external environment, vibrations of the equipment itself, or even electrical interferences in the power system. Although traditional wavelet transform and empirical mode decomposition can process non-stationary signals, for signals with a high-noise background, the sideband features extracted by them may be severely affected by the noise, thus reducing the accuracy and reliability of fault diagnosis.

[0006] Difficulty in dealing with complex multi-fault modes: Motor fault modes are complex and diverse, often involving multiple fault causes or multiple faults in different components. In this case, traditional signal processing methods may not be able to capture the sideband changes in multiple frequency bands simultaneously, resulting in incomplete or overlapping features, thus affecting the discrimination of different fault modes. Inability to dynamically adapt to signal changes: Motor power signals are usually non-linear and non-stationary, and the spectral features of the signals may change over time.

[0007] In the prior art, many signal processing methods do not have the adaptive ability and cannot automatically adjust the analysis parameters according to the real-time changes of the signals. This makes it difficult for traditional methods to obtain accurate diagnostic results in a dynamically changing working environment. High complexity of spectral analysis: Although existing spectral analysis methods can extract the sideband features of motor power signals, due to the calculation involving multiple frequency bands and multiple parameters, they often require complex calculation processes and long processing times, resulting in poor real-time performance and difficulty in meeting the requirements of rapid response in fault diagnosis.

[0008] Therefore, a mechanical fault diagnosis method and system for GIS disconnectors based on adaptive filtering are proposed to solve the above problems. Summary of the Invention

[0009] In view of this, the purpose of the present invention is to provide a mechanical fault diagnosis method and system for GIS disconnectors based on adaptive filtering to solve at least the above problems.

[0010] The technical solution adopted in the first aspect of the present invention is as follows:

[0011] A mechanical fault diagnosis method for GIS disconnectors based on adaptive filtering, the method comprising the following steps:

[0012] S1. Data acquisition and preprocessing: Obtain the power signal of the driving motor of the GIS disconnector through a sensor, and perform denoising, normalization processing, and general frequency analysis;

[0013] S2. Multi-scale distribution modulation decoupling and enhancement processing: Using multi-scale adaptive modulation decomposition, non-linear generalized inverse modulation compensation, dynamic feature adaptive spectrum enhancement, and variational spectrum reconstruction steps with adaptive energy constraint to extract and enhance the sideband features of the preprocessed signal;

[0014] S3. Fault diagnosis: Based on the enhanced sideband features, combined with rule-based methods or machine learning models for fault diagnosis.

[0015] Furthermore, the multi-scale adaptive modulation decomposition step includes:

[0016] Using a multi-scale joint modulation projection operator to decompose the power signal and extract the modulation envelope and phase information under different operating states. Specifically:

[0017] Using a multi-scale joint modulation projection operator to decompose the signal:

[0018]

[0019] Where A k (t) is the modulation envelope of the k-th level, corresponding to different sideband components; θ k (t) is the instantaneous phase, and n(t) is the noise, defined as:

[0020]

[0021] ω k (τ) represents the time-dependent angular frequency, τ is the time parameter, which is a dynamic representation of t, d is the differential operator. To ensure accurate decomposition at different scales, by introducing a time-scale joint modulation projection operator:

[0022]

[0023] Where, W k (t) is optimized by the maximum energy criterion:

[0024]

[0025] By optimizing the weights, the sideband features under specific states are strengthened.

[0026] Furthermore, the non-linear generalized inverse modulation compensation step includes:

[0027] Using a generalized inverse transform operator to perform inverse modulation compensation on the decomposed signal to eliminate non-linear phase distortion. Specifically:

[0028] Using the calculation formula:

[0029]

[0030] Among them: the generalized inverse transform operator Adopt a non - linear mapping:

[0031]

[0032] Among them, λ is a scale parameter adjusted according to experience. To avoid misjudgment of the decomposed signal, Approximately represent the modulation mode of the original signal. By dividing the numerator by the denominator, the modulation components in the power signal are separated to retain the sideband components related to faults.

[0033] Furthermore, the dynamic feature adaptive spectrum enhancement step includes:

[0034] Calculate the short - time Fourier transform of the input signal, design an adaptive gain operator to dynamically adjust the gain, and restore the enhanced time - domain signal through the inverse short - time Fourier transform;

[0035] Calculate the short - time Fourier transform of the input signal, specifically:

[0036] Obtain the time - frequency representation of the signal by performing the short - time Fourier transform STFT on the compensated signal:

[0037]

[0038] Among them, P comp (t) is the signal after non - linear generalized inverse modulation compensation processing; w(τ - t) is a short - time window function used to localize the time - frequency characteristics of the signal; f is the frequency variable, t is the time variable, and X(f, t) is the result of the short - time Fourier transform, representing the spectral distribution of the signal at different time points;

[0039] Design an adaptive gain operator to dynamically adjust the gain, specifically:

[0040] By introducing a time - transformation gain operator:

[0041] H(f, t)=(1 + βf m e -t / τ )H0(f)(0.23)

[0042] Among them: β is the gain intensity control parameter, m is the frequency amplification exponent, τ is the time scale parameter, and H0(f) is the initial filtering operator.

