Mechanical action characteristic intelligent monitoring system and method for direct-current miniature circuit breaker

Abstract

The invention relates to a mechanical action characteristic intelligent monitoring system and method for a direct-current miniature circuit breaker, and relates to the technical field of electrical equipment monitoring, and the system comprises a signal collection module which is used for sensing the mechanical action of the circuit breaker, collecting mechanical stress signals of different dimensions, and sending the mechanical stress signals to a control module; recording a corresponding signal timestamp; the signal processing module is used for cleaning the mechanical stress signal according to the signal frequency characteristic, and performing gain amplification and digital-to-analog conversion to a preset signal amplitude to obtain a corresponding digital gain signal; the intelligent analysis module is used for synchronizing a signal timestamp, calculating an action characteristic parameter according to the digital gain signal, monitoring and identifying the mechanical state of the circuit breaker in combination with a fault characteristic library, and outputting fault early warning information; through multi-source cooperative vibration, current and position, the intelligent analysis module is combined to calculate multiple characteristic parameters such as opening time, over travel time and closing time, and multi-dimensional quantitative monitoring of mechanical action characteristics of the circuit breaker is realized.

Description

Technical Field

[0001] This application relates to the field of electrical equipment monitoring technology, and in particular to an intelligent monitoring system and method for the mechanical action characteristics of DC miniature circuit breakers. Background Technology

[0002] Miniature circuit breakers (MCBs) are the most widely used terminal protection devices in building electrical distribution systems. They are used for protection against short circuits, overloads, and overvoltages in single-phase and three-phase circuits up to 125A, and are available in four types: single-pole 1P, two-pole 2P, three-pole 3P, and four-pole 4P. A miniature circuit breaker consists of an operating mechanism, contacts, protection devices (various trip units), and an arc-extinguishing system. Its main contacts are closed manually or electrically. After the main contacts are closed, the free-trip mechanism locks them in the closed position. The coil of the overcurrent trip unit and the thermal element of the thermal trip unit are connected in series with the main circuit, while the coil of the undervoltage trip unit is connected in parallel with the power supply. When a short circuit or severe overload occurs, the armature of the overcurrent trip unit engages, causing the free-trip mechanism to operate and disconnecting the main circuit. When the circuit is overloaded, the thermal element of the thermal trip unit heats up, causing the bimetallic strip to bend and pushing the free-trip mechanism to operate. When the circuit is undervoltageed, the armature of the undervoltage trip unit releases. This also causes the free release mechanism to operate.

[0003] Existing patents disclose a circuit breaker simulation device and method based on electrical and dynamic characteristics. In the simulation device, a human-machine interface unit sends parameter setting instructions to a microprocessor unit, and a precise time base unit provides the operating clock for the microprocessor unit. The microprocessor unit transmits the parameters set by the human-machine interface unit to the electrical characteristic unit and the dynamic characteristic unit. When the microprocessor unit receives a trigger signal, the electrical characteristic unit and the dynamic characteristic unit respectively simulate the opening and closing functions of the circuit breaker and the movement of the circuit breaker's transmission mechanism, and send the electrical characteristic parameters and dynamic characteristic parameters back to the microprocessor unit. The technical solution of the above invention has the advantages of complete functionality, simple operation, and ease of use. It also features strong operability, with time, stroke, speed, and current that can be arbitrarily set. It can serve as an important supporting equipment for relay protection testing and as a verification standard for high-voltage circuit breaker dynamic characteristic testers.

[0004] The existing technical solutions mentioned above have the following drawbacks: 1. Existing monitoring equipment generally relies on host computers for data processing and analysis, which cannot perform intelligent diagnosis, resulting in large data transmission delays, poor real-time performance, and diagnostic accuracy limited by the performance of external equipment. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the purpose of this application is to generate a fault feature library through a lightweight fault diagnosis model, monitor and identify action characteristic parameters, and output fault warning information. It does not rely on a host computer and achieves full-process embedded intelligent monitoring locally on the circuit breaker. The data processing latency is controlled at the millisecond level, and mechanical faults can be captured in real time. This completely solves the pain point of poor real-time performance of traditional solutions and ensures the operational reliability of DC power systems.

[0006] This was achieved using the following technical solutions: In a first aspect, this application provides an intelligent monitoring system for the mechanical operating characteristics of a DC miniature circuit breaker; comprising: The signal acquisition module is connected to the mechanical action generating mechanism of the circuit breaker at its input end. It is used to sense the mechanical action of the circuit breaker, collect mechanical stress signals of different dimensions, and record the corresponding signal timestamps. The signal processing module, whose input end is connected to the output end of the signal acquisition module, is used to clean the mechanical stress signal according to the signal frequency characteristics, and to amplify and convert the signal to a preset signal amplitude to obtain the corresponding digital gain signal. The intelligent analysis module, whose input end is connected to the output end of the signal processing module, is used to synchronize signal timestamps, calculate action characteristic parameters based on digital gain signals, monitor and identify the mechanical status of the circuit breaker in conjunction with the fault feature database, and output fault warning information.

