A mechanical fault diagnosis system and method based on motor current multi-feature fusion

Abstract

The application discloses a kind of mechanical fault diagnosis system and method based on motor current multi-feature fusion, it is related to high-voltage switchgear state monitoring technical field, the system includes current sensor, signal conditioning circuit, digital signal processor and host computer.Current sensor acquires the current signal of driving motor, and single polarity signal is converted by signal conditioning circuit.Digital signal processor extracts four dimensions of characteristic quantity of current rising slope, steady-state current amplitude, closing operation time and side frequency component, constitutes multidimensional feature vector, is input to pre-trained random forest classification model and carries out fusion analysis and fault classification, and diagnosis result is uploaded to host computer by communication module and is shown and alarm.The application only needs single current sensor, can realize the online, accurate diagnosis of multiple mechanical faults such as mechanism jamming, closing not in place, spring failure, while ensuring the accuracy of diagnosis, significantly reduces system cost and complexity.

Description

Technical Field

[0001] This invention relates to the field of condition monitoring technology for high-voltage switchgear, and in particular to a mechanical fault diagnosis system and method based on the fusion of multiple characteristics of motor current. Background Technology

[0002] High-voltage disconnect switches are the most numerous and widely used high-voltage switchgear in traction power supply systems. Exposed year-round to extreme outdoor environments, they are highly susceptible to damage from dust, fog, wind, rain, and snow, making them prone to malfunctions due to mechanical wear, poor contact, and insulation aging, such as contact overheating and operational jamming. As a key device in establishing electrical safety isolation boundaries in traction power supply systems, the number of high-voltage disconnect switches far exceeds that of circuit breakers, load switches, and other switching devices. They are fundamental to ensuring system maintainability and operational flexibility. Typical mechanical faults such as mechanical jamming, incomplete opening and closing, and spring failure caused by corrosion, lubrication failure, and component fatigue have become the primary hidden dangers threatening the safe operation of traction power supply systems. If these hidden dangers are not detected in time, they may lead to cascading accidents such as bus short circuits and transformer damage, resulting in large-scale power outages and huge economic losses, such as industrial shutdowns and business interruptions. Therefore, real-time and accurate fault monitoring and diagnosis of the high-voltage disconnect switch mechanism has become an urgent need to improve the intelligent operation and maintenance level of the power grid and ensure power supply reliability.

[0003] Fault diagnosis technology for high-voltage disconnect switches can monitor key parameters such as equipment temperature, mechanical vibration, and insulation performance in real time. Through data analysis, it provides early warnings of potential faults, preventing sudden equipment failures. This not only prevents accidents such as equipment burnout and fires, but also quickly locates fault points, shortens fault handling time, and significantly reduces the number of unplanned power outages, minimizing economic losses for users. Simultaneously, online monitoring devices enable 24-hour automated monitoring, reducing the frequency of manual inspections, lowering labor costs, and accurately locating faulty components, avoiding blind repairs, reducing waste of spare parts, and optimizing resource allocation.

[0004] To ensure system reliability, fault diagnosis technology for high-voltage disconnecting switches has always been a key research focus in the industry. Currently, domestic and international methods for condition monitoring and fault diagnosis of disconnecting switches mainly focus on the following aspects: Infrared thermometry, which detects poor contact or overheating faults by monitoring abnormal contact temperatures; however, this method is susceptible to interference from external heat sources and cannot reflect purely mechanical faults. Vibration signal analysis, which diagnoses mechanical abnormalities by analyzing vibration signals during operation; however, it requires the installation of additional vibration sensors, which are difficult to install and maintain in harsh field environments, and reliability is easily affected. Contact pressure monitoring, which uses technologies such as fiber optic gratings to detect contact finger pressure; however, it also faces the challenges of long-term sensor stability and cost in high-voltage, strong electromagnetic environments. Current signal analysis, which reflects mechanical characteristics by monitoring the amplitude, frequency, and other basic features of the motor current; however, the features are limited, information utilization is insufficient, and it is easily affected by load fluctuations and power grid interference; interference bands in the spectrum can easily lead to misjudgments.