[0043] Calculate the enhanced spectrum:

[0044] The enhanced signal spectrum X'(f, t) is given by:

[0045] X′(f, t)=H(f, t)·X(f, t)(0.24)

[0046] Among them, H(f,t) acts on the original spectrum to dynamically amplify the target sideband signal;

[0047] Restore the enhanced time-domain signal through the inverse short-time Fourier transform, specifically:

[0048] Transform the enhanced time-frequency signal back to the time domain through the inverse short-time Fourier transform to obtain the enhanced time signal:

[0049]

[0050] Furthermore, the steps of the variational spectrum reconstruction with adaptive energy constraint include:

[0051] Perform a Fourier transform on the enhanced signal, set the variational optimization objective, and iteratively solve it through the gradient descent method to optimize the energy distribution and resolution of the sideband features, specifically:

[0052] Perform a Fourier transform on the enhanced signal to obtain the spectrum before optimization:

[0053]

[0054] Make the sideband features clearer through the optimization formula 1.11, and at the same time ensure that the energy of the entire signal is consistent with the original signal:

[0055]

[0056] Among them, E represents the total energy of the signal;

[0057] Set the variational optimization objective:

[0058]

[0059] ρ is the regularization coefficient, used to adjust the balance between the energy constraint and the smoothing constraint;

[0060] Adopt the gradient descent method for iterative solution:

[0061]

[0062] Among them, the gradient of the loss function:

[0063]

[0064] Among them, J is the loss function, and η is the learning rate, which controls the update step size.

[0065] Furthermore, the rule-based method includes:

[0066] Set thresholds based on the sideband amplitude ratio, instantaneous frequency change rate, or power spectral entropy to determine poor contact, contact jitter, or load abnormal faults.

[0067] Further, the machine learning model includes a support vector machine (SVM), a random forest (RF), a long short-term memory network (LSTM), or a convolutional neural network (CNN).

[0068] The technical solution adopted in the second aspect of the present invention is as follows:

[0069] A GIS disconnector mechanical fault diagnosis system based on adaptive filtering, which is used to execute a GIS disconnector mechanical fault diagnosis method based on adaptive filtering. The system includes:

[0070] A data acquisition module: used to acquire the power signal of the driving motor of the GIS disconnector;

[0071] A data preprocessing module: used to perform denoising, normalization processing, and frequency spectrum analysis on the acquired signal;

[0072] A feature enhancement module: adopting multi-scale distributed modulation decoupling enhancement processing, including multi-scale adaptive modulation decomposition, non-linear generalized inverse modulation compensation, dynamic feature adaptive spectrum enhancement, and a variational spectrum reconstruction unit with adaptive energy constraint;

[0073] A fault diagnosis module: based on the enhanced sideband features, combined with a rule-based method or a machine learning model for fault diagnosis.

[0074] The fault diagnosis module supports online real-time diagnosis and integrates a visualization interface for displaying the time-domain waveforms, spectrum comparison before and after fault feature enhancement, and diagnosis results.

[0075] Compared with the prior art, the beneficial effects of the present invention are:

[0076] The present invention adopts an adaptive filtering sideband enhancement algorithm, which can adaptively identify and enhance the sideband features in the motor power signal in a dynamically changing signal, thereby effectively improving the reliability and accuracy of fault diagnosis. Through this algorithm, it is possible to effectively cope with a complex noise environment, extract more discriminative frequency-domain features, and provide more accurate data support for fault diagnosis. Description of the Drawings

[0077] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only the preferred embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0078] Figure 1 It is a schematic diagram of the overall process of a GIS disconnector mechanical fault diagnosis method based on adaptive filtering proposed in the embodiments of the present invention.

[0079] Figure 2 It is a schematic diagram for signal processing comparison in a mechanical fault diagnosis method for GIS disconnectors based on adaptive filtering proposed in an embodiment of the present invention.

[0080] Figure 3 It is a schematic diagram of the overall structure of a mechanical fault diagnosis system for GIS disconnectors based on adaptive filtering proposed in an embodiment of the present invention. Detailed implementation manners

[0081] In order to make the objectives, technical solutions and advantages of the present invention more apparent, exemplary embodiments according to the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments of the present invention. It should be understood that the present invention is not limited by the exemplary embodiments described herein. Based on the embodiments of the present invention described herein, all other embodiments obtained by those skilled in the art without creative efforts shall fall within the protection scope of the present invention.

