A method and system for identifying contact wear of disconnecting switches

Abstract

This invention discloses a method and system for identifying contact wear in disconnecting switches, relating to the field of power equipment fault diagnosis technology. The method for identifying contact wear in disconnecting switches includes the following steps: during the operation of the disconnecting switch, multi-source sensor data is collected, and the collected raw data is preprocessed to obtain a standardized multi-dimensional time-series monitoring dataset; based on the standardized multi-dimensional time-series monitoring dataset, wear-sensitive feature vectors characterizing the contact wear state are extracted; a contact wear situational awareness model is constructed, and the current contact wear state is identified based on the wear-sensitive feature vectors, and the wear trend is predicted; based on the wear trend prediction results, the remaining contact life is assessed and graded early warning is achieved. This invention, by extracting wear-sensitive feature vectors, quantifies the complex physical process of contact wear into trajectory points in a multi-dimensional feature space, enabling subtle changes in the wear state to be effectively characterized and quantified.

Description

Technical Field

[0001] This invention relates to the field of power equipment fault diagnosis technology, and in particular to a method and system for identifying wear on disconnector switch contacts. Background Technology

[0002] As a crucial switching device in power systems, the health of the contact system of disconnecting switches directly affects the operational safety and power supply reliability. Statistics show that overheating and burnout caused by poor contact in disconnecting switches has become one of the main causes of power equipment failures.

[0003] Currently, the detection and identification of contact wear in disconnecting switches mainly rely on the following technical means: First, periodic inspections during offline maintenance, which involve manually observing the contact surface and measuring contact resistance to determine the degree of wear. However, this method requires power outages, has a long detection cycle, and is difficult to detect sudden increases in wear in a timely manner. Second, online monitoring based on a single physical parameter, such as infrared thermography to monitor temperature rise and vibration sensors to monitor the impact of opening and closing. However, single parameters are easily affected by environmental interference, resulting in a high false alarm rate. Third, vibration signal analysis, which involves extracting the specific frequency amplitude ratio of the vibration signal from the GIS housing to determine the contact state. However, this method is limited to GIS equipment and has a single parameter. Fourth, multi-source information fusion technology, which inputs parameters such as contact temperature, contact resistance, and micro-displacement into an AI model for intelligent diagnosis. However, existing technologies are still limited to "point-like judgments" of the current state and fail to achieve dynamic perception of wear trends and prediction of remaining life.

[0004] Therefore, we urgently need to develop a method and system for identifying disconnector contact wear by enabling full lifecycle situational awareness, trend prediction, and early warning. Summary of the Invention

[0005] This invention aims to achieve full lifecycle situational awareness and hierarchical early warning of disconnector switch contact wear status through multi-source sensor data fusion and a dual-model driven architecture. This addresses problems in existing technologies such as single monitoring parameters, lack of trend prediction capabilities, significant environmental interference, inability to distinguish wear sources, and difficulty in extracting early, weak features. To achieve the above objectives, the technical solution adopted by this invention is as follows: A method for identifying contact wear of a disconnector switch includes the following steps: During the operation of the disconnecting switch, multi-source sensor data is collected, and the collected raw data is preprocessed to obtain a standardized multi-dimensional time-series monitoring dataset. Based on the standardized multi-dimensional time-series monitoring dataset, a wear-sensitive feature vector characterizing the wear state of the contact is extracted; A contact wear situation awareness model is constructed to identify the current contact wear state based on the wear-sensitive feature vector and predict the wear trend. Based on the wear trend prediction results, the remaining life of the contacts can be assessed and classified for early warning.

[0006] As a preferred embodiment of the present invention, the multi-source sensing data includes: vibration signals collected by an accelerometer, motor current signals collected by a current sensor, contact surface image signals collected by an image acquisition device, contact resistance signals collected by a micro-resistance meter, and environmental parameter signals collected by an environmental parameter sensor.

[0007] As a preferred embodiment of the present invention, the preprocessing includes: The vibration signal is subjected to detrending, bandpass filtering, and wavelet thresholding for noise reduction. The current signal is processed by removing the DC component and applying a moving average filter. The image data is processed by distortion correction, illumination normalization, and region of interest cropping. The contact resistance signal is denoised and the contact stability is verified to eliminate abnormal data points caused by contact jitter or poor contact in the measurement circuit. Outlier removal and linear interpolation are performed on environmental parameters.

[0008] As a preferred embodiment of the present invention, the wear-sensitive feature vector includes: Vibration impact features extracted from vibration signals include high-frequency detail energy extracted after wavelet packet decomposition of the vibration signal, as well as the maximum amplitude, root mean square value, and impulse factor obtained from time-domain analysis. Load variation features extracted from motor current signals, including peak current, rise time, and current fluctuation coefficient extracted after segmenting and identifying the motor current waveform; The coating appearance features extracted from the contact surface image include the average chromaticity and chromaticity standard deviation extracted after HSV space transformation of the contact image, and the percentage of coating peeling area obtained by chromaticity threshold segmentation. Electrical contact features extracted from preprocessed contact resistance signals, wherein the electrical contact features are contact resistance values ​​measured by a micro resistance meter under no-load conditions when the circuit is closed; Environmental compensation factors extracted from environmental parameters, including temperature, relative humidity, and pollution level; Together with the cumulative number of operations, they form a multidimensional feature vector.