[0007] By adopting the above technical solution, the signal acquisition module acquires multi-dimensional stress signals generated by the mechanical action of the circuit breaker in real time and records the timestamps (such as signal sensing based on vibration sensors or displacement sensors). Combined with the signal processing module, the noise is cleaned by bandpass filtering algorithm, the amplitude is optimized by automatic gain control (AGC), and digital-to-analog conversion is achieved by high-precision AD conversion chip (such as signal conditioning and AD conversion technology mentioned in the search results). Finally, the intelligent analysis module synchronizes the timestamps and extracts the action characteristic parameters based on time-frequency analysis (such as FFT or wavelet transform). Combined with classification algorithms such as SVM or random forest and fault feature library, the status is monitored, thereby realizing high-precision, real-time fault early warning of the mechanical status of the circuit breaker, improving system reliability and maintenance efficiency.

[0008] This application further specifies that the signal acquisition module includes: The vibration sensing unit is used to collect the mechanical stress signal of the circuit breaker's operating mechanism during the opening and closing process, obtain the mechanical vibration signal, and record the vibration timestamp. The current sensing unit is used to monitor the mechanical stress signal when the contact state of the circuit breaker changes, sense the current change signal, and record the change timestamp. The position sensing unit is used to detect the mechanical stress signal generated by the status of the position indication component of the circuit breaker, capture the position trigger signal, and record the trigger timestamp.

[0009] Furthermore, the signal processing module includes: The signal filtering unit is used to filter and denoise mechanical vibration signals, current surge signals and position trigger signals according to the signal frequency characteristics to obtain the corresponding key feature signals. The signal amplification unit is used to amplify the key feature signal according to the preset signal amplitude to obtain the corresponding gain analog signal; The analog-to-digital conversion unit is used to sample and convert the gain analog signal according to the Nyquist theorem to obtain the corresponding digital gain signal.

[0010] Furthermore, the intelligent analysis module includes: The time synchronization unit is used to synchronize the vibration timestamp and the trigger timestamp according to the abrupt change timestamp, and align the corresponding digital gain signals to obtain the corresponding synchronization gain signal. The parameter calculation unit is used to calculate the corresponding operating characteristic parameters based on the timestamp difference between different mechanical states of the circuit breaker during the opening and closing process, combined with the synchronous gain signal. The intelligent diagnostic unit is used to build a fault diagnosis model based on historical characteristic parameters, generate a fault feature library, monitor and identify action characteristic parameters, and output fault warning information. The functional calibration unit is used to periodically calibrate the signal acquisition module and the signal processing module according to the built-in standard signal source or external calibration instructions; The data management unit is used for local storage and retrieval of action characteristic parameters and fault warning information.

[0011] By adopting the above technical solution, the mechanical stress signals during the opening and closing process of the circuit breaker are collected from multiple dimensions by vibration, current change and position trigger sensing units and the timestamps are recorded (such as vibration signal analysis, change detection and position trigger detection algorithms). Combined with bandpass filtering (such as wavelet denoising or frequency domain filtering algorithms), automatic gain control (AGC) and high-precision AD conversion based on Nyquist theorem in the signal processing module, multi-source signals are synchronized by time alignment algorithm (such as dynamic time warping DTW), and a fault diagnosis model is constructed by time domain / frequency domain parameter calculation (such as root mean square, peak detection or FFT analysis) and classification algorithms such as SVM / random forest. The stability of signal acquisition and processing is ensured by periodic calibration algorithm (such as self-calibration based on standard signal source), and the local storage and backtracking of parameters and early warning information are realized through the data management unit. In this way, the accuracy and real-time performance of the mechanical condition monitoring of the circuit breaker are improved under complex working conditions, the false alarm rate is reduced and the system reliability is enhanced.