[0082] In the following description, numerous specific details are given to provide a more thorough understanding of the present invention. However, it is obvious to those skilled in the art that the present invention may be implemented without one or more of these details. In other instances, some well-known technical features are not described in order to avoid confusion with the present invention.

[0083] It should be understood that the present invention can be implemented in different forms and should not be construed as limited to the embodiments presented herein. On the contrary, providing these embodiments will make the disclosure thorough and complete, and will fully convey the scope of the present invention to those skilled in the art.

[0084] The purpose of the terms used herein is only to describe specific embodiments and is not a limitation of the present invention. When used herein, the singular forms "a", "an" and "the" are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms "comprising" and / or "including", when used in this specification, determine the presence of the described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups. When used herein, the term "and / or" includes any and all combinations of the related listed items.

[0085] To thoroughly understand the present invention, detailed structures will be presented in the following description to illustrate the technical solutions proposed by the present invention. The optional embodiments of the present invention are described in detail below. However, in addition to these detailed descriptions, the present invention may also have other implementation manners.

[0086] Example 1

[0087] Refer to Figure 1 and Figure 2 , a mechanical fault diagnosis method for GIS disconnectors based on adaptive filtering, the method comprising the following steps:

[0088] S1. Data acquisition and preprocessing: Obtain the power signal of the driving motor of the GIS disconnector through a sensor, and perform denoising, normalization processing, and general frequency analysis;

[0089] S2. Multi-scale distribution modulation decoupling enhancement (MSDMDE) processing: Adopt steps of multi-scale adaptive modulation decomposition (MS-AMD), non-linear generalized inverse modulation compensation (NL-GIMC), dynamic feature adaptive spectrum enhancement (DFASE), and adaptive energy-constrained variational spectrum reconstruction (AEVSR) to extract and enhance the sideband features of the preprocessed signal;

[0090] S3. Fault diagnosis: Based on the enhanced sideband features, perform fault diagnosis in combination with a rule-based method or a machine learning model.

[0091] Exemplarily, in step S1, standard data acquisition and preprocessing techniques are adopted to ensure the quality of the input signal and provide a reliable data basis for subsequent sideband feature enhancement. Signal acquisition can use existing sensors such as current transformers, power measurement modules, etc. to obtain the power signal of the driving motor of the GIS disconnector. The sampling frequency is matched with the operating characteristics of the equipment to ensure that key spectral information is not lost. For signal denoising, since the acquired signal is usually affected by environmental noise, electromagnetic interference, etc., conventional signal processing methods such as low-pass filtering, wavelet denoising, etc. are used to basically clean the signal to improve the signal-to-noise ratio. For signal normalization, to ensure the comparability of signals under different devices and different working conditions, the power signal is normalized or standardized so that its amplitude is within a certain range and the DC bias component is removed. For preliminary spectral analysis, through preliminary FFT analysis, the basic spectral structure of the signal is observed to provide reference information for sideband feature enhancement.

[0092] The multi-scale adaptive modulation decomposition (MS-AMD) step includes:

[0093] Decompose the power signal using a multi-scale joint modulation projection operator to extract the modulation envelope and phase information under different operating states, specifically:

[0094] Use a multi-scale joint modulation projection operator to decompose the signal:

[0095]

[0096] where A k(t) is the modulation envelope of the k-th level, corresponding to different sideband components; θ k (t) is the instantaneous phase, and n(t) is the noise, defined as:

[0097]

[0098] ω k ω(τ) represents the time-dependent angular frequency, τ is the time parameter, which is a dynamic representation of t, and d is the differential operator. To ensure accurate decomposition at different scales, the time-scale joint modulation projection operator is introduced:

[0099]

[0100] where W k (t) is optimized by the maximum energy criterion:

[0101]

[0102] By optimizing the weights, the sideband features in a specific state are strengthened.

[0103] Exemplarily, during the operation of a GIS disconnector, different states (closing, opening, contact jitter, etc.) correspond to different sideband features. These sidebands are usually composed of non-stationary modulation signals, and it is difficult for traditional single-scale decomposition methods to simultaneously analyze the sideband characteristics in different states. Therefore, we use the multi-scale joint modulation projection operator to decompose the signal:

[0104]

[0105] where A k (t) is the modulation envelope of the k-th level, corresponding to different sideband components; θ k (t) is the instantaneous phase, defined as:

[0106]

[0107] Here, ω k (τ) represents the time-dependent angular frequency.