[0009] As a preferred technical solution of the present invention, the contact wear situation perception model includes a physical model, which is constructed based on Hertz contact theory and Holm contact resistance theory, and establishes a quantitative relationship between contact resistance and contact material resistivity, contact spot equivalent radius, surface film tunneling resistance coefficient and effective conductive area. The cumulative number of operations is introduced as a wear process variable, and an evolution model of the contact resistance changing with the number of operations is established. The model parameters are calibrated by friction and wear tests. The physical model outputs a theoretical contact resistance value predicted based on the current number of operations, and an equivalent remaining coating thickness derived from the measured contact resistance.

[0010] As a preferred technical solution of the present invention, the contact wear situation perception model includes a data model, which is constructed using a CNN-LSTM fusion neural network architecture and is trained using multiple sets of wear-sensitive feature vectors and corresponding wear state labels collected during historical operation as training sample sets. The CNN part is used to extract the local coupling relationship between the components within the feature vector, and the LSTM part is used to capture the temporal dependency relationship of the feature vector as it evolves over time. The model input is a sequence of feature vectors from multiple consecutive operation moments, and the output is the wear state classification result and wear degree quantification value at the current moment. The wear status classification includes four levels: normal, slight wear, moderate wear, and severe wear. The wear degree quantification value is a continuous value between 0 and 100%.

[0011] As a preferred embodiment of the present invention, the contact wear situation perception model includes a fusion diagnostic mechanism: Input the feature vector at the current moment into the physical model to obtain the equivalent remaining coating thickness and the predicted theoretical contact resistance; Input a continuous feature vector sequence, including the current time, into the data model to obtain the wear state classification result and the wear degree quantification value; When the deviation between the wear degree quantification value output by the data model and the wear degree value calculated by the physical model based on the equivalent remaining coating thickness is less than a preset threshold, it is determined that the outputs of the two models are consistent, and the weighted average of the two is taken as the final wear degree assessment result. When the deviation exceeds the preset threshold, the deep diagnosis mode is triggered, and the current feature vector is matched with the historical fault case library for similarity. The most similar fault case and its subsequent evolution path are retrieved, and diagnostic suggestions and warning information are output.

[0012] As a preferred embodiment of the present invention, the wear trend prediction includes: A wear trend curve is constructed based on the historical wear degree sequence. The wear degree sequence is then subjected to three exponential smoothing processes to remove short-term fluctuations and obtain a monotonically increasing curve that reflects the long-term wear trend. The gray prediction model GM(1,1) or the ARIMA time series model is used to extrapolate and predict the trend curve to obtain the predicted wear level under the future number of operations. The remaining life assessment includes: setting a severe wear threshold, calculating the remaining number of operations required to reach the severe wear threshold based on the current wear level and the slope of the wear trend curve, and using this as a quantitative value of the remaining life.

[0013] As a preferred embodiment of the present invention, the graded early warning includes four warning levels based on the degree of wear and remaining life: When the wear level is greater than or equal to 30% and the remaining number of operations is greater than or equal to 2000, a warning level alert is issued, prompting an increase in the frequency of monitoring. When the wear level is greater than or equal to 50% and the remaining number of operations is greater than or equal to 500, an early warning level warning will be issued to prompt the scheduling of an upcoming maintenance plan. When the wear level is greater than or equal to 70% and the remaining number of operations is greater than or equal to 100, an alarm-level warning will be output, indicating that maintenance must be arranged during the next power outage window; When the wear level is greater than or equal to 85% or the remaining number of operations is less than 100, an emergency warning will be issued, prompting an immediate power outage for maintenance.

[0014] A disconnector switch contact wear identification system, comprising: The data acquisition module is used to collect multi-source sensor data during the operation of the disconnecting switch, including vibration signals collected by the accelerometer, motor current signals collected by the current sensor, contact surface image signals collected by the image acquisition device, contact resistance signals collected by the micro-resistance meter, and environmental parameter signals collected by the environmental parameter sensor. The preprocessing module receives multi-source sensor data collected by the data acquisition module, performs detrending, bandpass filtering and wavelet threshold denoising on the vibration signal, removes DC component and performs moving average filtering on the current signal, performs distortion correction, illumination normalization and region of interest cropping on the image data, performs denoising on the contact resistance signal and performs contact stability verification, removes abnormal data points caused by contact jitter or poor contact of the measurement circuit, and performs outlier removal and linear interpolation filling on the environmental parameters to obtain a standardized multi-dimensional time-series monitoring dataset. The feature extraction module, based on the standardized multi-dimensional time-series monitoring dataset output by the preprocessing module, extracts vibration impact features from vibration signals, load change features from motor current signals, plating appearance features from contact surface images, and electrical contact features from contact resistance measurements. The electrical contact features extracted from the preprocessed contact resistance signals are combined with compensation factors and cumulative operation counts in environmental parameters, and these feature parameters are fused in the feature layer to generate a wear-sensitive feature vector characterizing the wear state of the contact. The situational awareness module, with its built-in physical and data models, uses the wear-sensitive feature vector generated by the feature extraction module as input to identify the current wear state of the contact and predict the wear trend. The physical model is built upon Hertzian contact theory and Holm contact resistance theory, establishing a quantitative relationship between contact resistance and contact material resistivity, equivalent radius of the contact spot, surface film tunneling resistance coefficient, and effective conductive area. The data model employs a CNN-LSTM fusion neural network architecture, trained based on historical operating data. When the output deviation between the physical and data models is less than a preset threshold, the weighted average of the two is used as the evaluation result. When the deviation exceeds the threshold, a deep diagnostic mode is triggered, matching the current feature vector with a historical fault case database to output diagnostic suggestions and early warning information. The life assessment and early warning module calculates the remaining life of the contact based on the wear status identification results and wear trend prediction results output by the situational awareness module, and outputs graded early warning information according to the wear degree and remaining life. The graded early warning includes four levels: attention level, early warning level, alarm level and emergency level.

