Wind turbine gearbox rupture early warning method and device based on acoustic emission positioning technology

Abstract

The invention relates to the technical field of power generation equipment fault diagnosis, and discloses a fan gear box fracture early warning method and device based on an acoustic emission positioning technology, and the method comprises the steps: obtaining the internal structure parameters of a fan gear box, and constructing a sound wave propagation model based on the internal structure parameters of the fan gear box; performing sound wave propagation simulation at the positions of the plurality of sound emission sources by using the sound wave propagation model to obtain sound wave propagation simulation data; acquiring an acoustic emission real-time signal, and performing path separation on the acoustic emission real-time signal by using the graph convolutional network model to obtain classification probability data of a multipath signal path; and based on the sound wave propagation simulation data and the classification probability data of the multipath signal paths, fault signal detection is carried out by using a reinforcement learning agent model, and fan gearbox rupture early warning information is obtained. According to the invention, the false alarm rate is effectively reduced, the accuracy and reliability of the early warning system are greatly improved, and a more reliable guarantee is provided for the early warning of the rupture of the fan gearbox.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power generation equipment fault diagnosis, and particularly relates to a fan gear box rupture early warning method and device based on acoustic emission positioning technology. BACKGROUND

[0002] With the rapid development of the wind power industry, the reliability and stability of the fan as the core power generation equipment are crucial. The fan gear box as a key component of the fan is long-term in complex and variable operating conditions, and is prone to various faults. The internal rupture of the gear box is a relatively serious fault form, which may cause the fan to shut down and even cause safety accidents. Therefore, it is of great significance to accurately and timely warn the internal rupture of the fan gear box.

[0003] However, the related fan gear box rupture early warning method has problems such as multi-path interference, inaccurate signal processing, and high false positive rate, which cannot guarantee the accuracy and reliability of the fan gear box rupture early warning result. SUMMARY

[0004] Therefore, the present application provides a fan gear box rupture early warning method and device based on acoustic emission positioning technology to solve the problem that the accuracy and reliability of the fan gear box rupture early warning result cannot be guaranteed.

[0005] In a first aspect, the present application provides a fan gear box rupture early warning method based on acoustic emission positioning technology, which comprises:

[0006] Obtaining the internal structure parameters of the fan gear box, and constructing an acoustic wave propagation model based on the internal structure parameters of the fan gear box;

[0007] Using the acoustic wave propagation model to simulate acoustic wave propagation at multiple acoustic emission source positions to obtain acoustic wave propagation simulation data;

[0008] Obtaining acoustic emission real-time signals, and using a graph convolution network model to separate the acoustic emission real-time signals to obtain classification probability data of multi-path signal paths;

[0009] Based on the acoustic wave propagation simulation data and the classification probability data of the multi-path signal paths, using a reinforcement learning agent model to detect fault signals to obtain fan gear box rupture early warning information.

[0010] The fan gear box rupture early warning method based on the acoustic emission positioning technology provided in the embodiment can construct an acoustic wave propagation model based on internal structure parameters of the fan gear box, and then simulate acoustic wave propagation at multiple acoustic emission source positions by using the acoustic wave propagation model to obtain acoustic wave propagation simulation data, so that the propagation path and characteristics of the acoustic emission signal under multipath effect can be more accurately predicted, useful information and interference components in the signal can be more accurately identified, and the reliability of signal analysis is improved. Secondly, the path separation of the acoustic emission real-time signal is performed by using the graph convolution network model, which provides purer and more accurate signals for acoustic emission signal processing and fault diagnosis, and helps to improve the accuracy of fault diagnosis. Finally, based on the acoustic wave propagation simulation data and the classification probability data of the multipath signal path, the fault signal detection is performed by using the reinforcement learning agent model, so that the system can automatically optimize the fault detection strategy according to the actual working condition, effectively reduces the false positive rate, and greatly improves the accuracy and reliability of the early warning system, thereby providing more reliable protection for the fan gear box rupture early warning.

[0011] In an optional implementation, the internal structure parameters of the fan gear box are acquired, and the acoustic wave propagation model is constructed based on the internal structure parameters of the fan gear box, including:

[0012] The three-dimensional parameters of each component inside the fan gear box are acquired, and a three-dimensional physical model of the fan gear box is constructed based on the three-dimensional parameters of each component inside the fan gear box.

[0013] The acoustic parameters of each component inside the fan gear box are acquired, and the acoustic parameters of each component inside the fan gear box are given to the three-dimensional physical model of the fan gear box.

[0014] The three-dimensional physical model of the fan gear box to which the acoustic parameters of each component inside the fan gear box are given is subjected to finite element analysis to obtain the acoustic wave propagation model.

[0015] The fan gear box rupture early warning method based on the acoustic emission positioning technology provided in the embodiment can construct an acoustic wave propagation model based on internal structure parameters of the fan gear box, and then simulate acoustic wave propagation at multiple acoustic emission source positions by using the acoustic wave propagation model to obtain acoustic wave propagation simulation data, so that the propagation path and characteristics of the acoustic emission signal under multipath effect can be more accurately predicted, useful information and interference components in the signal can be more accurately identified, and the reliability of signal analysis is improved. Secondly, the path separation of the acoustic emission real-time signal is performed by using the graph convolution network model, which provides purer and more accurate signals for acoustic emission signal processing and fault diagnosis, and helps to improve the accuracy of fault diagnosis. Finally, based on the acoustic wave propagation simulation data and the classification probability data of the multipath signal path, the fault signal detection is performed by using the reinforcement learning agent model, so that the system can automatically optimize the fault detection strategy according to the actual working condition, effectively reduces the false positive rate, and greatly improves the accuracy and reliability of the early warning system, thereby providing more reliable protection for the fan gear box rupture early warning.

[0016] In an optional implementation, the path separation of the acoustic emission real-time signal is performed by using the graph convolution network model to obtain the classification probability data of the multipath signal path, including:

[0017] The acoustic emission real-time signal is subjected to signal preprocessing, and the acoustic emission real-time signal after the signal preprocessing is subjected to feature extraction to obtain signal propagation characteristics.

[0018] Based on the signal pre-processed acoustic emission real-time signal, the classification probability data of the multi-path signal path is determined by using a graph convolution network model.

[0019] The fan gearbox rupture early warning method based on the acoustic emission positioning technology provided in the embodiment uses a graph convolution network model to separate the acoustic emission real-time signal path, reduces the noise pollution to the features (such as amplitude and phase), and the features of the separated real signal path are closer to the actual fault signal, so that the "fault-noise" boundary during threshold adjustment is clearer.

[0020] In an optional implementation, based on the acoustic wave propagation simulation data and the classification probability data of the multi-path signal path, a reinforcement learning agent model is used to detect the fault signal to obtain the fan gearbox rupture early warning information, including:

[0021] The acoustic wave propagation simulation data is positioned and optimized to determine a multi-path interference degree index;

[0022] The classification probability data of the multi-path signal path is used to determine a signal separation accuracy;

[0023] The reinforcement learning agent model is used to determine a threshold adjustment strategy under multiple fan operating conditions;

[0024] The current fan operating parameters are obtained, and based on the current fan operating parameters, the multi-path interference degree index, and the signal path separation accuracy, the threshold adjustment strategy under the multiple fan operating conditions is used to determine a fault triggering threshold;

[0025] The signal propagation feature is compared with the fault triggering threshold, and if the signal propagation feature exceeds the fault triggering threshold, the fan gearbox rupture early warning information is generated.

[0026] The fan gearbox rupture early warning method based on the acoustic emission positioning technology provided in the embodiment can make timely threshold adjustment decisions through the reinforcement learning agent model as the operating conditions change, ensure that the fault signal can be accurately detected in various situations, not only improve the real-time performance of fault detection, but also ensure the accuracy of detection, and will not miss the real fault due to improper threshold setting, thereby providing more reliable protection for the safe operation of the equipment.

[0027] In an optional implementation, based on the acoustic wave propagation simulation data, a multi-path interference degree index is determined, including:

[0028] Based on the acoustic wave propagation simulation data, a positioning algorithm is used to determine an acoustic emission source position vector;

[0029] The propagation parameter simulation is performed based on the acoustic emission source position vector to obtain simulation basic parameters of acoustic wave propagation in the gearbox;

[0030] The actual measurement basic parameters of acoustic wave propagation in the gearbox are obtained, and the simulation basic parameters of acoustic wave propagation in the gearbox are compared with the actual measurement basic parameters of acoustic wave propagation in the gearbox to obtain a multipath interference degree index.

[0031] The fan gearbox rupture early warning method based on the acoustic emission positioning technology provided in the embodiment provides real-time environmental feedback for the reinforcement learning through the quantitative result of the positioning accuracy (i.e., the multipath interference degree index), so that the system can dynamically optimize the detection strategy according to the reliability of acoustic wave propagation, and improve the accuracy and robustness of the rupture early warning.

[0032] In an optional implementation, the threshold adjustment strategy under multiple fan operating conditions is determined by using a reinforcement learning agent model, including:

[0033] defining a state, an action and a reward function; wherein the state is historical false alarm data, fan operating parameters, a multipath interference degree index and a signal path separation accuracy corresponding to the multiple fan operating conditions; the action is an adjustment operation of a fault triggering threshold;

[0034] obtaining a current state, selecting a current action according to the current state, adjusting the fault triggering threshold based on the current action, and updating the current state to obtain an updated state;

[0035] calculating a current reward value based on the updated state;

[0036] calculating a current Q value based on the current state, the current action, the updated state and the current reward value;

[0037] comparing the current Q value with a target Q value, and iteratively updating the threshold adjustment strategy based on the comparison result until the current Q value is equal to the target Q value, to obtain the threshold adjustment strategy under the multiple fan operating conditions.

[0038] The fan gearbox rupture early warning method based on the acoustic emission positioning technology provided in the embodiment, the reinforcement learning agent model dynamically adjusts the fault triggering threshold according to the historical false alarm data and the current condition, and the acoustic emission signal features are quite different under different operating conditions during the operation of the fan. A fixed threshold value is easy to cause false alarms. Therefore, through the dynamic threshold adjustment mechanism, the system can automatically adapt to the operating condition changes, automatically increase the threshold value to avoid false alarms in the case of easy transient noise generation in the fan startup stage, and the reinforcement learning agent continuously learns and adjusts the threshold value in real time, so that the system always maintains the best fault detection performance.

[0039] In a second aspect, the present application provides a fan gearbox rupture early warning device based on acoustic emission positioning technology, which comprises:

[0040] a construction module configured to acquire internal structure parameters of the fan gearbox, and construct a sound wave propagation model based on the internal structure parameters of the fan gearbox;

[0041] a simulation module configured to perform sound wave propagation simulation at multiple acoustic emission source positions by using the sound wave propagation model, and obtain sound wave propagation simulation data;

[0042] a separation module configured to acquire acoustic emission real-time signals, and perform path separation on the acoustic emission real-time signals by using a graph convolution network model, and obtain classification probability data of multi-path signal paths;

[0043] a detection module configured to perform fault signal detection by using a reinforcement learning agent model based on the sound wave propagation simulation data and the classification probability data of the multi-path signal paths, and obtain fan gearbox cracking early warning information.

[0044] In a third aspect, the present application provides a computer device, comprising a memory and a processor, the memory and the processor are communicatively connected with each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the fan gearbox cracking early warning method based on acoustic emission positioning technology according to the first aspect or any one of the corresponding embodiments thereof.

[0045] In a fourth aspect, the present application provides a computer readable storage medium, which stores computer instructions, and the computer instructions are used to make a computer execute the fan gearbox cracking early warning method based on acoustic emission positioning technology according to the first aspect or any one of the corresponding embodiments thereof.

[0046] In a fifth aspect, the present application provides a computer program product, which comprises computer instructions, and the computer instructions are used to make a computer execute the fan gearbox cracking early warning method based on acoustic emission positioning technology according to the first aspect or any one of the corresponding embodiments thereof. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the drawings needed in the specific embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0048] Figure 1 is a flowchart of a fan gearbox cracking early warning method based on acoustic emission positioning technology according to an embodiment of the present application;

[0049] Figure 2is a flowchart of another fan gear box rupture early warning method based on acoustic emission positioning technology according to an embodiment of the present application;

[0050] Figure 3 is a flowchart of still another fan gear box rupture early warning method based on acoustic emission positioning technology according to an embodiment of the present application;

[0051] Figure 4 is a flowchart of yet another fan gear box rupture early warning method based on acoustic emission positioning technology according to an embodiment of the present application;

[0052] Figure 5 is a structural block diagram of a fan gear box rupture early warning device based on acoustic emission positioning technology according to an embodiment of the present application;

[0053] Figure 6 is a hardware structure schematic diagram of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0054] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application.

[0055] At present, acoustic emission positioning technology as an effective fault monitoring means has been applied in fan gear box fault early warning to a certain extent. However, in the actual application process, the technology faces many challenges:

[0056] 1) Multipath interference problem: the internal structure of the fan gear box is complex, and there are a large number of structure components such as partitions, bearing seats, etc.; when the acoustic waves generated by the acoustic emission source propagate in the box, these structures will cause the acoustic waves to be reflected and refracted, forming a multipath effect, which makes the received acoustic emission signal complex and distorted, seriously interfering with the accurate judgment of the acoustic emission source position.

[0057] For example, due to the superposition of reflected waves and direct waves, the signal received by the sensor may be abnormal in time and amplitude, resulting in a large error in the positioning algorithm based on time difference or amplitude, and in some complex structure gear boxes, the multipath effect may make the positioning error reach 20%-30% of the actual distance, greatly affecting the accuracy and reliability of the acoustic emission positioning technology.

[0058] 2) Complex structure positioning error: There are not only many components in the fan gearbox, but also great differences in shape, size and material properties of each component. This complex structure further exacerbates the difficulty of acoustic emission positioning; related acoustic emission positioning algorithms are usually based on simple geometric models and propagation assumptions, which cannot accurately adapt to the complex acoustic wave propagation characteristics inside the gearbox; for example, in some narrow spaces or irregularly shaped areas inside the gearbox, the propagation of acoustic waves may be significantly affected by boundary conditions, resulting in a large deviation between the calculated acoustic emission source location and the actual location by traditional algorithms. In addition, the influence of different materials on the absorption, scattering and propagation speed of acoustic waves also makes the propagation law of acoustic emission signals more complex, increasing the difficulty of positioning.

[0059] 3) Unreasonable sensor layout: The layout of acoustic emission sensors has a key influence on positioning accuracy and fault monitoring effect; in actual application, the layout of acoustic emission sensors in many fan gearboxes often lacks scientific planning, either the sensor distribution is too sparse, which cannot fully cover the potential fault areas inside the gearbox, resulting in that the acoustic emission signals generated by some faults cannot be effectively detected; or the sensor layout is too concentrated, causing monitoring redundancy in some areas and insufficient monitoring in other areas. Unreasonable sensor layout not only wastes monitoring resources, but also reduces the accuracy and timeliness of fault warning; for example, in some large fan gearboxes, due to unreasonable sensor layout, early fault signals of some key components may be missed, delaying the fault handling opportunity.