[0012] This application further specifies that the intelligent diagnostic unit includes: The data partitioning layer is used to divide historical characteristic parameters according to a preset data splitting ratio to obtain parameter training set and parameter test set; The data aggregation layer is used to classify the parameter training set according to the fault type, resulting in normal parameter blocks and several fault parameter blocks. The fault sequencing layer is used to assign priorities to fault parameter blocks according to the severity of the fault, thereby obtaining a parameter priority sequence for different faults. The cross-validation layer is used to perform cross-analysis on fault parameters of different degrees according to the parameter priority sequence, and to combine it with normal parameter blocks for verification and screening to determine the initial fault parameter range. The branch creation layer is used to construct branches for each layer of the neural network based on the number of fault types, and to determine the number of convolutional blocks and the hierarchical branch skeleton. The feature extraction layer is used to extract features from the normal parameter block based on the first extraction branch using N convolutional blocks and M pooling blocks to obtain the normal feature matrix; Based on the second extraction branch and the fault severity level k, feature extraction is performed on the fault parameter block using k*N convolutional blocks and 2M pooling blocks to obtain the fault feature matrix. The iterative training layer is used to perform several iterations on the normal feature matrix based on the first training branch using s residual blocks connected in parallel, 2s downsampling blocks and the tanh activation function, to generate the normal weight matrix. Based on the second training branch, the fault feature matrix is ​​iterated hierarchically using k*s InceptionV3 blocks, (k+1)*s upsampling blocks and ReLU activation function connected in series to generate a fault hierarchy weight matrix. Based on the third training branch, a flat expansion block is used to unify the dimensions of the normal weight matrix and the fault level weight matrix, and several iterations are performed in combination with the spatiotemporal attention mechanism to generate a hierarchical comprehensive weight matrix. The output layer is used to generate a single-class fault identifier based on the hierarchical branch skeleton, using 3 fully connected blocks, 2 random deactivated blocks and cross-entropy loss function, and to identify the fault feature type by combining the hierarchical comprehensive weight matrix on the parameter test set. The verification and localization layer is used to calculate the false positive rate and accuracy rate based on the fault feature type and the preset fault label, and generate an initial confusion matrix based on the preset accuracy threshold; at the same time, it locates and analyzes the corresponding action characteristic parameters according to the false positive rate, and generates a judgment optimization matrix based on the initial fault parameter range. The update optimization layer is used to transpose and correct the fault level weight matrix based on the confusion initial matrix, and to perform correction operations on the level comprehensive weight matrix with the decision optimization matrix to obtain the level corrected weight matrix. The model self-calibration layer is used to correct the weight matrix according to the hierarchy, perform parallel identification on the validation set, output the mechanical state of the circuit breaker, calculate the identification accuracy, and compare it with the accuracy threshold. If the recognition accuracy is greater than the accuracy threshold, the current level correction weight matrix is ​​determined to be a fault diagnosis model. If not, then based on the initial confusion matrix and the decision optimization matrix, calibration coefficients are generated, and the cross-entropy loss function is optimized and calibrated until the accurate threshold is reached; The rule generation layer is used to parse and aggregate the fault diagnosis model according to the rule format and the fault type to build a fault feature library; The rule application layer is used to match action characteristic parameters based on the fault feature library; If a match is successful, output a fault warning message containing the fault type and fault severity level; If the matching fails, the current action characteristic parameters are identified according to the fault diagnosis model, a new fault judgment rule is output and populated into the fault feature library.

[0013] By employing the above technical solution, historical characteristic parameters are split and classified through data partitioning and aggregation layers (e.g., dividing training / test sets and aggregating fault types according to preset ratios). Priority assignment by the fault ranking layer and validation filtering by the cross-validation layer (e.g., cross-analysis based on severity level) are combined. A branching layer is used to construct a neural network branch structure (e.g., combining InceptionV3 with residual blocks). In the feature extraction layer, normal and fault features are extracted through multiple convolutional / pooling blocks (e.g., N convolutional blocks and k*N InceptionV3 blocks working together). The iterative training layer uses tanh and ReLU activation functions (e.g., the division of residual blocks with InceptionV3 blocks). The model employs a multi-layered iterative approach, combining spatiotemporal attention mechanisms to optimize the weight matrix. The output layer generates a single-class fault identifier using fully connected layers and cross-entropy loss functions (e.g., random deactivation and loss function optimization strategies). The verification and localization layer dynamically corrects the model using confusion and decision optimization matrices (e.g., false positive rate and accuracy analysis). The model self-calibration layer ensures model convergence through parallel identification and calibration coefficient adjustment. Finally, the rule generation and application layer constructs and updates the fault feature library (e.g., rule-based parsing aggregation), forming a high-precision, robust fault diagnosis model. This significantly improves the accuracy and real-time performance of circuit breaker mechanical condition identification and reduces false alarm rates and enhances system adaptability and reliability through a self-calibration mechanism.

[0014] Secondly, this application also provides an intelligent monitoring method for the mechanical operating characteristics of DC miniature circuit breakers, employing the following technical solution: A method for intelligent monitoring of the mechanical operating characteristics of DC miniature circuit breakers, applied to an intelligent monitoring system for mechanical operating characteristics, comprising: The system senses the mechanical movement of the circuit breaker, collects mechanical vibration signals, current surge signals, and position trigger signals, and records the corresponding signal timestamps. The mechanical vibration signal, current surge signal, and position trigger signal are cleaned according to the signal frequency characteristics, and then amplified and converted to a preset signal amplitude to obtain the corresponding digital gain signal. The vibration timestamp and trigger timestamp are synchronized based on the mutation timestamp, and the corresponding digital gain signals are aligned to obtain the corresponding synchronization gain signal. Based on the timestamp difference between different mechanical states of the circuit breaker during the opening and closing process, the corresponding operating characteristic parameters are calculated in conjunction with the synchronization gain signal. Based on historical characteristic parameters, a fault diagnosis model is constructed, a fault feature library is generated, and action characteristic parameters are monitored and identified to output fault warning information.