[0005] In summary, existing fault diagnosis technologies for high-voltage disconnector mechanisms either rely on costly and difficult-to-maintain external sensors or remain at a superficial, single-feature analysis of motor current signals. These technologies suffer from a series of problems, including unintuitive diagnosis, incomplete feature set, weak anti-interference capabilities, and low accuracy in intelligent diagnosis. Therefore, there is an urgent need for a new method that can directly utilize readily available drive signals, deeply integrate multi-dimensional fault features, and possess high-precision intelligent diagnostic capabilities to achieve early, accurate, and reliable online diagnosis of faults in high-voltage disconnector mechanisms. Summary of the Invention

[0006] To address these issues, this invention provides a mechanical fault diagnosis system and method based on the fusion of multiple features of motor current, which solves the problems of high cost and difficult maintenance caused by relying on multiple sensors in the prior art, as well as incomplete analysis of single features and low diagnostic accuracy.

[0007] To address the aforementioned technical problems, embodiments of the present invention provide a mechanical fault diagnosis system based on multi-feature fusion of motor current, the system comprising: A current sensor is used to collect the current signal of the high-voltage disconnector drive motor. A signal conditioning circuit, connected to the current sensor, is used to convert the current signal into a unipolar signal suitable for processing. A digital signal processor, connected to the signal conditioning circuit, includes: The ADC sampling module is used to sample the conditioned current signal; The feature extraction module is used to extract at least four feature quantities from the sampled signal, including the current rise slope, steady-state current amplitude, closing operation time, and sideband component. The fault diagnosis module has a built-in pre-trained machine learning classification model, which is used to receive a feature vector composed of at least four feature quantities, and to perform fusion analysis and fault classification on the mechanical operating state of the high-voltage disconnect switch drive motor, and output the fault classification result. The communication module is used to transmit diagnostic results; The host computer is used to receive and display the diagnostic results and issue fault alarms.

[0008] Preferably, the signal conditioning circuit includes a level shifting circuit and a filtering circuit, used to convert the bipolar current signal into a unipolar signal and perform filtering processing.

[0009] Preferably, the current rise slope is obtained by calculating the ratio of the change in current to the time taken during the closing process, from the time the current signal exceeds the start-up threshold to the time taken; the steady-state current amplitude is the root mean square value of the current within at least one power frequency cycle before the closing is completed and after the current signal enters the stable operation phase; the closing operation time is the time interval from the moment the current signal first exceeds the start-up threshold to the moment it begins to drop sharply due to the action of the closing limit switch; the sideband component is the spectral amplitude feature extracted after performing spectral analysis on the current signal during the stable operation phase.

[0010] Preferably, the at least four feature quantities are extracted from four physical dimensions: transient response, steady-state amplitude, time process, and frequency domain modulation, respectively, to jointly constitute a multi-dimensional feature space describing the mechanical operating state.

[0011] Preferably, the machine learning classification model is a random forest classification model, which processes feature vectors in parallel through multiple decision trees. Each decision tree outputs a preliminary classification result, and the fault type is finally determined through a majority voting mechanism.

[0012] Preferably, the random forest classification model automatically learns the importance weights of the four feature quantities—current rise slope, steady-state current amplitude, closing operation time, and sideband components—in fault classification during the training process.

[0013] Preferably, the fault diagnosis module can identify at least one of the following fault types: normal state, mechanism jamming, incomplete closing, and spring failure.

[0014] This invention also provides a mechanical fault diagnosis method based on the fusion of multiple features of motor current. This method employs the aforementioned mechanical fault diagnosis system based on the fusion of multiple features of motor current, and includes the following steps: Acquire the current signal of the high-voltage disconnect switch drive motor; At least four feature quantities are extracted from the current signal, including the current rise slope, steady-state current amplitude, closing operation time, and sideband component. The feature vector consisting of the at least four features is input into a pre-trained machine learning classification model; Based on the output of the machine learning classification model, the mechanical fault diagnosis results of the high-voltage disconnect switch drive motor are obtained.

[0015] Preferably, the feature extraction process includes: Envelope extraction is performed on the current signal, and the start and end times are determined based on the envelope signal; Calculate the current rise slope during the startup phase; Calculate the steady-state current amplitude and sideband components during the steady-state phase; The operation time is calculated based on the start and end times.

[0016] Preferably, the machine learning classification model is a random forest classification model. During the training process, the random forest classification model splits nodes using the Gini impurity criterion and uses Bootstrap sampling to construct multiple decision trees to improve the model's generalization ability and robustness.

[0017] As can be seen from the above technical solutions, this invention application has the following beneficial effects: (1) This invention abandons the existing technology that relies on multiple sensors (such as vibration sensors and pressure sensors), and can complete the core data acquisition for fault diagnosis with only a single current sensor, without the need to add complex sensing equipment. On the one hand, it greatly reduces the hardware procurement cost and avoids the cumbersome process of multi-sensor collaborative calibration; on the other hand, the installation and maintenance of a single sensor is simpler, and it can be adapted to the extreme and harsh environment of high-voltage disconnect switches such as outdoor dust, fog, wind, rain, and strong electromagnetic fields, solving the pain points of traditional multi-sensor deployment and maintenance difficulties and poor long-term stability. At the same time, it simplifies the overall system structure and improves the feasibility of field application.

