Fault diagnosis method and diagnosis system for electrically operated valve actuating mechanism

Abstract

The invention discloses a fault diagnosis method and a fault diagnosis system for an electric valve actuating mechanism. The method comprises the following steps: synchronously acquiring signals through an anti-EMI (Electro-Magnetic Interference) multi-source sensor; adopting complex Morlet wavelet packet decomposition to extract a 1.2-2.4 kHz energy entropy minimum frequency band, and calculating a kurtosis index; separating the third harmonic of the current through variational mode decomposition, and calculating the total distortion rate of the third harmonic; a graph attention network with 12-dimensional features is constructed, and weighted fusion is carried out through a multi-head attention mechanism; the lightweight CNN outputs a fault type, and when the confidence coefficient is less than 0.9, a knowledge graph rule engine is triggered; and updating a threshold value based on a historical diagnosis clustering result, and aggregating edge model parameters by federal learning. The system comprises a wafer-level micro-strain sensing layer, an FPGA accelerated edge computing layer, a cloud platform supporting federated learning, and an AR maintenance guidance and block chain evidence storage module. The early fault detection rate is improved, the false alarm rate under strong EMI is reduced, and the average repair time is shortened.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial equipment fault diagnosis, and in particular to a fault diagnosis method and diagnosis system for electric valve actuator. BACKGROUND

[0002] The electric valve actuator is the core control unit of the industrial pipeline system, and its failure can cause production interruption and even safety accidents. The existing diagnosis technology has the following defects: 1. High early fault omission rate: traditional vibration analysis is not sensitive to early mechanical faults such as micro-cracks and worm gear pitting, and the acoustic emission technology is greatly disturbed by environmental noise. The single sensor scheme cannot capture mechanical vibration and electrical fault characteristics at the same time, resulting in a high false judgment rate of motor turn-to-turn short circuit.

[0003] 2. Serious false alarm of fixed threshold: when the equipment is aging or the load is suddenly changed, the fixed alarm threshold (such as kurtosis K>4.0) has a high false alarm rate under strong interference conditions; manual adjustment of the threshold requires shutdown, and the annual average loss of working hours is nearly 120 hours.

[0004] 3. Difficulty in cross-device knowledge transfer: the edge computing model is closed and independent, and it takes more than 30 days of training period for new devices to reach 90% accuracy (such as the SFC system of a certain power plant), and data privacy risks hinder multi-source information sharing.

[0005] 4. Low maintenance positioning efficiency: nuclear power maintenance records show that nearly 40% of unplanned shutdowns are caused by fault positioning errors, and the average repair time (MTTR) is as long as 4.5 hours. SUMMARY

[0006] This section aims to summarize some aspects of the embodiments of the present application and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract and title of the specification to avoid obscuring the purpose of this section, abstract and title, and such simplifications or omissions cannot be used to limit the scope of the present application.

[0007] Therefore, in order to solve the above technical problems, the present application provides the following technical solutions: a fault diagnosis method for electric valve actuator, comprising the following steps: S100: synchronously collecting mechanical vibration signals, motor current signals and acoustic emission signals through a multi-modal sensor array; S200: wavelet packet decomposition is performed on the mechanical vibration signals, the target frequency band with the minimum energy entropy is extracted, and the kurtosis index is calculated; S210: modal decomposition is performed on the motor current signals, the specific harmonic component is separated, and the harmonic distortion rate is calculated; S300: Construct a cross-modal correlation graph model, map the vibration frequency domain features, current harmonic features and acoustic emission features to graph nodes, and fuse the node relationships through attention mechanism weighting; S400: Classify the fusion features based on a lightweight convolutional neural network, and when the output confidence is lower than a preset threshold, trigger a knowledge graph rule engine for auxiliary diagnosis; S500: Cluster the historical diagnosis results, and dynamically update the fault judgment threshold according to the inter-class distance; S600: Update the global diagnosis model by aggregating the multi-edge device model parameters through federated learning.

[0008] As a preferred scheme of the fault diagnosis method for the electric valve actuator, the target frequency band with the minimum energy entropy in S200 includes: Three-layer wavelet packet decomposition is performed using a complex Morlet wavelet basis function; The energy distribution entropy value of each frequency band is calculated, and the 1.2-2.4 kHz frequency band with the lowest entropy value is selected.