[0015] By adopting the above technical solutions, multi-source signal fusion (such as vibration signal analysis, current surge sensing, and position trigger detection algorithms), combined with bandpass filtering (such as wavelet denoising or frequency domain filtering), automatic gain control (AGC), and high-precision AD conversion based on the Nyquist theorem, dynamic time warping (DTW) algorithm is used to synchronize and align multi-dimensional signal timestamps, and fault diagnosis models are constructed through time-domain / frequency-domain parameter calculation (such as root mean square, peak detection, or FFT analysis) and classification algorithms such as SVM / random forest. The cross-entropy loss function and fully connected network are combined to optimize the identification accuracy (such as parallel identification and calibration coefficient adjustment strategies of the model self-calibration layer), ultimately achieving high-precision, real-time monitoring and fault early warning of the circuit breaker's mechanical state, significantly reducing the false alarm rate and improving system reliability and adaptability.

[0016] In summary, the beneficial technical effects of this application are as follows: By coordinating multiple source sensors such as vibration, current, and position sensors, and combining them with an intelligent analysis unit to accurately calculate multiple characteristic parameters such as opening time, overtravel time, and closing time, multi-dimensional quantitative monitoring of the mechanical action characteristics of the circuit breaker is achieved, and the accuracy of fault judgment is improved compared with traditional solutions. By generating a fault feature library through a lightweight fault diagnosis model, monitoring and identifying action characteristic parameters, and outputting fault warning information, the system achieves full-process embedded intelligent monitoring locally on the circuit breaker. Data processing latency is controlled at the millisecond level, enabling real-time capture of mechanical faults. This completely solves the pain point of poor real-time performance in traditional solutions and ensures the operational reliability of DC power systems. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the intelligent monitoring system for mechanical motion characteristics in this application; Figure 2 This is a schematic diagram of the intelligent analysis module in this application; Figure 3 This is a schematic diagram of the intelligent diagnostic unit in this application; Figure 4 This is a schematic diagram of the intelligent monitoring method for mechanical motion characteristics in this application. Detailed Implementation

[0018] The present application will be further described in detail below with reference to the accompanying drawings.

[0019] Reference Figure 1 This application discloses an intelligent monitoring system for the mechanical operating characteristics of a DC miniature circuit breaker, comprising: The signal acquisition module is connected to the mechanical action generating mechanism of the circuit breaker at its input end. It is used to sense the mechanical action of the circuit breaker, collect mechanical stress signals of different dimensions, and record the corresponding signal timestamps. The signal processing module, whose input end is connected to the output end of the signal acquisition module, is used to clean the mechanical stress signal according to the signal frequency characteristics, and to amplify and convert the signal to a preset signal amplitude to obtain the corresponding digital gain signal. The intelligent analysis module, whose input end is connected to the output end of the signal processing module, is used to synchronize signal timestamps, calculate action characteristic parameters based on digital gain signals, monitor and identify the mechanical status of the circuit breaker in conjunction with the fault feature database, and output fault warning information.

[0020] The implementation principle of this embodiment is as follows: The signal acquisition module senses the mechanical action of the circuit breaker in real time, collects multi-dimensional mechanical stress signals such as vibration, current surges, and position triggering, and records timestamps (e.g., based on sensor signal sensing and timestamp recording technology); The signal processing module cleans and optimizes the signal amplitude and generates a digital gain signal by using bandpass filtering (e.g., wavelet denoising or frequency domain filtering algorithms), automatic gain control (AGC), and high-precision AD conversion chips (e.g., sampling conversion based on the Nyquist theorem); The intelligent analysis module uses the dynamic time warping (DTW) algorithm to synchronize the timestamps, combines time-frequency analysis (e.g., FFT or wavelet transform) to extract action characteristic parameters, and matches them with the fault feature database through classification algorithms such as SVM / random forest (e.g., classification strategy and cross-entropy loss function optimization), ultimately achieving high-precision monitoring and fault early warning of the circuit breaker's mechanical state.

[0021] Preferably, the signal acquisition module includes: The vibration sensing unit is used to collect the mechanical stress signal of the circuit breaker's operating mechanism during the opening and closing process, obtain the mechanical vibration signal, and record the vibration timestamp. The current sensing unit is used to monitor the mechanical stress signal when the contact state of the circuit breaker changes, sense the current change signal, and record the change timestamp. The position sensing unit is used to detect the mechanical stress signal generated by the status of the position indication component of the circuit breaker, capture the position trigger signal, and record the trigger timestamp.

[0022] In this embodiment, the vibration sensing unit is used to collect mechanical vibration signals of the circuit breaker operating mechanism during the opening and closing process. These signals include core information such as the force, timing, and component coordination of the operating mechanism's actions. This unit is installed on key vibration points of the operating mechanism, such as the moving contact linkage and spring assembly, using an adaptive installation method (e.g., adhesive or snap-fit ​​structure rigidly connected to the operating mechanism's housing). This ensures complete synchronization with the vibration characteristics of the operating mechanism, accurately capturing the vibration amplitude, frequency, and waveform changes at the moment of action.

[0023] Current sensing unit: Used to monitor the changes in current in the contacts of a DC miniature circuit breaker, serving as a reference point for mechanical action timing. This unit employs a non-contact design (e.g., clamped to the contact leads or sleeved onto conductive components), ensuring electrical insulation while achieving current monitoring and preventing interference with the normal switching operation of the circuit breaker. It accurately marks the time points of current zero-crossing or sudden changes by sensing abrupt current changes (such as a sudden drop in load current to zero during opening or a current increase from zero during closing).