[0018] (2) This invention creatively extracts features from four physical dimensions: transient response (current rise slope), steady-state amplitude (steady-state current amplitude), time process (closing operation time), and frequency domain modulation (sideband components), constructing a multi-dimensional feature space that comprehensively describes the mechanical operating state. This overcomes the shortcomings of traditional current signal analysis, which relies on only a single feature and has insufficient information utilization. At the same time, a pre-trained random forest classification model is used. Through parallel processing of multiple decision trees and a majority voting mechanism, the importance weights of each feature are automatically learned, effectively suppressing the influence of external factors such as load fluctuations and power grid interference. It can accurately identify various fault types and normal states, such as mechanism jamming, incomplete closing, and spring failure. The diagnostic accuracy is far superior to single feature analysis schemes, and the robustness is stronger. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Referring to the drawings will make the features and advantages of the present invention clearer. The drawings are illustrative and should not be construed as limiting the present invention in any way. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a block diagram of a mechanical fault diagnosis system based on the fusion of multiple features of motor current provided by the present invention; Figure 2 This is the equivalent circuit diagram of the asynchronous motor in this invention; Figure 3 This is a flowchart of the feature detection process in this invention; Figure 4 This is a flowchart of the decision tree fault judgment process in this invention; Figure 5 This is the random forest fault judgment process in this invention; Figure 6 This is a flowchart of a mechanical fault diagnosis method based on the fusion of multiple features of motor current provided by the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1: To address the problems of high cost and difficult maintenance due to reliance on multiple sensors in existing technologies, as well as incomplete single-feature analysis and low diagnostic accuracy. For example... Figure 1 As shown, this invention proposes a mechanical fault diagnosis system based on the fusion of multiple features of motor current. The system includes: A current sensor is used to collect the current signal of the high-voltage disconnector drive motor. A signal conditioning circuit, connected to the current sensor, is used to convert the current signal into a unipolar signal suitable for processing. A digital signal processor, connected to the signal conditioning circuit, includes: The ADC sampling module is used to sample the conditioned current signal; The feature extraction module is used to extract at least four feature quantities from the sampled signal, including the current rise slope, steady-state current amplitude, closing operation time, and sideband component. The fault diagnosis module has a built-in pre-trained machine learning classification model, which is used to receive a feature vector composed of at least four feature quantities, and to perform fusion analysis and fault classification on the mechanical operating state of the high-voltage disconnect switch drive motor, and output the fault classification result. The communication module is used to transmit diagnostic results; The host computer is used to receive and display the diagnostic results and issue fault alarms.

[0022] As can be seen from the above technical solution, this invention proposes a mechanical fault diagnosis system based on multi-feature fusion of motor current. It collects the current signal of the motor driven by the high-voltage disconnect switch using a current sensor, providing core raw data for fault diagnosis. Furthermore, only a single sensor is required to meet diagnostic needs, significantly reducing system hardware costs and deployment complexity. The signal conditioning circuit connected to the current sensor converts the collected bipolar current signal into a unipolar signal suitable for processing, effectively adapting to the processing requirements of the subsequent digital signal processor, while filtering out noise interference to ensure signal quality. In the digital signal processor connected to the signal conditioning circuit, the ADC sampling module samples the conditioned current signal, providing a regular digital signal for feature extraction. The feature extraction module extracts the current rise slope from the sampled signal. The system incorporates four multi-dimensional features: rate of change, steady-state current amplitude, closing operation time, and sideband components. These features comprehensively cover physical dimensions such as transient, steady-state, time-series, and frequency domain, overcoming the limitations of relying solely on single feature information and laying the foundation for accurate diagnosis. The fault diagnosis module has a built-in pre-trained machine learning classification model that receives feature vectors composed of the four features and performs fusion analysis and fault classification. This enables accurate identification of various mechanical faults such as mechanism jamming, incomplete closing, and spring failure, improving diagnostic accuracy and anti-interference capabilities. The communication module ensures efficient transmission of diagnostic results and real-time information synchronization. The host computer receives and displays the diagnostic results and issues fault alarms, providing visualized fault presentation and timely warnings. This facilitates rapid response by maintenance personnel and ensures the safe and stable operation of high-voltage disconnect switches and traction power supply systems.