[0009] As a preferred scheme of the fault diagnosis method for the electric valve actuator, the specific harmonic component in S210 is the 3rd harmonic, and the harmonic distortion rate is calculated by the ratio of the 3rd harmonic amplitude to the fundamental harmonic amplitude.

[0010] As a preferred scheme of the fault diagnosis method for the electric valve actuator, the construction of the cross-modal correlation graph model in S300 includes: Map 12-dimensional features such as vibration kurtosis, current harmonic distortion rate and acoustic emission count rate to graph nodes; Construct adjacent edges based on physical connection relationship and time sequence correlation; Learn the coupling weight between nodes through multi-head attention mechanism.

[0011] As a preferred scheme of the fault diagnosis method for the electric valve actuator, the triggering condition of the knowledge graph rule engine in S400 is that the confidence is lower than 0.9; The rule engine execution includes: Match the causal chain between fault entities; When the torque fluctuation coefficient is greater than 0.35 and the acoustic emission kurtosis is greater than 4.2, output the "gear tooth breakage" fault conclusion.

[0012] As a preferred scheme of the fault diagnosis method for the electric valve actuator, the dynamic threshold updating in S500 includes: K-means clustering is performed on the diagnosis results for 24 consecutive hours; Calculate the minimum inter-class distance of the health state cluster and the fault state cluster; Generate a new threshold according to the product of the minimum distance and an adjustment factor.

[0013] As a preferred scheme of the fault diagnosis method for the electric valve actuator, the federal learning aggregation period in S600 is 24 hours, and the model parameters are weighted and averaged according to the data amount of the edge device.

[0014] A fault diagnosis system for an electric valve actuator is applied to the fault diagnosis method for the electric valve actuator. The sensing layer includes micro strain gauges integrated on the surface of the worm shaft, MEMS triaxial accelerometers at the valve rod, Hall current sensors at the motor winding, and piezoelectric acoustic emission sensors on the worm box shell. The edge computing layer includes: The FPGA module accelerates wavelet packet decomposition and modal decomposition; The ARM processor deploys a lightweight CNN model and a dynamic threshold clustering algorithm; The cloud collaborative layer includes: The knowledge graph engine stores fault entity causal rules; The federal learning aggregator updates the global model according to the preset period; The maintenance interaction layer includes: The AR guidance module generates a three-dimensional maintenance animation when the CNN confidence is less than 0.9; The blockchain storage module records the diagnosis results and sensor data hash values.

[0015] An electronic device includes a memory, a processor, and a computer program stored on the memory, wherein the processor executes the computer program to implement the steps of the method.

[0016] A computer-readable storage medium stores a computer program, wherein the computer program is executed by a processor to implement the steps of the method.

[0017] The present application has the following advantages: Early fault detection sensitivity is improved: The application adopts acoustic emission sensors and vibration signal joint analysis to greatly improve the detection rate of micro cracks (≥0.05 mm); current harmonic monitoring can provide 24-hour early warning for motor short circuit, greatly reducing the false alarm rate. The adaptive threshold based on K-means clustering reduces the false alarm rate under sudden load conditions; federal learning technology improves the diagnosis accuracy of new equipment within 7 days and shortens the debugging cycle. AR maintenance guidance locates the fault point through three-dimensional animation, and MTTR is shortened from 4.5 hours to 1.2 hours; the blockchain storage module meets the requirements of nuclear power safety audit, and the risk of log tampering is reduced. The recognition rate of gear gap fault is improved, the spare parts inventory cost is reduced, the number of unplanned shutdowns is reduced, and the annual average increase of a single device is increased. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creating labor intensity on the premise of the drawings. Among them: Figure 1 The workflow diagram of the present application.

[0019] Figure 2 The internal structure diagram of the computer device of the present application. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0021] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited by the specific embodiments disclosed below.

[0022] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an independent or selective embodiment that excludes other embodiments.