[0024] Position sensing unit: Used to capture the position signals of the circuit breaker's opening and closing actions, thus determining the endpoint of the mechanical movement. This unit is located beside the movement trajectory of the opening and closing position indicator component (such as the side of the opening position indicator rod or the periphery of the closing latch). It can employ photoelectric or magnetic principles, and outputs high-level / low-level position trigger signals by detecting the "in position / out position" status of the position indicator component.

[0025] Preferably, the signal processing module includes: The signal filtering unit is used to filter and denoise mechanical vibration signals, current surge signals and position trigger signals according to the signal frequency characteristics to obtain the corresponding key feature signals. The signal amplification unit is used to amplify the key feature signal according to the preset signal amplitude to obtain the corresponding gain analog signal; The analog-to-digital conversion unit is used to sample and convert the gain analog signal according to the Nyquist theorem to obtain the corresponding digital gain signal.

[0026] Reference Figure 2 Preferably, the intelligent analysis module includes: The time synchronization unit is used to synchronize the vibration timestamp and the trigger timestamp according to the abrupt change timestamp, and align the corresponding digital gain signals to obtain the corresponding synchronization gain signal. The parameter calculation unit is used to calculate the corresponding operating characteristic parameters based on the timestamp difference between different mechanical states of the circuit breaker during the opening and closing process, combined with the synchronous gain signal. The intelligent diagnostic unit is used to build a fault diagnosis model based on historical characteristic parameters, generate a fault feature library, monitor and identify action characteristic parameters, and output fault warning information. The functional calibration unit is used to periodically calibrate the signal acquisition module and the signal processing module according to the built-in standard signal source or external calibration instructions; The data management unit is used for local storage and retrieval of action characteristic parameters and fault warning information.

[0027] The implementation principle of this embodiment is as follows: Mechanical stress signals during the opening and closing process of the circuit breaker are collected from multiple dimensions by vibration, current surge, and position trigger sensing units, and corresponding timestamps are recorded (e.g., vibration signal analysis, surge detection, and position trigger detection algorithms). This is combined with bandpass filtering (e.g., wavelet denoising or frequency domain filtering algorithms), automatic gain control (AGC), and high-precision AD conversion based on the Nyquist theorem in the signal processing module. Dynamic time warping (DTW) algorithms are used to synchronize and align the timestamps of multiple signal sources. Action characteristic parameters are extracted through time-domain / frequency-domain parameter calculations (e.g., root mean square, peak detection, or FFT analysis). A fault diagnosis model and fault feature library are constructed based on classification algorithms such as SVM / random forest and historical characteristic parameters. Periodic calibration algorithms (e.g., self-calibration based on a built-in standard signal source) ensure the stability of signal acquisition and processing. Finally, data management is achieved through local storage and backtracking technology. This improves the accuracy and real-time performance of circuit breaker mechanical condition monitoring under complex operating conditions, reduces false alarm rates, and enhances system reliability.

[0028] Reference Figure 3 Preferably, the intelligent diagnostic unit includes: The data partitioning layer is used to divide historical characteristic parameters according to a preset data splitting ratio to obtain parameter training set and parameter test set; The data aggregation layer is used to classify the parameter training set according to the fault type, resulting in normal parameter blocks and several fault parameter blocks. The fault sequencing layer is used to assign priorities to fault parameter blocks according to the severity of the fault, thereby obtaining a parameter priority sequence for different faults. The cross-validation layer is used to perform cross-analysis on fault parameters of different degrees according to the parameter priority sequence, and to combine it with normal parameter blocks for verification and screening to determine the initial fault parameter range. The branch creation layer is used to construct branches for each layer of the neural network based on the number of fault types, and to determine the number of convolutional blocks and the hierarchical branch skeleton. The feature extraction layer is used to extract features from the normal parameter block based on the first extraction branch using N convolutional blocks and M pooling blocks to obtain the normal feature matrix; Based on the second extraction branch and the fault severity level k, feature extraction is performed on the fault parameter block using k*N convolutional blocks and 2M pooling blocks to obtain the fault feature matrix. The iterative training layer is used to perform several iterations on the normal feature matrix based on the first training branch using s residual blocks connected in parallel, 2s downsampling blocks and the tanh activation function, to generate the normal weight matrix. Based on the second training branch, the fault feature matrix is ​​iterated hierarchically using k*s InceptionV3 blocks, (k+1)*s upsampling blocks and ReLU activation function connected in series to generate a fault hierarchy weight matrix. Based on the third training branch, a flat expansion block is used to unify the dimensions of the normal weight matrix and the fault level weight matrix, and several iterations are performed in combination with the spatiotemporal attention mechanism to generate a hierarchical comprehensive weight matrix. The output layer is used to generate a single-class fault identifier based on the hierarchical branch skeleton, using 3 fully connected blocks, 2 random deactivated blocks and cross-entropy loss function, and to identify the fault feature type by combining the hierarchical comprehensive weight matrix on the parameter test set. The verification and localization layer is used to calculate the false positive rate and accuracy rate based on the fault feature type and the preset fault label, and generate an initial confusion matrix based on the preset accuracy threshold; at the same time, it locates and analyzes the corresponding action characteristic parameters according to the false positive rate, and generates a judgment optimization matrix based on the initial fault parameter range. The update optimization layer is used to transpose and correct the fault level weight matrix based on the confusion initial matrix, and to perform correction operations on the level comprehensive weight matrix with the decision optimization matrix to obtain the level corrected weight matrix. The model self-calibration layer is used to correct the weight matrix according to the hierarchy, perform parallel identification on the validation set, output the mechanical state of the circuit breaker, calculate the identification accuracy, and compare it with the accuracy threshold. If the recognition accuracy is greater than the accuracy threshold, the current level correction weight matrix is ​​determined to be a fault diagnosis model. If not, then based on the initial confusion matrix and the decision optimization matrix, calibration coefficients are generated, and the cross-entropy loss function is optimized and calibrated until the accurate threshold is reached; The rule generation layer is used to parse and aggregate the fault diagnosis model according to the rule format and the fault type to build a fault feature library; The rule application layer is used to match action characteristic parameters based on the fault feature library; If a match is successful, output a fault warning message containing the fault type and fault severity level; If the matching fails, the current action characteristic parameters are identified according to the fault diagnosis model, a new fault judgment rule is output and populated into the fault feature library.