[0023] The mechanical fault diagnosis system based on multi-feature fusion of motor current described in this invention is primarily used for online mechanical fault diagnosis of high-voltage disconnector drive motors. Specifically, it includes a current sensor, signal conditioning circuit, digital signal processor (DSP), communication module, and host computer. These modules work collaboratively to achieve fully automated processing from signal acquisition to fault alarm. The system architecture corresponds to the attached diagram. Figure 1 As shown.

[0024] In this embodiment, the current sensor serves as the core component for signal acquisition. It is selected to match the operating current range of the high-voltage disconnect switch drive motor (preferably a three-phase asynchronous motor). Specifically, a Hall effect sensor is used, installed in the motor's stator current circuit, to acquire the stator current signal during real-time operation of the drive motor. This sensor must meet the long-term stability requirements under high-voltage, strong electromagnetic environments. The acquired signal is a bipolar AC current signal with an amplitude range matching the motor's rated current to ensure undistorted signal quality and provide reliable raw data for subsequent feature extraction.

[0025] In this embodiment, the signal conditioning circuit is directly connected to the output terminal of the current sensor and consists of a level shifting circuit and a filter circuit connected in series. The specific implementation is as follows: Level shifting circuit: An operational amplifier is used to build an adder circuit to convert the bipolar current signal output by the sensor into a unipolar signal that can be received by the DSP. During the shifting process, the linearity error of the signal is guaranteed to be ≤0.5% to avoid signal distortion affecting feature calculation. Filtering circuit: A second-order RC low-pass filter circuit is adopted, with the cutoff frequency set at 500Hz. It is used to filter out high-frequency noise such as power grid interference and motor electromagnetic noise. The ripple coefficient of the filtered signal is ≤1%, ensuring signal stability and adapting to the sampling requirements of the DSP. It realizes the function of "converting bipolar current signals into unipolar signals and performing filtering processing".

[0026] In this embodiment, the digital signal processor is the core computing unit of the system, and it integrates an ADC sampling module, a feature extraction module, a fault diagnosis (algorithm judgment) module, and a communication module. Specific implementation details are as follows: ADC sampling module: The DSP uses a built-in 12-bit high-speed ADC converter to sample the conditioned unipolar current signal at equal intervals. The sampling frequency is set to 1kHz. The sampled data is stored in the DSP's internal buffer in real time (buffer capacity ≥ 1024 bytes) to ensure that the sampled data is continuously stored without loss. The sampling accuracy error is ≤ 0.1%, which meets the data density requirements of feature extraction. Feature extraction module: Based on DSP hardware computing resources, it automatically extracts four-dimensional features through software algorithms. The features are constructed from four physical dimensions: transient response, steady-state amplitude, time process, and frequency domain modulation, collectively forming a multi-dimensional feature space. The specific extraction process is detailed in the appendix. Figure 3 As shown; Fault diagnosis module: Built-in pre-trained random forest classification model. The model processes feature vectors in parallel through multiple decision trees. Each decision tree is a binary tree structure, containing internal decision nodes and leaf classification nodes. Decision nodes store feature indexes and split thresholds, and leaf nodes store fault category identifiers. Communication module: It adopts RS485 communication protocol or Ethernet interface, with baud rate set to 9600bps (RS485) or 100Mbps (Ethernet), to transmit fault diagnosis results to the host computer in real time, with communication delay ≤100ms, to ensure timely feedback of diagnosis results.

[0027] In this embodiment, the host computer is an industrial-grade computer, equipped with dedicated fault diagnosis and monitoring software, and has the following functions: Data reception: Receives diagnostic results transmitted by the DSP via the communication module, including fault type, characteristic data, diagnostic confidence level, etc. Display function: The graphical interface displays the operating status (normal / fault), fault type, characteristic value curves, historical diagnostic records, etc. of the high-voltage disconnect switch drive motor in real time. Alarm function: When the diagnostic result indicates a fault, an audible and visual alarm (buzzer + red indicator light) will be triggered to alert maintenance personnel. At the same time, the fault occurrence time, characteristic data, and other information will be recorded for easy traceability and analysis.

[0028] Furthermore, taking an asynchronous motor as an example, we analyze the significance and principle of characteristic quantity selection. Most high-voltage disconnect switches have electrically operated operating mechanisms. When a fault occurs in the operating mechanism, the motor torque often changes, and consequently, the motor stator current will also change differently from the normal operating current. Therefore, we can first derive the relationship between operating torque and motor current.

