Primary helium fan fault evolution trend prediction method in combination with fault mechanism analysis

By constructing a multi-scale fault mechanism model and an active transfer learning strategy, combined with digital twin technology, the problem of integrating physical mechanisms and data-driven methods in the fault diagnosis of the main helium blower was solved, achieving accurate fault prediction and visualization, and improving the reliability and safety of the equipment.

CN118934705APending Publication Date: 2024-11-12HUANENG NUCLEAR ENERGY TECH RES INST CO LTD +1
View PDF 0 Cites 11 Cited by

Patent Information

Application Number
CN202411107081.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-13
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

Existing fault diagnosis methods for main helium blowers lack guidance from physical mechanisms, making it difficult to accurately grasp the fault development pattern. Furthermore, data-driven methods are easily affected by data quality and changes in operating conditions, and there is a lack of effective intelligent diagnostic technologies that integrate physical mechanisms and data-driven approaches.

Method used

A multi-scale fault mechanism model was constructed. Based on the fault mechanism analysis, component-level and system-level physical models were established by monitoring data from vibration, temperature, pressure, acoustic emission, and displacement sensors. An active transfer learning strategy was used to train the prediction model, and a digital twin of the main helium blower was constructed for fault tracing and visualization.

Benefits of technology

It improves the accuracy and interpretability of fault prediction, enhances the model's generalization ability and sample efficiency, realizes three-dimensional dynamic visualization of faults, assists maintenance personnel in quickly diagnosing root causes, and improves equipment reliability and safety.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118934705A_ABST
    Figure CN118934705A_ABST
Patent Text Reader

Abstract

The invention discloses a main helium fan fault evolution trend prediction method combined with fault mechanism analysis, and relates to the technical field of nuclear power station key equipment state monitoring and fault diagnosis, and the method comprises the steps: obtaining the operation monitoring data of a main helium fan, and constructing a multi-scale fault mechanism model of the main helium fan based on the operation monitoring data; performing feature extraction on operation monitoring data by using the multi-scale fault mechanism model to obtain fault features; based on the fault features, an active transfer learning strategy is adopted to train a fault prediction model, and an evolution trend prediction result of the main helium fan fault is obtained; and combining the evolution trend prediction result with a three-dimensional model of the primary helium fan, constructing a digital twinborn body of the primary helium fan, and realizing fault traceability analysis and visual display. According to the method, the accuracy, generalization and interpretability of fault prediction of the main helium fan are remarkably improved, a reliable basis is provided for predictive maintenance decisions, and the method has important significance in improving the safety and economy of a nuclear power station.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of state monitoring and fault diagnosis of key equipment in nuclear power plants, and in particular to a method for predicting the fault evolution trend of a main helium blower combined with fault mechanism analysis. Background Art

[0002] The main helium blower is one of the key equipment in a high-temperature gas-cooled reactor nuclear power plant. Its main function is to transfer the heat generated by the heat exchanger to the power conversion system to drive the steam turbine to generate electricity. Various mechanical failures will inevitably occur in the main helium blower during operation. If these failures develop to a certain extent and are not discovered and handled in time, they may cause unplanned shutdown of the main helium blower and even cause serious safety accidents, affecting the safe and stable operation of the nuclear power plant.

[0003] Traditional main helium blower fault diagnosis methods mainly rely on threshold judgment and trend analysis of monitoring parameters such as vibration and temperature. Although these methods are simple and easy to implement, they lack in-depth consideration of the complex fault mechanisms inside the equipment and it is difficult to accurately grasp the occurrence and development of faults. In recent years, with the development of advanced sensing technology and artificial intelligence algorithms, some scholars have tried to introduce data-driven machine learning methods into the field of main helium blower fault diagnosis. These methods improve the intelligent level of diagnosis to a certain extent by analyzing a large amount of historical operating data, mining fault characteristics, and establishing fault classification or prediction models. However, due to the lack of guidance from physical mechanisms, these methods are easily affected by data quality and changes in operating conditions, and their generalization ability and interpretability need to be improved.