[0023] Reference Figure 1 For the first embodiment of the present application, a fault diagnosis method for electric valve actuator is provided, which integrates multi-source perception, graph attention feature modeling, lightweight neural network judgment and dynamic threshold adaptive strategy to improve environmental adaptability and fault recognition accuracy in complex industrial sites. The specific steps are as follows: Step S100: Multi-source sensor signal acquisition; According to the structure and installation environment of the actuator, a multi-modal sensor array is arranged at the key part of the valve rod driving mechanism for synchronous acquisition of mechanical vibration signals, electrical signals and acoustic signals; Further, the sensor array comprises: MEMS three-axis accelerometer, bandwidth 0-10kHz, sensitivity≥100mV / g; Piezoelectric acoustic emission sensor, resonant frequency about 150kHz; Closed-loop Hall current sensor, measurement accuracy 0.5%FS; Distributed clock synchronization unit, through PTPv2 protocol to ensure that the sampling clock deviation of each channel is less than 1μs; It should be noted that the deployment position and number of the multi-source sensor can be optimized and adjusted according to the structure size of the actuator and the on-site environment, so as to ensure the signal coverage rate and data synchronization.

[0024] Step S200: Mechanical vibration feature extraction; The vibration original signal collected in step S100 is subjected to three-layer wavelet packet decomposition, and the energy entropy and kurtosis index of each frequency band are calculated to represent the mechanical state; 1. Wavelet packet decomposition; Select complex Morlet wavelet base function , three-layer decomposition is carried out according to the following formula, and eight frequency band nodes are obtained; ; Among them, : Morlet wavelet base function, used for time-frequency localization analysis of signals; (bandwidth parameter): control the decay rate of wavelet in time domain, in this embodiment, take 1.5, so that the wavelet has moderate time resolution; : (center frequency): determines the main oscillation frequency of the wavelet in the frequency domain, in this embodiment, take 1.0, to cover the main vibration component of about 1kHz.

[0025] In this formula, the three-layer wavelet packet decomposition method is used for the collected vibration signal, and the complex Morlet wavelet is selected as the mother wavelet (the vibration working condition of the electric valve actuator has obvious non-stationary characteristics (switching impact, load mutation), the complex Morlet wavelet can be transformed by translation-stretching, and it can efficiently decompose the energy change in different time scales, which is helpful to extract the impact spectrum and harmonic information; 2. For the i-th wavelet packet node (frequency band node) signal , calculate its energy entropy , use entropy to measure the uniformity of frequency band signal energy distribution, select the main frequency band that causes the impact, and the calculation formula is as follows: ; in, , represents the normalized energy ratio of the jth sampling point in the frequency band, and n represents the total number of wavelet coefficients after frequency band decomposition; represents the jth wavelet coefficient at the i-th frequency band node; In this formula, when a fault shock or resonance occurs, the signal energy will be concentrated in a specific frequency band, and the energy entropy to reduce; to The smallest frequency band is used for subsequent analysis, which is beneficial to improving the accuracy of fault feature extraction; 3. Select the three nodes with the smallest energy entropy (typical frequency range is about 1.2-2.4kHz) for bandpass filtering, and calculate the kurtosis value of the filtered signal (used to measure the peak mutation degree of the impact fault signal relative to the steady vibration signal). The calculation formula is as follows: ; , ; in, Indicates averaging the vibration signal sample sequence after bandpass filtering in the brackets; Represents the time domain vibration signal within the selected frequency band after bandpass filtering; 、 They are signal mean (reflecting the average level of signal data) and standard deviation (used to measure the degree of dispersion of signal data); when When , it indicates that the signal has obvious impact components, and it is judged that there is an impact type fault (mechanical collision or wear); In the existing research on bearing and gearbox faults, when mechanical impact faults occur, the signal kurtosis value exceeds the range of 4.0~4.5. This solution takes the median value of 4.2 to enhance robustness, and the kurtosis threshold Based on a company's bearing failure database from 2020 to 2023 (sample size: 1,203 impact failures); It should be noted that the above-mentioned frequency bands and thresholds can be adjusted according to different models and operating conditions of the actuators to take into account both fault sensitivity and noise suppression.