[0029] refer to Figure 2 Describe the equivalent circuit diagram of an asynchronous motor, as shown in the figure: It is the input voltage of the asynchronous motor; It is the current flowing through the stator windings; The equivalent resistance for iron loss; This refers to the rotor phase current referred to the stator side; It is the phase resistance of the stator winding of the asynchronous motor; It is the leakage reactance of the stator winding of the asynchronous motor; This is the phase resistance of each phase winding of the motor rotor after conversion, and This is the converted leakage reactance; and These are the stator potential and the converted rotor potential, respectively. The excitation current required to create a magnetic field for the motor; The main magnetic flux reactance of the excitation branch; This is for simulating resistance.

[0030] Since the leakage impedance is much smaller than the excitation impedance, the excitation impedance branch in the diagram can be considered as an open circuit, thus yielding the relationship between current and slip: ; The total mechanical power of the isolating switch asynchronous motor Equal to electromagnetic torque With rotor mechanical angular velocity Product: , Electromagnetic power and synchronous angular velocity for: , , in, This refers to the number of stator phases of the motor. This is the synchronous speed.

[0031] From the various equations, we can obtain: .

[0032] Therefore, the electromagnetic torque output by the motor is related to the stator current of the motor. A functional relationship exists.

[0033] The torque balance equation for an asynchronous motor is: , in, The electromagnetic torque is the output torque of the motor under no-load conditions. Since the additional and mechanical losses are small and negligible, it is approximately assumed that the electromagnetic torque is equal to the output torque. Therefore, when a mechanical failure occurs in the disconnecting switch, the resulting change in operating torque will inevitably lead to a change in the stator current.

[0034] Furthermore, any mechanical fault in the high-voltage disconnector mechanism will alter its dynamic characteristics, thereby modulating the stator current of the drive motor. To achieve accurate diagnosis, this invention abandons the limitations of a single feature quantity and creatively extracts features from four complementary physical dimensions: transient, steady-state, time-series, and frequency domain, constructing a feature vector that comprehensively describes the mechanism's operating state, specifically including: (1) Current rise slope, this characteristic quantity characterizes the dynamic process of the motor electromagnetic torque overcoming the static friction and inertia of the system at the moment of closing and starting.

[0035] At the instant the motor starts, the system's equation of motion can be written as: , because Differentiating both sides, we get: , The two equations combined yield: , Substituting the function relating current and slip, we get: , Under normal conditions, the load torque It conforms to the design curve. The motor follows its inherent... Curve acceleration, The current decreases steadily from 1. The starting current then smoothly decreases. The shape and slope of the current rise are fixed. Mechanism jamming increases starting resistance, causing the current rise curve to slow down; spring failure may reduce the load and change the rise slope. This characteristic is a key indicator for diagnosing abnormalities in the starting process.

[0036] (2) Steady-state current amplitude, which is a characteristic quantity that represents the maintaining torque required during the uniform and stable operation of the mechanism.

[0037] When the motor is running stably, the electromagnetic torque is equal to the load torque: , Incomplete closing, mechanism jamming, spring failure, and discrepancy between electromagnetic torque and load torque can lead to... Changes in this characteristic are a direct indicator for diagnosing abnormal operating loads.

[0038] (3) Closing operation time, which is a macroscopic measure of the total time taken for a complete closing operation.

[0039] The kinematic equation for the rotation of the motor is: ; From the startup location To the closing position The total displacement is constant, and the angular velocity is constant. The entire time-domain waveform is composed of net torque Decide: , When the mechanism jams or the spring fails. Increasing the angular acceleration leads to a decrease in the angular acceleration, until the same displacement is achieved. The required time is extended; conversely, when the spring fails, Decreasing the angular acceleration leads to an increase in the angular acceleration, achieving the same displacement. The required time is reduced; when the circuit breaker fails to close properly, The time required for the change also changes; this characteristic is a comprehensive indicator for diagnosing the overall performance degradation of the diagnostic system.

[0040] (4) Sideband components: When jamming or incomplete opening / closing occurs during the opening and closing of the high-voltage disconnector, the asymmetrical air gap magnetic field will cause deformation of the rotating magnetic field in the air gap. The most direct impact on the motor is that it will cause torque fluctuations. Torque fluctuations cause speed fluctuations. Therefore, the rotor angular velocity changes with the torque, and the frequency of speed fluctuations is also increasing. Also for The modulated signal is: , After transformation, we get: , Therefore, the fundamental frequency and frequency of the motor current can be used as... The sideband components serve as the basis for diagnosing fault types in the system. An increase in the energy of the sideband components directly indicates the presence of local, periodic impulse-type faults, which is a precise indicator for diagnosing system faults.