[0060] 4) High false alarm rate: Related acoustic emission monitoring systems usually use a single trigger condition, such as only judging whether a fault occurs according to signal amplitude or frequency. However, the fan operating environment is complex, and there are various noise disturbances. These interference signals may trigger the monitoring system, leading to false alarms; for example, when the fan starts, stops or is subjected to external impact, it may produce some transient high-amplitude or specific frequency signals, but these signals are not generated by internal gearbox faults. According to actual operation data statistics, the false alarm rate of acoustic emission monitoring systems with single trigger condition can reach 30%-40%, which not only increases the workload of maintenance personnel, but also may lead to neglect of real faults, reducing the credibility of the warning system.

[0061] In summary, the fan gearbox fault warning method based on acoustic emission positioning technology has obvious deficiencies in multi-path interference suppression, complex structure positioning accuracy, sensor layout optimization and reducing false alarm rate, and needs to be improved.

[0062] In the application of fan gearbox acoustic emission positioning technology for crack warning, in order to overcome the problems of multipath interference, inaccurate signal processing and high false alarm rate, the embodiment of the present application provides a fan gearbox crack warning method based on acoustic emission positioning technology, which improves from three key aspects of multipath interference suppression, deep learning signal path separation based on physical constraints and dynamic threshold adjustment mechanism based on reinforcement learning, specifically including:

[0063] 1) Multipath interference suppression: the internal structure of the fan gearbox is complex, and the multipath interference seriously affects the acoustic emission positioning accuracy during the sound wave propagation; for this, an accurate sound wave propagation model is established:

[0064] a. Model construction method: use professional modeling tools to carry out detailed three-dimensional modeling work on the internal structure of the fan gearbox, accurately depict the shape, size and relative position relationship of various structural components such as gearboxes, bearing seats, gears and shafts, and construct a three-dimensional model that highly matches the actual situation; on this basis, with the help of advanced technical means such as finite element analysis, the sound wave propagation process in the gearbox is accurately simulated, which comprehensively covers the reflection, refraction phenomenon of sound wave and the interaction with different structural components.

[0065] b. Parameter assignment and significance: fully consider the acoustic characteristic differences of the materials of each component of the gearbox, and assign accurate sound wave propagation parameters such as sound speed and attenuation coefficient to each component in the model; different materials have different effects on sound wave propagation, and accurate parameter setting can more realistically reflect the propagation law of sound wave in the actual structure, through this way, the propagation path and characteristics of the acoustic emission signal under multipath effect can be more accurately predicted, providing a solid and reliable foundation for the subsequent positioning algorithm, effectively improving the accuracy and reliability of positioning.

[0066] 2) Deep learning signal path separation based on physical constraints: in related signal processing methods, the separation of multipath signals is limited to simple geometric models, and it is often difficult to achieve ideal results when processing reflection and refraction signals in complex structures such as fan gearboxes, therefore the embodiment of the present application innovatively proposes a deep learning signal path separation method based on physical constraints:

[0067] a. Introduce three-dimensional physical model: in the signal processing flow, introduce the three-dimensional physical model of the gearbox structure established by finite element simulation, which details the internal structure of the gearbox, completely presents the spatial position and geometric shape of each component and their mutual relationship, and provides accurate physical basis for signal propagation simulation.

[0068] b. Combined with the graph convolutional network (GCN): the graph convolutional network is used for path separation of the multi-path signal. In the specific implementation process, the virtual training data containing the sound wave propagation characteristics under different paths are generated by using the box three-dimensional model. These data cover various characteristics of the sound wave after reflection and refraction between different structural components. The GCN model is trained by using the data. In this way, the GCN model can learn the unique characteristics of the reflection / refraction signal, significantly enhance the recognition ability of the model to the reflection / refraction signal, and then realize more accurate separation of the multi-path signal, and improve the overall accuracy of the acoustic emission signal processing.

[0069] c. Innovation: the related algorithm mainly relies on a simple geometric model to process the signal. When facing the complex structure of the fan gear box, the sound wave propagation process cannot be comprehensively and accurately described, resulting in poor signal processing effect. The embodiment of the application innovatively combines physical simulation and data-driven, and uses the prior knowledge provided by the physical model to guide the training of the deep learning model, breaking the dependence of the related algorithm on the simple geometric model. This innovative method can more effectively solve the multi-path signal processing problem in the complex structure, significantly improve the accuracy and reliability of the signal processing, and provide a more advanced technical means for the acoustic emission signal processing of the fan gear box.

[0070] 3) Dynamic threshold adjustment mechanism based on reinforcement learning: the related fault triggering threshold is usually fixed, however, the running conditions of the fan are complex and changeable, and the acoustic emission signal characteristics change accordingly. The fixed threshold is easy to cause false alarms, affecting the accuracy of the early warning system. The embodiment of the application proposes a dynamic threshold adjustment mechanism based on reinforcement learning to solve this problem:

[0071] a. Design reinforcement learning agent: design a reinforcement learning agent based on deep Q network (DQN). The agent has strong learning and decision-making ability, and can dynamically adjust the fault triggering threshold according to historical false alarm data and current working condition information (such as wind speed, load, etc.).

[0072] b. Dynamic adjustment strategy: in the actual application scene, there is transient noise during the start-up stage of the fan. At this time, the acoustic emission signal characteristics are quite different from those during normal operation. Based on the perception of the current working condition, the reinforcement learning agent automatically increases the fault triggering threshold, effectively avoiding false alarms caused by transient noise. As the working condition of the fan continues to change, the agent continues to learn and adjust the threshold in real time, ensuring that the system always maintains the best fault detection performance.

[0073] The embodiment of the application reflects the following innovation: the fixed threshold mode cannot adapt to the complex and changeable operation conditions of the fan, resulting in a high false alarm rate, and the embodiment of the application realizes intelligent dynamic adjustment of the threshold by means of the reinforcement learning technology, so that the system can automatically optimize the fault detection strategy according to the actual working condition, the innovative mechanism effectively reduces the false alarm rate, greatly improves the accuracy and reliability of the early warning system, and provides more reliable protection for the early warning of the fan gearbox rupture.

[0074] The embodiment of the application provides a fan gearbox rupture early warning method based on acoustic emission positioning technology, and it should be noted that the execution subject of the fan gearbox rupture early warning method based on acoustic emission positioning technology provided by the embodiment of the application can be a fan gearbox rupture early warning device based on acoustic emission positioning technology, the fan gearbox rupture early warning device based on acoustic emission positioning technology can be realized by software, hardware or a combination of software and hardware to become part or all of an electronic device, and the electronic device can be a server or a terminal, wherein the server in the embodiment of the application can be a server, or a server cluster composed of multiple servers, and the terminal in the embodiment of the application can be a smart phone, a personal computer, a tablet computer, a wearable device, a smart robot and other smart hardware devices. In the following method embodiment, the execution subject is taken as an example to be an electronic device.

[0075] According to the embodiment of the application, a fan gearbox rupture early warning method based on acoustic emission positioning technology is provided, and it should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that here.

[0076] In the embodiment, a fan gearbox rupture early warning method based on acoustic emission positioning technology is provided, which can be used for the above-mentioned electronic device, Figure 1 The flowchart of the fan gearbox rupture early warning method based on acoustic emission positioning technology according to the embodiment of the application is shown in Figure 1 The flowchart includes the following steps:

[0077] In step S101, the internal structure parameters of the fan gearbox are acquired, and an acoustic wave propagation model is constructed based on the internal structure parameters of the fan gearbox.

[0078] In step S102, the acoustic wave propagation model is used to simulate acoustic wave propagation at multiple acoustic emission source positions, and acoustic wave propagation simulation data is obtained.

[0079] In step S103, real-time acoustic emission signals are acquired, and a graph convolution network model is used to separate the real-time acoustic emission signals to obtain classification probability data of the multipath signal paths.

[0080] Specifically, before using the graph convolutional network model to perform path separation on the acoustic emission real-time signal, the graph convolutional network model needs to be trained, and the steps include:

[0081] 1) Three-dimensional physical model establishment and data generation:

[0082] Finite element simulation modeling: using finite element analysis software such as ANSYS or ABAQUS, based on the three-dimensional model of the gearbox, an accurate finite element model is established, and reasonable meshing is performed on the finite element model to ensure that both the acoustic wave propagation can be accurately simulated and the computational load is not too large; according to the actual material parameters, the corresponding elastic modulus, density and other physical properties of each component in the model are assigned to simulate the propagation characteristics of acoustic waves in different materials.

[0083] Virtual training data generation: multiple virtual acoustic emission source positions are set in the finite element model to simulate acoustic emission events at different positions. Through finite element simulation, the signal characteristics of acoustic waves propagating inside the gearbox to each sensor position are calculated, including sound pressure, phase, propagation time, etc. Collect the above data to generate a virtual training dataset containing acoustic wave propagation characteristics under different paths; preprocess the data in the virtual training dataset, such as normalization, to ensure data consistency and usability.

[0084] 2) Graph Convolutional Network (GCN) model construction and training:

[0085] Model construction: based on a deep learning framework such as TensorFlow or PyTorch, a graph convolutional network model is constructed; define the nodes and edges of the graph, the nodes can represent the sensor positions or specific positions inside the gearbox, and the edges represent the acoustic wave propagation paths or the connection relationships between the nodes; set appropriate graph convolutional layers, pooling layers and fully connected layers to construct a network model that can effectively process graph-structured data.

[0086] Model training: input the generated virtual training dataset into the GCN model for training, set appropriate training parameters such as learning rate, iteration number, loss function, etc. During the training process, the model learns the acoustic wave propagation characteristics in the virtual training data and gradually masters the characteristics of reflected / refracted signals. Use the validation set to monitor the training process to prevent model overfitting, adjust the model parameters to optimize the model performance; the following loss function is used to measure the difference between the model prediction results and the true labels:

[0087]

[0088] wherein, represents the loss function value, which is used to measure the difference between the predicted result and the true label of the deep learning model based on physical constraints in the signal path separation task. By minimizing the loss function, the model parameters are optimized to make the model better separate the multipath signals; P is the number of samples in the virtual training data set, which is collected by simulating different fault scenarios and normal operation states for multiple times in the research of fan gearbox rupture early warning based on acoustic emission positioning technology, and is used to train the deep learning model. The number of samples affects the training effect and generalization ability of the model; y p is the true label of the sample p, y p ∈{0,1},0 represents a noise signal path, and 1 represents a true acoustic emission signal path, which is determined based on artificial labeling of acoustic emission signals or known fault simulation conditions, and is used to supervise model training, so that the model learns to distinguish between true signal paths and noise signal paths; σ(x) is a Sigmoid function that maps the model output to the [0, 1] interval. It is often used in neural networks to convert the results of linear transformations into probability form to determine the authenticity of the signal path, so that the model output is more consistent with the actual probability distribution, facilitating the classification task; K is the number of graph convolution layers in the GCN model, which is determined by comparing the influence of different layers on the performance of the model through multiple experiments. K determines the complexity and feature extraction ability of the model. An appropriate number of layers can achieve a good balance between computational complexity and model effectiveness, effectively extracting features related to acoustic emission signal paths; W k is the weight matrix of the k-th layer of graph convolution, which is constantly adjusted during model training, and is used to transform node features, learn the relationship and feature representation between different nodes, and optimize the model's ability to extract signal features and classification; W k (m,n) represents the element in the m-th row and n-th column of the matrix W k , which participates in the specific transformation calculation of node features. Its value is adjusted during the training process according to the feedback of the loss function to optimize the performance of the model. represents the mean value of element (m, n) in the weight matrix W k of the k-th layer during the training process, which is used for weight regularization to prevent model overfitting. By statistically calculating the updated weight matrix elements in each iteration, the mean value is obtained to keep the weight value within a reasonable range and avoid the model relying too much on certain features; ReLU(x) is a rectified linear unit function that introduces nonlinear features to enable the model to learn more complex signal feature relationships. When dealing with the combined features of different frequency components in multipath signals, the ReLU function can highlight useful features, suppress useless or negative features, and enhance the expression ability of the model; A pqis the connection weight between node p and node q in the graph structure adjacency matrix, reflecting the correlation between nodes, which is determined according to the spatial position relationship of the nodes in the gearbox structure and the possible path of sound wave propagation, for example, the connection weight between nodes with closer distance and unobstructed sound wave propagation is larger, helping the model to capture the relationship information between nodes and better process the graph structure data; q is the feature vector of node q, containing features related to sound wave propagation paths extracted from virtual training data, such as propagation time, amplitude, phase, etc., providing input features for the model, enabling the model to learn the feature representation of different path signals; and are the feature vectors of sample p based on simulation and actual measurement respectively, containing simulated sound wave propagation path related features, used for comparison with the feature vector obtained by actual measurement to evaluate the accuracy of model simulation, assisting model training, obtained by actual measurement, containing real sound wave propagation path related features as supervision information for model training, guiding the model to learn accurate signal features; λ1, λ2, λ3, λ4 are coefficients for balancing different terms, determined through multiple cross-validation experiments, used to balance the contributions of weight regularization, simulation and measured feature difference, correlation and energy difference in the loss function, so that the performance of the model in different aspects is reasonably optimized; is the Kullback-Leibler divergence, used to measure the difference between the simulated feature vector and the actual measurement feature vector By calculating the KL divergence, the similarity between the simulated features and the actual measurement features is evaluated, helping the model to adjust the parameters to reduce the difference; is the correlation coefficient, used to measure the linear correlation between the simulated feature vector and the actual measurement feature vector , and are the mean values of and respectively, used to evaluate the closeness of the linear relationship between the simulated features and the actual features, providing a reference for model optimization; is the energy of the simulated feature vector, is the energy of the actual measurement feature vector; ∈ is a very small positive number, such as ∈ = 1e-6, used to avoid the case where the denominator is zero, ensuring the mathematical reasonableness of the formula when calculating the energy difference ratio.

[0089] wherein, the calculation formula of the KL divergence is as follows:

[0090]

[0091] wherein I represents the dimension of the feature vector, and it is assumed that the feature vector dimension I = 10 in this application scenario, covering the amplitude, phase and other key features of different frequency bands.

[0092] correlation coefficient The calculation formula is:

[0093]

[0094] The calculation formula of the energy of the simulated and actually measured feature vectors is:

[0095]

[0096] wherein J represents the feature dimension used for calculating the energy, and it is assumed that J = 5, the key feature dimension related to the energy is selected for calculation, the accuracy of the model simulation is evaluated from the energy angle by calculating the energy of the simulated feature vector, and the model training is assisted; the energy of the actually measured feature vector corresponds to the energy of the simulated feature vector, and by comparing the energy difference between the two, the model can better learn the real signal features.