[0015] The present invention has the following advantages: This invention integrates multi-source sensor data such as vibration, current, images, contact resistance, and environmental parameters, and performs targeted preprocessing on various types of data to construct a multi-dimensional monitoring dataset that comprehensively reflects the operating status of the contact. This invention can capture early signs of the contact wear process from different physical dimensions, overcome the one-sidedness of single-parameter evaluation, and provide a rich and reliable input basis for subsequent wear status identification.

[0016] This invention extracts wear-sensitive feature vectors that include vibration and impact characteristics, load change characteristics, coating appearance characteristics, electrical contact characteristics, environmental compensation factors, and cumulative number of operations. This invention can quantify the complex physical process of contact wear into trajectory points in a multi-dimensional feature space, enabling subtle changes in wear state to be effectively characterized and quantified, and significantly improving the sensitivity and accuracy of state identification.

[0017] This invention constructs a contact wear situation awareness model that includes a physical model and a data model, and designs a fusion diagnostic mechanism. When the output deviation between the physical model and the data model is less than a preset threshold, the weighted average of the two is taken as the evaluation result. When the deviation exceeds the threshold, a deep diagnostic mode based on a historical fault case library is triggered. This invention achieves the complementary advantages of theoretical mechanism and data-driven approach, ensuring the physical interpretability of the evaluation results and enabling adaptive learning of complex wear patterns under different working conditions. At the same time, the fault case matching mechanism provides decision support for root cause analysis of abnormal states.

[0018] This invention constructs a wear trend curve based on historical wear degree sequences and uses a time series prediction model to extrapolate and predict the wear trend, thereby calculating the remaining number of operations required to reach the severe wear threshold and realizing a quantitative assessment of the remaining life of the contacts. This invention completes the leap from current state identification to future trend prediction, enabling the operation and maintenance strategy to shift from post-event handling to pre-event early warning, significantly improving the scientific nature and foresight of equipment operation and maintenance.

[0019] This invention classifies warning levels into four grades—attention, early warning, alarm, and emergency—based on wear and remaining lifespan, and matches specific thresholds and maintenance recommendations for each grade. This invention transforms complex lifespan assessment results into intuitive and operable operation and maintenance instructions, providing clear decision-making basis for on-site personnel, effectively avoiding the risks of over-maintenance or neglect, and optimizing the allocation of maintenance resources while ensuring the safe operation of the power grid. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only schematic diagrams of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort. Figure 1 This is a flowchart illustrating a method for identifying contact wear of a disconnector switch as used in an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of a disconnector contact wear identification system used in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0023] Example 1: A method for identifying contact wear of a disconnector switch, such as... Figure 1 As shown, it includes the following steps: Step S1: During the operation of the disconnecting switch, multi-source sensor data is collected, and the collected raw data is preprocessed to obtain a standardized multi-dimensional time-series monitoring dataset. The multi-source sensor data includes the following five types of signals: Vibration signal: Acquired by an accelerometer, which is installed on the stationary contact side of the disconnector or near the transmission mechanism to sense the vibration waveform generated by mechanical impact, friction, jamming, etc. during the contact opening and closing operation; The sampling frequency of the vibration signal should satisfy the Nyquist sampling theorem, and is usually set to 10kHz~50kHz to cover the high-frequency components of contact impact events. Motor current signal: Acquired by a current sensor, which is installed at the power input terminal of the motor operated by the disconnect switch to monitor load changes during the movement of the motor drive contacts; the current signal sampling frequency is usually set to 1kHz~20kHz to capture the transient characteristics of the current waveform. Contact surface image signal: Acquired by an image acquisition device, which is installed at a suitable position to capture the meshing area of ​​the moving and stationary contacts. The image is usually triggered after the circuit is closed. The image resolution should meet the requirements of subsequent image processing, generally not less than 640×480 pixels. Contact resistance signal: acquired by a micro resistance meter, which uses a four-wire Kelvin measurement method to eliminate the influence of lead resistance; the measurement circuit is connected to the terminals on both sides of the moving and stationary contacts, and the measurement is performed when the disconnecting switch is closed and no load current flows through, so as to obtain a static contact resistance value that can reflect the true state of the contact surface. Environmental parameter signals: collected by environmental parameter sensors, including temperature and humidity sensors; environmental parameters are used to compensate and correct other sensor data, eliminating the interference of environmental factors on wear condition assessment.

[0024] The collected raw data is preprocessed to obtain a standardized multi-dimensional time-series monitoring dataset. Differentiated preprocessing methods are used for different types of data, including: The vibration signal is subjected to detrending, bandpass filtering, and wavelet thresholding for noise reduction. Detrending: Subtracting the mean or linear trend term of the signal eliminates the influence of sensor zero-point drift; Bandpass filtering: Design a bandpass filter to filter out power frequency interference and high-frequency noise, while retaining the effective frequency band components related to the mechanical impact of the contacts.