[0041] Furthermore, the core of the feature extraction module is to accurately extract four feature quantities from the sampled signal: current rise slope, steady-state current amplitude, closing operation time, and sideband components. The extraction process is based on the physical characteristics of motor operation and is automated by combining software algorithms. The specific implementation steps are as follows: (I) Signal preprocessing and envelope extraction 1. First, the digital signal obtained by the ADC sampling... Preprocessing was performed to remove abnormal impulse interference (outliers deviating from the mean by 3σ criteria were removed). 2. Extract the signal envelope using the DSP's internal software algorithm. The specific steps are as follows: Preprocessed AC current signal Take the absolute value to obtain the full-wave rectified signal. ,Right now ; Using a first-order infinite impulse response (IIR) filter For digital low-pass filtering, the difference equation is: ,in The filter coefficients (values ​​ranging from 0.1 to 0.3, calibrated according to actual operating conditions) are used to smooth the signal and obtain an envelope that reflects the overall trend of current amplitude variation. .

[0042] (ii) Determining the start and end times of circuit breaker closing 1. Start-up timing determination: Real-time comparison of envelope amplitude. Compared with the preset starting threshold (the threshold is 10%~20% of the motor's rated current, calibrated using historical normal operation data), when When the starting threshold is exceeded for three consecutive sampling cycles, the closing operation is determined to have started, and the start time is recorded. ; 2. End time determination: Continuously monitor the envelope during the closing process. The rate of change, when When a sudden drop occurs due to the activation of the closing limit switch (drop ≥ 50% and lasting for 2 sampling periods), the closing operation is considered complete, and the end time is recorded. .

[0043] (III) Calculation and Implementation of Each Characteristic Quantity 1. Current rise slope: Select envelope From the start time Until the peak time Data segment ( ; The rising edge line of the data segment is fitted using the least squares method, and the slope of the line is calculated. That is, the slope of the current rise. ,in The sampling period is 1ms. This characteristic quantity represents the dynamic process of the motor overcoming static friction and inertia at the moment of closing and starting.

[0044] 2. Steady-state current amplitude: When envelope When the rate of change is ≤5% and lasts for ≥1 power frequency cycle (20ms, corresponding to 50Hz power frequency), the motor is judged to have entered the stable operation stage; Select the original sampled signal of at least one power frequency cycle within this steady-state stage, and calculate its root mean square value as the steady-state current amplitude. The calculation formula is: , in, Number of sampling points in the steady-state phase ( , here ≥20), and These represent the start and end times of the steady-state phase, respectively.

[0045] 3. Closing operation time: Based on the recorded start time and end time Calculate the time interval , which is the total time of the closing operation, reflects the time required for a complete closing operation.

[0046] 4. Sideband components: A Fast Fourier Transform (FFT) was performed on the current signal in the steady state phase, with 1024 transformation points, to obtain the spectrum. Extracting the fundamental frequency from the spectrum Sideband components on both sides (frequency is ,in (For motor slip), the ratio of the amplitude of the sideband component to the amplitude of the fundamental frequency is calculated as the characteristic value of the sideband component. This characteristic value reflects the speed modulation effect caused by motor torque fluctuation.

[0047] Furthermore, the training and diagnosis process of the random forest classification model built into the fault diagnosis module is implemented as follows: (I) Model Training Implementation 1. Sample collection: Collect N sets of historical fault data samples (N≥1000) from the actual substation operating environment, covering four typical operating conditions: normal state, mechanism jamming, incomplete closing, and spring failure. Each set of samples contains the values ​​of four characteristic quantities and the corresponding fault type label. 2. Training subset construction: Using the Bootstrap sampling technique, n sets of samples (n≈0.7N) are randomly selected with replacement from the original sample set as the training subset of a single decision tree. This process is repeated to construct multiple decision trees (the number of decision trees is set to 50~100, which can be adjusted according to the diagnostic accuracy requirements) to ensure the difference between the base learners of each decision tree. 3. Decision Tree Construction: A single decision tree uses Gini impurity as the node splitting criterion. The formula for calculating Gini impurity is: , in, The sample set for the current node. Number of fault types ( ), For the sample set belonging to the first The proportion of class failures; when splitting at each node, select the feature that maximizes the reduction of Gini impurity and the corresponding splitting threshold from four features to construct a binary decision tree; 4. Weight Learning: During model training, the importance weights of the four features in fault classification are automatically learned. The importance weights are determined by calculating the contribution of each feature to the splitting of all decision tree nodes. The higher the contribution, the greater the weight, and the higher the decision priority of the corresponding feature in subsequent diagnosis.