[0004] At present, the field of main helium blower fault diagnosis still lacks an intelligent diagnosis technology that effectively integrates physical mechanism knowledge and data-driven methods. How to establish accurate physical models at multiple scales for different components and failure modes? How to extract features that are both in line with physical laws and can sensitively reflect fault signs? How to make full use of a small amount of scattered data under different working conditions to build a robust and generalized prediction model? How to use new technologies such as digital twins to achieve intelligent fault tracing, accurate early warning and three-dimensional visualization? Solving these problems is expected to further improve the accuracy, reliability and practicality of main helium blower fault diagnosis, but it still faces many technical challenges and urgently needs to carry out systematic research. Summary of the invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides a method for predicting the evolution trend of main helium blower faults combined with fault mechanism analysis, which can solve the problem that the field of main helium blower fault diagnosis still lacks an intelligent diagnosis technology that effectively integrates physical mechanism knowledge and data-driven methods.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions, which are combined with the main helium blower fault evolution trend prediction method based on the fault mechanism analysis, including:

[0008] The operation monitoring data of the main helium blower is obtained, and a multi-scale fault mechanism model of the main helium blower is constructed based on the operation monitoring data; the multi-scale fault mechanism model is used to extract features of the operation monitoring data to obtain fault features; based on the fault features, an active transfer learning strategy is adopted to train a fault prediction model to obtain an evolution trend prediction result of the main helium blower fault; the evolution trend prediction result is combined with the three-dimensional model of the main helium blower to construct a digital twin of the main helium blower to realize fault tracing analysis and visualization display.

[0009] As a preferred solution of the method for predicting the main helium blower fault evolution trend combined with fault mechanism analysis described in the present invention, the operation monitoring data includes real-time synchronous collection of monitoring data through vibration, temperature, pressure, acoustic emission and displacement sensors, and time synchronization, denoising and dimensionality reduction processing.

[0010] As a preferred solution of the main helium fan fault evolution trend prediction method combined with fault mechanism analysis described in the present invention, the multi-scale fault mechanism model includes a component-level refined physical model and a system-level simplified coupling model; the component-level refined physical model includes modeling the aerodynamic force of the main helium fan blades, considering the gas compressibility and viscosity effects, and establishing the following control equation based on the NS equation:

[0011]

[0012] Among them, W is the conserved variable vector, F is the flux vector, S is the source term vector, is the divergence operator, which represents the spatial rate of change of flux, To represent the time partial derivative, describe the rate of change of the variable with time; by solving the equation, the pressure distribution and velocity field on the blade surface are obtained, and then the aerodynamic load and stress distribution of the blade are calculated to provide boundary conditions for crack propagation analysis.

[0013] As a preferred solution of the method for predicting the main helium blower fault evolution trend combined with fault mechanism analysis described in the present invention, the component-level refined physical model also includes, for the electromagnetic bearing coil winding short-circuit fault, establishing a model that considers the dynamic characteristics of the coil circuit, as follows:

[0014]

[0015] Among them, u c is the coil voltage, R c is the coil resistance, L c is the coil inductance, ic is the coil current, e c is the coil back electromotive force, is the rate of change of current over time (A / s), which represents the time derivative of current. By solving the equation and combining it with the electromagnetic force model, it is possible to determine whether the coil is short-circuited and evaluate the force and motion state of the electromagnetic bearing.

[0016] As a preferred solution of the main helium blower fault evolution trend prediction method combined with fault mechanism analysis described in the present invention, the system-level simplified coupling model includes using a lumped parameter method, considering the coupling effect between the rotor, bearing, and frame, and establishing a rotor-bearing-frame coupling vibration equation:

[0017]

[0018] Among them, M, C, G, K are mass, damping, gyro, and stiffness matrices respectively, [x] is the generalized displacement vector, [F] is the external excitation vector, Ω is the rotation speed, is the generalized velocity vector, representing the first-order derivative of [x] with respect to time; is the generalized acceleration vector, which represents the second-order derivative of [x] with respect to time; the stiffness and damping parameters of the bearing come from the component-level model;

[0019] Data assimilation technology is used to realize the fusion of two-scale models. The Kalman filtering algorithm is used to take the component-level model output as the input of the system-level model. The system-level model output is fed back to correct the component-level model parameters, and iterative optimization is performed to achieve parameter matching and collaborative solution of multi-scale models.

[0020] As a preferred solution of the main helium fan fault evolution trend prediction method combined with fault mechanism analysis described in the present invention, wherein: the feature extraction of the operation monitoring data includes building a corresponding relationship library between the main helium fan fault mode and the monitoring parameters, and for the rotor imbalance fault, extracting the 1X component amplitude A in the vibration spectrum diagram 1x , establish A 1x Quantitative relationship with unbalance m, radius r, speed ω and phase α:

[0021] A 1x =mrω 2 cosα

[0022] After extracting the initial feature set, the Relief-F algorithm is used to screen the optimal feature subset in combination with the mapping relationship library;

[0023] For blade cracking failure, the impact component E in the blade tip timing signal is extracted, and the quantitative relationship between the impact component E and the crack length a and crack depth b is established as shown below:

[0024] E=f(a,b)=k·a m b n

[0025] Among them, k, m, and n are unknown coefficients, which are obtained through finite element simulation and experimental calibration.