[0025] Wavelet thresholding denoising: Wavelet decomposition is used to decompose the signal into detail coefficients and approximation coefficients at different scales. Thresholds are set on the detail coefficients for soft or hard thresholding to filter out noise components. Then, wavelet reconstruction is performed to obtain the denoised vibration signal. The principle of wavelet thresholding denoising is to utilize the difference in amplitude distribution between signal and noise in the wavelet domain. By thresholding, the wavelet coefficients corresponding to noise are suppressed, while the wavelet coefficients corresponding to the signal are retained, thereby achieving signal-to-noise separation.

[0026] The current signal is processed by removing the DC component and applying a moving average filter. Remove DC component: Subtract the steady-state average value of the initial segment of the current signal to eliminate the zero bias of the current sensor; Moving average filtering: A sliding window is used to average the current signal, smoothing high-frequency glitches in the current waveform and highlighting the macroscopic trend of current changes.

[0027] The image data is processed by distortion correction, illumination normalization, and region of interest cropping. Distortion correction: Based on camera calibration parameters, distortion correction is performed on the image to eliminate geometric distortion caused by lens distortion; Illumination normalization: Histogram equalization and other methods are used to adjust the brightness distribution of the image and eliminate the influence of changes in illumination conditions during different shooting. Region of Interest (ROI) Cropping: Based on a preset coordinate range, crop a local image containing the meshing area of ​​the moving and stationary contacts from the original image, reducing the computational load of subsequent image processing and eliminating background interference.

[0028] The contact resistance signal is denoised and the contact stability is verified to eliminate abnormal data points caused by contact jitter or poor contact in the measurement circuit. Noise reduction processing: The raw voltage and current signals output by the micro resistance meter are digitally low-pass filtered to eliminate high-frequency electromagnetic interference in the measurement circuit; Contact stability verification: The stability of contact resistance values ​​measured multiple times is judged. If the fluctuation range of multiple measurements exceeds the preset threshold, the data of this measurement is determined to be invalid due to contact jitter or poor contact of the measurement circuit. The data is then discarded and the next operation cycle is triggered for retesting to ensure that the contact resistance values ​​used for subsequent analysis truly reflect the stable contact state of the contact.

[0029] Environmental parameters are processed by outlier removal and linear interpolation. Outlier removal: Statistical methods are used to identify outliers in the environmental parameter sequence. If the temperature or humidity value at a certain moment exceeds the range of ±3 times the historical mean, it is determined to be an outlier caused by sensor mismeasurement or interference and is removed. Linear interpolation imputation: For the locations of outliers that have been removed, linear interpolation of the nearest valid data points is used to imput them, ensuring the continuity of the environmental parameter sequence.

[0030] Through the above preprocessing, a standardized multi-dimensional time-series monitoring dataset is obtained, providing input for subsequent feature extraction.

[0031] Step S2: Based on the standardized multi-dimensional time-series monitoring dataset, extract wear-sensitive feature vectors that characterize the wear state of the contacts; The wear-sensitive feature vector includes: Vibration impact features extracted from vibration signals include high-frequency detail energy extracted after wavelet packet decomposition of the vibration signal, as well as the maximum amplitude, root mean square value, and impulse factor obtained from time-domain analysis. Wavelet packet decomposition is performed on the vibration signal to obtain high-frequency detail energy. Wavelet packet decomposition can divide the signal into low-frequency and high-frequency parts at the same time, and obtain more refined frequency band information. Taking 3-level wavelet packet decomposition as an example, 8 frequency bands of equal width can be obtained. The energy or energy ratio of the high-frequency band is selected as a feature. This feature reflects the strength of the impact component generated by events such as contact impact and friction. Maximum amplitude: The maximum absolute value of the vibration waveform, reflecting the severity of the impact; Root mean square (RMS): The effective value of the vibration waveform, reflecting the overall energy level of the vibration; Pulse factor: The ratio of the maximum amplitude to the rectified average value, used to detect whether there is a significant impulse component in the signal; the higher the pulse factor, the more significant the impulse component in the signal.

[0032] Load variation features extracted from motor current signals, including peak current, rise time, and current fluctuation coefficient extracted after segmenting and identifying the motor current waveform; Peak current: The maximum value during the rising phase of the motor current, reflecting the peak torque required by the drive mechanism to overcome resistance such as friction and jamming; Rise time: The time it takes for the current to rise from 10% to 90% of its peak value, reflecting the smoothness of the mechanism's operation; a prolonged rise time may indicate increased resistance or decreased driving force. Current fluctuation coefficient: During the current steady-state period, the ratio of the standard deviation to the mean of the current waveform is calculated; this coefficient reflects the stability of the current, and increased fluctuation may indicate abnormal vibration or intermittent jamming in the mechanism.

[0033] The coating appearance features extracted from the contact surface image include the average chromaticity and chromaticity standard deviation extracted after HSV space transformation of the contact image, and the percentage of coating peeling area obtained by chromaticity threshold segmentation. HSV space conversion: Converting RGB images to the HSV color space; the HSV space decomposes color into hue, saturation, and lightness, where hue is less sensitive to changes in lighting and is more suitable for color analysis; coatings have a specific hue range in the HSV space. Average chromaticity and chromaticity standard deviation: Extract the H component of the contact area in the image and calculate its mean and standard deviation; the average chromaticity reflects the overall color tendency of the coating, and the chromaticity standard deviation reflects the uniformity of the coating color; coating peeling or oxidation will cause changes and dispersion in chromaticity.