[0048] (II) Real-time Diagnosis Implementation 1. Feature vector input: The four feature quantities obtained by the feature extraction module are arranged in a fixed order (current rise slope → steady-state current amplitude → closing operation time → side frequency component) to form a feature vector, which is then input into the random forest classification model; 2. Parallel Decision Making: Feature vectors are simultaneously input into all decision trees via the parallel architecture of the DSP. Each decision tree performs inference independently: starting from the root node, the corresponding feature value is obtained based on the feature index stored in the node, and compared with the splitting threshold. If it is less than the threshold, the left child node is traversed; otherwise, the right child node is traversed, until the leaf node is reached. The preliminary classification result of the tree is then output (corresponding to the appendix). Figure 4 (Decision tree fault diagnosis process) 3. Majority Voting: The system has a built-in voting statistics module that counts the preliminary classification results of all decision trees and uses a majority voting mechanism to determine the final fault type, i.e., the fault type with the most votes is the diagnosis result; if the confidence level (number of votes / total number of decision trees) of the fault type with the highest number of votes is ≥90%, the result is output directly; if the confidence level is between 70% and 90%, it is marked as a suspected fault; if the confidence level is <70%, a diagnosis anomaly is output, prompting for re-collection of data; 4. Result Output: The diagnostic results (fault type / normal state, confidence level) are transmitted to the host computer via the communication module, completing the fault diagnosis process (see attached diagram). Figure 5 (Random forest fault diagnosis process).

[0049] Example 2: Figure 6 As shown, this invention provides a mechanical fault diagnosis system based on the fusion of multiple features of motor current. This system is used to implement the mechanical fault diagnosis method based on the fusion of multiple features of motor current in Embodiment 1 above, and specifically includes: Acquire the current signal of the high-voltage disconnect switch drive motor; At least four feature quantities are extracted from the current signal, including the current rise slope, steady-state current amplitude, closing operation time, and sideband component. The feature vector consisting of the at least four features is input into a pre-trained machine learning classification model; Based on the output of the machine learning classification model, the mechanical fault diagnosis results of the high-voltage disconnect switch drive motor are obtained.

[0050] Furthermore, the mechanical fault diagnosis method of the present invention is implemented using the above-mentioned system, as detailed below: First, the high-voltage disconnect switch is closed. The current sensor collects the stator current signal of the drive motor in real time, and the collection time covers the entire closing process (from motor start-up to closing). The amplitude range of the collected bipolar current signal matches the sensor range to ensure no saturation distortion.

[0051] Then, the acquired bipolar current signal is input to the signal conditioning circuit, converted into a unipolar signal by the level shifting circuit, and then filtered out high-frequency interference by the low-pass filter circuit to obtain the conditioned signal adapted to the DSP. The DSP's ADC sampling module samples the conditioned signal at equal intervals at a sampling frequency of 1kHz. The sampled data is stored in the buffer area. The sampling process is synchronized with the closing operation to ensure data integrity.

[0052] Next, following the feature extraction process of the feature extraction module in Example 1, envelope extraction and start / end time determination are completed, and four feature quantities are calculated: current rise slope, steady-state current amplitude, closing operation time, and sideband component, to construct a four-dimensional feature vector.

[0053] Secondly, the four-dimensional feature vector is input into the pre-trained random forest classification model. Through the real-time diagnosis process of the fault diagnosis module in Example 1, the fault classification results are obtained, including one of the following: normal state, mechanism jamming, incomplete closing, spring failure, or suspected fault, and abnormal diagnosis prompt.

[0054] Finally, the communication module transmits the diagnostic results to the host computer, which displays the fault type (or normal state), characteristic quantity curves, diagnostic confidence level, and fault occurrence time in a graphical interface. If the diagnostic result is a fault type, the host computer immediately triggers an audible and visual alarm and stores the fault data in the local database for subsequent traceability and analysis. If it is a normal state, only the operating status and characteristic quantity data are displayed. If it is a suspected fault or diagnostic anomaly, the system prompts maintenance personnel to check the equipment or re-execute the closing operation for secondary diagnosis.