[0029] In this embodiment, N convolutional blocks are used to perform feature mapping on normal parameter blocks using convolutional kernels to output feature maps. Max pooling is then used to reduce the dimensionality of the output to obtain a normal feature matrix. S residual blocks and 2S downsampling blocks are input in parallel. The Tanh activation function is used to normalize the output to [−1,1] to generate a normal weight matrix. The space is gradually compressed and gradient vanishing is prevented by using dimension skip connections. By using KN convolutional blocks, the number of convolutional kernels is increased (K times that of normal branches) to capture finer-grained fault features of fault parameter blocks. 2M pooling blocks are used for higher-frequency dimensionality reduction, outputting a fault feature matrix. KS InceptionV3 blocks and (K+1)S upsampling blocks are input (serially). Positive features are preserved using the ReLU activation function, generating a fault level weight matrix. Features are fused through multi-scale convolutions (e.g., parallel 1×1, 3×3, 5×5 convolutions) and spatial resolution is gradually restored to match dimensionality requirements. The normal weight matrix and the fault level weight matrix are flattened into vectors, and feature maps are generated by 1×1 convolution dimensionality reduction, Tanh activation and attention weights are generated by Sigmoid. The weighted output is a comprehensive weight matrix of the level. The test set features and hierarchical integrated weight matrix are fitted by matrix multiplication using three fully connected blocks. Combined with two random deactivation blocks and the cross-entropy loss function, the error between the predicted probability distribution and the true label is calculated, and the probability vector of the fault type is output. Based on the statistical model prediction results, a C×C confusion matrix (rows: true class, columns: predicted class) is constructed. The accuracy and misclassification rate are calculated, and the decision threshold is adjusted based on the confusion matrix to generate an optimization matrix to reduce high-cost errors (such as misclassifying severe faults as normal). The hierarchical comprehensive weight matrix is ​​transposed and corrected to adjust the feature weight distribution. The gradient is calculated based on the cross-entropy loss, the optimizer is updated, and the hierarchical corrected weight matrix is ​​output.

[0030] The implementation principle of this embodiment is as follows: the data partitioning layer splits historical characteristic parameters into training set and test set according to a preset ratio; the data aggregation layer classifies normal parameter block and fault parameter block according to fault type; the fault sorting layer assigns priority sequence according to severity level; the cross-validation layer determines the initial fault parameter range through cross-analysis and verification with normal parameter block; and the branch creation layer constructs neural network branch skeleton according to the number of fault types. The feature extraction layer uses the first extraction branch (N convolutional blocks and M pooling blocks) to extract the normal feature matrix, and the second extraction branch (KN convolutional blocks and 2M pooling blocks) to extract the fault feature matrix. The iterative training layer generates the normal weight matrix iteratively through the first training branch (parallel S residual blocks, 2S downsampling blocks and Tanh activation function), and generates the fault-level weight matrix hierarchically through the second training branch (serial KS InceptionV3 blocks, (K+1)*S upsampling blocks and ReLU activation function). The third training branch generates the hierarchical comprehensive weight matrix after unifying the dimensions by combining flattened blocks and spatiotemporal attention mechanism. The recognition output layer generates a single-class fault recognizer based on the hierarchical branch skeleton (3 fully connected blocks, 2 random deactivation blocks and cross-entropy loss function), and recognizes the feature type of the test set by combining the hierarchical comprehensive weight matrix. The verification and localization layer calculates the false positive rate and accuracy through fault labels, generates the initial confusion matrix and the decision optimization matrix, and the update optimization layer performs transpose correction and operation on the weight matrix to obtain the hierarchical corrected weight matrix. The model self-calibration layer calculates accuracy by identifying the validation set in parallel. If the accuracy is not reached, the calibration coefficient is used to optimize the cross-entropy loss function until convergence. Finally, the rule generation layer analyzes the model to build a fault feature library. When the rule application layer matches successfully, it outputs fault warning information. If the match fails, it generates new rules to expand the feature library, thus achieving high-precision adaptive optimization for circuit breaker fault diagnosis.