[0108] By decomposing the power signal, we can separately extract the modulation modes in different operating states. For example, during the closing process, the low-frequency sideband is more obvious, while during the contact jitter stage, the high-frequency sideband dominates the signal characteristics. Therefore, using the multi-scale method can accurately identify the modulation effects in different states and improve the diagnostic ability. Further optimization: To ensure accurate decomposition at different scales, we introduce the time-scale joint modulation projection operator:

[0109]

[0110] Among them, W is the weight matrix or projection coefficient matrix in the time-scale domain. Its core role is to select or enhance the most significant components of the signal in the time-scale domain while suppressing noise or irrelevant components. W k (t) is optimized by the maximum energy criterion:

[0111]

[0112] By optimizing the weights, we can extract the main modulation components of the target state to the greatest extent and further enhance the sideband features under specific states.

[0113] The non-linear generalized inverse modulation compensation (NL-GIMC) step includes:

[0114] Use the generalized inverse transform operator to perform inverse modulation compensation on the decomposed signal to eliminate non-linear phase distortion. Specifically:

[0115] Use the calculation formula:

[0116]

[0117] Among them: the generalized inverse transform operator Use non-linear mapping:

[0118]

[0119] Among them, λ is the scale parameter adjusted according to experience. To avoid misjudgment of the decomposed signal, Approximately represent the modulation mode of the original signal. By dividing the numerator by the denominator, the modulation components in the power signal are separated to retain the sideband components related to faults.

[0120] Exemplarily, after the envelope is decomposed, since the motor power signal usually has non-linear phase distortion, we must perform inverse modulation compensation. Traditional methods usually directly use envelope normalization, but in the non-linear case, this method is prone to misjudgment. Therefore, we use:

[0121]

[0122] Among them: the generalized inverse transform operator Use non-linear mapping:

[0123]

[0124] In the process of mechanical fault diagnosis of GIS disconnectors, multi-scale adaptive modulation decomposition (MS-AMD) and non-linear generalized inverse modulation compensation (NL-GIMC) are two closely related steps that jointly act on the extraction and enhancement of the sideband features of the power signal P(t). The goal is to gradually remove the non-linear distortion in the signal and enhance the sideband components that can be used for fault discrimination.

[0125] It realizes the further optimization of the MS-AMD results. Through inverse modulation compensation, the non-linear distortion in the signal is eliminated.

[0126] Due to the complex actual operating environment of GIS disconnectors, the motor drive process may be affected by electromagnetic interference, transient fluctuations, etc., resulting in the distortion of sideband features. If the signal decomposed by MS-AMD is directly used for diagnosis, misjudgment may be introduced. Specifically:

[0127] First use to approximately represent the modulation mode of the original signal. Then, by dividing the numerator by the denominator, the modulation components in the power signal are separated, so that only the sideband components related to faults are retained on the left side of the work. The generalized inverse transform operator ensures that the finally restored signal will not lose key information due to non-linear distortion.

[0128] The role and goal of MS-AMD:

[0129] Since the power signal P(t) of the drive motor of GIS disconnectors is affected by complex modulation effects in different states (such as closing, opening, contact jitter, etc.), it exhibits multi-scale sideband features. Therefore, directly analyzing P(t) may cause key features to be masked by noise or other irrelevant components.

[0130] MS-AMD decomposes P(t) into modulation components of multiple scales through adaptive modulation decomposition, that is, a series of signal components with different modulation envelopes Ak(t) and instantaneous phases θk(t). These components respectively correspond to the characteristic signals in different operating states, enabling us to independently analyze the characteristics of each state.

[0131] In this way, MS-AMD decomposes the complex signal into multiple interpretable components and lays the foundation for the next modulation compensation.

[0132] The role and goal of NL-GIMC:

[0133] Although MS-AMD has decomposed the modulation envelope Ak(t) and the instantaneous phase θk(t), due to the non-linear distortion problem in the actual signal, simply decomposing is not sufficient to ensure the reliability of the sideband features.

[0134] NL-GIMC eliminates the nonlinear phase distortion through generalized inverse modulation compensation, further restoring the clean sideband features and avoiding the influence of unnecessary phase and amplitude changes in the signal on the diagnostic results.

[0135] Intuitively, MS-AMD is responsible for decomposing the signal to make it clearer, while NL-GIMC further optimizes the decomposed signal to make it more accurate.

[0136] Specifically, NL-GIMC adopts a generalized inverse transform operator to perform inverse modulation compensation on the signal components obtained by MS-AMD, ensuring that the decomposed signal does not generate incorrect sideband features due to nonlinear effects.

[0137] In other words, MS-AMD is the process of feature separation, while NL-GIMC is the process of feature optimization. Without MS-AMD, we cannot accurately separate the sideband features of different states; without NL-GIMC, we cannot ensure that the decomposed signal is not affected by nonlinear distortion.