[0026] As a preferred solution of the method for predicting the main helium blower fault evolution trend combined with fault mechanism analysis described in the present invention, the active transfer learning strategy includes constructing a prediction model based on a long short-term memory network LSTM:

[0027] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0028] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0029] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0030]

[0031] h t =o t ·tanh(c t )

[0032] Among them, i t 、f t , o t are the states of the input gate, forget gate, and output gate at time step t, respectively. t is the memory cell state at time step t, is the candidate memory cell state at time step t, h t is the hidden layer state at time step t, x t The input at time step t, W i , W f , W o , W c Input x t The corresponding weight matrices are used for the input gate, forget gate, output gate and candidate memory unit respectively; b i ,b f ,b o ,b c: corresponding bias vector; σ: sigmoid activation function. tanh: hyperbolic tangent activation function. ⊙: Hadamard product (element-wise multiplication); t: represents the current time step. t-1: represents the previous time step. ; Introduce the attention mechanism to the hidden layer state h at different times t Assign weights;

[0033] The parameter migration method is used to copy the LSTM model parameters trained under the source working condition to the target working condition to achieve cold start.

[0034] As a preferred solution of the method for predicting the fault evolution trend of the main helium blower combined with the fault mechanism analysis described in the present invention, wherein: the construction of the digital twin of the main helium blower includes:

[0035] Initialize the graph structure and use the key components of the main helium blower as nodes of the graph, wherein the key components include blades, bearings, rotors, electromagnetic bearings, and seals;

[0036] According to the mechanical structure and physical connection relationship, the connection between nodes is initially established, and according to the physical characteristics and possible failure modes of each node, the corresponding failure mechanism model is established to determine the physical impact path between nodes and update the edges of the graph;

[0037] Apply the conditional independence test:

[0038] For any three nodes X, Y, and Z, calculate the conditional mutual information I(X; Y|Z) between X and Y:

[0039]

[0040] If I(X;Y|Z) is significantly less than the unconditional mutual information I(X;Y), then X and Y are considered to be independent under the given Z condition; the significantly less means (I(X;Y)-I(X;Y|Z)) / I(X;Y)>τ; wherein τ is a preset threshold, and τ=0.2 is taken in the present invention. That is, when the conditional mutual information I(X;Y|Z) is reduced by more than 20% compared with the unconditional mutual information I(X;Y), X and Y are considered to be independent under the given Z condition.

[0041] Where I(X; Y|Z) represents the conditional mutual information between X and Y given Z; X, Y, Z represent three random variables; x, y, z represent the specific values ​​of the three random variables X, Y, and Z respectively; p(x, y, z) represents the joint probability of X=x, Y=y, and Z=z occurring simultaneously; p(x, y|z) represents the conditional probability of X=x and Y=y occurring simultaneously under the condition of Z=z; p(x|z) represents the conditional probability of X=x occurring under the condition of Z=z; p(y|z) represents the conditional probability of Y=y occurring under the condition of Z=z;

[0042] Update the graph structure, add newly discovered causal relationships, and delete insignificant connections; insignificant means that when the difference between I(X; Y|Z) and I(X; Y) does not meet the condition of being significantly less than , the connection is considered insignificant;

[0043] Using the updated graph structure, a graph traversal algorithm is applied, starting from the node where the fault is detected and tracing back in the reverse direction of the edge until the possible root node is reached, generating a fault propagation path.

[0044] A computer device includes a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the processor executes the computer program, steps of a method for predicting the fault evolution trend of a main helium blower combined with fault mechanism analysis are implemented.

[0045] 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 a method for predicting the fault evolution trend of a main helium blower combined with fault mechanism analysis are implemented.

[0046] Beneficial effects of the present invention: 1. The present invention fully considers the physical mechanisms of different components and failure modes of the main helium blower by constructing a multi-scale failure mechanism model, thereby improving the accuracy and interpretability of failure prediction. Component-level refined modeling reveals the intrinsic relationship between local stress strain, vibration and failure of key components, while system-level simplified modeling depicts the dynamic coupling effect between components. The two work together to improve the overall fidelity of modeling.

[0047] 2. The present invention integrates domain knowledge to guide data-driven feature extraction, overcoming the limitations of traditional data-driven methods that are easily affected by data quality and operating conditions. By building a fault-monitoring parameter mapping relationship library, using expert experience to pre-screen candidate feature sets with clear physical meanings, and then using machine learning algorithms to select the most sensitive feature subsets, the "mechanism + data" dual optimization of feature extraction is achieved.