[0034] Coating peeling area ratio: Set the threshold range of the H component to divide the image into normal coating area and peeling area; count the proportion of pixels in the peeling area to the total pixels in the contact area as the quantitative value of coating peeling area ratio.

[0035] Electrical contact features extracted from preprocessed contact resistance signals, wherein the electrical contact features are contact resistance values ​​measured by a micro resistance meter under no-load conditions when the circuit is closed; Contact resistance value: The static contact resistance stability value measured by a micro resistance meter under no-load conditions is taken; this value is the core electrical parameter that directly reflects the contact state of the contacts; according to Holm contact theory, the magnitude of the contact resistance is closely related to the contact pressure, the resistivity of the coating material, the effective conductive area, and the state of the surface oxide film.

[0036] Environmental compensation factors extracted from environmental parameters, including temperature, relative humidity, and pollution level; Temperature: Ambient temperature affects the resistivity and thermal expansion of materials, which in turn affects the measured value of contact resistance. It needs to be included as a compensation factor in the feature vector. Relative humidity: Ambient humidity affects the oxidation and corrosion rate of contact surfaces and is an important factor influencing the wear process; Contamination level: Based on local environmental conditions or historical data, the degree of contamination is quantified into a level value; the contamination level reflects the degree of pollution of the contact surface by salt spray, dust, etc., and affects the contact reliability.

[0037] Together with the cumulative number of operations, they form a multidimensional feature vector.

[0038] The cumulative number of operations since commissioning is read from the equipment ledger records; the number of operations is the core variable driving the wear process of the contacts. As the number of operations increases, the coating gradually thins and the morphology of the contact surface changes.

[0039] After normalizing all the above feature parameters, they are combined to form a multidimensional wear-sensitive feature vector for the current moment, which is used for subsequent wear state identification.

[0040] Step S3: Construct a contact wear situation awareness model, identify the current contact wear state based on the wear-sensitive feature vector, and predict the wear trend; The contact wear situation perception model includes a physical model, which is constructed based on Hertz contact theory and Holm contact resistance theory, and establishes a quantitative relationship between contact resistance and contact material resistivity, contact spot equivalent radius, surface film tunneling resistance coefficient and effective conductive area. Establish a quantitative relationship between contact resistance and key physical parameters:

[0041] in, Rc For contact resistance, ρ The resistivity of the contact material. a The equivalent radius of the contact spot. σ The surface film tunneling resistivity is... A This represents the effective conductive area.

[0042] The cumulative number of operations is introduced as a wear process variable, and an evolution model of the contact resistance changing with the number of operations is established. The model parameters are calibrated by friction and wear tests. Introducing cumulative operation count N As a wear process variable, establish the contact spot radius a With the number of operations N Degradation model:

[0043] in, a 0 is the initial contact radius. f ( N The function is a monotonically decreasing function, and its specific form can be determined through friction and wear tests; Through laboratory friction and wear tests, contact samples of the same type and material were subjected to different numbers of opening and closing operations to simulate the process. The contact resistance was measured at intervals and the contact radius was calculated in reverse, thereby calibrating the parameters to be determined in the degradation model.

[0044] The physical model outputs a theoretical contact resistance value predicted based on the current number of operations, and an equivalent remaining coating thickness derived from the measured contact resistance. Theoretical contact resistance prediction: Based on the current number of operations N, substitute the degradation model and contact resistance formula to calculate the predicted theoretical contact resistance value; Equivalent remaining coating thickness: Based on the measured contact resistance value, combined with the initial coating thickness and initial contact resistance, the current equivalent remaining coating thickness is obtained through inversion calculation; this thickness is an intuitive physical quantity characterizing the degree of contact wear.

[0045] The contact wear situation awareness model includes a data model, which is constructed using a CNN-LSTM fusion neural network architecture and trained using multiple sets of wear-sensitive feature vectors and corresponding wear state labels collected during historical operation phases as training sample sets. The CNN part is used to extract the local coupling relationship between the components within the feature vector, and the LSTM part is used to capture the temporal dependency relationship of the feature vector as it evolves over time. The CNN part is used to extract the local coupling relationships between the components within a feature vector. The input is a sequence of feature vectors from multiple consecutive operation times. The CNN performs sliding convolutions along the feature dimension using one-dimensional convolutional layers to automatically learn the combination patterns between different feature components. Pooling layers are usually followed by convolutional layers for dimensionality reduction. The LSTM part is used to capture the temporal dependencies of feature vectors as they evolve over time. The feature sequences extracted by the CNN are input into the LSTM layer, and the LSTM learns the long-term dependency patterns in the sequence through its gating mechanism, that is, the evolution of wear and tear as the number of operations increases. The model input is a sequence of feature vectors from multiple consecutive operation moments, and the output is the wear state classification result and wear degree quantification value at the current moment. The wear status classification includes four levels: normal, slight wear, moderate wear, and severe wear. The wear degree quantification value is a continuous value between 0 and 100%.