[0055] This embodiment provides a mechanical fault diagnosis system based on the fusion of multiple features of motor current, which is used to implement the aforementioned mechanical fault diagnosis method based on the fusion of multiple features of motor current. Therefore, the specific implementation of the mechanical fault diagnosis system based on the fusion of multiple features of motor current can be found in the previous embodiment section of the mechanical fault diagnosis method based on the fusion of multiple features of motor current. To avoid redundancy, it will not be repeated here.

[0056] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0057] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0058] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0059] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A mechanical fault diagnosis system based on multi-feature fusion of motor current, characterized in that, include: A current sensor is used to collect the current signal of the high-voltage disconnector drive motor. A signal conditioning circuit, connected to the current sensor, is used to convert the current signal into a unipolar signal suitable for processing. A digital signal processor, connected to the signal conditioning circuit, includes: The ADC sampling module is used to sample the conditioned current signal; The feature extraction module is used to extract at least four feature quantities from the sampled signal, including the current rise slope, steady-state current amplitude, closing operation time, and sideband component. The fault diagnosis module has a built-in pre-trained machine learning classification model, which is used to receive a feature vector composed of at least four feature quantities, and to perform fusion analysis and fault classification on the mechanical operating state of the high-voltage disconnect switch drive motor, and output the fault classification result. The communication module is used to transmit diagnostic results; The host computer is used to receive and display the diagnostic results and issue fault alarms.

2. The mechanical fault diagnosis system based on multi-feature fusion of motor current according to claim 1, characterized in that, The signal conditioning circuit includes a level shifting circuit and a filtering circuit, used to convert bipolar current signals into unipolar signals and perform filtering processing.

3. The mechanical fault diagnosis system based on multi-feature fusion of motor current according to claim 1, characterized in that, The current rise slope is obtained by calculating the ratio of the change in current to the time taken during the closing process, from the time the current signal exceeds the start-up threshold to the time taken to reach the peak value; the steady-state current amplitude is the root mean square value of the current within at least one power frequency cycle before the closing is completed and after the current signal enters the stable operation phase; the closing operation time is the time interval from the moment the current signal first exceeds the start-up threshold to the moment when it begins to drop sharply due to the action of the closing limit switch; the sideband component is the spectral amplitude feature extracted after performing spectral analysis on the current signal during the stable operation phase.

4. The mechanical fault diagnosis system based on multi-feature fusion of motor current according to claim 1, characterized in that, The at least four feature quantities are extracted from four physical dimensions: transient response, steady-state amplitude, time process, and frequency domain modulation, respectively, and together they constitute a multi-dimensional feature space describing the mechanical operating state.

5. The mechanical fault diagnosis system based on multi-feature fusion of motor current according to claim 1, characterized in that, The machine learning classification model is a random forest classification model, which processes feature vectors in parallel through multiple decision trees. Each decision tree outputs a preliminary classification result, and the fault type is finally determined through a majority voting mechanism.

6. The mechanical fault diagnosis system based on multi-feature fusion of motor current according to claim 5, characterized in that, During training, the random forest classification model automatically learns the importance weights of four features—current rise slope, steady-state current amplitude, closing operation time, and sideband components—in fault classification.

7. The mechanical fault diagnosis system based on multi-feature fusion of motor current according to claim 1 or 5, characterized in that, The fault diagnosis module can identify at least one of the following fault types: normal state, mechanism jamming, incomplete closing, and spring failure.

8. A mechanical fault diagnosis method based on multi-feature fusion of motor current, characterized in that, The method employs the mechanical fault diagnosis system based on multi-feature fusion of motor current as described in any one of claims 1 to 7, and includes the following steps: Acquire the current signal of the high-voltage disconnect switch drive motor; At least four feature quantities are extracted from the current signal, including the current rise slope, steady-state current amplitude, closing operation time, and sideband component. The feature vector consisting of the at least four features is input into a pre-trained machine learning classification model; Based on the output of the machine learning classification model, the mechanical fault diagnosis results of the high-voltage disconnect switch drive motor are obtained.

9. The mechanical fault diagnosis method based on multi-feature fusion of motor current according to claim 8, characterized in that, The feature extraction process includes: Envelope extraction is performed on the current signal, and the start and end times are determined based on the envelope signal; Calculate the current rise slope during the startup phase; Calculate the steady-state current amplitude and sideband components during the steady-state phase; The operation time is calculated based on the start and end times.

10. The mechanical fault diagnosis method based on multi-feature fusion of motor current according to claim 8, characterized in that, The machine learning classification model is a random forest classification model. During the training process, the random forest classification model splits nodes using the Gini impurity criterion and uses Bootstrap sampling to construct multiple decision trees to improve the model's generalization ability and robustness.

Images