[0048] 3. The present invention adopts active transfer learning to train the prediction model, making full use of scattered monitoring data under different working conditions, significantly improving the generalization and sample efficiency of the model. The model parameters of the source working condition are transferred to the target working condition to realize knowledge reuse, and the training process is interactively optimized through active learning strategies, focusing on boundary sample mining. A small number of inquiries each time can quickly improve the model effect, breaking through the bottleneck of scarce fault data.

[0049] 4. The present invention introduces digital twin technology to construct a virtual image of the main helium blower, realizing three-dimensional dynamic visualization of fault tracing, evolution, and early warning. Through fault-component mapping and high-fidelity physical simulation, the whole process of how the fault is generated, propagated, and developed at multiple scales in time and space is intuitively demonstrated, assisting operation and maintenance personnel to quickly diagnose the root cause and formulate predictive maintenance strategies.

[0050] 5. In the actual application of nuclear power engineering, the present invention can effectively guide the status monitoring, fault warning and maintenance decision of the main helium blower, and improve the reliability, safety and economy of the equipment. By predicting the degradation trend of key components in advance and reasonably arranging maintenance time and projects, unplanned downtime can be minimized, the service life of the equipment can be extended, and the safety of the nuclear power plant can be protected.

[0051] 6. This invention sets a new paradigm for the intelligent operation and maintenance of complex rotating machinery and equipment. The proposed methods of multi-scale mechanism modeling, feature extraction optimization, active transfer learning, digital twin presentation, etc. have certain universality and demonstration significance, and can be extended to other large rotating machinery such as steam turbines and compressors, to help the intelligent management of equipment health in the era of industrial Internet. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0053] Figure 1 A schematic flow chart of a method for predicting the main helium blower fault evolution trend in combination with fault mechanism analysis provided in one embodiment of the present invention.

[0054] Figure 2 A schematic diagram of main helium blower monitoring data collection and preprocessing for a main helium blower fault evolution trend prediction method combined with fault mechanism analysis provided in one embodiment of the present invention.

[0055] Figure 3 A multi-scale mechanism model construction and fusion flow chart of a method for predicting the main helium blower fault evolution trend combined with fault mechanism analysis provided in one embodiment of the present invention.

[0056] Figure 4 A flow chart of an active learning strategy iterative optimization model of a method for predicting the main helium blower fault evolution trend combined with fault mechanism analysis provided in one embodiment of the present invention.

[0057] Figure 5A digital twin-based fault tracing analysis and visualization flowchart of a main helium blower fault evolution trend prediction method combined with fault mechanism analysis provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0058] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0059] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0060] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0061] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0062] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0063] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0064] Example 1, reference Figure 1-Figure 5 , which is the first embodiment of the present invention, provides a method for predicting the main helium blower fault evolution trend in combination with fault mechanism analysis, including:

[0065] Step S1: Obtain the operation monitoring data of the main helium blower. Figure 2 As shown in the figure, various types of sensors are arranged on the key components of the main helium blower, such as bearings, blades, rotors, and frames, including vibration, temperature, pressure, acoustic emission, and displacement sensors. Among them, the sampling frequency of the vibration sensor is 10kHz, the accuracy of the temperature sensor is 0.1℃, the range of the pressure sensor is 0-10MPa, the frequency range of the acoustic emission sensor is 50kHz-400kHz, and the resolution of the displacement sensor is 1μm. The sensors collect monitoring data in real time and synchronously, and perform preprocessing such as time synchronization, denoising, and dimensionality reduction. The wavelet threshold method is used for denoising, with 3 times the standard deviation as the threshold; the principal component analysis method is used for dimensionality reduction, and the principal components with a cumulative contribution rate greater than 95% are retained.

[0066] Step S2: Based on the operation monitoring data, a multi-scale fault mechanism model of the main helium blower is constructed. Corresponding physical models are constructed for typical failure modes of the main helium blower, such as impeller imbalance, impeller crack, electromagnetic bearing coil short circuit, electromagnetic bearing clearance abnormality, and auxiliary bearing sealing surface wear.

[0067] At the component level, a refined model is established. Taking the aerodynamic modeling of the main helium fan blade as an example, considering the gas compressibility and viscosity effects, the following control equation is established based on the NS equation:

[0068]

[0069] Among them, W is the conserved variable vector, F is the flux vector, S is the source term vector, is the divergence operator, which represents the spatial rate of change of flux, To represent the time partial derivative, describe the rate of change of the variable over time. By solving this equation, the pressure distribution and velocity field on the blade surface can be obtained, and then the aerodynamic load and stress distribution of the blade can be calculated, providing boundary conditions for crack growth analysis.

