GCS drawer unit mechanical fault diagnosis method and system based on multi-modal analysis

Abstract

The invention discloses a GCS drawer unit mechanical fault diagnosis method and system based on vibration-current multi-mode analysis, and the method comprises the steps: firstly, synchronously collecting multi-source state data in the mechanical operation and operation process through a vibration sensor, a temperature sensor and a current transformer which are disposed on a drawer guide rail and a circuit breaker contact arm; secondly, an improved dynamic time warping algorithm is adopted for the vibration signals to extract the difference degree between the vibration signals and a standard operation waveform, and contact resistance and temperature rise characteristics of the electric signals are calculated; the extracted multi-dimensional feature vectors such as the vibration difference degree, the contact resistance and the temperature rise rate are input into a support vector machine multi-classifier for fault mode recognition, and specific fault types and confidence coefficients are output; finally, the diagnostic model can be optimized according to maintenance feedback. The problem that hidden faults such as poor contact and jamming caused by abrasion and deformation of mechanical parts of a GCS drawer cabinet are difficult to find in time is solved, and early-stage, accurate and online diagnosis of mechanical faults is achieved.

Description

Technical Field

[0001] This invention relates to the field of electrical equipment condition monitoring and fault diagnosis technology, and in particular to a method and system for diagnosing mechanical faults in GCS drawer units based on multimodal analysis. Background Technology

[0002] Low-voltage withdrawable switchgear (GCS) is widely used in power distribution systems due to its compact structure and ease of maintenance. Its core component, the drawer unit, achieves circuit connection through a plug-in / plug-out method. However, frequent operation, mechanical stress, and thermal effects can easily lead to mechanical failures such as drawer rail deformation, contact system wear, and operating mechanism spring fatigue. These failures initially manifest as increased contact resistance and difficulty in plugging and unplugging; if not addressed promptly, they can develop into overheating, arcing, or even short-circuit accidents, threatening power supply safety.

[0003] Existing monitoring methods mostly focus on monitoring electrical parameters such as operating current and temperature, making it difficult to effectively capture purely mechanical performance degradation. For example, contact wear may not cause a significant temperature rise when the current is low, but it already poses a safety hazard. Vibration signals can directly reflect the operating state of mechanical components, but currently there is a lack of effective solutions that deeply integrate vibration analysis with electrical analysis and apply it to GCS drawer units for accurate fault classification.

[0004] Therefore, there is an urgent need for an intelligent method that can diagnose mechanical faults in GCS drawer units early and accurately. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a mechanical fault diagnosis method and system for GCS drawer units based on vibration-current multimodal analysis, so as to realize online identification and early warning of typical mechanical faults such as guide rail deformation, contact wear, and spring fatigue.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A method for mechanical fault diagnosis of GCS drawer units based on vibration-current multimodal analysis includes the following steps: S100. Data Synchronization Acquisition: When the drawer unit performs a plugging / unplugging operation or is in operation, the following data is synchronously triggered and acquired: vibration signal generated by the vibration sensor deployed on the drawer guide rail; temperature signal generated by the wireless temperature sensor deployed on the circuit breaker contact arm; current signal generated by the current transformer in the power distribution circuit. S200. Multimodal Feature Extraction: Process the raw data collected in step S100 to extract feature vectors representing the mechanical health status: a. Vibration feature extraction: The vibration signal of a complete insertion and removal operation is extracted and dynamically time-warped with the preset standard healthy vibration waveform. The cumulative minimum path distance is used as the vibration waveform difference feature. b. Electrical feature extraction: Based on the real-time current value and the temperature rise of the contact arm, the dynamic contact resistance is estimated using a thermal balance model; at the same time, the temperature rise rate under unit current is calculated. c. Combine the vibration waveform difference, dynamic contact resistance, and temperature rise rate to form a multi-dimensional feature vector; S300. Fault Mode Recognition: Input the multi-dimensional feature vector obtained in step S200 into a pre-trained support vector machine multi-classification model for fault classification. The output of the multi-classification model includes the fault type and the corresponding diagnostic confidence. The fault type includes at least: normal state, guide rail deformation, contact wear, and spring fatigue.