[0097] Further, the calculation of the loss function value aims to measure the difference between the model prediction and the actual situation by comprehensively considering multiple aspects to optimize the performance of the deep learning model based on physical constraints in the acoustic emission signal path separation task; first, the cross-entropy loss term (containing the Sigmoid function) is used to measure the accuracy of the model in classifying the signal path, i.e., determining whether the signal path is the real acoustic emission signal or noise; then, the weight regularization term (based on the difference between the weight matrix elements and the mean value) is used to prevent the model from overfitting, so that the model has better generalization ability; then, the KL divergence term and the correlation coefficient term are introduced to evaluate the consistency of the simulated features and the actual features from the distribution difference and linear correlation angles, respectively, to further optimize the model's learning of the features; finally, the energy difference term measures the difference between the simulated and actual features from the energy angle, comprehensively improving the model's fitting ability for signal features.

[0098] Further, The construction idea of the calculation formula is that the Sigmoid function has the characteristic of mapping any real number to the [0, 1] interval, which meets the value range of probability, and is commonly used in deep learning to convert the linear output of the neural network into a probability form, so as to classify the signal path and determine the probability of belonging to the real acoustic emission signal path or the noise signal path; converting the linear output of the model into a probability value that can be used for classification solves the problem of how to combine the neural network output with the actual signal path classification task, so that the model can effectively distinguish the real acoustic emission signal path and the noise signal path.

[0099] Further, the construction idea of the calculation formula of ReLU(x) = max(0, x) is: the function can effectively introduce nonlinear transformation by setting all negative inputs to 0 and retaining positive inputs. In processing complex nonlinear data such as acoustic emission signals, the ReLU function (Linear rectification function) can highlight useful features and suppress useless or negative features, so that the model can learn more complex feature relationships; solve the problem of how to learn more complex feature relationships when the neural network processes complex nonlinear data such as acoustic emission signals, improve the expression ability of the model, and make it better process the complex features of multipath signals.

[0100] Further, The construction idea of the calculation formula is: KL divergence is used to measure the difference between two probability distributions; the simulated feature vector and the actual measured feature vector are regarded as two distributions, and the difference between the two is measured by calculating the weighted sum of the logarithmic ratio of their corresponding elements, solving the problem of how to quantify the distribution difference between the simulated features and the actual measured features, helping the model understand the deviation of the simulated features from the real features, so as to adjust the parameters to make the simulated features closer to the real features, and improve the learning accuracy of the model on the features of the acoustic emission signal.

[0101] Further, The construction idea of the calculation formula is: the correlation coefficient measures the linear correlation between two variables (simulated feature vector and actual measured feature vector) by calculating the ratio of the covariance of the two variables to the product of their standard deviations. In the calculation process, the feature vectors are first centered (subtract the mean), then the product is calculated, and then normalized by dividing the standard deviation product; solve the problem of how to evaluate the closeness of the linear relationship between the simulated features and the actual measured features, so that the model can optimize the learning of the features of the acoustic emission signal from the perspective of linear correlation, improve the fitting ability of the model to the real signal features, and further improve the accuracy of multipath signal separation.

[0102] Further, in the research on fan gearbox rupture early warning based on acoustic emission positioning technology, multipath signals interfere with each other, making it difficult to accurately separate signals of different paths. The calculation of the loss function value helps the deep learning model to learn accurate signal path features by comprehensively measuring the loss from multiple aspects, improves the accuracy of multipath signal separation, and provides support for accurately identifying the propagation path of the acoustic emission source, which helps to more accurately judge whether the gearbox has a rupture or other faults.

[0103] Further, the above loss function value range is a non-negative real number, the smaller the value, the better the learning effect of the GCN model on different path signal features, and the more accurate the identification of real and noise signal paths. By comprehensively considering weight regularization, simulated and measured feature difference, correlation and energy difference and other aspects, the signal path separation effect based on physical constraint deep learning is comprehensively improved. The larger the value, the worse the learning effect of the model, and the lower the accuracy of signal path identification.

[0104] Further, in the training process, the model adjusts the parameters of the weight matrix W k by minimizing the loss function, so that the learning effect of the model on different path signal features is better and better; the training process is monitored by using the validation set to prevent the model from overfitting; if the value of the loss function on the validation set starts to rise, it means that the model may have overfitting, at which time the training parameters need to be adjusted or regularization methods are used to optimize the model performance.

[0105] Further, before actual application, a large amount of data is needed to train the deep learning model based on physical constraints. The specific steps include:

[0106] 1) Data preparation: Collect a large amount of acoustic emission signal data of the fan gearbox under different working conditions (normal, light fault, heavy fault), generate a training data set, determine the sample number P, and pre-process the data, including normalization and other operations, to ensure the consistency and availability of the data. At the same time, according to the actual situation, determine the dimension I of the feature vector, the feature dimension J used to calculate the energy and other parameters.

[0107] 2) Model parameter initialization: Set the number of graph convolution layers K, initialize the weight matrix W k , and determine the weight regularization coefficients λ1, λ2, λ3, λ4. These coefficients need to be determined through multiple cross-validation experiments to balance the contributions of different terms in the loss function.

[0108] 3) Training process: input the samples in the training data set into the model one by one, calculate the loss function value according to the current model parameters, adjust the parameters of the weight matrix W k according to the loss function value through the back propagation algorithm, so that the loss function value gradually decreases. In the training process, the loss function value on the validation set is monitored in real time. If the validation set loss starts to rise, it means that the model may have overfitting, and the training parameters (such as learning rate) need to be adjusted or regularization methods are used to optimize the model.

[0109] Step S104, based on the sound wave propagation simulation data and the classification probability data of the multi-path signal path, using the reinforcement learning agent model to detect the fault signal, and obtaining the fan gearbox rupture early warning information.

[0110] Specifically, the original acoustic emission signal is processed by a sound wave propagation model and a GCN model to separate the real signal path from the noise path, improve the signal purity and positioning accuracy; the preprocessed signal features are input into the reinforcement learning agent model, and the reinforcement learning agent model judges whether the signal belongs to the fault signal according to the current threshold value: if the signal features (such as amplitude, phase) exceed the dynamically adjusted threshold value and meet other state conditions (such as low false alarm rate, high signal path separation accuracy), the system triggers the rupture early warning; if the threshold value is not exceeded, it is considered as normal operation or noise interference, and no early warning is triggered.

[0111] The fan gearbox rupture early warning method based on acoustic emission positioning technology provided in the embodiment can construct a sound wave propagation model based on the internal structure parameters of the fan gearbox, and then simulate sound wave propagation at multiple acoustic emission source positions by using the sound wave propagation model to obtain sound wave propagation simulation data, so that the propagation path and characteristics of the acoustic emission signal under multipath effect can be more accurately predicted, the useful information and interference components in the signal can be more accurately identified, and the reliability of signal analysis can be improved; secondly, the graph convolution network model is used to separate the paths of the acoustic emission real-time signal, which provides purer and more accurate signals for acoustic emission signal processing and fault diagnosis, and helps to improve the accuracy of fault diagnosis; finally, based on the sound wave propagation simulation data and the classification probability data of the multipath signal path, the reinforcement learning agent model is used for fault signal detection, so that the system can automatically optimize the fault detection strategy according to the actual working condition, effectively reduces the false alarm rate, and greatly improves the accuracy and reliability of the early warning system, providing more reliable protection for fan gearbox rupture early warning.

[0112] In the embodiment, a fan gearbox rupture early warning method based on acoustic emission positioning technology is provided, which can be used for the electronic device described above, Figure 2 is a flowchart of the fan gearbox rupture early warning method based on acoustic emission positioning technology according to the embodiment of the application, as Figure 2 shown, the flowchart includes the following steps:

[0113] In step S201, the internal structure parameters of the fan gearbox are obtained, and a sound wave propagation model is constructed based on the internal structure parameters of the fan gearbox.

[0114] Specifically, in the complex environment of the fan gearbox, multipath interference seriously affects the accuracy of acoustic emission positioning. In order to effectively suppress multipath interference, the sound wave propagation process needs to be accurately simulated first, and then the positioning algorithm is optimized based on the simulation results.

[0115] The above step S201 includes:

[0116] In step S2011, the three-dimensional parameters of each component inside the fan gearbox are obtained, and a three-dimensional physical model of the fan gearbox is constructed based on the three-dimensional parameters of each component inside the fan gearbox.

[0117] Specifically, professional three-dimensional modeling software such as SolidWorks, ANSYS Design Modeler, etc. is used to accurately model the internal structure of the fan gearbox according to the design drawings and actual dimensions of the fan gearbox. During modeling, each structural component is described in detail, including but not limited to the shape, size, and relative position of gears, shafts, bearing seats, partitions, etc. to ensure that the model closely matches the actual gearbox structure.

[0118] Step S2012, obtaining the acoustic parameters of each component inside the fan gearbox, and assigning the acoustic parameters of each component inside the fan gearbox to the three-dimensional physical model of the fan gearbox.

[0119] Specifically, by consulting relevant material manuals, acoustic databases or through experimental testing, the acoustic parameters of the materials of each component of the gearbox are obtained, such as sound speed, density, attenuation coefficient, etc. For components of different materials, accurate parameters corresponding to them are assigned respectively; for example, for metal gears, the corresponding sound speed and attenuation coefficient are determined according to their specific alloy composition; for plastic or rubber seals, appropriate acoustic parameters are also set according to their material properties.

[0120] Step S2013, performing finite element analysis on the three-dimensional physical model of the fan gearbox after assigning the acoustic parameters of each component inside the fan gearbox, to obtain a sound wave propagation model.

[0121] Specifically, the built three-dimensional model is imported into a finite element analysis software such as ANSYS or COMSOL Multiphysics, and an acoustic analysis module is defined in the software, with appropriate boundary conditions set, such as assuming the gearbox shell as an acoustic hard boundary to simulate the reflection of sound waves at the boundary. At the same time, set the excitation source to simulate the sound waves generated by the sound emission source, and adjust the analysis parameters such as grid division accuracy, time step, etc. to ensure the accuracy and reliability of the simulation results.

[0122] Step S202, using the sound wave propagation model to simulate sound wave propagation at multiple sound emission source positions to obtain sound wave propagation simulation data.

[0123] Specifically, the finite element analysis software is run to simulate the propagation process of sound waves inside the gearbox. Analyze the simulation results to observe the reflection, refraction path of sound waves, and the sound pressure distribution, phase change, etc. at different positions. By simulating sound emission sources at different positions multiple times, comprehensive sound wave propagation data is obtained to provide detailed propagation model data support for subsequent positioning algorithms.

[0124] Step S203, obtaining the real-time acoustic emission signal, and using the graph convolution network model to separate the real-time acoustic emission signal to obtain classification probability data of the multipath signal path. For details, please refer to Figure 1Step S103 of the illustrated embodiment will not be described here again.

[0125] Step S204, based on the sound wave propagation simulation data and the classification probability data of the multipath signal path, using a reinforcement learning agent model for fault signal detection, obtaining the fan gear box cracking early warning information. For details, please refer to Figure 1 Step S104 of the illustrated embodiment will not be described here again.

[0126] The fan gear box cracking early warning method based on acoustic emission positioning technology provided in this embodiment can more accurately predict the propagation path and characteristics of the acoustic emission signal under multipath effect by performing detailed three-dimensional modeling on the complex structure inside the fan gear box, and assigning accurate sound wave propagation parameters to each component. The sound wave propagation process is accurately simulated using finite element analysis. Based on the sound wave propagation model, the subsequent positioning algorithm can obtain more reliable data basis, thereby effectively reducing the positioning error caused by multipath interference and greatly improving the positioning accuracy of the acoustic emission source.

[0127] In this embodiment, a fan gear box cracking early warning method based on acoustic emission positioning technology is provided, which can be used in the electronic device described above, Figure 3 is a flowchart of a fan gear box cracking early warning method based on acoustic emission positioning technology according to an embodiment of the present application, as Figure 3 The flowchart includes the following steps:

[0128] Step S301, obtain the internal structure parameters of the fan gear box, and construct a sound wave propagation model based on the internal structure parameters of the fan gear box. For details, please refer to Figure 2 Step S201 of the illustrated embodiment will not be described here again.

[0129] Step S302, use the sound wave propagation model to simulate sound wave propagation at multiple acoustic emission source positions, and obtain sound wave propagation simulation data. For details, please refer to Figure 2 Step S202 of the illustrated embodiment will not be described here again.

[0130] Step S303, obtain the acoustic emission real-time signal, and use a graph convolution network model to separate the acoustic emission real-time signal, and obtain classification probability data of the multipath signal path.

[0131] Specifically, the above step S303 includes:

[0132] Step S3031, signal preprocessing is performed on the acoustic emission real-time signal, and feature extraction is performed on the acoustic emission real-time signal after signal preprocessing, to obtain signal propagation features.

[0133] Specifically, a plurality of sensors are reasonably arranged around the fan gearbox to ensure that acoustic emission signals from different directions and positions can be collected. When an acoustic emission event occurs, each sensor synchronously collects signals, and records the initial data such as the time and intensity of signals received by each sensor. The collected original signals are preprocessed, including removing noise interference, for example, using filtering algorithms such as low-pass filters to remove high-frequency noise and high-pass filters to remove low-frequency noise to improve the quality of the signals. At the same time, the signals are normalized to the same scale range to facilitate subsequent analysis.

[0134] Further, features related to the propagation path are extracted from the preprocessed signals. For example, according to the time difference of signal arrival, the time of signals from different paths arriving at different sensors will be different. By accurately measuring and calculating this time difference, the propagation direction of the signal can be initially inferred. The phase change of the signal is analyzed. Since the sound wave may experience different reflections and refractions during propagation in different paths, the phase changes. Extracting phase features helps to distinguish different paths. The frequency component features of the signal can also be extracted. Different paths may have different effects on the frequency of the signal, and certain frequency components may be enhanced or weakened on a specific path.

[0135] Step S3032, based on the acoustic emission real-time signal after signal preprocessing, the classification probability data of the multipath signal path is determined by using the graph convolution network model.

[0136] Specifically, after the GCN model is trained, the model is applied to separate the signal path during the real-time operation of the fan gearbox. The specific steps include:

[0137] 1) Real-time signal preprocessing: real-time acoustic emission signals are collected, and filtering, denoising and other preprocessing operations are performed to remove obvious noise interference, so that the signals input into the model are more pure.

[0138] 2) Signal path separation and analysis: the preprocessed signals are input into the trained model, the model separates the multipath signals according to the learned features, and determines the possible propagation path of the acoustic emission source by analyzing the signal components of different paths output by the model and combining the information provided by the accurate propagation model; for example, if the energy, phase and other features of a certain path signal output by the model match the known fault signal features, and the path is consistent with the possible propagation path of the fault in the accurate propagation model, it can be further determined that the acoustic emission signal corresponding to the path may be related to the fault.