[0070] For the short-circuit fault of the electromagnetic bearing coil winding, considering the circuit dynamic characteristics of the coil, the following coil voltage equation is established:

[0071]

[0072] Among them, u c is the coil voltage, R c is the coil resistance, L c is the coil inductance, i c is the coil current, e c is the coil back electromotive force, is the rate of change of current over time (A / s), which represents the time derivative of current. By solving this equation and combining it with the electromagnetic force model, it can be determined whether the coil is short-circuited, and then the force and motion state of the electromagnetic bearing can be evaluated.

[0073] At the system level, the lumped parameter method is used to establish a simplified model. Considering the coupling effect between the rotor, bearings, and frame, the following vibration equation is established:

[0074]

[0075] Among them, M, C, G, K are mass, damping, gyro and stiffness matrices respectively, [x] is the generalized displacement vector, [F] is the external excitation vector, and Ω is the rotation speed. The stiffness and damping parameters of the bearing come from the component-level model.

[0076] The data assimilation technique is used to achieve the fusion of two scale models, such as Figure 3 Using the Kalman filter algorithm, the output of the component-level model is used as the input of the system-level model, and the output of the system-level model is fed back to correct the parameters of the component-level model, and iterative optimization is performed to achieve parameter matching and collaborative solution of the multi-scale model.

[0077] Step S3: Use the multi-scale fault mechanism model and domain knowledge to extract features from the operation monitoring data. Build a corresponding relationship library between the main helium blower fault mode and the monitoring parameters. For example, for a rotor imbalance fault, extract the 1X component amplitude A in the vibration spectrum diagram. 1x , establish its quantitative relationship with the unbalance amount m, radius r, speed ω and phase α:

[0078] A 1x =mrω 2 cosα

[0079] For another example, for blade cracking failure, the impact component E in the blade tip timing signal is extracted, and its quantitative relationship with the crack length a and depth b is established:

[0080] E=f(a,b)=k·a m b n

[0081] Among them, k, m, and n are unknown coefficients, which can be obtained through finite element simulation and experimental calibration.

[0082] After extracting the initial features in the time domain, frequency domain, and time-frequency domain, combined with the corresponding relationship library between fault modes and monitoring parameters, the features are screened and sorted using algorithms such as Relief-F to automatically obtain the optimal feature subset that is highly correlated with the fault mode. The specific implementation steps are as follows:

[0083] A total of 1000 initial features are extracted from the time domain, frequency domain, and time-frequency domain.

[0084] The initial features are preliminarily screened based on the pre-established library of correspondence between fault modes and monitoring parameters. For example, for blade cracks, the features related to the blade tip timing signal are retained. The degree of crack can be expressed as:

[0085] E=k·a m b n

[0086] Among them, E is the impact component in the blade tip timing signal, a is the crack length, b is the crack depth, and k, m, and n are unknown coefficients. This step can initially reduce the number of features to about 200.

[0087] Apply the Relief-F algorithm. The core idea of ​​the Relief-F algorithm is to evaluate the distinguishing ability of features. For feature f, the simplified formula for updating its weight can be expressed as:

[0088]

[0089] Among them, X is a randomly selected sample, Hit is the nearest sample of the same type as X, Miss is the nearest sample of a different type as X, and k is the number of neighbors.

[0090] The top 50 features with the highest weights are selected as feature subsets, and the selected feature subsets are manually reviewed to ensure that each feature has a clear physical meaning.

[0091] Through the above steps, the feature dimension is finally reduced from 1000 to 50, and each feature is related to a specific fault mode and has a clear physical meaning. This feature selection method that combines domain knowledge and data-driven not only ensures the physical interpretability of the features, but also improves the efficiency and accuracy of the subsequent fault prediction model.

[0092] Step S4: Based on the fault characteristics, the active transfer learning strategy is used to train the fault prediction model. Build a prediction model based on the long short-term memory network (LSTM):

[0093] i t =σ(W i ·[h t-1,x t ]+b i )

[0094] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0095] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0096]

[0097] h t =o t ·tanh(c t )

[0098] Among them, i t ,f,o t They are input gate, forget gate and output gate respectively, c t is the memory unit state, h t is the hidden layer state, W and b are the weight matrix and bias vector to be learned. The attention mechanism is introduced to the hidden layer state h at different times t Assign weights to highlight the impact of key time steps.

[0099] In order to solve the problem of insufficient training samples caused by frequent switching of the main helium blower working conditions, the parameter migration method is used to copy the LSTM model parameters trained under the source working condition to the target working condition to achieve cold start. A small number of samples are used for fine-tuning under the target working condition.