[0031] Reference Figure 4 A method for intelligent monitoring of the mechanical operating characteristics of DC miniature circuit breakers, applied to an intelligent monitoring system for the mechanical operating characteristics of DC miniature circuit breakers, comprising: S1: Sensing the mechanical action of the circuit breaker, collecting mechanical vibration signals, current surge signals and position trigger signals, and recording the surge timestamp, vibration timestamp and trigger timestamp; S2: Clean the mechanical vibration signal, current surge signal and position trigger signal according to the signal frequency characteristics, and amplify and convert the gain to the preset signal amplitude to obtain the corresponding digital gain signal; S3: Synchronize the vibration timestamp and trigger timestamp according to the mutation timestamp, and align the corresponding digital gain signals to obtain the corresponding synchronization gain signal; S4: Calculate the corresponding operating characteristic parameters based on the timestamp difference between different mechanical states of the circuit breaker during the opening and closing process, combined with the synchronization gain signal. S5: Based on historical characteristic parameters, construct a fault diagnosis model, generate a fault feature library, monitor and identify action characteristic parameters, and output fault warning information.

[0032] The implementation principle of this embodiment is as follows: by synchronously collecting the vibration, current change and position trigger signals of the circuit breaker and recording the timestamp, after signal cleaning and gain conversion, the time series is aligned to generate a synchronous gain signal, and then the action characteristic parameters are calculated according to the state time difference of the opening and closing process. Finally, a fault model is constructed by combining historical parameters for real-time monitoring and early warning.

[0033] The embodiments described in this specific implementation are preferred embodiments of this application and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. An intelligent monitoring system for the mechanical operating characteristics of a DC miniature circuit breaker, comprising: The signal acquisition module is connected to the mechanical action generating mechanism of the circuit breaker at its input end. It is used to sense the mechanical action of the circuit breaker, collect mechanical stress signals of different dimensions, and record the corresponding signal timestamps. The signal processing module, with its input end connected to the output end of the signal acquisition module, is used to clean the mechanical stress signal according to the signal frequency characteristics, and to amplify and convert the signal amplitude to a preset value to obtain the corresponding digital gain signal. The intelligent analysis module, with its input end connected to the output end of the signal processing module, is used to synchronize the signal timestamp, calculate the action characteristic parameters based on the digital gain signal, monitor and identify the mechanical state of the circuit breaker in conjunction with the fault feature database, and output fault warning information.

2. The intelligent monitoring system for the mechanical action characteristics of DC miniature circuit breakers according to claim 1, characterized in that, The signal acquisition module includes: The vibration sensing unit is used to collect the mechanical stress signal of the circuit breaker's operating mechanism during the opening and closing process, obtain the mechanical vibration signal, and record the vibration timestamp. The current sensing unit is used to monitor the mechanical stress signal when the contact state of the circuit breaker changes, sense the current change signal, and record the change timestamp. The position sensing unit is used to detect the mechanical stress signal generated by the status of the position indication component of the circuit breaker, capture the position trigger signal, and record the trigger timestamp.

3. The intelligent monitoring system for the mechanical action characteristics of DC miniature circuit breakers according to claim 1, characterized in that, The signal processing module includes: The signal filtering unit is used to filter and denoise mechanical vibration signals, current surge signals and position trigger signals according to the signal frequency characteristics to obtain the corresponding key feature signals. The signal amplification unit is used to amplify the key feature signal according to a preset signal amplitude to obtain a corresponding gain analog signal. The analog-to-digital conversion unit is used to sample and convert the gain analog signal according to the Nyquist theorem to obtain the corresponding digital gain signal.

4. The intelligent monitoring system for the mechanical action characteristics of DC miniature circuit breakers according to claim 1, characterized in that: The intelligent analysis module includes: The time synchronization unit is used to synchronize the vibration timestamp and the trigger timestamp according to the abrupt change timestamp, and align the corresponding digital gain signals to obtain the corresponding synchronization gain signal. The parameter calculation unit is used to calculate the corresponding action characteristic parameters based on the timestamp difference between different mechanical states of the circuit breaker during the opening and closing process, combined with the synchronous gain signal. The intelligent diagnostic unit is used to construct a fault diagnosis model based on historical characteristic parameters, generate a fault feature library, monitor and identify the action characteristic parameters, and output fault warning information.