[0139] Further, the GCN model separates the input multipath signals according to the learned reflection / refraction signal characteristics, and outputs the signal components of different paths. By analyzing these signal components, the possible propagation paths of the acoustic emission source can be determined. Combined with the information provided by the sound wave propagation model, the judgment of the signal path is further optimized, and the accuracy of the multipath signal separation is improved.

[0140] Further, using the established three-dimensional physical model, the propagation of sound waves in different possible paths is simulated according to the geometric shape, component position and material acoustic characteristics of the internal structure of the gearbox. The simulated signal characteristics, such as propagation time, phase change, frequency characteristics, etc., are compared with the actual collected and extracted signal characteristics. Based on the comparison results, a matching algorithm such as correlation analysis is used to find the simulated propagation path that best matches the actual signal characteristics. For each signal received by a sensor, such matching analysis is performed, and then the results of multiple sensors are integrated to determine the possible propagation path of the acoustic emission source. If the matching results of multiple sensors all point to some similar paths, then these paths are more likely to be the real propagation paths of the acoustic emission source. After determining the possible propagation paths, the likelihood of these paths can be further sorted according to signal intensity and other information, providing a more accurate basis for subsequent positioning algorithms.

[0141] Further, the final output layer of the GCN model is activated by a Sigmoid function to map the linear transformation results of the model to the [0, 1] interval, outputting the probability value of each sample (signal path) belonging to the real acoustic emission signal path. If the output value is close to 1, it means that the model considers the path to be a real signal path with high probability; if it is close to 0, it is determined to be a noise or interference path.

[0142] Further, in the acoustic emission signals of the gearbox, the signals generated by real faults usually have specific propagation paths (such as direct propagation through the gear meshing surface), and their characteristics match well with the simulated regular reflection / refraction patterns in the virtual training data, with the model output probability close to 1. The propagation paths of noise signals (such as environmental vibrations and electrical interference) are irregular, and their characteristics match the noise patterns in the training data, with the model output probability close to 0. Through the above methods, the GCN model converts the propagation rules simulated by the physical model (such as finite element analysis results) into calculable probability judgments, providing reliable signal preprocessing results for subsequent acoustic emission source positioning and fault warning.

[0143] Further, after the deep learning method based on physical constraints separates the multipath signals through the graph convolution network (GCN), the following key information is output: signal path classification result: the probability of distinguishing real acoustic emission signal paths from noise paths (output by the Sigmoid function); separation accuracy (A sep，t) : Reflects the separation effect of the model on the multipath signal (such as the proportion of correctly classified paths).

[0144] Step S304, based on the sound wave propagation simulation data and the classification probability data of the multipath signal path, a reinforcement learning agent model is used for fault signal detection to obtain the wind turbine gearbox rupture early warning information. For details, please refer to Figure 2 Step S204 of the embodiment shown, which will not be repeated here.

[0145] The wind turbine gearbox rupture early warning method based on acoustic emission positioning technology provided in this embodiment uses a graph convolution network model to separate the acoustic emission real-time signal path, reduces the noise pollution to the features (such as amplitude and phase), and the features of the separated real signal path are closer to the actual fault signal, making the "fault-noise" boundary clearer when adjusting the threshold.

[0146] In this embodiment, a wind turbine gearbox rupture early warning method based on acoustic emission positioning technology is provided, which can be used in the electronic device described above, Figure 4 is a flowchart of a wind turbine gearbox rupture early warning method based on acoustic emission positioning technology according to an embodiment of the present application, as shown in Figure 4 The flowchart includes the following steps:

[0147] Step S401, obtain the internal structure parameters of the wind turbine gearbox, and construct a sound wave propagation model based on the internal structure parameters of the wind turbine gearbox. For details, please refer to Figure 3 Step S301 of the embodiment shown, which will not be repeated here.

[0148] Step S402, use the sound wave propagation model to simulate sound wave propagation at multiple acoustic emission source positions to obtain sound wave propagation simulation data. For details, please refer to Figure 3 Step S302 of the embodiment shown, which will not be repeated here.

[0149] Step S403, obtain the acoustic emission real-time signal, and use a graph convolution network model to separate the acoustic emission real-time signal path to obtain classification probability data of the multipath signal path. For details, please refer to Figure 3 Step S303 of the embodiment shown, which will not be repeated here.

[0150] Step S404, based on the sound wave propagation simulation data and the classification probability data of the multipath signal path, a reinforcement learning agent model is used for fault signal detection to obtain the wind turbine gearbox rupture early warning information.

[0151] Specifically, the above step S404 includes:

[0152] Step S4041, position optimization is performed on the sound wave propagation simulation data to determine the multipath interference degree index.

[0153] In some optional embodiments, the step S4041 comprises:

[0154] Step a1, based on the sound wave propagation simulation data, the position of the sound emission source vector is determined by using the positioning algorithm.

[0155] Specifically, after obtaining the sound wave propagation simulation data, the following formula is used to optimize the positioning accuracy:

[0156]

[0157] Wherein, is the estimated position vector of the sound emission source That is, the estimated position of the sound emission source in three-dimensional space, which is used to determine the position of the possible fault such as crack occurring inside the gearbox; r is the variable vector of the sound emission source position (r=(x, y, z)), which is used to find the variable of the sound emission source position to minimize the objective function, and the optimal sound emission source position estimate is searched by changing its value; N is the number of sensors, representing the number of sensors actually arranged around the fan gearbox for receiving the acoustic emission signal, which is used to obtain the time, amplitude and other information of the arrival of the acoustic emission signal, providing data support for positioning the sound emission source; is the time of the sound wave reaching the i-th sensor based on the sound wave propagation model simulation; is the actually measured time of the sound wave reaching the i-th sensor, which is obtained by actually collecting the sensor, and is used to compare with the simulation time to evaluate the accuracy of the simulation and determine the position of the sound emission source; is the standard deviation of the time measurement of the i-th sensor, reflecting the uncertainty of the sensor measuring the arrival time of the sound wave, which is used to measure the fluctuation degree of the time measurement value, and is used for normalization in the formula to make the time error of different sensors comparable; λ is the coefficient for balancing the influence of time and amplitude, which can change the relative importance of time difference and amplitude difference in the objective function by adjusting the coefficient, to adapt to different measurement environments and requirements, and optimize the estimation of the position of the sound emission source; M is the number of signal amplitude sampling points, which is used to quantify the number of sampling points of the acoustic emission signal amplitude, and through the analysis of the amplitude of multiple sampling points, the amplitude characteristics of the acoustic emission signal are more comprehensively described; is the amplitude of the acoustic emission signal at the j-th sampling point based on the sound wave propagation model simulation; is the actually measured amplitude of the acoustic emission signal at the j-th sampling point, which is obtained by actually measuring, and is used to compare with the simulation amplitude to assist in determining the position of the sound emission source; The standard deviation of the amplitude measurement of the jth sampling point, reflecting the uncertainty of the amplitude measurement of the sampling point, is used for normalizing the amplitude difference and enhancing the comparability of the amplitude error of different sampling points; μ is a coefficient for balancing the phase influence, used for balancing the contribution of the phase difference in the objective function, so that the time, amplitude and phase information can be considered comprehensively when positioning the acoustic emission source, and the positioning precision is improved; L is the number of phase sampling points of the signal, used for quantifying the number of phase sampling points of the acoustic emission signal, and the phase characteristics of the acoustic emission signal are more comprehensively described through the analysis of multiple phase sampling points; The phase of the acoustic emission signal at the lth sampling point simulated based on the sound wave propagation model; The phase of the acoustic emission signal at the lth sampling point actually measured, obtained through actual measurement, used for comparison with the simulated phase to assist in positioning the acoustic emission source; The standard deviation of the phase measurement of the lth sampling point, reflecting the uncertainty of the phase measurement, used for normalizing the phase difference to make the phase error of different sampling points comparable.

[0158] Wherein, The calculation formula of is used to accurately simulate the time required for the sound wave to propagate in the complex structure of the fan gear box to each sensor, considering the influence of different material regions on the sound speed, so as to more accurately reflect the actual propagation situation. The time of the sound wave reaching the ith sensor simulated based on the sound wave propagation model The calculation formula of is as follows:

[0159]

[0160] Wherein, s i (r) is the propagation path from the acoustic emission source r to the ith sensor, and v(s) is the sound speed on the path s, which reflects the influence of material characteristics on the sound speed by assigning corresponding sound speed values to different material regions.

[0161] Further, The calculation process of considers the influence of factors such as attenuation, distance and frequency on the amplitude in the signal propagation process, accurately simulates the amplitude variation, and the amplitude of the acoustic emission signal at the jth sampling point simulated based on the sound wave propagation model The calculation formula of is as follows:

[0162]

[0163] Wherein, A0 is the initial amplitude of the acoustic emission signal, α is the attenuation coefficient, l j (r) is the propagation distance from the acoustic emission source r to the jth sampling point, ω is the angular frequency of the acoustic emission signal, Δt j (r) is the propagation time difference from the acoustic emission source r to the jth sampling point.

[0164] Further, For accurate simulation of the phase change of the acoustic emission signal during propagation, the phase of the acoustic emission signal at the lth sampling point simulated based on the acoustic wave propagation model The calculation formula is as follows:

[0165]

[0166] Wherein, is the propagation time from the acoustic emission source r to the lth sampling point based on the acoustic wave propagation model simulation, is the initial phase.

[0167] Further, the core idea of the calculation formula of the acoustic emission source position vector is to determine the most likely acoustic emission source position by minimizing the difference between the simulated time, amplitude and phase and the actual measured values Specifically, the first term in the formula measures the difference between the simulated time and the actual measured time, the second term measures the difference between the simulated amplitude and the actual measured amplitude, and the third term measures the difference between the simulated phase and the actual measured phase. The time, signal amplitude and phase information of the acoustic wave propagation to the sensor are considered, the propagation of the above parameters in the complex structure inside the gearbox is simulated using the acoustic wave propagation model, and compared with the actual measured values. By adjusting the weight coefficients λ and μ of the time, amplitude and phase difference terms, the importance of the three differences in positioning can be flexibly balanced according to the actual situation; for example, in some cases, time measurement may be more accurate, at which time λ can be appropriately increased to make the time difference play a more critical role in positioning.

[0168] Further, in the research of fan gearbox rupture early warning based on acoustic emission positioning technology, multi-path interference will cause the acoustic emission signal propagation path to be complex, making it difficult for the positioning method to accurately determine the acoustic emission source position. The above formula (6) effectively suppresses the influence of multi-path interference on positioning by accurately simulating acoustic wave propagation and comparing multiple parameters, improves the acoustic emission source positioning accuracy, and thus more accurately determines the position of the possible rupture inside the gearbox, providing a basis for timely warning and maintenance.

[0169] Further, The construction idea of the calculation formula is to consider that the sound speed will change when the acoustic wave propagates in the complex structure inside the gearbox and passes through different material regions. The function accurately calculates the acoustic wave propagation time by integrating the propagation path s i (r) from the acoustic emission source r to the ith sensor, combining the sound speed v(s) at each point on the path, to simulate the actual propagation situation; accurately simulates the propagation time of the acoustic wave in the complex structure, solves the problem of difficult accurate determination of the propagation time under multi-path interference, and provides accurate time parameters for acoustic emission source positioning.

[0170] Further, The construction idea of the calculation formula is that the amplitude of the acoustic emission signal is affected by the propagation distance, attenuation and frequency in the propagation process, the function considers the propagation distance l j (r) caused attenuation (represented by A(r) = A0e ), and the frequency ω and the propagation time difference Δt j (r) modulation of the amplitude (represented by φ(r) = φ0+ ωrΔt ), accurately simulates the amplitude change of the signal, solves the problem that the signal amplitude is difficult to accurately simulate under the multi-path interference, and provides accurate amplitude parameters for acoustic emission source positioning.

[0171] Further, The construction idea of the calculation formula is that the phase of the acoustic emission signal is related to the propagation time and the initial phase, the function considers the acoustic emission signal angular frequency ω multiplied by the propagation time , plus the initial phase φ0 Accurately simulates the phase change of the signal, accurately simulates the phase change of the acoustic emission signal in the propagation process, solves the problem that the signal phase is difficult to accurately simulate under the multi-path interference, and provides accurate phase parameters for acoustic emission source positioning.

[0172] Step a2, based on the acoustic emission source position vector, the propagation parameter simulation is carried out to obtain the simulation basic parameters of the sound wave propagation in the gearbox.

[0173] Specifically, the obtained acoustic emission source position vector is the basic parameter of simulating the sound wave propagation in the gearbox, and the simulation basic parameters of the sound wave propagation in the gearbox include: the simulation propagation time Based on and the gearbox structure model, the theoretical time of the sound wave reaching the i-th sensor is calculated; the simulation signal amplitude Based on the amplitude attenuation after the signal propagation; the simulation signal phase Based on the phase change after the signal propagation.

[0174] Step a3, the actual measurement basic parameters of the sound wave propagation in the gearbox are obtained, the simulation basic parameters of the sound wave propagation in the gearbox are compared with the actual measurement basic parameters of the sound wave propagation in the gearbox, and the multi-path interference degree index is obtained.

[0175] Specifically, the above simulation basic parameters are compared with the actual measurement value , the multi-path interference degree index can be obtained, and the calculation formula of the multi-path interference degree index is:

[0176]

[0177] Error mp，t Error is a comprehensive deviation index of the simulation value and the measured value in multipath interference suppression, used to evaluate the influence degree of multipath interference on the propagation simulation of acoustic emission signals; respectively, the time of the acoustic wave reaching the i-th sensor based on simulation and actual measurement at t time; Error is the standard deviation of the time measurement of the i-th sensor at t time; respectively, the amplitude of the acoustic emission signal at the j-th sampling point based on simulation and actual measurement at t time; Error is the standard deviation of the amplitude measurement of the j-th sampling point at t time; respectively, the phase of the acoustic emission signal at the l-th sampling point based on simulation and actual measurement at t time; Error is the standard deviation of the phase measurement of the l-th sampling point at t time; N, M, L are respectively the number of sensors, the number of amplitude sampling points, and the number of phase sampling points, Error mp，t The smaller Error is, the better the multipath interference suppression effect is, and the greater the positive contribution to the reward value is.

[0178] Further, considering the differences between the simulation value and the measured value of the time, the signal amplitude, and the phase of the acoustic wave propagating to the sensor under multipath interference, the deviations of the time, the amplitude, and the phase are calculated respectively, and a comprehensive deviation index is obtained by averaging according to the number of sensors and the number of sampling points, which comprehensively measures the influence degree of multipath interference on the propagation simulation of acoustic emission signals; accurately evaluates the influence of multipath interference on the propagation simulation of acoustic emission signals, and provides a quantitative index reflecting the multipath interference suppression effect for the reward function, helping the reinforcement learning agent to understand the influence of the current multipath interference on fault detection, so as to make more reasonable threshold adjustment decisions.