[0100] Active learning strategies are adopted during fine-tuning, such as Figure 4 As shown in the figure. Initially, a small number of samples are randomly selected to train LSTM, and then the trained model is used to predict all samples and calculate the uncertainty (such as entropy) of each sample. The top 5% of samples with the largest uncertainty are selected and handed over to experts for annotation to expand the training set. The above process is repeated until the model performance converges. Active learning allows only the most valuable samples to be annotated each time, which significantly improves the sample utilization efficiency and model generalization performance.

[0101] Active learning strategies are adopted during fine-tuning, such as Figure 4 As shown. Initially, a small number of samples are randomly selected to train LSTM, and then the trained model is used to predict all samples and calculate the uncertainty (such as entropy) of each sample. The specific calculation steps are as follows:

[0102] For each unlabeled sample, the current LSTM model is used to make predictions to obtain the probability distribution P = {p1, p2, ..., pn} of the sample belonging to each possible fault category, where n is the number of fault categories.

[0103] Calculate the predicted probability entropy H of the sample as an uncertainty indicator:

[0104]

[0105] Where H is the uncertainty of the sample, n is the total number of fault categories, and p i is the predicted probability that the sample belongs to the i-th fault category; when p i When =0, define 0*log(0)=0.

[0106] According to the calculated entropy value H, all unlabeled samples are sorted in descending order.

[0107] The first 5% samples with the largest uncertainty are selected from the sorted samples and handed over to experts for labeling to expand the training set.

[0108] Repeat the above process until the model performance converges.

[0109] The model outputs the prediction results of the evolution trend of the fault quantitative indicators (such as imbalance, crack length, etc.) of the key components of the main helium blower (such as rotors, bearings, blades, etc.) in the future.

[0110] Step S5: Combine the prediction results with the three-dimensional model of the main helium blower to construct its digital twin to achieve fault tracing analysis and visualization. Figure 5 As shown in the figure, a high-fidelity three-dimensional geometric model of the main helium blower is first constructed, including the rotor, bearings, blades, frame, seals and other components, with a geometric accuracy of 0.1mm. On this basis, the multi-physics finite element method is used to solve the physical fields such as stress and strain, vibration displacement, fluid pressure, temperature distribution, etc. of the components to obtain the physical state of the components.

[0111] Construct a component-level fault propagation directed graph in the digital twin. The specific implementation steps are as follows:

[0112] Initialize the graph structure and use the key components of the main helium blower as nodes of the graph, wherein the key components include blades, bearings, rotors, electromagnetic bearings, and seals;

[0113] According to the mechanical structure and physical connection relationship, the connection between nodes is initially established, and according to the physical characteristics and possible failure modes of each node, the corresponding failure mechanism model is established to determine the physical impact path between nodes and update the edges of the graph;

[0114] Apply the conditional independence test, and for any three nodes X, Y, and Z, calculate the conditional mutual information I(X; Y|Z) of X and Y:

[0115]

[0116] If I(X;Y|Z) is significantly less than the unconditional mutual information I(X;Y), then X and Y are considered to be independent under the given Z condition; the significantly less means (I(X;Y)-I(X;Y|Z)) / I(X;Y)>τ; wherein τ is a preset threshold, and τ=0.2 is taken in the present invention. That is, when the conditional mutual information I(X;Y|Z) is reduced by more than 20% compared with the unconditional mutual information I(X;Y), X and Y are considered to be independent under the given Z condition.

[0117] Where I(X; Y|Z) represents the conditional mutual information between X and Y given Z; X, Y, Z represent three random variables; x, y, z represent the specific values ​​of the three random variables X, Y, and Z respectively; p(x, y, z) represents the joint probability of X=x, Y=y, and Z=z occurring simultaneously; p(x, y|z) represents the conditional probability of X=x and Y=y occurring simultaneously under the condition of Z=z; p(x|z) represents the conditional probability of X=x occurring under the condition of Z=z; p(y|z) represents the conditional probability of Y=y occurring under the condition of Z=z;

[0118] Update the graph structure, add newly discovered causal relationships, and delete insignificant connections; insignificant means that when the difference between I(X; Y|Z) and I(X; Y) does not meet the condition of being significantly less than , the connection is considered insignificant;

[0119] Using the updated graph structure, a graph traversal algorithm is applied, starting from the node where the fault is detected and tracing back in the reverse direction of the edge until the possible root node is reached, generating a fault propagation path.

[0120] Through the above steps, we can build a dynamically updated fault propagation directed graph in the digital twin. This graph is not only based on the physical model, but also integrates data-driven causal discovery, and can adaptively reflect the actual operating status of the equipment. The final fault propagation path can intuitively reveal the generation and evolution process of the fault, help operation and maintenance personnel quickly locate the root cause of the fault, and formulate targeted maintenance strategies. This method combines the interpretability of the mechanism model and the adaptability of the data-driven method, greatly improving the accuracy and efficiency of fault diagnosis.