[0007] S400. Diagnostic Result Output and Model Optimization: Output the fault diagnosis result obtained in step S300; after the on-site maintenance personnel confirm the diagnosis result, use the feature vector used in this diagnosis and the confirmed real fault label as new samples to perform incremental learning on the support vector machine multi-classification model and optimize the model parameters.

[0008] The present invention further specifies that, in step S200a, the specific process of the dynamic time warping calculation is as follows: Let the standard healthy vibration waveform sequence be... The vibration waveform sequence to be measured is Construct an m×n cumulative cost matrix D, where the element D(i,j) is calculated using the following formula: , ,in The distance is the point-to-point Euclidean distance; the value of the lower right element D(m,n) of the matrix is ​​the minimum cumulative path distance from sequence R to sequence T, which is defined as the vibration waveform difference.

[0009] The present invention further provides that, in step S200b, the estimation formula for the dynamic contact resistance is: , where is the difference between the measured temperature of the contact arm and the ambient temperature, A is the comprehensive heat dissipation coefficient related to heat dissipation conditions, I is the real-time effective value of the current, and t is the continuous energizing time.

[0010] The present invention further specifies that the support vector machine multi-classification model is constructed using a one-to-one strategy, and its kernel function is a radial basis function.

[0011] In a further embodiment of the present invention, in step S400, the incremental learning adopts an error-driven adaptive weight adjustment strategy, that is, when the model diagnoses an error, the weight of the new sample is increased and it participates in the model update with a higher learning rate.

[0012] The present invention also provides a GCS drawer unit mechanical fault diagnosis system for implementing the method, characterized in that it comprises: Data acquisition module: includes vibration sensor, temperature sensor, current transformer and data acquisition unit, used to synchronously acquire multi-source status data; Edge computing unit: With built-in processor and memory, it is configured to execute the following programs: Feature extraction program: Implements the multimodal feature extraction steps described above; Fault diagnosis procedure: Load and run the support vector machine multi-classification model to achieve fault mode recognition; Self-learning optimization procedure: Incremental learning of the model is achieved based on maintenance feedback; Human-computer interaction and communication module: used to display diagnostic results, issue warnings, and upload data to the superior monitoring system.

[0013] The present invention further provides that the vibration sensor is a broadband accelerometer with a frequency response range covering 10Hz to 10kHz.

[0014] In a further embodiment of the present invention, the temperature sensor is a wireless passive surface acoustic wave temperature sensor or a battery-powered wireless temperature sensor.

[0015] The beneficial effects of the present invention are as follows: Compared with the prior art, the beneficial effects of the present invention are: 1. Early warning: Vibration analysis can provide early warnings in the early stages of mechanical failure, before serious electrical problems occur.

[0016] 2. Precise diagnosis: By integrating multi-dimensional information such as vibration, temperature, and current, and using the SVM model, different types of mechanical faults can be distinguished, guiding precise maintenance.

[0017] 3. High adaptability: It has a self-learning function and can adaptively adjust the diagnostic threshold as the equipment ages, reducing the false alarm rate.

[0018] 4. Non-invasive: The sensor is easy to install and does not affect the original structure and electrical performance of the equipment, making it suitable for the intelligent transformation of existing equipment. Attached Figure Description

[0019] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating an embodiment of the present invention. Detailed Implementation

[0020] The following will describe in detail the implementation of this application with reference to the accompanying drawings and embodiments, so that the implementation process of how this application uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.