[0138] The dynamic feature adaptive spectrum enhancement (DFASE) step includes:

[0139] Calculating the short-time Fourier transform of the input signal, designing an adaptive gain operator to dynamically adjust the gain, and restoring the enhanced time-domain signal through the inverse short-time Fourier transform;

[0140] Calculating the short-time Fourier transform of the input signal, specifically:

[0141] By performing the short-time Fourier transform STFT on the compensated signal, obtaining the time-frequency representation of the signal:

[0142]

[0143] where, P comp (t) is the signal after nonlinear generalized inverse modulation compensation processing; w(τ - t) is a short-time window function used to localize the time-frequency characteristics of the signal; f is the frequency variable, t is the time variable, and X(f, t) is the result of the short-time Fourier transform, representing the spectral distribution of the signal at different time points;

[0144] Designing an adaptive gain operator to dynamically adjust the gain, specifically:

[0145] By introducing a time-varying gain operator:

[0146] H(f, t) = (1 + βf m e -t / τ )H0(f)(0.39)

[0147] where: β is the gain intensity control parameter, m is the frequency amplification index, τ is the time scale parameter, and H0(f) is the initial filtering operator.

[0148] Calculate the enhanced spectrum:

[0149] The enhanced signal spectrum X'(f,t) is given by:

[0150] X′(f,t) = H(f,t)·X(f,t)(0.40)

[0151] where H(f,t) acts on the original spectrum to dynamically amplify the target sideband signal;

[0152] Restore the enhanced time-domain signal through the inverse short-time Fourier transform, specifically:

[0153] Transform the enhanced time-frequency signal back to the time domain through the inverse short-time Fourier transform to obtain the enhanced time signal:

[0154]

[0155] Exemplarily, GIS disconnectors exhibit different sideband characteristics in different states (such as closing, opening, contact jitter). These sideband characteristics are usually affected by noise, making it difficult to directly extract these characteristics from the power signal P(t). Therefore, after the first two steps of multi-scale adaptive modulation decomposition (MS-AMD) and non-linear generalized inverse modulation compensation (NL-GIMC), we still need to further optimize the identifiability of the sideband signal to ensure its applicability to actual fault diagnosis.

[0156] DFASE mainly consists of three parts:

[0157] 1. Calculate the short-time Fourier transform (STFT) of the input signal to extract the time-frequency characteristics of the sideband signal.

[0158] 2. Design an adaptive gain operator H(f,t) to dynamically adjust the gain according to time and frequency, so that the target sideband signal is enhanced while the interference at other frequencies is suppressed.

[0159] 3. Restore the enhanced time-domain signal Penhanced(t) through the inverse STFT (ISTFT).

[0160] Short-time Fourier transform (STFT)

[0161] First, we perform the short-time Fourier transform (STFT) on the compensated signal to obtain the time-frequency representation of the signal:

[0162]

[0163] where, Pcomp (t) is the signal after NL-GIMC processing; w(τ - t) is a short-time window function (such as Hanning window or Gaussian window) used to localize the time-frequency characteristics of the signal; f is the frequency variable and t is the time variable. X(f, t) is the result of the short-time Fourier transform, representing the spectral distribution of the signal at different time points. This transform converts the time-domain signal to the time-frequency domain, making the spectral energy distribution of the sideband signal visible at different time points, facilitating subsequent enhancement processing.

[0164] Design a dynamic adaptive gain operator

[0165] In the time-frequency domain, we hope to adaptively enhance the sideband features under different states (closing, opening, jitter, etc.). Therefore, we introduce a time-varying gain operator:

[0166] H(f, t) = (1 + βf m e -t / τ )H0(f)

[0167] where: β is the gain intensity control parameter (determining the magnitude of the enhancement). m is the frequency amplification exponent (controlling the enhancement degree of the high-frequency part). τ is the time-scale parameter (controlling the variation of the gain with time). H0(f) is the initial filtering operator, usually a band-pass filter, used to preliminarily screen the sideband frequency range. This formula ensures that the key sideband signals are enhanced while other frequency components are not over-amplified. In particular, the gain is time-dependent and it changes gradually over time, thus better adapting to the sideband features under different states of the GIS disconnector. For example, at the moment of closing, some sideband features may be stronger, while during contact jitter, other sideband features may dominate the signal. This gain operator can be dynamically adjusted to ensure that signals under different states can be enhanced.

[0168] Calculate the enhanced spectrum

[0169] The enhanced signal spectrum X'(f, t) is given by:

[0170] X′(f, t) = H(f, t)·X(f, t)

[0171] where: H(f, t) acts on the original spectrum, dynamically amplifying the target sideband signal; this step ensures that the sideband signals corresponding to specific fault states are maximally enhanced, thereby improving the accuracy of subsequent fault diagnosis.

[0172] For example, in the state of contact jitter, some high-frequency sideband signals may be weak, while during closing, the low-frequency sidebands are weak. Through this enhancement strategy, we can better highlight the key sideband components, enabling the diagnostic model to more clearly distinguish different states.