[0026] S210: Harmonic feature extraction of current signal; 1. Perform modal decomposition on the current signal collected in step S100 and extract the specified harmonic distortion rate: Specifically, the motor current signal With the variational mode decomposition (VMD) model (from Dragomiretskiy 2014), VMD splits the current signal into several modes, separates the harmonic components, extracts specific harmonic components in the current, suppresses mode aliasing, and improves the accuracy of harmonic feature extraction. The optimization objective function (from the above literature) is expressed as follows: ; The constraint condition is: wherein, represents the original current signal; wherein, represents the kth modal component signal obtained after signal decomposition, reflecting its change over time t; represents the center frequency corresponding to the kth modal component; is a generalized function, which takes infinite value at , and 0 at other positions, and the integral over the entire domain is 1; represents the time derivative; 2、Separate the 3rd harmonic component from it, and calculate the proportion of the 3rd component in the total harmonic distortion rate , use the harmonic distortion measure to judge the health status of the motor winding, and the rise of high-order harmonics indicates local short circuit or asymmetry, and the calculation formula is as follows: ; wherein, is the 3rd harmonic amplitude, is the fundamental amplitude; The relative contribution of the third harmonic to the total harmonic energy is measured by , and the third harmonic is often related to motor winding imbalance or mechanical jamming, which is an important feature of electrical state; In this embodiment, when , it indicates that there may be a stator winding asymmetry or short circuit fault; The threshold value 7.8% is derived from laboratory bench tests: 20 same type actuators are injected with stator inter-turn short circuit faults, and statistics show that It rises to 8.2% ± 0.4% on average 24 hours before the fault occurs, so take the value of one standard deviation from the mean as the early warning threshold (i.e. 7.8%); S300: Cross-modal feature correlation; Based on the frequency domain features, harmonic features and acoustic emission count rates extracted in steps S200 and S210, a graph structure is constructed and coupled modeling is performed using a graph neural network (GNN) or attention mechanism: 1、Graph construction; Map the 12-dimensional modal features into graph nodes, with physical connections and timing dependencies as adjacent edges; Modeling multimodal signal data nodes as a graph structure ; in is the signal mode node set, is the inter-modal correlation edge; The Graph Attention Network (GAT) mechanism is used in this graph structure to achieve weight distribution between modalities through attention coefficients, strengthen the influence of key modalities, suppress noise, and model the coupling relationship between modalities.

[0027] 2. GNN update; Node feature updates are performed according to the following formula: Attention weight The calculation formula is as follows: ; in, : Learnable attention vector; : Leaky ReLU to avoid the "dead node" phenomenon; : Normalization function, ensuring that the sum of attention of the same node i to all neighbors is 1; : weight matrix; : represents the i-th node feature; This formula can adaptively distribute the importance of neighbor features and improve the GNN's ability to express different modal coupling relationships; The weighted updated node (new representation of the node generated by integrating neighborhood information for subsequent classification or decision-making) is expressed as: ; in, represents the set of adjacent nodes of node i, represents the activation function, using the exponential linear unit (ELU) as the activation function to introduce nonlinear factors; The network configuration parameters are as follows: The number of attention heads is 8; The output dimension of each layer is 128; The learning rate is 0.001; The dropout coefficient is set to 0.3; When a worm gear meshing fault occurs in the device, the attention weight between the "vibration node" and the "current node" in the figure can reach 0.82, indicating that the two types of signals are highly coupled; Step S400: Mixed intelligent fault decision-making; 1. This step uses a lightweight convolutional neural network (CNN) for fault classification. The network structure includes: Three layers of convolutional and pooling layers to extract modal fusion features; One fully connected output layer with a class dimension of 47; Using ReLU activation function and Softmax output layer; The model is pruned and quantized, with an inference delay of less than 30 milliseconds, and can be deployed in edge controllers; 2. When the confidence is lower than the preset threshold (e.g. 0.9), the system automatically calls the expert rule system based on the knowledge graph for auxiliary diagnosis; It is worth noting that the overall misdiagnosis and missed diagnosis rate of the fusion model on 47 types of faults is evaluated by ROC curve, and the inflection point that balances the sensitivity and specificity is about 0.9 (i.e. the preset threshold is set to 0.9).