[0021] like Figure 1 As shown, this invention provides a mechanical fault diagnosis method for GCS drawer units based on vibration-current multimodal analysis, comprising the following steps: S100. Data Synchronization Acquisition: When the drawer unit performs a plugging / unplugging operation or is in operation, the following data is synchronously triggered and acquired: vibration signal generated by the vibration sensor deployed on the drawer guide rail; temperature signal generated by the wireless temperature sensor deployed on the circuit breaker contact arm; current signal generated by the current transformer in the power distribution circuit. S200. Multimodal Feature Extraction: Process the raw data collected in step S100 to extract feature vectors representing the mechanical health status: a. Vibration feature extraction: The vibration signal of a complete insertion and removal operation is extracted and dynamically time-warped with the preset standard healthy vibration waveform. The cumulative minimum path distance is used as the vibration waveform difference feature. b. Electrical feature extraction: Based on the real-time current value and the temperature rise of the contact arm, the dynamic contact resistance is estimated using a thermal balance model; at the same time, the temperature rise rate under unit current is calculated. c. Combine the vibration waveform difference, dynamic contact resistance, and temperature rise rate to form a multi-dimensional feature vector; S300. Fault Mode Recognition: Input the multi-dimensional feature vector obtained in step S200 into a pre-trained support vector machine multi-classification model for fault classification. The output of the multi-classification model includes the fault type and the corresponding diagnostic confidence. The fault type includes at least: normal state, guide rail deformation, contact wear, and spring fatigue.

[0022] S400. Diagnostic Result Output and Model Optimization: Output the fault diagnosis result obtained in step S300; after the on-site maintenance personnel confirm the diagnosis result, use the feature vector used in this diagnosis and the confirmed real fault label as new samples to perform incremental learning on the support vector machine multi-classification model and optimize the model parameters.

[0023] In step S200a, the specific process of the dynamic time warping calculation is as follows: Let the standard healthy vibration waveform sequence be... The vibration waveform sequence to be measured is Construct an m×n cumulative cost matrix D, where the element D(i,j) is calculated using the following formula: , ,in The distance is the point-to-point Euclidean distance; the value of the lower right element D(m,n) of the matrix is ​​the minimum cumulative path distance from sequence R to sequence T, which is defined as the vibration waveform difference.

[0024] The present invention further provides that, in step S200b, the estimation formula for the dynamic contact resistance is: , where is the difference between the measured temperature of the contact arm and the ambient temperature, A is the comprehensive heat dissipation coefficient related to heat dissipation conditions, I is the real-time effective value of the current, and t is the continuous energizing time.

[0025] The present invention further specifies that the support vector machine multi-classification model is constructed using a one-to-one strategy, and its kernel function is a radial basis function.

[0026] In a further embodiment of the present invention, in step S400, the incremental learning adopts an error-driven adaptive weight adjustment strategy, that is, when the model diagnoses an error, the weight of the new sample is increased and it participates in the model update with a higher learning rate.

[0027] The present invention also provides a GCS drawer unit mechanical fault diagnosis system for implementing the method, characterized in that it comprises: Data acquisition module: includes vibration sensor, temperature sensor, current transformer and data acquisition unit, used to synchronously acquire multi-source status data; Edge computing unit: With built-in processor and memory, it is configured to execute the following programs: Feature extraction program: Implements the multimodal feature extraction steps described above; Fault diagnosis procedure: Load and run the support vector machine multi-classification model to achieve fault mode recognition; Self-learning optimization procedure: Incremental learning of the model is achieved based on maintenance feedback; Human-computer interaction and communication module: used to display diagnostic results, issue warnings, and upload data to the superior monitoring system.

[0028] The present invention further provides that the vibration sensor is a broadband accelerometer with a frequency response range covering 10Hz to 10kHz.

[0029] In a further embodiment of the present invention, the temperature sensor is a wireless passive surface acoustic wave temperature sensor or a battery-powered wireless temperature sensor.

[0030] Specifically, the overall system architecture of this invention is as follows: This system adopts a three-tier architecture of "cloud-edge-device" to achieve distributed data processing and centralized management.

[0031] Terminal layer: Vibration monitoring unit: It adopts an IEPE type accelerometer with a frequency response range of 10Hz-10kHz and a sensitivity of 100mV / g. It is installed on the guide rail support of each drawer unit to capture mechanical vibrations during insertion and removal operations and operation.