[0121] Finally, the fault prediction results are mapped to the corresponding components of the main helium blower three-dimensional model. A dynamic distribution diagram of the physical parameters (such as stress, vibration, etc.) of the component within the prediction time range is generated to intuitively display its fault evolution process. Through the digital twin platform, the fault quantitative indicators of key components and the evolution of the physical parameter field over time are presented in the form of three-dimensional animation, which assists operation and maintenance personnel to intuitively understand and predict the health status of the main helium blower.

[0122] In summary, the method for predicting the evolution trend of main helium blower faults combined with fault mechanism analysis proposed in the present invention is based on multi-source heterogeneous monitoring big data, integrates domain knowledge and machine learning technology, constructs a multi-scale mechanism model combining component level and system level, and extracts sensitive fault features with physical significance; adopts active transfer learning strategy, makes full use of different working condition data, and significantly improves the generalization and sample efficiency of the prediction model; introduces digital twins to reveal the generation, propagation and evolution mechanism of faults. The present invention can significantly improve the accuracy and explainability of main helium blower fault prediction, provide a reliable basis for its predictive maintenance decision, extend its service life, and improve the safety, economy and environmental protection of nuclear power steam turbine units.

[0123] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention 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 invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

[0124] Embodiment 2, the second embodiment of the present invention, is different from the previous embodiment in that:

[0125] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0126] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0127] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0128] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0129] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0130] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for predicting the failure evolution trend of the main helium blower combined with the failure mechanism analysis is characterized by: include, Acquire operation monitoring data of the main helium blower, and construct a multi-scale fault mechanism model of the main helium blower based on the operation monitoring data; Using the multi-scale fault mechanism model to extract features from the operation monitoring data, and obtain fault features; Based on the fault characteristics, an active transfer learning strategy is adopted to train the fault prediction model and obtain the evolution trend prediction result of the main helium blower fault; The evolution trend prediction results are combined with the three-dimensional model of the main helium blower to construct a digital twin of the main helium blower to achieve fault tracing analysis and visualization display.

2. The method for predicting the main helium blower failure evolution trend combined with failure mechanism analysis according to claim 1, characterized in that: The operation monitoring data includes real-time synchronous collection of monitoring data through vibration, temperature, pressure, acoustic emission and displacement sensors, and time synchronization, denoising and dimensionality reduction processing.

3. The method for predicting the main helium blower failure evolution trend combined with failure mechanism analysis according to claim 2 is characterized in that: The multi-scale fault mechanism model includes a component-level refined physical model and a system-level simplified coupling model; the component-level refined physical model includes modeling the aerodynamic force of the main helium fan blades, considering the gas compressibility and viscosity effects, and establishing the following control equation based on the NS equation: Where W is the conserved variable vector, F is the flux vector, and S is the source term vector. is the divergence operator, which represents the spatial rate of change of flux, To represent the time partial derivative, describe the rate of change of the variable with time; by solving the equation, the pressure distribution and velocity field on the blade surface are obtained, and then the aerodynamic load and stress distribution of the blade are calculated to provide boundary conditions for crack propagation analysis.

4. The method for predicting the main helium blower failure evolution trend combined with failure mechanism analysis according to claim 3 is characterized by: The component-level refined physical model also includes establishing a model that considers the dynamic characteristics of the coil circuit for the electromagnetic bearing coil winding short-circuit fault, as follows: Among them, u c is the coil voltage, R c is the coil resistance, L c is the coil inductance, i c is the coil current, e c is the coil back electromotive force, is the rate of change of current over time, representing the time derivative of current; by solving the equation and combining it with the electromagnetic force model, it is possible to determine whether the coil is short-circuited and evaluate the force and motion state of the electromagnetic bearing.

5. The method for predicting the main helium blower failure evolution trend combined with failure mechanism analysis according to claim 4, characterized in that: The system-level simplified coupling model includes using a lumped parameter method, considering the coupling effect between the rotor, bearings, and frame, and establishing a rotor-bearing-frame coupling vibration equation: Among them, M, C, G, K are mass, damping, gyro, and stiffness matrices respectively, [x] is the generalized displacement vector, [F] is the external excitation vector, Ω is the rotation speed, is the generalized velocity vector, representing the first-order derivative of [x] with respect to time; is the generalized acceleration vector, which represents the second-order derivative of [x] with respect to time; the stiffness and damping parameters of the bearing come from the component-level model; Data assimilation technology is used to realize the fusion of two-scale models. The Kalman filtering algorithm is used to take the component-level model output as the input of the system-level model. The system-level model output is fed back to correct the component-level model parameters, and iterative optimization is performed to achieve parameter matching and collaborative solution of multi-scale models.