[0179] Further, the multipath interference index is used as the state input of the reinforcement learning agent model to represent the reliability of the current acoustic emission signal propagation, for example, if Error mp，t The larger the value is, the more serious the multipath interference is, and the lower the positioning result reliability is, so the fault detection threshold needs to be improved to avoid false positives; if Error mp，t The smaller the value is, the higher the positioning accuracy is, and the more reliable the signal characteristics are, so the threshold can be appropriately reduced to improve the fault detection sensitivity; the multipath interference index is embedded in the reward function to guide the threshold optimization, and then drive the dynamic threshold adjustment decision.

[0180] Step S4042, determining the signal separation accuracy based on the classification probability data of the multipath signal path.

[0181] Step S4043, determining the threshold adjustment strategy under multiple fan operating conditions by using the reinforcement learning agent model.

[0182] Specifically, the wind turbine operating conditions (such as wind speed, load, temperature), the multi-path interference degree (Error_{mp,t}), the signal path separation accuracy (A_{sep,t}) and other state information are analyzed in real time by a deep Q network (DQN) agent to dynamically adjust the fault triggering threshold. The threshold is a key parameter for determining whether the acoustic emission signal is a fault signal. When the acoustic emission signal characteristics (such as amplitude and frequency) collected by the sensor exceed the threshold, the system determines that there may be a gearbox rupture or other fault, and then triggers a warning.

[0183] In some optional embodiments, the above step S4043 comprises:

[0184] Step b1, defining state, action and reward function; wherein the state is the historical false alarm data corresponding to multiple wind turbine operating conditions, wind turbine operating parameters, multi-path interference degree indicators and signal path separation accuracy; the action is the adjustment operation of the fault triggering threshold.

[0185] Specifically, the state space of the DQN (Deep Q Network, deep reinforcement learning) agent model is determined, and the state includes historical false alarm data and current operating condition information. The historical false alarm data includes the number of false alarms, false alarm time, signal characteristics at the time of false alarm and other information in the past period of time. The current operating condition information includes wind speed, load, temperature and other wind turbine operating parameters, as well as multi-path interference degree indicators and signal path separation accuracy. The above information is encoded to form a state vector that can be understood by the DQN agent model.

[0186] Further, the action space of the DQN agent model is defined, and the action represents the adjustment operation of the fault triggering threshold. For example, a series of discrete threshold adjustment steps are set, such as increasing or decreasing 5dB, 10dB, etc. The DQN agent model adjusts the fault triggering threshold by selecting different actions.

[0187] Further, the reward function takes the number of real fault detections as positive reward and the number of false alarms as negative reward, guiding the reinforcement learning agent model to optimize the threshold to balance the detection sensitivity and reliability.

[0188] Further, the reward function is used to guide the DQN agent model to learn the optimal threshold adjustment strategy. The design principle of the reward function is: when the threshold adjustment operation made by the agent can reduce false alarms and not miss real faults, a positive reward is given; when there are false alarms or missed alarms, a negative reward is given. For example, if in a certain state, the number of false alarms is reduced and all real faults are successfully detected after the agent adjusts the threshold, a larger positive reward is given; if there are still false alarms or missed alarms, a corresponding negative penalty is given. Wherein, the reward value r t The reward value r is calculated according to the following formula:

[0189]

[0190] wherein, r t is the reward value obtained after taking action a t in state s t , which is determined by comprehensively considering multiple factors to guide the reinforcement learning agent to learn the optimal threshold adjustment strategy, and the higher the reward value, the more beneficial the action is to optimize the fault detection performance; is the number of real faults detected at time t, which is obtained by comparing with the actual fault records and is an important indicator to measure the accuracy of fault detection, and the more real faults detected, the greater the positive contribution to the reward value; is the number of false alarms at time t, which is determined by comparing with the actual situation, and the fewer the number of false alarms, the greater the positive contribution to the reward value, and reducing the number of false alarms is one of the important goals of optimizing the fault warning system; β1 and β2 are weight coefficients, respectively used to measure the influence of real fault detection and false alarm on the reward value; Error mp,t is the comprehensive deviation index of the simulated value and the measured value in multi-path interference suppression (i.e. the multi-path interference degree index); β3 represents the contribution weight of multi-path interference suppression effect to the reward value; A sep,t is the deep learning signal path separation accuracy based on physical constraints, which is calculated by comparing the classification results of the processed signal path at the current time with the real label, and the proportion of correct classification is taken as the accuracy, the higher the accuracy, the better the signal path separation effect, which has a positive contribution to the reward value and helps to improve the accuracy of fault detection; β4 is the influence of signal path separation effect on the reward value; Noise int，t is the interference noise index in the engine room; β5 is the effect of interference noise on the reward value; State f，t and State gb，t are the running state indicators of the fan and the gearbox, respectively; β6 and β7 represent the weights of the fan and the gearbox running state on the reward value.

[0191] wherein, the calculation formula of the interference noise index Noise int，t in the engine room is:

[0192]

[0193] In the formula, N(t, f) is the interference noise signal at time t and frequency f, S(t, f) is the acoustic emission signal at time t and frequency f, f1 and f2 are the frequency range of interest, assuming f1 = 100 Hz, f2 = 1000 Hz, the values of f1 and f2 are determined according to the main frequency range of the acoustic emission signal of the fan gearbox; by calculating the proportion of interference noise energy and total signal energy in this frequency range, the influence of the interference noise in the cabin on the acoustic emission signal is evaluated, the smaller the index value, the greater the contribution to the reward value, which means that the interference noise has less influence on the acoustic emission signal, which helps to improve the accuracy of fault detection.

[0194] The cabin interference noise index Noise int，t In the calculation process, the influence of the interference noise in the cabin on the acoustic emission signal is quantified by calculating the proportion of interference noise energy and total signal energy in a specific frequency range, f1 and f2 corresponding to the main frequency range of the acoustic emission signal of the fan gearbox are selected, so that this index can reflect the actual influence of the interference noise on the fault detection; the interference degree of the interference noise in the cabin on the acoustic emission signal is accurately evaluated, a quantitative index reflecting the influence of the interference noise is provided for the reward function, which helps the reinforcement learning agent to consider the interference noise factor when adjusting the threshold, and improves the accuracy of fault detection in complex noise environment.

[0195] where State f，t and State gb，t The calculation formula is:

[0196]

[0197] In the formula, State f，t is the fan operating state index; F i (t) is the i-th characteristic parameter of the fan operating state at time t, such as wind speed, rotating speed, etc. is the normal mean value of the characteristic parameter; I f is the number of fan operating state characteristic parameters; assuming I f = 3, corresponding to wind speed, rotating speed and power respectively; by calculating the relative deviation of the current characteristic parameter from the normal mean value, the operating state of the fan is evaluated, the closer the operating state to the normal, the greater the contribution to the reward value, which helps to reasonably adjust the fault triggering threshold under different operating states of the fan; State gb，t is the gearbox operating state index; GB j (t) is the j-th characteristic parameter of the gearbox operating state at time t, such as the wear degree of the gear, oil temperature, etc. is the normal mean value of the characteristic parameter; I gb is the number of gearbox operating state characteristic parameters; assuming I gb= 4, respectively corresponding to the gear wear, oil temperature, oil quality index and vibration amplitude, by calculating the relative deviation of the current characteristic parameter and the normal mean, the running state of the gearbox is measured, the more stable the running state, the greater the positive influence on the reward value, which helps to reasonably adjust the fault triggering threshold in different running states of the gearbox.

[0198] State f，t In the calculation process, the key characteristic parameters of the fan running state (such as wind speed, rotating speed, power) are selected, the running state of the fan is evaluated by calculating the relative deviation of the current characteristic parameter and the normal mean, which can intuitively reflect the deviation degree of the fan running state from the normal state, and provide the quantitative information of the fan running state for the reinforcement learning agent, that is, quantize the fan running state, so that the reinforcement learning agent can adjust the fault triggering threshold according to the different running states of the fan, and solve the problem of how to reasonably adjust the threshold to ensure the accuracy of fault detection when the running state of the fan changes.

[0199] State gb，t In the calculation process, the key characteristic parameters of the gearbox running state (such as gear wear, oil temperature, oil quality index, vibration amplitude) are selected, the running state of the gearbox is measured by calculating the relative deviation of the current characteristic parameter and the normal mean, and the quantitative information of the gearbox running state is provided for the reinforcement learning agent, that is, the gearbox running state is quantized, so that the reinforcement learning agent can adjust the fault triggering threshold according to the different running states of the gearbox, and solve the problem of how to reasonably adjust the threshold to ensure the accuracy of fault detection when the running state of the gearbox changes.

[0200] Further, by defining the above reward function, the threshold adjustment operation made by the agent model can reduce false positives, accurately detect real faults, improve multi-path interference suppression effect, improve signal path separation accuracy, reduce interference noise influence in the cabin, and adapt to the running state of the fan and the gearbox, so as to obtain a higher reward value, thereby guiding the agent to learn the optimal threshold adjustment strategy.

[0201] Further, considering the key aspects of fan and gearbox fault early warning, real fault detection, false alarm control, multi-path interference suppression, signal path separation, interference noise influence and the running state of the fan and the gearbox are quantified as components of the reward value, and the importance of each factor is balanced through weight coefficients β1-β7, so that the reward function can comprehensively and reasonably reflect the influence of different actions on the performance of the fault early warning system, and guide the reinforcement learning agent to learn the optimal threshold adjustment strategy; solve the problem of how to comprehensively evaluate the influence of different threshold adjustment actions on the performance of the fault early warning system under complex and variable working conditions, and provide a clear learning direction for the reinforcement learning agent, so that it can dynamically adjust the threshold according to different working conditions and improve the accuracy of fault detection.

[0202] Step b2, obtaining the current state, selecting the current action according to the current state, adjusting the fault triggering threshold based on the current action, and updating the current state to obtain the updated state.

[0203] Step b3, calculating the current reward value based on the updated state.

[0204] Specifically, during the operation of the fan, the DQN agent model continuously observes the current state, selects an action (adjusts the threshold), executes the action, and then observes the new state and the reward obtained. The state, action, reward and new state of each step are stored in the experience replay buffer. When the data stored in the buffer reaches a certain amount, a batch of data is randomly sampled from the buffer for training.

[0205] Step b4, calculating the current Q value based on the current state, the current action, the updated state and the current reward value.

[0206] Step b5, comparing the current Q value with the target Q value, and updating the threshold adjustment strategy based on the comparison result until the current Q value and the target Q value, obtaining the threshold adjustment strategy under multiple fan operating conditions.

[0207] Specifically, the error between the target Q value and the current Q value is calculated using the sampled data. The neural network parameters of the DQN agent model are updated through the back propagation algorithm, so that the current Q value gradually approaches the target Q value. This process is repeated continuously, and as the training progresses, the DQN agent model gradually learns the optimal threshold adjustment strategy.

[0208] Further, the Q value update formula is:

[0209]

[0210] Wherein, Q(s t , a t ) is the Q value of taking action a t in state s t In the research of fan gearbox rupture early warning based on acoustic emission positioning technology, it comprehensively reflects the expectation of long-term cumulative reward by taking a specific action in the current state, and is an important indicator for reinforcement learning agent to evaluate the value of action; s t is the state of deep Q network at time t, which includes historical false alarm data, current operating condition information, multi-path interference degree index Error mp,t , signal path separation accuracy A sep,t , cabin internal interference noise index Noise int,t , fan operating state index State f,t , gear box operating state index State gb，tand a comprehensive reflection of the various relevant information of the system at time t, providing the basis for the agent to make action decisions; t For the action to be taken in state s t , i.e. the adjustment operation of the fault triggering threshold, assuming the action space is a t ∈{Δθ1, Δθ2, …, Δθ5}, corresponding to an increase of 5%, an increase of 10%, a decrease of 5%, a decrease of 10%, and no change, respectively, the fault triggering threshold is dynamically adjusted by selecting different actions; α is the learning rate, controlling the step size of each update, through experimental testing, setting α = 0.1, affecting the speed and stability of model learning, a suitable learning rate can make the model effectively adjust the Q value in the training process and gradually learn the optimal strategy; γ is the discount factor, determining the importance of future rewards, taking a value between 0 and 1, so that the reinforcement learning agent considers not only the current reward but also the possible rewards in the future when making decisions, a larger γ value indicates that more emphasis is placed on future rewards, and the agent will tend to choose actions that bring long-term high returns, a smaller γ value makes the agent more concerned about the current reward. In this study, a suitable γ value helps the agent find the optimal threshold adjustment strategy under different working conditions to maximize the long-term cumulative reward; For the maximum Q value among all possible actions in state s t+1 .

[0211] Where, the Q value update is based on the idea of time difference learning, i.e. Q(s t , a t ) represents the value estimate of taking action a t in state s t , r t is the reward obtained immediately after executing action a t , reflecting the influence of the action on the system performance at the current time, is the estimate of the future reward obtained by taking the optimal action from the next state s t+1 , multiplied by the discount factor γ to reflect the discounting of future rewards, by calculating the difference between the current estimate and the target value containing the current reward and the estimate of the future optimal reward (i.e. ), and multiplying it by the learning rate α to update Q(s t , a tAs time goes by and experience accumulates, the agent can continuously adjust its value assessment of each state-action pair, gradually approaching the true value under the optimal strategy. In the dynamic threshold adjustment mechanism based on reinforcement learning, in order for the agent to learn the optimal threshold adjustment strategy under different operating conditions, an effective method is needed to update the value assessment of each state-action pair. The Q-value update process provides an iterative update mechanism, which continuously uses the current reward and the expectation of future rewards to correct the current value estimate, enabling the agent to learn the optimal threshold adjustment action sequence in complex environments (i.e., various operating conditions of wind turbine gearboxes). This solves the problem of how to enable the reinforcement learning agent model to learn an effective strategy in a dynamic environment to optimize fault detection performance.

[0212] Furthermore, by continuously repeating the process of experience replay and network update, as training progresses, the DQN agent model gradually learns the optimal threshold adjustment strategy, enabling it to dynamically adjust the fault trigger threshold according to different operating conditions in order to maximize long-term cumulative rewards.

[0213] Furthermore, through iterative learning, the reinforcement learning agent model gradually masters the optimal threshold strategy under different operating conditions, ensuring that early warnings are triggered in a timely manner when real fault signals are detected, while suppressing false alarms caused by noise.

[0214] Step S4044: Obtain the current wind turbine operating parameters. Based on the current wind turbine operating parameters, multipath interference level index and signal path separation accuracy, determine the fault trigger threshold using threshold adjustment strategies under various wind turbine operating conditions.

[0215] Specifically, the operating conditions of the wind turbine are monitored in real time, including changes in parameters such as wind speed, load, and temperature. The current operating condition information is fed back to the DQN agent model in a timely manner as a basis for the agent to adjust the threshold.