[0032] Temperature monitoring unit: It adopts a surface acoustic wave wireless passive temperature sensor with a center frequency of 433MHz, a temperature measurement range of -40℃ to +125℃, and an accuracy of ±0.5℃. It is directly installed on the moving and stationary contacts of the circuit breaker.

[0033] Electrical monitoring unit: It adopts an open-type Rogowski coil current sensor with an accuracy of 0.5%, which is installed on the drawer outlet cable to monitor the load current.

[0034] Data acquisition unit: Based on the STM32H7 series microcontroller, it integrates a multi-channel synchronous ADC and is responsible for the acquisition, preliminary filtering, and packetization of sensor signals. It communicates with the edge computing unit via a CAN bus.

[0035] Edge computing layer: Hardware: Employs NVIDIA Jetson Nano or similar embedded AI computing platforms.

[0036] Functionality: Deploys core diagnostic algorithms responsible for real-time feature extraction, fault diagnosis, and model self-learning from data in one or more drawers. Features local storage capabilities, caching at least 30 days of raw data.

[0037] Cloud platform / monitoring center: Functions: Receives diagnostic results and feature data uploaded by each edge computing unit, performs global status visualization, historical data analysis, report generation, and cross-cabinet horizontal comparison. Used for deploying and updating the global diagnostic model.

[0038] The data acquisition and preprocessing process includes: Main trigger signal: Auxiliary limit switch on the drawer unit. When the handle is turned, the switch state changes, generating a rising / falling edge signal as the absolute starting point for data acquisition.

[0039] Auxiliary trigger signal: The current transformer detects the change in current from zero to positive (power is on when the drawer is inserted) or from positive to zero (power is off when the drawer is removed).

[0040] The system uses the limit switch signal as a reference to synchronously start the recording window for vibration, temperature (recording the trend of change), and current data.

[0041] Data acquisition parameters Vibration signal: Sampling frequency: 20kHz (satisfying the Nyquist sampling theorem for a 10kHz frequency component).

[0042] Sampling duration: Record continuously for 10 seconds from the start of the trigger to ensure coverage of the entire insertion and removal process.

[0043] Preprocessing: Real-time bandpass filtering (50Hz - 8kHz) is performed at the MCU to remove power frequency interference and high-frequency noise.

[0044] Current signal: Sampling frequency: 2kHz.

[0045] Sampling duration: synchronized with the vibration signal.

[0046] Temperature signal: Sampling frequency: 1Hz (operational status monitoring).

[0047] During the event, temperature data is recorded for one minute before and after the event to analyze the impact of the operation on the temperature.

[0048] Vibration characteristics: Implementation using an improved DTW algorithm Establishment of a health template library: After new equipment is put into operation or undergoes major overhaul, vibration signals from N consecutive normal and smooth insertion and removal operations (e.g., N=20) are collected. These signals are aligned and averaged to generate a "gold standard template" R. The DTW distance between these N operations and the template R is calculated, and their average value μ and standard deviation σ are taken to set the initial alarm threshold. .

[0049] Real-time DTW computation optimization: Step 1: Data Normalization. To prevent the influence of small fluctuations in signal amplitude, the real-time signal T and template R are first normalized using Z-score:

[0050] in, To test the mean of the sequence, To test the standard deviation of the sequence, The mean of the reference sequence. Standard deviation of the reference sequence Step 2: Construct the cumulative cost matrix D. The matrix has dimensions M×N.

[0051] initialization: Then recursively fill in the blanks.

[0052]

[0053]

[0054] The recursive process can be explained as follows: This recursive formula means that the optimal path to point (i, j) is to choose the path with the minimum cumulative cost among the three neighbors on the left (i-1, j), below (i, j-1), or bottom left (i-1, j-1), and add the local cost of the current point (i, j).

[0055] Step 3: Extract vibration features: .

[0056] For electrical characteristics: First, estimate the dynamic contact resistance: First, calculate the temperature rise. Calculate the rate of temperature rise ,in Let be the average current from the start of power-on to the current moment. Let be the duration from the start of power-on to the current moment, and according to the formula: Estimate the dynamic contact resistance, where A is the comprehensive heat dissipation coefficient related to heat dissipation conditions, I is the real-time effective value of the current, and t is the continuous energizing time.