[0028] For example: If the torque fluctuation coefficient is greater than 0.35 and the kurtosis of the acoustic emission signal is greater than 4.2, the fault type is "gear tooth breakage" with a confidence of 0.93; In the fault laboratory bench test, 14 gear tooth breakage samples have torque fluctuation coefficients greater than 0.35; under normal working conditions and slight wear, the coefficient is less than 0.30, so 0.35 is taken as the boundary; Combined with curve fitting and expert experience, the rule coverage rate is 95% under tooth breakage, the accuracy is 91%, and the comprehensive score is about 0.93, which is taken as the internal confidence of the rule; If the high-frequency energy of the vibration is concentrated near 2.4 kHz and there is an acoustic emission signal spectrum anomaly at the same time, the inference result is "worm crack"; 1. When the CNN output confidence in S400 is less than 0.9, activate the AR maintenance guidance module; (a) Call the device three-dimensional digital twin model and superimpose highlighted markers on components with a fault probability greater than 70%; (b) Generate disassembly animation and torque tightening parameter indication (such as gear tooth breakage fault needs to mark the disassembly sequence: valve cover bolt → worm box positioning pin); (c) Project the maintenance guidance interface through AR glasses; S500: Dynamic threshold updating mechanism; 1. Online clustering: K-means clustering is performed on multiple diagnosis results to classify health status categories; The minimum distance D between classes (measuring the similarity between state clusters for adaptive threshold calculation) is expressed as: ; wherein, represents the i-th type of sample center; 2, alarm dynamic threshold (Clustering results into alarm judgment threshold, dynamically adapt to equipment state changes) According to the following formula self-adaptive adjustment: ; wherein, is an adjustment factor, the value range is 0.3 to 0.4; Through historical fault data set optimization determination: after clustering 100 groups of health / fault state, calculate the false positive rate (FPR) and false negative rate (FNR) under different Value, select the minimum FPR+FNR=0.35 (typical value) as the reference, allow ± 0.05 floating to adapt to different equipment models; Each 24 hours to sample re-clustering, and adjust the threshold, effectively inhibit false alarm phenomenon; Step S600: federated learning model aggregation strategy; Adopt federated learning mechanism to aggregate the local model of different edge devices, the aggregation period is 24 hours; It is worth mentioning that the aggregation period is set to 24 hours, based on the characteristics of industrial field device operation cycle: About 50,000 valid data samples are generated per day, meeting the model update requirements; Avoid frequent communication to increase the load of edge devices (when the measured period is less than 12h, the device CPU occupancy rate is greater than 35%, affecting real-time diagnosis).

[0029] The center server updates the global model by weighted average according to the data volume of each edge device, and the update formula is as follows: ; wherein: represents the model parameter (weight) of the k-th device after t rounds of training; is the sample quantity of the device, , represents the sum of all device sample quantities; This strategy not only protects the data privacy of each device, but also maintains the stability of model convergence; Diagnosis results and sensor original data hash value are written into the alliance chain: Use SHA-256 to generate data digest; Through the smart contract, the digest is stored in the Hyperledger Fabric channel; The storage timestamp is bound with the device ID, which can be traced back during auditing.

[0030] The application adopts acoustic emission and vibration combined analysis technology, and can accurately identify early-stage faults of a micro level. The technology can capture extremely subtle abnormal signals in the equipment, has extremely high sensitivity to early-stage fault hidden dangers such as micro cracks, and can discover the faults in time in the initial stage of fault development, thereby gaining valuable time for subsequent maintenance work. Meanwhile, the application also introduces current harmonic analysis technology, and when the total harmonic distortion of the current harmonic exceeds a specific threshold, an early warning signal of motor short circuit can be sent in advance. Compared with the traditional method, the early warning method can discover potential faults in advance for a long time, effectively avoids further deterioration of the faults, and guarantees stable operation of the equipment.

[0031] The application has a dynamic threshold mechanism, can maintain a low false alarm rate in a strong interference environment, and can accurately determine the running state of the equipment even in the presence of strong interference factors such as electromagnetic interference, thereby avoiding invalid maintenance and resource waste caused by false alarms. In addition, the application realizes knowledge transfer across devices by using federated learning technology. By sharing and learning diagnostic experience between different devices, a new device can achieve a high diagnostic accuracy rate in a short time after being put into use, thereby greatly shortening the adaptation period of the new device and improving the overall diagnostic efficiency.

[0032] The application introduces AR maintenance guidance technology to provide intuitive and accurate maintenance guidance for maintenance personnel. The maintenance personnel can obtain the fault information and maintenance steps of the equipment in real time through the AR device, quickly locate the fault point and repair it, thereby significantly shortening the average repair time and improving the maintenance efficiency. Meanwhile, the application uses blockchain storage technology to ensure the security and non-tamperability of the device operation log. The technology meets the needs of industries such as nuclear power that have strict safety audit requirements, effectively reduces the risk of log tampering, and provides a strong guarantee for the safe operation of the equipment.