5. The intelligent monitoring system for the mechanical action characteristics of DC miniature circuit breakers according to claim 4, characterized in that: The intelligent diagnostic unit includes: The data partitioning layer is used to divide historical characteristic parameters according to a preset data splitting ratio to obtain parameter training set and parameter test set; The data aggregation layer is used to classify the parameter training set according to the fault type, and obtain normal parameter blocks and several fault parameter blocks; The fault sorting layer is used to assign priorities to the fault parameter blocks according to the severity level of the fault, so as to obtain a parameter priority sequence for different faults. The cross-validation layer is used to perform cross-analysis on fault parameters of different degrees according to the parameter priority sequence, and to perform verification and screening in conjunction with the normal parameter block to determine the initial fault parameter range. The branch creation layer is used to construct branches for each layer of the neural network based on the number of fault types, and to determine the number of convolutional blocks and the layer branch skeleton.

6. The intelligent monitoring system for the mechanical action characteristics of DC miniature circuit breakers according to claim 5, characterized in that: The intelligent diagnostic unit also includes: The feature extraction layer is used to extract features from the normal parameter block based on the first extraction branch using N convolutional blocks and M pooling blocks to obtain the normal feature matrix; Based on the second extraction branch and the fault severity level k, feature extraction is performed on the fault parameter block using k*N convolutional blocks and 2M pooling blocks to obtain the fault feature matrix. An iterative training layer is used to perform several iterations on the normal feature matrix based on the first training branch using s residual blocks connected in parallel, 2s downsampling blocks and the tanh activation function, to generate a normal weight matrix. Based on the second training branch, the fault feature matrix is ​​iterated hierarchically using k*s InceptionV3 blocks, (k+1)*s upsampling blocks and ReLU activation function connected in a serial manner to generate a fault hierarchy weight matrix. The third training branch uses a flattened expansion block to unify the dimensions of the normal weight matrix and the fault level weight matrix, and combines a spatiotemporal attention mechanism to perform several iterations to generate a hierarchical comprehensive weight matrix. The identification output layer is used to generate a single-class fault identifier based on the hierarchical branch skeleton, using 3 fully connected blocks, 2 random deactivated blocks, and the cross-entropy loss function. It also combines the hierarchical comprehensive weight matrix to identify the parameter test set and determine the fault feature type.

7. The intelligent monitoring system for the mechanical action characteristics of DC miniature circuit breakers according to claim 1, characterized in that: The intelligent diagnostic unit also includes: The verification and localization layer is used to calculate the false positive rate and accuracy rate based on the fault feature type and the preset fault label, and generate an initial confusion matrix based on the preset accuracy threshold; at the same time, it locates and analyzes the corresponding action characteristic parameters according to the false positive rate, and generates a judgment optimization matrix based on the initial fault parameter range. An update optimization layer is used to transpose and correct the fault level weight matrix according to the confusion initial matrix, and to perform a correction operation on the level comprehensive weight matrix with the judgment optimization matrix to obtain the level corrected weight matrix. The model self-calibration layer is used to correct the weight matrix according to the layer, perform parallel identification on the validation set, output the mechanical state of the circuit breaker, calculate the identification accuracy, and compare it with the accuracy threshold. If the recognition accuracy is greater than the accuracy threshold, then the current level correction weight matrix is ​​determined to be a fault diagnosis model; If not, then based on the initial confusion matrix and the decision optimization matrix, calibration coefficients are generated, and the cross-entropy loss function is optimized and calibrated until the accurate threshold is reached; The rule generation layer is used to parse and aggregate the fault diagnosis model according to the rule format and the fault type to build a fault feature library; The rule application layer is used to match action characteristic parameters based on the fault feature library; If a match is successful, output a fault warning message containing the fault type and fault severity level; If the matching fails, the current action characteristic parameters are identified according to the fault diagnosis model, a new fault judgment rule is output, and it is added to the fault feature library.

8. The intelligent monitoring system for the mechanical action characteristics of DC miniature circuit breakers according to claim 1, characterized in that: The intelligent analysis module also includes: The functional calibration unit is used to periodically calibrate the signal acquisition module and the signal processing module according to the built-in standard signal source or external calibration instructions; The data management unit is used for local storage and retrieval of action characteristic parameters and fault warning information.

9. An intelligent monitoring method for the mechanical operating characteristics of a DC miniature circuit breaker, applied to the system described in any one of claims 1-8, characterized in that, include: The system senses the mechanical movement of the circuit breaker, collects mechanical vibration signals, current surge signals, and position trigger signals, and records the corresponding signal timestamps. The mechanical vibration signal, the current surge signal, and the position trigger signal are cleaned according to the signal frequency characteristics, and then amplified and converted to a preset signal amplitude to obtain the corresponding digital gain signal. The vibration timestamp and trigger timestamp are synchronized based on the mutation timestamp, and the corresponding digital gain signals are aligned to obtain the corresponding synchronization gain signal. Based on the timestamp difference between different mechanical states of the circuit breaker during the opening and closing process, the corresponding operating characteristic parameters are calculated in conjunction with the synchronous gain signal. Based on historical characteristic parameters, a fault diagnosis model is constructed, a fault feature library is generated, and the action characteristic parameters are monitored and identified to output fault warning information.

Images