[0057] Then feature vectors are constructed. The kernel function, also known as the radial basis function, is: ; The decision function is: ; Diagnosis is based on the following filtering criteria: The confidence threshold is

[0058] if: If the condition is not met, the output will be "Status abnormal, type undetermined".

[0059] The present invention will now be further described with reference to specific embodiments: Data Acquisition: Sampling Frequency Setting. Vibration Signal: 20kHz; Temperature: 1Hz; Current: 2kHz. When the drawer operating handle is moved, the limit switch is triggered, and the system begins recording all data for the next 10 seconds.

[0060] Feature extraction: Vibration Difference: The main vibration waveform from 1.5 to 4.5 seconds after the start of the operation is extracted and compared with the healthy template for DTW calculation. The healthy template is taken from the average waveform of the first 10 normal operations after the drawer is installed and debugged. The calculated difference is within the normal range of less than 50; if it is greater than 100, it indicates a significant abnormality.

[0061] Contact resistance: Take the data after the drawer is fully inserted and the power is stable (e.g., after 5 minutes), and substitute it into the formula. The heat dissipation coefficient A is calibrated through factory testing.

[0062] Fault identification: The SVM model was trained using 500 samples from the historical database (approximately 100 samples per fault type). The real-time extracted feature vector [vibration difference = 135, contact resistance = 45 μΩ, temperature rise rate = 0.8℃ / s / A] was input into the model, and the model output was "guide rail deformation" with a confidence level of 87%.

[0063] Optimization: After maintenance personnel inspect and confirm that the guide rail is indeed slightly deformed and perform lubrication adjustments, they confirm the diagnostic result on the HMI. The system stores this set of features and the "guide rail deformation" label into the training library and automatically starts the model incremental learning process during off-peak hours at night.

[0064] For example, when the following fault characteristics appear: local temperature rise: the temperature of a single contact reaches 45℃, while the temperature of adjacent points is only 25℃; trend of change: temperature rise rate α = 1.2℃ / min; load correlation: δT = 3.5 (significantly abnormal); the diagnostic result can be: poor contact of the contacts, which requires tightening.

[0065] When the following fault characteristics occur: Overall temperature rise: all monitoring points rise synchronously, with an average temperature increase of 32℃; Ambient temperature: the ambient temperature inside the cabinet reaches 45℃; Load condition: the current is only 60% of the rated value. Diagnosis: Ventilation system malfunction; filter cleaning required. When the following fault characteristics appear: slow increase: the baseline temperature rises from 15℃ to 22℃ within 30 days. Trend stability: Daily variation pattern disappears; Harmonic characteristics: Current harmonic content increases; Diagnosis: Thermal aging of insulation material; preventative maintenance recommended. This invention achieves the following functions: Early warning: Vibration analysis can issue early warnings in the initial stages of mechanical failures, before serious electrical problems arise. Precise diagnosis: By integrating multi-dimensional information such as vibration, temperature, and current, and utilizing an SVM model, different types of mechanical failures can be distinguished, guiding precise maintenance. Strong adaptability: Possessing self-learning capabilities, it can adaptively adjust diagnostic thresholds as equipment ages, reducing false alarm rates. Non-invasive: The sensors are easy to install, do not affect the original structure and electrical performance of the equipment, and are suitable for the intelligent transformation of in-service equipment.

[0066] As used in the specification and claims, certain terms are used to refer to specific components. Those skilled in the art will understand that hardware manufacturers may use different terms to refer to the same component. This specification and claims do not distinguish components based on differences in name, but rather on differences in function. As mentioned throughout the specification and claims, "within the error range of packet reception, those skilled in the art can solve the technical problem within a certain error range and basically achieve the technical effect."