[0033] In terms of fault identification, the application greatly improves the identification rate of gear gap faults, has a significant improvement compared with the traditional scheme, can more accurately discover equipment faults, and reduces losses caused by missed faults. In terms of spare parts inventory management, the application optimizes the spare parts inventory strategy by accurate fault prediction and diagnosis, reduces the cost of spare parts inventory, and improves the capital utilization rate of enterprises. In terms of equipment running stability, the application effectively reduces the number of unplanned shutdowns, guarantees the continuous and stable operation of the equipment, improves the production efficiency, and brings considerable economic benefits to enterprises.

[0034] Embodiment 2 Reference Figure 2For the second embodiment of the application, which is different from the first embodiment, the embodiment further discloses a fault diagnosis system for an electric valve actuator, applied to the fault diagnosis method for the electric valve actuator, comprising the following hardware architecture: 1. Sensing layer: 1.1 Multi-modal sensor array: Worm shaft surface: wafer-level micro strain gauge (SC801, TSV package), monitoring axial micro deformation; Valve stem part: MEMS triaxial accelerometer (bandwidth 0-10 kHz), collecting mechanical vibration signals; Motor winding: closed-loop Hall current sensor (precision 0.5% FS), extracting current harmonics; Worm box shell: piezoelectric acoustic emission sensor (resonant frequency 150 kHz), capturing crack propagation waves.

[0035] 1.2 Anti-interference design: Differential transmission circuit (CMRR> 120 dB) suppresses common-mode noise; Electromagnetic compatibility certification: passed ISO 11452-8 IV level radiation immunity test (actual measurement position error <0.1% under 200V / m EMI).

[0036] 2. Edge computing layer: 2.1 Hardware platform: Master control: ARM Cortex-M7 (600 MHz) + FPGA coprocessor; FPGA real-time acceleration: wavelet packet decomposition, VMD current decomposition, dynamic threshold calculation; Memory: ≥1MB SRAM (store lightweight CNN model and feature cache).

[0037] 2.2 Performance binding method flow: Multi-source feature extraction time ≤100ms→ realized by FPGA pipeline; CNN fault classification delay ≤30ms→ model pruned and quantized to 480KB; Emergency shutdown response ≤15ms→ interrupt priority preemption dynamic threshold output.

[0038] 3. Cloud diagnosis layer: 3.1 Hybrid decision support: Knowledge graph engine: store 47 types of fault rules (such as "gear tooth breakage: torque fluctuation greater than 0.35 and acoustic emission kurtosis > 4.2"), activate reasoning when CNN confidence <0.9; Federal learning aggregator: according to the formula Update global model every 24 hours.

[0039] 3.2 Life prediction module: Input: health index on the edge layer; Prediction model: LSTM network, output remaining useful life (RUL) and confidence interval; 4, maintenance and evidence layer: 4.1 AR maintenance guidance module: Trigger condition: CNN confidence <0.9 or knowledge graph output high-confidence root cause.

[0040] Function: Superimpose three-dimensional markers on AR glasses (such as the location of the worm crack); Dynamically generate disassembly animation (torque parameters and order are bound to corresponding rules).

[0041] 4.2 Blockchain evidence: Evidence content: Original sensor data hash value (SHA-256); Diagnosis results and confidence; Generated evidence timestamp + device ID; Audit support: in line with nuclear power safety specifications (traceable and tamper-proof).

[0042] 5, communication layer: Edge to cloud: upload feature vectors and diagnosis results based on MQTT protocol asynchronous; Local anti-interference: ZigBee 3.0 TDMA mechanism ensures real-time data transmission in the sensing layer.

[0043] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, which should be covered by the scope of the claims of the present application.

[0044] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can be in the form of a computer program product implemented on one or more computer usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer usable program code. The solutions in the embodiments of the present application can be implemented in various computer languages, such as object-oriented programming languages Java and interpreted scripting language JavaScript, etc.