[0173] Inverse Short-Time Fourier Transform (ISTFT)

[0174] Finally, we transform the enhanced time-frequency signal back to the time domain through the Inverse Short-Time Fourier Transform (ISTFT) to obtain the enhanced time signal:

[0175]

[0176] The steps of the variational spectrum reconstruction with adaptive energy constraint include:

[0177] Perform Fourier transform on the enhanced signal, set the variational optimization objective, and iteratively solve it through the gradient descent method to optimize the energy distribution and resolution of the sideband features, specifically:

[0178] Perform Fourier transform on the enhanced signal to obtain the spectrum before optimization:

[0179]

[0180] Make the sideband features clearer through the optimization formula 1.11, while ensuring that the energy of the entire signal is consistent with the original signal:

[0181]

[0182] where E represents the total energy of the signal;

[0183] Set the variational optimization objective:

[0184]

[0185] ρ is the regularization coefficient, used to adjust the balance between energy constraint and smoothing constraint;

[0186] Adopt the gradient descent method for iterative solution:

[0187]

[0188] where the gradient of the loss function:

[0189]

[0190] where J is the loss function and η is the learning rate, controlling the update step size.

[0191] Exemplarily, different states of the GIS disconnector will cause changes in the energy distribution of the sideband characteristics of the driving motor power signal P(t) in the frequency spectrum. Under the processing of multi-scale adaptive modulation decomposition (MS-AMD) and non-linear generalized inverse modulation compensation (NL-GIMC), we have successfully extracted the key sideband signals in different states and enhanced them specifically through dynamic feature adaptive spectrum enhancement (DFASE). However, in practical applications, due to noise, signal loss, and the complexity of equipment operating conditions, the enhanced sideband signals may still be affected by uneven energy distribution and insufficient resolution. This will result in unclear sideband signal morphology during fault discrimination, thus reducing the accuracy of diagnosis.

[0192] Adaptive energy-constrained variational spectrum reconstruction (AEVSR) aims to further optimize the spectral characteristics of the signal, make the sideband energy distribution more stable, the spectrum sharper, improve the feature contrast, and make the identification of fault features more accurate.

[0193] First, we perform a Fourier transform on the enhanced signal to obtain the spectrum before optimization:

[0194]

[0195] We hope to optimize Equation 1.11 to make the sideband features clearer while ensuring that the energy of the entire signal remains the same as the original signal:

[0196]

[0197] where E represents the total energy of the signal; this constraint ensures that the optimized spectrum will not cause an imbalance in the energy distribution of the signal due to over-enhancement or compression.

[0198] For optimization, we set the following variational optimization objective:

[0199]

[0200] The first term on the right side of Equation 1.13 ensures that the energy distribution of the optimized signal is consistent with the target energy; the second term introduces high-order derivative regularization to prevent excessive oscillation of the spectrum and make the sideband features smoother; ρ is the regularization coefficient used to adjust the balance between energy constraint and smoothing constraint. This optimization objective ensures that the sideband features will not be distorted due to over-amplification or filtering; the introduction of the high-order derivative term ensures that there will be no violent fluctuations in the spectrum, thus improving the resolution and diagnostic reliability of the signal.

[0201] To solve the above optimization problem, we use the gradient descent method for iterative solution:

[0202]

[0203] Among them, the loss function gradient:

[0204]

[0205] η is the learning rate, which controls the update step size. This iterative process can adaptively adjust the spectrum to make it more compliant with the energy constraint and improve the resolution of fault features. The optimized signal is obtained through the inverse Fourier transform (same as Equation 1.10).

[0206] Time-domain analysis: The original signal contains the main signal component and multiple sideband components. At the same time, it is affected by noise and external interference, and the overall waveform is relatively complex, making it difficult to directly extract features. Some features are masked by noise due to their low amplitude, resulting in a low overall contrast of the signal.

[0207] Frequency-domain analysis: In the spectrum, several main frequency components can be seen. However, due to the large noise, some features are relatively blurred, especially the weaker sideband information, which is covered by the background noise, making it difficult to directly analyze the signal state.

[0208] After multi-scale adaptive modulation decomposition

[0209] Time-domain analysis: After the decomposition process, the overall shape of the signal becomes smoother, and it is possible to clearly see the changes in certain fluctuation patterns. The different components of the signal are split, enabling the hidden sideband features to be better observed at different time scales.

[0210] Frequency-domain analysis: Compared with the original signal, the clarity of some sideband features has been improved, and the background noise has been weakened. However, certain non-linear effects can still be seen, resulting in slight distortion of some features.

[0211] After non-linear generalized inverse modulation compensation

[0212] Time-domain analysis: After the compensation process, the fluctuations of the signal become more stable, avoiding abnormal amplitude changes caused by non-linear effects during the operation of the device. Compared with the previous processing results, the signal shows more regular performance during key time periods, and the amplitude changes tend to be smoother.