[0046] Wear status classification results: The Softmax activation function is used to output the probability of the current wear status belonging to four levels: normal, slight wear, moderate wear, and severe wear. The category with the highest probability is taken as the classification result. Wear degree quantification value: The Sigmoid activation function is used to output a continuous value between 0 and 1. Multiplying it by 100% gives the wear degree quantification value between 0% and 100%, which is used to quantitatively characterize the degree of contact degradation. The model training uses multiple sets of wear-sensitive feature vectors and corresponding wear status labels collected during historical operation phases as the training sample set. Wear status labels can be obtained through historical maintenance records, contact resistance threshold determination, expert experience annotation, etc.

[0047] The contact wear situational awareness model includes a fusion diagnostic mechanism: Input the feature vector at the current moment into the physical model to obtain the equivalent remaining coating thickness. And the predicted value of theoretical contact resistance; Based on the equivalent remaining coating thickness, the wear level value from the perspective of the physical model can be calculated. ,in This represents the initial coating thickness. Input a continuous feature vector sequence, including the current time, into the data model to obtain the wear state classification result and the wear degree quantification value; Calculate the wear level of the physical model With the wear and tear of the data model Deviation between:

[0048] Set deviation threshold δ When the deviation between the quantified wear level output by the data model and the wear level calculated by the physical model based on the equivalent remaining coating thickness... Less than the preset threshold δ When the outputs of the two models are consistent, the weighted average of the two models is taken as the final wear assessment result; the weighting coefficient can be set according to the model confidence level.

[0049] when deviation Exceeding the preset threshold δ When a significant discrepancy is found between the outputs of the two models, a deep diagnostic mode is triggered. The current feature vector is matched with the historical fault case library for similarity. The historical fault case library stores various typical fault cases that have occurred in the past operation. Each case records the feature vector sequence before the fault occurred, the fault type, the evolution path and the final maintenance conclusion. The similarity between the current feature vector and the feature vector of each case in the case library is calculated using a similarity measurement method. The most similar fault case and its subsequent evolution path are retrieved, and diagnostic suggestions and warning information are output.

[0050] Step S4: Based on the wear trend prediction results, the remaining life of the contacts is assessed and graded for early warning.

[0051] The wear trend prediction includes: A wear trend curve is constructed based on the historical wear degree sequence. The wear degree sequence is then subjected to three exponential smoothing processes to remove short-term fluctuations and obtain a monotonically increasing curve that reflects the long-term wear trend. The final wear values ​​obtained from each operation assessment are arranged in chronological order of operation time to form a sequence. W 1, W 2, ..., Wn The corresponding number of operations N 1, N 2, ..., Nn ; The wear degree series is subjected to triple exponential smoothing. Triple exponential smoothing can smooth and predict series with trends and seasonality. For wear degree series, which usually shows a monotonically increasing trend, smoothing can remove short-term random fluctuations and extract a monotonically increasing trend curve that reflects the long-term evolution of wear. The gray prediction model GM(1,1) or the ARIMA time series model is used to extrapolate and predict the trend curve to obtain the predicted wear level under the future number of operations. The grey prediction model GM(1,1) is suitable for sequences with a small amount of data and a clear exponential growth trend. It generates a first-order grey differential equation by accumulating the original sequence, estimates the model parameters, and then predicts the future value. ARIMA model: Applicable to time series with certain statistical regularities; by identifying the autocorrelation and partial autocorrelation functions of the series, the model order is determined, and the ARIMA model is fitted to achieve the prediction of future values.

[0052] The remaining life assessment includes: setting a severe wear threshold, calculating the remaining number of operations required to reach the severe wear threshold based on the current wear level and the slope of the wear trend curve, and using this as a quantitative value of the remaining life.

[0053] Severe wear threshold: A critical value for the degree of wear is set based on equipment technical specifications, operating experience, or test data. When the wear reaches or exceeds this threshold, the contact is considered to be in a state of severe wear and needs to be repaired or replaced. Based on the current level of wear Instantaneous slope of the wear trend curve k The remaining number of operations is calculated using a linear approximation:

[0054] Among them, slope k This indicates the wear rate, expressed as % per cycle.

[0055] The tiered early warning system includes four warning levels based on the degree of wear and remaining lifespan: When the wear level is greater than or equal to 30% and the remaining number of operations is greater than or equal to 2000, a warning level alert is issued, prompting an increase in the frequency of monitoring. When the wear level is greater than or equal to 50% and the remaining number of operations is greater than or equal to 500, an early warning level warning will be issued to prompt the scheduling of an upcoming maintenance plan. When the wear level is greater than or equal to 70% and the remaining number of operations is greater than or equal to 100, an alarm-level warning will be output, indicating that maintenance must be arranged during the next power outage window; When the wear level is greater than or equal to 85% or the remaining number of operations is less than 100, an emergency warning will be issued, prompting an immediate power outage for maintenance.

[0056] The above thresholds can be adjusted adaptively according to different voltage levels, different equipment models, and different operating environments.