[0045] The computer program instructions can 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 such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0046] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0047] The computer program instructions can 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 such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0048] While the preferred embodiments of the application have been described, additional variations and modifications can be employed, as will be appreciated by those of ordinary skill in the art, once armed with the foregoing disclosure. Therefore, the following claims are intended to include all such modifications and variations as falling within the scope of the present application.

[0049] Obviously, numerous modifications and variations of the present application are possible in light of the above teachings. It is therefore to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. A fault diagnosis method for an electric valve actuator, characterized in that: The following steps are involved: S100: Synchronously collect mechanical vibration signals, motor current signals, and acoustic emission signals through a multimodal sensor array; S200: performing wavelet packet decomposition on the mechanical vibration signal, extracting the target frequency band with the minimum energy entropy and calculating the kurtosis index; S210: performing modal decomposition on the motor current signal, separating specific subharmonic components and calculating harmonic distortion rate; S300: Construct a cross-modal association graph model, map vibration frequency domain features, current harmonic features, and acoustic emission features into graph nodes, and weightedly fuse node relationships through an attention mechanism; S400: Fault classification is performed based on fused features using a lightweight convolutional neural network. When the output confidence level falls below a preset threshold, the knowledge graph rule engine is triggered to assist in diagnosis. S500: Clustering historical diagnosis results and dynamically updating the fault judgment threshold based on the distance between clusters; S600: Aggregate model parameters of multiple edge devices through federated learning to update the global diagnosis model.

2. A fault diagnosis method for an electric valve actuator according to claim 1, characterized in that: The extraction of the target frequency band with the minimum energy entropy in S200 includes: The complex Morlet wavelet basis function is used to perform three-layer wavelet packet decomposition; Calculate the entropy of the energy distribution in each frequency band and select the 1.2-2.4kHz frequency band with the lowest entropy.

3. A fault diagnosis method for an electric valve actuator according to claim 2, characterized in that: The specific subharmonic component in S210 is the third harmonic, and the harmonic distortion rate is calculated by the ratio of the third harmonic amplitude to the fundamental wave amplitude.

4. A fault diagnosis method for an electric valve actuator according to claim 3, characterized in that: The construction of the cross-modal association graph model in S300 includes: Mapping 12-dimensional features including vibration kurtosis, current harmonic distortion rate, and acoustic emission count rate into graph nodes; Construct adjacent edges based on physical connection relationships and temporal correlations; Learning the coupling weights between nodes through a multi-head attention mechanism.

5. A fault diagnosis method for an electric valve actuator according to claim 4, characterized in that: The triggering condition of the knowledge graph rule engine in S400 is that the confidence level is lower than 0.9; The rule engine execution includes: Matching causal chains between faulty entities; When the torque fluctuation coefficient is greater than 0.35 and the acoustic emission peak value is greater than 4.2, the "gear tooth breakage" fault conclusion is output.

6. A fault diagnosis method for an electric valve actuator according to claim 5, characterized in that: The dynamic threshold update in S500 includes: K-means clustering was performed on the continuous 24-hour diagnosis results; Calculate the minimum inter-cluster distance between the healthy state cluster and the fault state cluster; Generates a new threshold based on the product of the minimum distance and the adjustment factor.

7. A fault diagnosis method for an electric valve actuator according to claim 6, characterized in that: The federated learning aggregation cycle in S600 is 24 hours, and the model parameters are weighted averaged according to the amount of edge device data.

8. A fault diagnosis system for an electric valve actuator, applied to the fault diagnosis method for an electric valve actuator according to any one of claims 1 to 7, characterized in that: The system comprises: Sensing layer: micro strain gauge integrated on the surface of the worm gear shaft, MEMS triaxial accelerometer at the valve stem, Hall current sensor at the motor winding, and piezoelectric acoustic emission sensor on the worm gear housing; Edge computing layer: FPGA module to accelerate wavelet packet decomposition and modal decomposition; ARM processor, deploying lightweight CNN model and dynamic threshold clustering algorithm; Cloud collaboration layer: Knowledge graph engine, which stores the causal rules of fault entities; Federated learning aggregator, which updates the global model at a preset period; Maintenance interaction layer: AR guidance module, which generates 3D maintenance animation when the CNN confidence level is less than 0.9; The blockchain evidence storage module records the diagnostic results and sensor data hash values.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images