[0213] Frequency-domain analysis: The sideband features become more symmetric in the spectrum, indicating that the non-linear distortion of the signal has been effectively suppressed. The main signal component is more concentrated, and the originally scattered low-frequency interference has been weakened, making the features in the sideband part more clearly visible.

[0214] After dynamic feature adaptive spectrum enhancement

[0215] Time-domain analysis: The waveform of the signal is further enhanced at key positions, and the originally weak sideband components become more prominent in the time domain. The overall contrast of the signal increases, and the amplitudes of the sideband features become more significant, making them easier to identify in subsequent analyses.

[0216] Frequency-domain analysis: The amplitudes of the sideband features are significantly increased, the main signal part remains stable, and at the same time, the energy of non-target components is appropriately weakened, enhancing the contrast of the target sidebands. The energy distribution between sidebands is more balanced and is no longer affected by random noise.

[0217] The rule-based method includes:

[0218] Setting thresholds based on sideband amplitude ratio, instantaneous frequency change rate, or power spectral entropy to determine poor contact, contact jitter, or abnormal load faults.

[0219] Exemplarily, for some typical fault modes, we can directly perform fault determination based on empirical rules.

[0220] For example:

[0221] If the sideband amplitude ratio exceeds a certain threshold (e.g., 3%), it can be considered that there is poor contact in the GIS disconnector;

[0222] If the change rate of the instantaneous frequency exceeds a certain threshold (e.g., 5 Hz / ms), it may be contact jitter;

[0223] If the power spectral entropy is much higher than the normal level, it may mean abnormal load of the drive motor.

[0224] This type of method is applicable to scenarios where the fault mode is relatively clear and the signal feature distribution is relatively fixed.

[0225] The machine learning model includes Support Vector Machine (SVM), Random Forest (RF), Long Short-Term Memory Network (LSTM), or Convolutional Neural Network (CNN).

[0226] Exemplarily, when the fault mode is relatively complex, or there is a large overlap in the features of different faults, using data-driven methods can improve the accuracy of diagnosis. Common methods include:

[0227] Support Vector Machine (SVM): Suitable for fault classification in the case of small samples, especially suitable for cases where there are obvious boundaries between sideband features;

[0228] Random Forest (RF): Can automatically select the most discriminative features and construct multiple decision trees,

[0229] Improving the robustness of diagnosis;

[0230] Long Short-Term Memory Network (LSTM): Suitable for time series signal analysis, it can capture the dynamic change characteristics during the operation of GIS disconnectors;

[0231] Convolutional Neural Network (CNN): It can directly input time-frequency images (such as images after STFT or wavelet transform), and automatically extract features and classify through deep learning.

[0232] Embodiment 2

[0233] Refer to Figure 3 , a GIS disconnector mechanical fault diagnosis system based on adaptive filtering, the system is used to execute the GIS disconnector mechanical fault diagnosis method based on adaptive filtering, and the system includes:

[0234] Data acquisition module: Used to acquire the power signal of the GIS disconnector drive motor;

[0235] Data preprocessing module: Used to perform denoising, normalization processing and general frequency analysis on the acquired signal;

[0236] Feature enhancement module: Adopt multi-scale distribution modulation decoupling enhancement processing, including multi-scale adaptive modulation decomposition, non-linear generalized inverse modulation compensation, dynamic feature adaptive spectrum enhancement and variational spectrum reconstruction unit with adaptive energy constraint;

[0237] Fault diagnosis module: Based on the enhanced sideband features, combine the rule-based method or machine learning model to perform fault diagnosis.

[0238] The fault diagnosis module supports online real-time diagnosis and integrates a visualization interface for displaying the time-domain waveforms, spectrum comparison before and after fault feature enhancement, and diagnosis results.

[0239] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A mechanical fault diagnosis method for GIS disconnectors based on adaptive filtering, characterized in that, The method includes the following steps: S1. Data acquisition and preprocessing: Obtain the power signal of the GIS disconnector drive motor through a sensor, and perform denoising, normalization processing, and general frequency analysis; S2. Multi-scale distribution modulation decoupling and enhancement processing: Adopt steps of multi-scale adaptive modulation decomposition, non-linear generalized inverse modulation compensation, dynamic feature adaptive spectrum enhancement, and variational spectrum reconstruction with adaptive energy constraint to extract and enhance the sideband features of the preprocessed signal; S3. Fault diagnosis: Based on the enhanced sideband features, perform fault diagnosis in combination with the rule-based method or machine learning model.