[0216] Furthermore, the DQN agent model selects an action based on the learned threshold adjustment strategy according to the current state, that is, whether to adjust the fault trigger threshold and the adjustment range. For example, during the wind turbine startup phase, when a rapid change in wind speed and a large fluctuation in the amplitude of the acoustic emission signal are detected, the agent automatically increases the threshold according to the learned strategy to avoid false alarms caused by transient noise. As the wind turbine's operating conditions stabilize, the agent adjusts the threshold according to the real-time state to ensure that the system always maintains the best fault detection performance.

[0217] Step S4045: Compare the signal propagation characteristics with the fault triggering threshold. If the signal propagation characteristics exceed the fault triggering threshold, generate a wind turbine gearbox rupture warning message.

[0218] The fan gearbox rupture early warning method based on acoustic emission positioning technology provided in the embodiment is based on the principle of reinforcement learning, and through defining states, actions and reward functions, a reinforcement learning agent model (a deep Q network) learns how to adjust a fault triggering threshold under different working conditions to maximize long-term cumulative rewards. The states include various information related to the operation of the fan gearbox, the actions are specific adjustment operations on the fault triggering threshold, the reward function comprehensively considers multiple factors such as the accuracy of fault detection (the number of real fault detections and the number of false alarms), the multi-path interference suppression effect, the signal path separation accuracy, the interference noise inside the engine room and the running state of the fan and the gearbox, and the Q value update formula continuously adjusts the value evaluation of the agent for each state-action pair according to the current reward and the estimation of future rewards, so that the agent gradually learns the optimal strategy. In the research on the fan gearbox rupture early warning based on acoustic emission positioning technology, the fixed threshold fault detection method is difficult to adapt to the complex and variable working conditions of the fan, and is prone to false alarms or missed alarms. Therefore, the threshold is dynamically adjusted through reinforcement learning, so that the fault detection system can automatically optimize the fault triggering threshold according to the real-time working condition information, effectively solving the limitations of the fixed threshold detection method, improving the accuracy and real-time performance of fault detection, and more accurately warning the fan gearbox rupture.

[0219] The specific steps of the fan gearbox rupture early warning method based on acoustic emission positioning technology will be described below through specific embodiments.

[0220] Embodiment 1

[0221] The specific steps of the fan gearbox rupture early warning method based on acoustic emission positioning technology include:

[0222] Step 1: In the complex environment of the fan gearbox, multi-path interference seriously affects the accuracy of acoustic emission positioning. To effectively suppress multi-path interference, the core idea is to first accurately simulate the sound wave propagation process, and then optimize the positioning algorithm based on the simulation results. The specific steps are as follows:

[0223] I. Sound wave propagation model establishment steps:

[0224] 1) Three-dimensional modeling:

[0225] Use a three-dimensional modeling software (such as SolidWorks, ANSYSDesignModeler, etc.) to accurately reproduce the internal structure of the fan gearbox according to the detailed design drawings and actual measurement dimensions. This step is crucial because only by accurately reproducing the shapes, sizes and relative positions of the gears, shafts, bearing seats, partitions and other components inside the gearbox can an accurate physical model basis be provided for subsequent sound wave propagation simulation.

[0226] Further, in the modeling process, each detail needs to be carefully handled, such as the tooth profile parameters of the gear, the diameter and length of the shaft, the thickness and position of the partition, etc., to ensure that the model is highly consistent with the actual structure. The above steps are like building a precise virtual gearbox to create a realistic "stage" for sound wave propagation simulation.

[0227] 2) Material parameter determination:

[0228] By consulting relevant material manuals, acoustic databases, or conducting special experimental tests, the sound speed, density, attenuation coefficient, and other key parameters of the materials of each component of the gearbox are obtained.

[0229] Further, since different materials have significantly different effects on sound wave propagation, such as the acoustic characteristics of metal components and plastic components being completely different, therefore, for each specific material of a component, it is crucial to assign accurate acoustic parameters, which is like setting unique "performance attributes" for each "actor" in the virtual gearbox, making them perform realistically in the "show" of sound wave propagation.

[0230] 3) Finite element analysis settings:

[0231] The built three-dimensional model (i.e. the fan gearbox three-dimensional physical model after assigning acoustic parameters to each component inside the fan gearbox) is imported into professional finite element analysis software (such as ANSYS or COMSOL Multiphysics). In the software, the acoustic analysis module is used to set boundary conditions; for example, assuming that the gearbox shell is an acoustic hard boundary, the reflection of sound waves at the boundary is simulated, which is more close to the reflection characteristics of sound waves on the gearbox wall in the actual situation.

[0232] Further, the excitation source is set to simulate the sound waves generated by the sound emission source: by adjusting the analysis parameters, such as reasonably dividing the model into grids, the grid density can accurately capture the details of sound wave propagation without causing excessive computational load; selecting appropriate time steps to accurately simulate the propagation process of sound waves over time, etc., to ensure the accuracy and reliability of the simulation results; this step is like setting rules and rhythm for the "show" in the virtual gearbox, allowing the sound wave propagation simulation to proceed in an orderly manner.

[0233] 4) Simulation calculation and result analysis:

[0234] Run the finite element analysis software to simulate the propagation of sound waves inside the gearbox. During the simulation process, the propagation position, reflection and refraction path of sound waves at different times, as well as the sound pressure distribution, phase change and other detailed characteristics at different positions are calculated.

[0235] Further, by simulating the acoustic emission source at different positions multiple times, comprehensive acoustic wave propagation data is obtained, which will provide detailed propagation model data support for subsequent positioning algorithms; for example, acoustic wave propagation data includes the time, amplitude, and phase information of acoustic waves from a certain acoustic emission source, after a series of reflections and refractions, reaching each sensor. In-depth analysis of the above simulation results is like repeatedly studying the "performance video", and useful information is extracted from it to lay the foundation for accurately determining the acoustic emission source position.

[0236] II. Positioning algorithm optimization based on simulation results:

[0237] After obtaining the acoustic wave propagation data simulated by the acoustic wave propagation model, it is applied to the positioning algorithm. By accurately simulating the acoustic wave propagation process and optimizing the positioning algorithm based on the simulation results, multipath interference is effectively suppressed, and the positioning accuracy of the acoustic emission source is improved.

[0238] Step 2: When processing acoustic emission signals of fan gearboxes, accurate separation of multipath signals is a key problem, and the deep learning method based on physical constraints provides an effective solution. The basic idea is to integrate the physical structure information of the gearbox into the deep learning model to improve the accuracy of signal path separation. The specific implementation steps are as follows:

[0239] I. Establishment of three-dimensional physical model and data generation:

[0240] 1) Finite element simulation modeling:

[0241] First, use finite element analysis software (such as ANSYS or ABAQUS) to build an accurate finite element model based on the three-dimensional model of the gearbox. During modeling, reasonable meshing is crucial. Too coarse meshing may not accurately capture the details of acoustic wave propagation, while too fine meshing will result in a sharp increase in computational load. Therefore, according to the structural characteristics of the gearbox and the computing resources, choose an appropriate mesh density.

[0242] At the same time, according to the actual material parameters, assign appropriate elastic modulus, density and other physical properties to each component in the model. This step is to accurately simulate the propagation characteristics of acoustic waves in different materials, because different materials have a significant impact on the propagation speed and attenuation of acoustic waves. For example, metal components and plastic components have a significant difference in the propagation of acoustic waves. Accurate assignment of physical properties can make the simulation more realistic.

[0243] 2) Virtual training data generation:

[0244] In the established finite element model, set multiple virtual acoustic emission source positions to simulate acoustic emission events at different positions. Through finite element simulation, calculate the signal characteristics of sound waves propagating inside the gearbox to various sensor positions, including sound pressure, phase, propagation time, etc. The above information will form the basis of virtual training data.

[0245] To make the virtual training data more representative, various different working conditions and fault scenarios need to be considered; for example, simulate acoustic emission signals under normal operating conditions, and acoustic emission signals under different degrees of gear wear, bearing failure, etc. After collecting the above data, generate a virtual training dataset containing sound wave propagation characteristics under different paths.

[0246] 3) Preprocess the generated data, such as normalization, to ensure data consistency and usability. Normalization can make data of different features have the same scale, which helps improve the training efficiency and accuracy of deep learning models; for example, for sound pressure data and propagation time data, map them to the same numerical range through normalization.

[0247] II. Graph Convolutional Network (GCN) model construction and training:

[0248] 1) Model construction:

[0249] Based on a deep learning framework (such as TensorFlow or PyTorch), construct a graph convolutional network model; in the construction process, first define the nodes and edges of the graph, the nodes can represent sensor positions or specific positions inside the gearbox, and the edges represent sound wave propagation paths or connections between nodes. The definition of the above graph structure can effectively integrate the physical structure information of the gearbox into the model.

[0250] Set appropriate graph convolutional layers, pooling layers, and fully connected layers to construct a network model that can effectively process graph structure data. Graph convolutional layers are used for feature extraction of graph structure data, pooling layers are used to reduce data dimensions, and fully connected layers are used to integrate and classify the extracted features. By reasonably designing the parameters and connection methods of the above layers, the model can learn the characteristics of multi-path signals.

[0251] 2) Model training:

[0252] Input the generated virtual training dataset into the GCN model for training. Before training, set appropriate training parameters such as learning rate, iteration number, loss function, etc. The learning rate determines the step size of the model in each parameter update. A too large learning rate may cause the model to fail to converge, and a too small learning rate will make the training process extremely slow. The number of iterations determines the number of times the model learns the training data, which needs to be adjusted according to the actual situation to avoid overfitting or underfitting.

[0253] The following loss function is used to measure the difference between the model's predictions and the true labels:

[0254]

[0255] Where P is the number of samples in the virtual training dataset. Assuming that in an actual wind turbine gearbox monitoring scenario, P = 1000 samples are collected for training by simulating different fault scenarios and normal operating conditions multiple times; y p For the true label of sample p, y p ∈{0, 1}, where 0 represents the noise signal path and 1 represents the real acoustic emission signal path, y p It is determined based on manual annotation of acoustic emission signals or known fault simulation conditions; W is the sigmoid function, which maps the model output to the interval [0, 1]. K is the number of graph convolutional layers in the GCN model. After multiple experiments comparing the impact of different numbers of layers on model performance, it was determined that K=5 layers can achieve a good balance between computational complexity and model performance. k W is the weight matrix of the k-th graph convolutional layer; k (m,n) represents matrix W k The element in the m-th row and n-th column, W represents the weight matrix of the k-th layer. k The mean of the elements (m, n) during training is used for weight regularization; ReLU(x) = max(0, x) is the modified linear unit function, introducing non-linear characteristics; A pq X represents the connection weight between node p and node q in the adjacency matrix of a graph structure, reflecting the correlation between nodes; q Let be the feature vector of node q, containing features related to the sound wave propagation path extracted from the virtual training data. and λ1, λ2, λ3, and λ4 are the feature vectors of sample p obtained based on simulation and actual measurement, respectively; λ1, λ2, λ3, and λ4 are the coefficients that weigh different terms. After multiple cross-validation experiments, λ1 = 0.01, λ2 = 0.1, λ3 = 0.05, and λ4 = 0.2 were determined. Kullback-Leibler divergence is used to measure the simulated eigenvectors. With actual measured feature vector The difference between them is I, which is the dimension of the feature vector involved in its calculation. Assume I = 10. The correlation coefficient measures the simulated feature vector. With actual measured feature vector linear correlation, and They are respectively and the mean of and are the energy of the simulated and actual measured feature vectors respectively, where J represents the feature dimension used to calculate the energy, assuming J = 5, ∈ is a very small positive number, such as ∈ = 1e-6, to avoid the case of zero denominator.

[0256] During the training process, the model continuously adjusts the parameters of the weight matrix W k so that the model learns the characteristics of different path signals better and better; the validation set is used to monitor the training process to prevent overfitting of the model; if the value of the loss function on the validation set starts to rise, it means that the model may have overfitting, at which time the training parameters need to be adjusted or regularization methods are used to optimize the model performance.

[0257] III. Multi-path signal separation and processing:

[0258] 1) Real-time signal input: In practical applications, the real-time collected acoustic emission signals are first preprocessed. Since the actual collected signals often contain various noises, it is necessary to remove obvious noise interference through filtering, denoising and other operations to make the signals input into the model more pure; for example, a band-pass filter can be used to filter out noise in other frequency bands according to the frequency range of the acoustic emission signal.

[0259] 2) Signal path separation: The preprocessed signal is input into the trained GCN model, and the GCN model separates the input multi-path signal according to the reflection / refraction signal characteristics learned in the training process. The model outputs the signal components of different paths, which can be used to determine the possible propagation path of the acoustic emission source. Combined with the information provided by the previously established acoustic wave propagation model, the judgment of the signal path can be further optimized; for example, the acoustic wave propagation model can provide prior information such as the propagation time and amplitude of the acoustic wave at different positions to help the GCN model more accurately identify the true acoustic emission signal path, thereby improving the accuracy of multi-path signal separation.

[0260] The deep learning signal path separation method based on physical constraints can effectively utilize the physical structure information of the gearbox and improve the accuracy of multi-path signal separation, providing a more reliable signal basis for subsequent fault diagnosis.

[0261] Step 3: The fixed threshold fault detection method is difficult to adapt to the complex and variable operating conditions of the fan, which may lead to false positives or false negatives; the dynamic threshold adjustment mechanism based on reinforcement learning aims to dynamically adjust the fault triggering threshold according to different operating conditions through intelligent learning to improve the accuracy of fault detection. Its implementation process mainly includes three key steps: reinforcement learning agent design, training, and dynamic threshold adjustment, as follows:

[0262] I. Deep Q Network (DQN) Agent Model Design:

[0263] State Definition: First, define the state space of the DQN agent model. The state contains rich information, including not only historical false alarm data but also current operating condition information. Historical false alarm data records the false alarm situation in the past period, such as the number of false alarms in the past n time steps , which helps the agent model understand the system's past performance and better adjust the strategy; current operating condition information includes wind speed v t , load L t , temperature T t , and other fan operating parameters, as well as multi-path interference degree indicators Error mp，t , signal path separation accuracy A sep，t , and other indicators related to acoustic emission signal processing; encode the above information to form a state vector that the DQN agent model can understand The above state definition comprehensively reflects the system's history and current situation, providing sufficient basis for the agent model to make reasonable decisions.

[0264] Action Definition: Define the action space of the DQN agent model, which represents the adjustment operation of the fault trigger threshold; considering flexibility and operability in practical applications, set a series of discrete threshold adjustment steps, such as where Δθ i is the different threshold adjustment amount, such as increasing the threshold by 5%, increasing it by 10%, decreasing it by 5%, decreasing it by 10%, or keeping it unchanged; the agent model dynamically adjusts the fault trigger threshold by selecting different actions to adapt to different operating conditions.

[0265] Reward Function Design: Setting a reasonable reward function is the key to guiding the DQN agent model to learn the optimal strategy. The design principle of the reward function is to encourage the agent to make decisions that can reduce false alarms and not miss real faults.