[0067] The foregoing description illustrates and describes several preferred embodiments of the present invention. However, as previously stated, it should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the inventive concept described herein through the foregoing teachings or techniques or knowledge in related fields. Any modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for mechanical fault diagnosis of GCS drawer unit based on vibration-current multimodal analysis, characterized in that, Includes the following steps: S100. Data Synchronization Acquisition: When the drawer unit is performing a plugging / unplugging operation or is in operation, the following data is synchronously triggered and acquired: vibration signal generated by the vibration sensor deployed on the drawer rail; temperature signal generated by the wireless temperature sensor deployed on the circuit breaker contact arm; current signal generated by the current transformer in the power distribution circuit. S200. Multimodal Feature Extraction: Process the raw data collected in step S100 to extract feature vectors representing the mechanical health status: a. Vibration feature extraction: The vibration signal of a complete insertion and removal operation is extracted and dynamically time-warped with the preset standard healthy vibration waveform. The cumulative minimum path distance is used as the vibration waveform difference feature. b. Electrical feature extraction: Based on the real-time current value and the temperature rise of the contact arm, the dynamic contact resistance is estimated using a thermal balance model; at the same time, the temperature rise rate under unit current is calculated. c. Combine the vibration waveform difference, dynamic contact resistance, and temperature rise rate to form a multi-dimensional feature vector; S300. Fault Mode Recognition: Input the multi-dimensional feature vector obtained in step S200 into a pre-trained support vector machine multi-classification model for fault classification. The output of the multi-classification model includes the fault type and the corresponding diagnostic confidence. The fault type includes at least: normal state, guide rail deformation, contact wear, and spring fatigue.

2. S400. Diagnostic Result Output and Model Optimization: Output the fault diagnosis result obtained in step S300; after the on-site maintenance personnel confirm the diagnosis result, use the feature vector used in this diagnosis and the confirmed real fault label as new samples to perform incremental learning on the support vector machine multi-classification model and optimize the model parameters.

3. The method according to claim 1, characterized in that, In step S200a, the specific process of the dynamic time warping calculation is as follows: Let the standard healthy vibration waveform sequence be... The vibration waveform sequence to be measured is Construct an m×n cumulative cost matrix D, whose element D(i,j) is calculated using the following formula: , ,in The distance is the point-to-point Euclidean distance; the value of the lower right element D(m,n) of the matrix is ​​the minimum cumulative path distance from sequence R to sequence T, which is defined as the vibration waveform difference.

4. The method according to claim 1, characterized in that, In step S200b, the dynamic contact resistance The estimation formula is: ,in, The difference between the temperature measured by the contact arm and the ambient temperature is given by , A is the comprehensive heat dissipation coefficient related to heat dissipation conditions, I is the real-time effective value of the current, and t is the continuous energizing time.

5. The method according to claim 1, characterized in that, In step S300, the support vector machine multi-classification model is constructed using a one-to-one strategy, and its kernel function is a radial basis function.

6. The method according to claim 1, characterized in that, In step S400, the incremental learning adopts an error-driven adaptive weight adjustment strategy, that is, when the model diagnoses an error, the weight of the new sample is increased and it participates in the model update with a higher learning rate.

7. A GCS drawer unit mechanical fault diagnosis system for implementing the method of any one of claims 1-5, characterized in that, include: Data acquisition module: includes vibration sensor, temperature sensor, current transformer and data acquisition unit, used to synchronously acquire multi-source status data; Edge computing unit: With built-in processor and memory, it is configured to execute the following programs: Feature extraction program: Implements the multimodal feature extraction steps described in claim 1; Fault diagnosis procedure: Load and run the support vector machine multi-classification model to achieve fault mode recognition; Self-learning optimization procedure: Incremental learning of the model is achieved based on maintenance feedback; Human-computer interaction and communication module: used to display diagnostic results, issue warnings, and upload data to the superior monitoring system.

8. The system according to claim 6, characterized in that, The vibration sensor is a broadband accelerometer with a frequency response range covering 10Hz to 10kHz.

9. The system according to claim 6, characterized in that, The temperature sensor is a wireless passive surface acoustic wave temperature sensor or a battery-powered wireless temperature sensor.

Images