2. The method according to claim 1, characterized in that, The multi-scale adaptive modulation decomposition step includes: Decompose the power signal using the multi-scale joint modulation projection operator to extract the modulation envelope and phase information under different operating states, specifically: Use the multi-scale joint modulation projection operator to decompose the signal: where A k (t) is the modulation envelope of the k-th level, corresponding to different sideband components; θ k (t) is the instantaneous phase, and n(t) is the noise, defined as: ω k (τ) represents the time-dependent angular frequency, where τ is the time parameter and is a dynamic representation of t, and d is the differential operator. To ensure accurate decomposition at different scales, introduce the time-scale joint modulation projection operator: Among them, W k (t) is optimized by the maximum energy criterion: Optimize the weights to strengthen the sideband features under specific states.

3. The method according to claim 2, wherein The non-linear generalized inverse modulation compensation step includes: Perform inverse modulation compensation on the decomposed signal using the generalized inverse transform operator to eliminate non-linear phase distortion, specifically: Use the calculation formula: Among them: the generalized inverse transformation operator Adopt a non-linear mapping: Among them, λ is a scale parameter adjusted according to experience. To avoid misjudging the decomposed signal, is approximately used to represent the modulation mode of the original signal. By dividing the numerator by the denominator, the modulation components in the power signal are separated to retain the sideband components related to faults.

4. The method according to claim 3, wherein The dynamic feature adaptive spectrum enhancement step includes: Calculate the short-time Fourier transform of the input signal, design an adaptive gain operator to dynamically adjust the gain, and restore the enhanced time-domain signal through the inverse short-time Fourier transform; Calculate the short-time Fourier transform of the input signal, specifically: Obtain the time-frequency representation of the signal by performing the short-time Fourier transform STFT on the compensated signal; Among them, P comp (t) is the signal after non-linear generalized inverse modulation compensation processing; w(τ - t) is a short-time window function used to localize the time-frequency characteristics of the signal; f is the frequency variable, t is the time variable, and X(f, t) is the result of the short-time Fourier transform, representing the spectral distribution of the signal at different time points; Design an adaptive gain operator to dynamically adjust the gain, specifically: By introducing the time-transform gain operator: H(f,t) = (1 + βf m e -t / τ )H0(f)(0.8) Where: β is the gain intensity control parameter, m is the frequency amplification exponent, τ is the time-scale parameter, and H0(f) is the initial filtering operator. Calculate the enhanced spectrum: The enhanced signal spectrum X'(f,t) is given by: X′(f,t) = H(f,t)·X(f,t)(0.9) where H(f,t) acts on the original spectrum to dynamically amplify the target sideband signal; Restore the enhanced time-domain signal through the inverse short-time Fourier transform, specifically: Transform the enhanced time-frequency signal back to the time domain through the inverse short-time Fourier transform to obtain the enhanced time signal:

5. The method according to claim 4, characterized in that The variational spectrum reconstruction step with adaptive energy constraint includes: Perform Fourier transform on the enhanced signal, set the variational optimization objective, and iteratively solve through the gradient descent method to optimize the energy distribution and resolution of the sideband features, specifically: Perform Fourier transform on the enhanced signal to obtain the spectrum before optimization: Make the sideband features clearer through optimizing formula 1.11 while ensuring that the total energy of the entire signal is consistent with the original signal: Where E represents the total energy of the signal; Set the variational optimization objective: ρ is the regularization coefficient used to adjust the balance between energy constraint and smoothing constraint; Perform iterative solution using the gradient descent method: Where the gradient of the loss function: Where J is the loss function and η is the learning rate to control the update step size.

6. The method according to claim 5, wherein The rule-based method includes: Set thresholds based on sideband amplitude ratio, instantaneous frequency change rate, or power spectrum entropy to determine faults such as poor contact, contact jitter, or abnormal load.

7. The diagnostic method according to claim 6, characterized in that The machine learning model includes Support Vector Machine (SVM), Random Forest (RF), Long Short-Term Memory Network (LSTM), or Convolutional Neural Network (CNN).

8. A mechanical fault diagnosis system for GIS disconnectors based on adaptive filtering, characterized in that The system is used to execute the method according to any one of claims 1-7, and the system includes: Data acquisition module: used to acquire the power signal of the GIS disconnector drive motor; Data preprocessing module: used to perform denoising, normalization processing, and frequency analysis on the acquired signal; Feature enhancement module: adopts multi-scale distribution modulation decoupling enhancement processing, including multi-scale adaptive modulation decomposition, non-linear generalized inverse modulation compensation, dynamic feature adaptive spectrum enhancement, and variational spectrum reconstruction unit with adaptive energy constraint; Fault diagnosis module: based on the enhanced sideband features, combines rule-based methods or machine learning models for fault diagnosis.

9. The system according to claim 8, wherein The fault diagnosis module supports online real-time diagnosis and integrates a visualization interface for displaying the time-domain waveforms, spectrum comparison before and after fault feature enhancement, and diagnosis results.

Images