[0266] II. DQN Agent Training:

[0267] Experience Replay: During the operation of the fan, the DQN agent continuously observes the current state s t , selects an action a t according to the current policy, executes the action, and the environment will feedback the new state s t+1 and the corresponding reward value r t , and the state s t , action a t , reward r t , and new state s t+1The experience replay buffer is used to break the correlation between data, improve the stability and efficiency of learning, and when the data stored in the buffer reaches a certain amount, a batch of data is randomly sampled from the buffer for training. This random sampling method can avoid the agent from over-relying on a part of data in the learning process, thereby better generalizing to various working conditions.

[0268] Network update: using the sampled data, calculate the error between the target Q value and the current Q value. The target Q value represents the estimated value of the long-term cumulative reward obtained by taking the optimal action in the current state, while the current Q value is the reward estimate value obtained by taking action under the current policy. Through the backpropagation algorithm, adjust the neural network parameters of the DQN agent, so that the current Q value gradually approaches the target Q value;

[0269] By continuously repeating the process of experience replay and network update, as the training progresses, the DQN agent model gradually learns the optimal threshold adjustment strategy, which can dynamically adjust the fault trigger threshold according to different working conditions to maximize the long-term cumulative reward.

[0270] III. Dynamic threshold adjustment

[0271] Real-time working condition monitoring: real-time monitoring of the operating conditions of the fan, including changes in wind speed, load, temperature and other parameters, as well as the degree of multi-path interference, signal path separation accuracy and other indicators related to acoustic emission signal processing. The above information is fed back to the DQN agent model in real time as the basis for the agent to adjust the threshold; for example, when the wind speed suddenly increases, it may introduce more environmental noise, affecting the detection of acoustic emission signals, at which point the agent needs to make a corresponding threshold adjustment decision based on this working condition change.

[0272] Threshold adjustment decision: the DQN agent model selects an action based on the learned policy according to the current state (including historical false alarm data and real-time working condition information), i.e. whether to adjust the fault trigger threshold and the adjustment amplitude; for example, during the fan startup phase, due to the presence of transient noise, the acoustic emission signal features are different from those during normal operation, and multi-path interference may be more severe, and signal path separation accuracy may be lower. At this point, the DQN agent model automatically raises the threshold according to the learned policy based on the current state information to avoid false alarms caused by transient noise. As the fan operating conditions change, the agent continues to learn and adjust the threshold in real time, so that the system always maintains the best fault detection performance, effectively reducing the false alarm rate and improving the real-time performance and accuracy of fault detection.

[0273] Through the above implementation of the dynamic threshold adjustment mechanism based on reinforcement learning, the fault detection system can intelligently adjust the fault triggering threshold according to the actual operating conditions of the fan, adapt to complex and variable working environments, and improve the reliability and stability of the entire fan gearbox rupture early warning system.

[0274] Embodiment 2:

[0275] In practical applications, the multi-path interference suppression formula provides a more accurate signal propagation model and initial data for deep learning signal path separation based on physical constraints, and accurate signal path separation provides reliable signal features and state information for dynamic threshold adjustment based on reinforcement learning. The specific application in the fan gearbox rupture early warning process includes:

[0276] I. Application of the multi-path interference suppression formula in the fan gearbox rupture early warning process includes:

[0277] 1) Application under normal operating conditions: When the fan gearbox is operating normally, the acoustic emission signal is relatively stable, and although multi-path interference exists, it is relatively mild. At this time, the multi-path interference suppression formula is as follows:

[0278]

[0279] The steps of multi-path interference suppression include:

[0280] Determine the initial parameters: First, determine the number of sensors N, the number of amplitude sampling points M, and the number of phase sampling points L according to the structure of the fan gearbox and the sensor layout. At the same time, based on historical data or preliminary tests, determine the standard deviation of each sensor time measurement Standard deviation of amplitude measurement Standard deviation of phase measurement For the balance coefficients λ and μ, the default empirical values can be used first, and they can be fine-tuned according to the actual situation in subsequent operation.

[0281] Real-time monitoring and calculation: Real-time acquisition of the actual time of sound waves arriving at each sensor Actual amplitude And the actual phase Use the accurate propagation model to simulate sound wave propagation in real time to obtain the simulated time Simulated amplitude And the simulated phase Substitute these values into the formula to calculate, and by continuously adjusting the sound emission source position variable vector r, find the That is, the estimated sound emission source position, which can be used as a reference for the characteristics of the acoustic emission signal during normal operation for comparison in subsequent abnormal situations.

[0282] 2) Mild fault condition application: When the fan gearbox has a mild fault, such as slight gear wear or component loosening, the acoustic emission signal will change, and the multi-path interference may increase. The steps for multi-path interference suppression include:

[0283] Parameter adjustment: Due to the change in signal characteristics caused by the fault, the standard deviation of time, amplitude, and phase measurements may need to be re-evaluated and adjusted; at the same time, the balance coefficients λ and μ are appropriately adjusted according to the type of fault and the focus of the signal impact; for example, if the fault mainly affects the signal amplitude, λ is appropriately increased.

[0284] Fault location and analysis: According to the real-time monitoring and calculation steps under normal conditions, the adjusted parameters are calculated to obtain the estimated acoustic emission source position Compare this position with the reference position under normal conditions. If the deviation exceeds a certain threshold (which needs to be determined according to the specific structure of the fan gearbox and historical data), the occurrence position of the fault can be preliminarily judged. Combined with the changes in acoustic emission signal characteristics, such as amplitude increase and phase anomaly, further analyze the possible causes and severity of the fault.

[0285] 3) Application in severe fault condition: In the case of severe fault, such as severe gear damage or shaft fracture, the acoustic emission signal will change significantly, and the multi-path interference will be more complex. The steps for multi-path interference suppression include:

[0286] Strengthen the model and optimize the parameters: At this time, the original accurate propagation model may need to be further refined or corrected to more accurately simulate the complex sound wave propagation situation; at the same time, the adjustment range of each parameter is larger, the standard deviation needs to be accurately measured again, and the balance coefficients λ and μ need to be optimized significantly according to the degree of fault impact on different signal characteristics.

[0287] Fault severity assessment: The acoustic emission source position estimated by formula calculation If the position deviates greatly from the normal condition, and the signal characteristics (such as amplitude far exceeding the normal range and phase seriously disturbed) also indicate that the fault is severe, the severity of the fault can be quickly judged according to the above information, combined with the pre-set fault severity assessment standard (based on historical fault data and expert experience), to provide a basis for timely maintenance measures.

[0288] II. Application of dynamic threshold adjustment based on reinforcement learning in the process of fan gearbox rupture warning includes:

[0289] 1) Fan startup phase: In the fan startup phase, the working condition changes greatly, the acoustic emission signal characteristics are unstable, and the multi-path interference and noise impact are large. The specific steps of dynamic threshold adjustment based on reinforcement learning include:

[0290] State initialization: Determine the initial state s0 of the deep Q network agent, including acquiring operating condition information such as wind speed, load, and temperature at startup, as well as the multipath interference level index Error based on previous simulations or empirical estimates. mp，0 Signal path separation accuracy A sep，0 Noise level inside the cabin int，0 Wind turbine operating status indicators f，0 Gearbox operating status indicators gb，0 Meanwhile, initialize historical false alarm data.

[0291] Action selection and threshold adjustment: Based on the current state s0, the agent selects an action a0 according to the current strategy (such as a random strategy or a strategy based on previous experience), which adjusts the fault trigger threshold. For example, due to the high noise during the startup phase, the threshold may be appropriately increased to avoid false alarms.

[0292] Reward Calculation and Q-Value Update: After executing action a0, obtain the new state s1, and calculate the reward value r0 according to the reward function:

[0293]

[0294] Then, update the Q value using the Q-value update formula:

[0295]

[0296] As the startup process continues, the above steps are repeated, and the agent gradually learns the threshold adjustment strategy suitable for the wind turbine startup phase.

[0297] 2) Stable operation phase of the wind turbine: When the wind turbine enters the stable operation phase, the operating conditions are relatively stable, but the threshold still needs to be dynamically adjusted based on real-time monitoring information. The specific steps of dynamic threshold adjustment based on reinforcement learning include:

[0298] Real-time status updates: Real-time monitoring of operating conditions such as wind speed, load, and temperature, as well as multipath interference level indicators (Error). mp，t Signal path separation accuracy A sep，t Noise level inside the cabin int，t Wind turbine operating status indicators f，t Gearbox operating status indicators gb，t Wait, update the proxy status s t At the same time, historical false alarm data is recorded.

[0299] Policy-based action selection: The agent selects actions based on the current state s. t Based on the learned strategies, select an action a.t For example, if the multipath interference degree indicator Error mp，t increases, and the signal path separation accuracy A sep，t decreases, the agent can choose to appropriately lower the threshold to improve the sensitivity of fault detection.

[0300] Reward calculation and Q-value update: After performing action a t , the new state s t+1 is obtained, the reward value r t is calculated, and the Q-value is updated, similar to the start-up phase. By continuously updating the Q-value, the agent continuously optimizes the threshold adjustment strategy to adapt to subtle operating condition changes that may occur during the stable operation phase.

[0301] 3) Fault occurrence phase: When the fan gearbox fails, the acoustic emission signal characteristics, multipath interference, noise, and the operating state of the fan and gearbox will change significantly. The specific steps of dynamic threshold adjustment based on reinforcement learning include:

[0302] State change and action adjustment: The occurrence of a fault will cause the agent's state s t to change significantly, such as a significant increase in the multipath interference degree indicator Error mp，t , a decrease in the signal path separation accuracy A sep，t , and a deviation of the fan and gearbox operating state indicators State f，t , State gb，t from the normal range, etc. The agent selects action a t based on the learned strategy according to these state changes, which may significantly adjust the threshold to ensure accurate detection of the fault.

[0303] Fault confirmation and reward feedback: If the real fault is successfully detected by adjusting the threshold (i.e. increases), the corresponding term in the reward function will give a positive reward, reinforcing the agent's decision. At the same time, according to the results after fault handling, such as the fan and gearbox returning to normal operation after repair, the agent's state and Q-value are further updated, allowing the agent to learn the optimal threshold adjustment strategy during the fault occurrence and handling process to better handle similar fault situations in the future.

[0304] 4) Re-operation phase after fault repair: After the fan gearbox is repaired and put back into operation, it is a transition phase, and the operating conditions gradually recover from the post-fault state to the normal stable state. The specific steps of dynamic threshold adjustment based on reinforcement learning include:

[0305] State Reset and Initialization: Partially reset and reinitialize the state of the agent. First, clean or adjust the historical false alarm data appropriately, as the system situation has changed after the fault is repaired. Based on the actual situation after the fault is repaired, reevaluate and set the fan operating state indicators State f，t , gearbox operating state indicators State gb，t . At the same time, based on the possible impact on the acoustic emission signal propagation path during the fault repair process, reestimate the multi-path interference degree indicators Error mp,t and signal path separation accuracy A sep,t to determine the initial state s t of the agent.

[0306] Step-by-step adjustment and adaptation: the agent selects action a t according to the current state s t , at this time the action may tend to adjust the threshold to a value close to the normal operating state, but will be relatively conservative to avoid false alarms or missed alarms due to the system not being completely stable; as the running time progresses, real-time monitoring of changes in each state indicator, such as the gradual recovery of multi-path interference to normal levels, the gradual improvement of signal path separation accuracy, etc., based on these changes, the agent continuously adjusts the action, so that the threshold gradually adapts to the new stable operating state.

[0307] Learning and optimization: in this process, the reward value r t is calculated through the reward function, and the Q-value is updated using the Q-value update formula; for example, if after adjusting the threshold, the system can accurately detect a small number of potential problems that may exist under the new operating state (confirmed by subsequent detection means), the reward function will give positive rewards to encourage the agent to further optimize the threshold adjustment strategy to adapt to the operating state after the fault is repaired more quickly and prepare for future stable operation.

[0308] It should be noted that the following points should be noted during the above application process:

[0309] 1) Continuous updating and maintenance of data: the fan operating environment and gearbox state will change over time, so new data needs to be continuously collected to update the model and parameters; for example, periodically reevaluate the propagation model parameters in the multi-path interference suppression formula, and the training data for the deep learning model based on physical constraints to ensure that the model can adapt to new working conditions and fault modes.

[0310] 2) Dynamic adjustment of parameters: the parameters in each formula are not fixed and need to be dynamically adjusted according to the actual operating conditions. For example, in the dynamic threshold adjustment based on reinforcement learning, the weight coefficients may need to be adjusted according to different seasons and different operating stages (such as the initial period of new equipment being put into use and the long-term operation afterwards) to balance the influence of each factor on the reward value.

[0311] Integration and optimization of the system: When integrating the algorithms involved in the three formulas into a complete monitoring system, attention should be paid to the interface and data interaction between the parts. For example, how the results of signal path separation are accurately transmitted to the dynamic threshold adjustment module, and how the dynamic threshold adjustment affects the parameter settings of multi-path interference suppression, etc. By continuously optimizing system integration, the performance and reliability of the entire fan gearbox rupture early warning system based on acoustic emission positioning technology are improved.

[0312] By discussing the application of the above method in practice, comprehensive technical guidance can be provided to relevant technical personnel, helping relevant technical personnel to better utilize the above technology to effectively warn of fan gearbox rupture.

[0313] Example 3:

[0314] The dynamic threshold adjustment mechanism based on reinforcement learning and the deep learning signal path separation method based on physical constraints logically form a complete link of "signal preprocessing → feature optimization → decision execution", and the two achieve overall optimization of fan gearbox rupture early warning through data interaction and target cooperation. The specific relationship is as follows:

[0315] I. Data interaction: Signal path separation provides key input features for threshold adjustment:

[0316] 1) The output of signal path separation as the state parameter of threshold adjustment: After separating the multi-path signal based on the physical constraint deep learning method through the graph convolution network (GCN), the following key information is output: signal path classification result: the probability of distinguishing between real acoustic emission signal path and noise path (output through the Sigmoid function); separation accuracy (A sep,t ): reflects the separation effect of the model on the multi-path signal (such as the proportion of correctly classified paths); the above information is directly used as the state input (s t ) of the reinforcement learning agent, to evaluate the reliability of the current signal processing.

[0317] For example: If A sep，t is high (such as > 90%), it means that the signal purity is high, and the threshold adjustment can be more sensitive to detect early faults; if A sep，t is low (such as < 60%), it means that the multi-path interference has not been effectively suppressed, and the threshold needs to be increased to avoid false positives.