6. The method for predicting the main helium blower failure evolution trend combined with failure mechanism analysis according to claim 5, characterized in that: The feature extraction of the operation monitoring data includes constructing a corresponding relationship library between the main helium blower failure mode and the monitoring parameters, and extracting the 1X component amplitude A in the vibration spectrum diagram for the rotor imbalance failure. 1x , establish A 1x Quantitative relationship with unbalance m, radius r, speed ω and phase α: A 1x mrω 2 cosα After extracting the initial feature set, the Relief-F algorithm is used to screen the optimal feature subset in combination with the mapping relationship library; For blade cracking failure, the impact component E in the blade tip timing signal is extracted, and the quantitative relationship between the impact component E and the crack length a and crack depth b is established as shown below: E=f(a,b)=k·a m ·b n Among them, k, m, and n are unknown coefficients, which are obtained through finite element simulation and experimental calibration.

7. The method for predicting the main helium blower failure evolution trend combined with failure mechanism analysis according to claim 6, characterized in that: The active transfer learning strategy includes building a prediction model based on the long short-term memory network LSTM: i t =σ(W i ·[h t-1 ,x t ]+b i ) f t =σ(W f ·[h t-1 ,x t ]+b f ) the t =σ(W o ·[h t-1 ,x t ]+b o ) h t =o t ·tanh(c t ) Among them, i t 、f t , o t are the states of the input gate, forget gate, and output gate at time step t, respectively. t is the memory cell state at time step t, is the candidate memory cell state at time step t, h t is the hidden layer state at time step t, x t The input at time step t, W i , W f , W o , W c Input x t The corresponding weight matrices are used for the input gate, forget gate, output gate and candidate memory unit respectively; b i ,b f ,b o ,b c : corresponding bias vector; σ: sigmoid activation function. tanh: hyperbolic tangent activation function. ⊙: Hadamard product (element-wise multiplication); t: represents the current time step. t-1: represents the previous time step. ; Introduce the attention mechanism to the hidden layer state h at different times t Assign weights; The parameter migration method is used to copy the LSTM model parameters trained under the source working condition to the target working condition to achieve cold start.

8. The method for predicting the main helium blower failure evolution trend combined with failure mechanism analysis according to claim 7, characterized in that: The construction of the digital twin of the main helium blower includes: Initialize the graph structure and use the key components of the main helium blower as nodes of the graph, wherein the key components include blades, bearings, rotors, electromagnetic bearings, and seals; According to the mechanical structure and physical connection relationship, the connection between nodes is initially established, and according to the physical characteristics and possible failure modes of each node, the corresponding failure mechanism model is established to determine the physical impact path between nodes and update the edges of the graph; Apply the conditional independence test: For any three nodes X, Y, and Z, calculate the conditional mutual information I(X; Y|Z) between X and Y: If I(X; Y|Z) is significantly less than the unconditional mutual information I(X; Y), then X and Y are considered independent under the given condition of Z; the significantly less than means (I(X; Y)-I(X; Y|Z)) / I(X; Y)>τ; where τ is a preset threshold; Where I(X; Y|Z) represents the conditional mutual information between X and Y given Z; X, Y, Z represent three random variables; x, y, z represent the specific values ​​of the three random variables X, Y, and Z respectively; p(x, y, z) represents the joint probability of X=x, Y=y, and Z=z occurring simultaneously; p(x, y|z) represents the conditional probability of X=x and Y=y occurring simultaneously under the condition of Z=z; p(x|z) represents the conditional probability of X=x occurring under the condition of Z=z; p(y|z) represents the conditional probability of Y=y occurring under the condition of Z=z; Update the graph structure, add newly discovered causal relationships, and delete insignificant connections; insignificant means that when the difference between I(X; Y|Z) and I(X; Y) does not meet the condition of being significantly less than , the connection is considered insignificant; Using the updated graph structure, a graph traversal algorithm is applied, starting from the node where the fault is detected and tracing back in the reverse direction of the edge until the possible root node is reached, generating a fault propagation path.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 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 8 are implemented.

Citation Information

Cited By

  • Game agent input and output boundary positioning method for time series data inversion

    CN119378695A

  • A Method for Locating the Input-Output Boundary of a Game Agent for Inverting Time-Series Data

    CN119378695B

  • Intelligent cell gate fault early warning system based on combinatorial algorithm

    CN119474886A

  • Life prediction method, device and system for chain transmission system for multi-component coupling

    CN119578127A

  • Electromagnetic water meter error compensation method and system combining conductivity and temperature

    CN120609427A