[0057] Example 2: A disconnector contact wear identification system, such as... Figure 2 As shown, it includes: The data acquisition module is used to collect multi-source sensor data during the operation of the disconnecting switch, including vibration signals collected by the accelerometer, motor current signals collected by the current sensor, contact surface image signals collected by the image acquisition device, contact resistance signals collected by the micro-resistance meter, and environmental parameter signals collected by the environmental parameter sensor. The preprocessing module receives multi-source sensor data collected by the data acquisition module, performs detrending, bandpass filtering and wavelet threshold denoising on the vibration signal, removes DC component and performs moving average filtering on the current signal, performs distortion correction, illumination normalization and region of interest cropping on the image data, performs denoising on the contact resistance signal and performs contact stability verification, removes abnormal data points caused by contact jitter or poor contact of the measurement circuit, and performs outlier removal and linear interpolation filling on the environmental parameters to obtain a standardized multi-dimensional time-series monitoring dataset. The feature extraction module, based on the standardized multi-dimensional time-series monitoring dataset output by the preprocessing module, extracts vibration impact features from vibration signals, load change features from motor current signals, plating appearance features from contact surface images, and electrical contact features from contact resistance measurements. The electrical contact features extracted from the preprocessed contact resistance signals are combined with compensation factors and cumulative operation counts in environmental parameters, and these feature parameters are fused in the feature layer to generate a wear-sensitive feature vector characterizing the wear state of the contact. The situational awareness module, with its built-in physical and data models, uses the wear-sensitive feature vector generated by the feature extraction module as input to identify the current wear state of the contact and predict the wear trend. The physical model is built upon Hertzian contact theory and Holm contact resistance theory, establishing a quantitative relationship between contact resistance and contact material resistivity, equivalent radius of the contact spot, surface film tunneling resistance coefficient, and effective conductive area. The data model employs a CNN-LSTM fusion neural network architecture, trained based on historical operating data. When the output deviation between the physical and data models is less than a preset threshold, the weighted average of the two is used as the evaluation result. When the deviation exceeds the threshold, a deep diagnostic mode is triggered, matching the current feature vector with a historical fault case database to output diagnostic suggestions and early warning information. The life assessment and early warning module calculates the remaining life of the contact based on the wear status identification results and wear trend prediction results output by the situational awareness module, and outputs graded early warning information according to the wear degree and remaining life. The graded early warning includes four levels: attention level, early warning level, alarm level and emergency level.

[0058] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying wear on disconnector switch contacts, characterized in that, Includes the following steps: During the operation of the disconnecting switch, multi-source sensor data is collected, and the collected raw data is preprocessed to obtain a standardized multi-dimensional time-series monitoring dataset. Based on the standardized multi-dimensional time-series monitoring dataset, a wear-sensitive feature vector characterizing the wear state of the contact is extracted; A contact wear situation awareness model is constructed to identify the current contact wear state based on the wear-sensitive feature vector and predict the wear trend. Based on the wear trend prediction results, the remaining life of the contacts can be assessed and classified for early warning.

2. The method for identifying contact wear of a disconnector switch according to claim 1, characterized in that, The multi-source sensing data includes: vibration signals acquired by an accelerometer, motor current signals acquired by a current sensor, contact surface image signals acquired by an image acquisition device, contact resistance signals acquired by a micro-resistance meter, and environmental parameter signals acquired by an environmental parameter sensor.

3. The method for identifying contact wear of a disconnector switch according to claim 2, characterized in that, The preprocessing includes: The vibration signal is subjected to detrending, bandpass filtering, and wavelet thresholding for noise reduction. The current signal is processed by removing the DC component and applying a moving average filter. The image data is processed by distortion correction, illumination normalization, and region of interest cropping. The contact resistance signal is denoised and the contact stability is verified to eliminate abnormal data points caused by contact jitter or poor contact in the measurement circuit. Outlier removal and linear interpolation are performed on environmental parameters.

4. The method for identifying contact wear of a disconnector switch according to claim 1, characterized in that, The wear-sensitive feature vector includes: Vibration impact features extracted from vibration signals include high-frequency detail energy extracted after wavelet packet decomposition of the vibration signal, as well as the maximum amplitude, root mean square value, and impulse factor obtained from time-domain analysis. Load variation features extracted from motor current signals, including peak current, rise time, and current fluctuation coefficient extracted after segmenting and identifying the motor current waveform; The coating appearance features extracted from the contact surface image include the average chromaticity and chromaticity standard deviation extracted after HSV space transformation of the contact image, and the percentage of coating peeling area obtained by chromaticity threshold segmentation. Electrical contact features extracted from preprocessed contact resistance signals, wherein the electrical contact features are contact resistance values ​​measured by a micro resistance meter under no-load conditions when the circuit is closed; Environmental compensation factors extracted from environmental parameters, including temperature, relative humidity, and pollution level; Together with the cumulative number of operations, they form a multidimensional feature vector.

5. The method for identifying contact wear of a disconnector switch according to claim 1, characterized in that, The contact wear situation perception model includes a physical model, which is constructed based on Hertz contact theory and Holm contact resistance theory, and establishes a quantitative relationship between contact resistance and contact material resistivity, contact spot equivalent radius, surface film tunneling resistance coefficient and effective conductive area. The cumulative number of operations is introduced as a wear process variable, and an evolution model of the contact resistance changing with the number of operations is established. The model parameters are calibrated by friction and wear tests. The physical model outputs a theoretical contact resistance value predicted based on the current number of operations, and an equivalent remaining coating thickness derived from the measured contact resistance.

6. The method for identifying contact wear of a disconnector switch according to claim 1, characterized in that, The contact wear situation awareness model includes a data model, which is constructed using a CNN-LSTM fusion neural network architecture and trained using multiple sets of wear-sensitive feature vectors and corresponding wear state labels collected during historical operation phases as training sample sets. The CNN part is used to extract the local coupling relationship between the components within the feature vector, and the LSTM part is used to capture the temporal dependency relationship of the feature vector as it evolves over time. The model input is a sequence of feature vectors from multiple consecutive operation moments, and the output is the wear state classification result and wear degree quantification value at the current moment. The wear status classification includes four levels: normal, slight wear, moderate wear, and severe wear. The wear degree quantification value is a continuous value between 0 and 100%.

7. The method for identifying contact wear of a disconnector switch according to claim 1, characterized in that, The contact wear situational awareness model includes a fusion diagnostic mechanism: Input the feature vector at the current moment into the physical model to obtain the equivalent remaining coating thickness and the predicted theoretical contact resistance; Input a continuous feature vector sequence, including the current time, into the data model to obtain the wear state classification result and the wear degree quantification value; When the deviation between the wear degree quantification value output by the data model and the wear degree value calculated by the physical model based on the equivalent remaining coating thickness is less than a preset threshold, it is determined that the outputs of the two models are consistent, and the weighted average of the two is taken as the final wear degree assessment result. When the deviation exceeds the preset threshold, the deep diagnosis mode is triggered, and the current feature vector is matched with the historical fault case library for similarity. The most similar fault case and its subsequent evolution path are retrieved, and diagnostic suggestions and warning information are output.

8. The method for identifying contact wear of a disconnector switch according to claim 1, characterized in that, The wear trend prediction includes: A wear trend curve is constructed based on the historical wear degree sequence. The wear degree sequence is then subjected to three exponential smoothing processes to remove short-term fluctuations and obtain a monotonically increasing curve that reflects the long-term wear trend. The gray prediction model GM(1,1) or the ARIMA time series model is used to extrapolate and predict the trend curve to obtain the predicted wear level under the future number of operations. The remaining life assessment includes: setting a severe wear threshold, calculating the remaining number of operations required to reach the severe wear threshold based on the current wear level and the slope of the wear trend curve, and using this as a quantitative value of the remaining life.

9. The method for identifying contact wear of a disconnector switch according to claim 1, characterized in that, The tiered early warning system includes four warning levels based on the degree of wear and remaining lifespan: When the wear level is greater than or equal to 30% and the remaining number of operations is greater than or equal to 2000, a warning level alert is issued, prompting an increase in the frequency of monitoring. When the wear level is greater than or equal to 50% and the remaining number of operations is greater than or equal to 500, an early warning level warning will be issued to prompt the scheduling of an upcoming maintenance plan. When the wear level is greater than or equal to 70% and the remaining number of operations is greater than or equal to 100, an alarm-level warning will be output, indicating that maintenance must be arranged during the next power outage window; When the wear level is greater than or equal to 85% or the remaining number of operations is less than 100, an emergency warning will be issued, prompting an immediate power outage for maintenance.

10. A system for identifying contact wear of a disconnector switch, characterized in that, The system employs a method for identifying contact wear of a disconnector as described in any one of claims 1 to 9, comprising: The data acquisition module is used to collect multi-source sensor data during the operation of the disconnecting switch, including vibration signals collected by the accelerometer, motor current signals collected by the current sensor, contact surface image signals collected by the image acquisition device, contact resistance signals collected by the micro-resistance meter, and environmental parameter signals collected by the environmental parameter sensor. The preprocessing module receives multi-source sensor data collected by the data acquisition module, performs detrending, bandpass filtering and wavelet threshold denoising on the vibration signal, removes DC component and performs moving average filtering on the current signal, performs distortion correction, illumination normalization and region of interest cropping on the image data, performs denoising on the contact resistance signal and performs contact stability verification, removes abnormal data points caused by contact jitter or poor contact of the measurement circuit, and performs outlier removal and linear interpolation filling on the environmental parameters to obtain a standardized multi-dimensional time-series monitoring dataset. The feature extraction module, based on the standardized multi-dimensional time-series monitoring dataset output by the preprocessing module, extracts vibration impact features from vibration signals, load change features from motor current signals, plating appearance features from contact surface images, and electrical contact features from contact resistance measurements. The electrical contact features extracted from the preprocessed contact resistance signals are combined with compensation factors and cumulative operation counts in environmental parameters, and these feature parameters are fused in the feature layer to generate a wear-sensitive feature vector characterizing the wear state of the contact. The situational awareness module, with its built-in physical and data models, uses the wear-sensitive feature vector generated by the feature extraction module as input to identify the current wear state of the contact and predict the wear trend. The physical model is built upon Hertzian contact theory and Holm contact resistance theory, establishing a quantitative relationship between contact resistance and contact material resistivity, equivalent radius of the contact spot, surface film tunneling resistance coefficient, and effective conductive area. The data model employs a CNN-LSTM fusion neural network architecture, trained based on historical operating data. When the output deviation between the physical and data models is less than a preset threshold, the weighted average of the two is used as the evaluation result. When the deviation exceeds the threshold, a deep diagnostic mode is triggered, matching the current feature vector with a historical fault case database to output diagnostic suggestions and early warning information. The life assessment and early warning module calculates the remaining life of the contact based on the wear status identification results and wear trend prediction results output by the situational awareness module, and outputs graded early warning information according to the wear degree and remaining life. The graded early warning includes four levels: attention level, early warning level, alarm level and emergency level.

Images