[0318] 2) The difference between simulated and measured features provides environmental feedback for threshold adjustment: In the signal path separation process, the features simulated by the physical model (X sim，p ) and the actual measured features (X meas,pthe difference (such as KL divergence, correlation coefficient, energy difference) of the signal features of the simulation and the measured data is integrated into the multi-path interference degree index (Error mp,t ). The index is used as part of the reinforcement learning state to quantify the accuracy of the signal propagation model: Error mp,t The smaller the Error mp，t The larger the Error

[0319] II. Target synergy: jointly optimize the sensitivity and reliability of fault detection:

[0320] 1) Signal path separation lays the physical foundation for threshold adjustment: the correlation threshold adjustment method is difficult to set a fixed threshold due to the distortion of signal features caused by multi-path interference. The physically constrained deep learning method improves signal quality in the following ways:

[0321] Suppress reflection / refraction interference: use the three-dimensional model of the gearbox and the GCN network to separate the real signal path, reducing the pollution of noise to the features (such as amplitude, phase).

[0322] Enhance feature distinguishability: the features of the separated real signal path are closer to the actual fault signal, making the "fault-noise" boundary clearer during threshold adjustment.

[0323] For example: in the gear wear fault, the separated signal may show a jump in the amplitude of a specific frequency component. At this time, the dynamic threshold can be accurately adjusted for this feature to avoid being triggered by similar features in the noise.

[0324] 2) Dynamic threshold adjustment amplifies the value of signal separation: even if signal path separation improves feature purity, dynamic threshold adjustment is still needed according to real-time working conditions. Reinforcement learning optimizes in the following ways:

[0325] Working condition adaptation: in working conditions such as fan startup and load mutation, signal feature distribution may deviate from the normal state (such as increased transient noise). Dynamic threshold adjustment can temporarily relax or tighten the threshold according to the current working condition (such as wind speed, load) and signal separation effect, avoiding the "one-size-fits-all" defect of a fixed threshold.

[0326] False alarm-miss alarm balance: the signal separation accuracy (A sep，t ) is used as the positive term (β4·A sep，t ) of the reward function, guiding the agent to lower the threshold to improve sensitivity when the separation effect is good, and to increase the threshold to reduce false alarms when the separation effect is poor, forming a closed loop of "high-quality signal → aggressive detection, low-quality signal → conservative decision".

[0327] III. Mathematical correlation: parameter coupling in key formulas:

[0328] 1) Reward function integrates signal separation results: The reward function of reinforcement learning directly includes the signal path separation accuracy (A). sep，t ) and simulation-measured feature difference (Error) mp，t ):

[0329]

[0330] β3 and β4 are weighting coefficients, indicating the accuracy of signal separation (A sep，t ) and model reliability (Error) mp，t This directly affects the agent's reward feedback, driving it to prioritize threshold strategies that match the signal separation results.

[0331] 2) Q-value update depends on signal processing quality: When a reinforcement learning agent optimizes its threshold adjustment strategy using the Q-value update formula, the state transition (S... t →S t+1 The core basis is the comparison result between signal characteristics and threshold.

[0332] For example: if, after signal path separation, a certain feature (such as amplitude) is determined to be a "fault signal" by a threshold, and it is actually a real fault ( If the threshold is increased, the corresponding Q value will receive a positive update, strengthening the priority of the threshold adjustment action; if the separated signal characteristics are misjudged as faults (due to residual interference), the number of false alarms will be used to determine the fault. The negative reward inhibits the action, forcing the agent to adjust the threshold.

[0333] IV. Summary: The closed loop from "signal purification" to "intelligent decision-making" is shown in Table 1 below:

[0334] Table 1:

[0335] Link Signal path separation based on physical constraints Dynamic threshold adjustment based on reinforcement learning Core objective Separate real signals from noise, improve feature purity Dynamically determine fault threshold according to signal characteristics and working conditions Key outputs Signal path classification, separation accuracy, feature difference index Dynamic threshold, fault warning decision Logical association Provide reliable input features for threshold adjustment Convert signal characteristics into specific warning actions Synergistic value Reduce "noise interference error" of threshold adjustment Amplify "fault feature enhancement effect" of signal separation

[0336] Both technologies constrain the signal processing process through physical models and combine data-driven optimization decision-making strategies to form a complete technical chain of "first purifying the signal, then making intelligent decisions." Together, they solve the core problems of "difficulty in suppressing interference and difficulty in adapting thresholds" in traditional acoustic emission early warning systems, ultimately achieving high precision and high reliability in wind turbine gearbox rupture early warning.

[0337] The above embodiments have the following beneficial effects:

[0338] 1) Multi-path interference suppression improves positioning accuracy: By establishing an accurate propagation model, a detailed three-dimensional model of the complex structure inside the fan gearbox is established, and accurate sound wave propagation parameters are assigned to each component. The sound wave propagation process is accurately simulated using finite element analysis, which can more accurately predict the propagation path and characteristics of the acoustic emission signal under multi-path effect. Based on this model, the subsequent positioning algorithm can obtain more reliable data basis, thereby effectively reducing the positioning error caused by multi-path interference and greatly improving the positioning accuracy of the acoustic emission source. In practical applications, the positioning error can be reduced by 30%-50%, and the position of the gearbox internal fault can be determined more accurately, providing strong support for timely maintenance and fault troubleshooting; Enhance the reliability of signal analysis: The accurate propagation model helps better understand the propagation law of acoustic emission signals inside the gearbox. When analyzing the collected acoustic emission signals, based on the model, the useful information and interference components in the signal can be more accurately identified, improving the reliability of signal analysis, which is of great significance for accurately judging the running state of the gearbox and discovering potential faults in advance; For example, when analyzing the frequency, amplitude and other characteristics of the signal, false characteristics caused by multi-path interference can be excluded, making the analysis results more truly reflect the fault condition inside the gearbox.

[0339] 2) Deep learning signal path separation based on physical constraints improves multi-path signal separation accuracy: A three-dimensional physical model of the gearbox structure is introduced, and a graph convolution network is used to separate the multi-path signal. Virtual training data is generated using the CAD model of the gearbox, enabling the GCN model to learn the characteristics of reflected / refracted signals, significantly enhancing the model's ability to identify and separate multi-path signals in complex structures. Compared with signal processing methods based on simple geometric models, the method can more accurately separate multi-path signals, increasing the accuracy of multi-path signal separation by 20%-30%, providing purer and more accurate signals for subsequent acoustic emission signal processing and fault diagnosis, and helping to improve the accuracy of fault diagnosis; Breakthrough traditional algorithm limitations: innovatively combining physical simulation with data-driven, the method breaks through the dependence of traditional algorithms on simple geometric models; This combination fully utilizes the prior knowledge provided by the physical model to guide the training of the deep learning model, enabling the model to better adapt to the complex internal structure of the fan gearbox; Even in the face of extremely complex sound wave propagation scenarios, the method can effectively process multi-path signals, improving the adaptability and robustness of the entire acoustic emission signal processing system; For example, when the internal structure of the gearbox changes slightly or there are some irregular components, traditional algorithms may fail, but this method can still accurately separate multi-path signals, ensuring the normal operation of the fault warning system.

[0340] 3) Dynamic threshold adjustment mechanism based on reinforcement learning to reduce false alarm rate: a reinforcement learning agent based on deep Q network is designed to dynamically adjust the fault triggering threshold according to historical false alarm data and current working conditions. In the process of fan operation, the characteristics of acoustic emission signals differ greatly under different working conditions, and traditional fixed threshold values are prone to false alarms. Through the dynamic threshold adjustment mechanism, it can automatically adapt to the changes of working conditions, and automatically increase the threshold value to avoid false alarms in the case of transient noise at the start-up stage of the fan. Practical application shows that this mechanism can reduce the false alarm rate by 40%-50%, greatly reducing the waste of manpower and material resources caused by false alarms, and improving the credibility of the early warning system; improve the real-time performance and accuracy of fault detection: the reinforcement learning agent continuously learns and adjusts the threshold value, so that the system always maintains the best fault detection performance. With the change of working conditions, the agent can make timely threshold adjustment decisions to ensure accurate detection of fault signals in various situations. This not only improves the real-time performance of fault detection and can issue an early warning at the first time of fault occurrence, but also ensures the accuracy of detection and will not miss real faults due to improper threshold setting. For example, when the load of the fan changes suddenly, the agent can quickly adjust the threshold value according to the current working conditions to accurately detect the fault signals that may be caused by load changes, providing more reliable protection for the safe operation of the equipment.

[0341] In this embodiment, a fan gearbox rupture early warning device based on acoustic emission positioning technology is also provided, which is used to realize the above-mentioned embodiments and preferred embodiments, and will not be described again. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware, or a combination of software and hardware is also possible and contemplated.

[0342] The present embodiment provides a fan gearbox rupture early warning device based on acoustic emission positioning technology, as shown in Figure 5 , comprising:

[0343] The construction module 501 is configured to obtain the internal structure parameters of the fan gearbox, and construct an acoustic wave propagation model based on the internal structure parameters of the fan gearbox;

[0344] The simulation module 502 is configured to perform acoustic wave propagation simulation at a plurality of acoustic emission source positions using the acoustic wave propagation model to obtain acoustic wave propagation simulation data;

[0345] The separation module 503 is configured to obtain real-time acoustic emission signals, and use a graph convolution network model to separate the real-time acoustic emission signals to obtain classification probability data of the multipath signal paths;

[0346] The detection module 504 is used to detect fault signals based on acoustic wave propagation simulation data and classification probability data of multipath signal paths, and to obtain early warning information of wind turbine gearbox rupture by using a reinforcement learning agent model.

[0347] The further functional descriptions of each of the above modules are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0348] In this embodiment, a wind turbine gearbox rupture early warning device based on acoustic emission positioning technology is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0349] This invention also provides a computer device having the above-described features. Figure 5 The image shows a wind turbine gearbox rupture early warning device based on acoustic emission positioning technology.

[0350] Please see Figure 6 , Figure 6 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 6 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 6 Take a processor 10 as an example.

[0351] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0352] The memory 20 stores instructions executable by the at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0353] The memory 20 can include a program storage area and a data storage area. The program storage area can store an operating system and application programs required by at least one function. The data storage area can store data created according to the use of the computer device, and the like. In addition, the memory 20 can include a high-speed random access memory, and can also include a non-transitory memory such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some optional embodiments, the memory 20 can optionally include a memory disposed remotely with respect to the processor 10, and these remote memories can be connected to the computer device through a network. Examples of the above network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0354] The memory 20 can include a volatile memory such as a random access memory, and can also include a non-volatile memory such as a flash memory, a hard disk, or a solid-state disk. The memory 20 can also include a combination of the above-mentioned kinds of memories.

[0355] The computer device further includes a communication interface 30 for communication of the computer device with other devices or communication networks.

[0356] The embodiments of the present application also provide a computer readable storage medium. The above method according to the embodiments of the present application can be implemented in hardware, firmware, or recorded in a storage medium, or implemented as computer code originally stored in a remote storage medium or a non-transitory machine readable storage medium and downloaded to a local storage medium, so that the method described herein can be processed by such software on a storage medium using a general purpose computer, a special purpose processor, or programmable or special purpose hardware. The storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid-state disk, and the like. Further, the storage medium can also include a combination of the above-mentioned kinds of memories. It can be understood that the computer, the processor, the microprocessor controller, or the programmable hardware includes a storage component that can store or receive software or computer code, when the software or computer code is accessed and executed by the computer, the processor, or the hardware, the method shown in the above embodiments is implemented.

[0357] Part of the present application can be applied as a computer program product, for example, computer program instructions, when executed by a computer, through the operation of the computer, can invoke or provide the method and / or technical solutions according to the present application. Those skilled in the art should understand that the form of computer program instructions in computer readable medium includes but is not limited to source files, executable files, installation package files and the like, and accordingly, the way of computer program instructions executed by computer includes but is not limited to: the computer directly executes the instructions, or the computer compiles the instructions and then executes the corresponding compiled program, or the computer reads and executes the instructions, or the computer reads and installs the instructions and then executes the corresponding installed program. Here, the computer readable medium can be any available computer readable storage medium or communication medium accessible to the computer.

[0358] Although the embodiments of the present application are described in conjunction with the drawings, various modifications and changes can be made by those skilled in the art without departing from the spirit and scope of the present application, and such modifications and changes fall within the scope defined by the appended claims.

Claims

1. A fan gear box rupture early warning method based on acoustic emission positioning technology, characterized in that, The method comprises: obtaining internal structure parameters of a fan gearbox, and constructing a sound wave propagation model based on the internal structure parameters of the fan gearbox; performing sound wave propagation simulation at multiple acoustic emission source positions by using the sound wave propagation model to obtain sound wave propagation simulation data; obtaining an acoustic emission real-time signal, performing path separation on the acoustic emission real-time signal by using a graph convolution network model to obtain classification probability data of multi-path signal paths; performing fault signal detection by using a reinforcement learning agent model based on the sound wave propagation simulation data and the classification probability data of the multi-path signal paths to obtain fan gearbox rupture early warning information.

2. The method of claim 1, wherein, The method comprises: obtaining three-dimensional parameters of each component inside the fan gearbox, and constructing a three-dimensional physical model of the fan gearbox based on the three-dimensional parameters of each component inside the fan gearbox; obtaining acoustic parameters of each component inside the fan gearbox, and assigning the acoustic parameters of each component inside the fan gearbox to the three-dimensional physical model of the fan gearbox; performing finite element analysis on the three-dimensional physical model of the fan gearbox to which the acoustic parameters of each component inside the fan gearbox are assigned to obtain the sound wave propagation model.

3. The method of claim 2, wherein, The method comprises: performing signal preprocessing on the acoustic emission real-time signal, and extracting features from the signal-preprocessed acoustic emission real-time signal to obtain signal propagation features; determining the classification probability data of the multi-path signal paths by using the graph convolution network model based on the signal-preprocessed acoustic emission real-time signal.

4. The method of claim 3, wherein, The method comprises: performing positioning optimization on the sound wave propagation simulation data to determine a multi-path interference degree index; determining a signal separation accuracy rate based on the classification probability data of the multi-path signal paths; determining a threshold adjustment strategy under multiple fan operating conditions by using the reinforcement learning agent model; obtaining current fan operating parameters, and determining a fault triggering threshold by using the threshold adjustment strategy under the multiple fan operating conditions based on the current fan operating parameters, the multi-path interference degree index, and the signal path separation accuracy rate; comparing the signal propagation features with the fault triggering threshold, and generating the fan gearbox rupture early warning information if the signal propagation features exceed the fault triggering threshold.

5. The method of claim 4, wherein, The method comprises: determining a sound emission source position vector by using a positioning algorithm based on the sound wave propagation simulation data; performing propagation parameter simulation based on the sound emission source position vector to obtain simulation basic parameters of sound wave propagation inside the gearbox; Obtaining actual measurement basis parameters of sound wave propagation inside the gearbox, comparing the simulation basis parameters of sound wave propagation inside the gearbox with the actual measurement basis parameters of sound wave propagation inside the gearbox, and obtaining the multi-path interference degree index.

6. The method of claim 4, wherein, The threshold adjustment strategy under multiple fan operating conditions is determined by using the reinforcement learning agent model, which includes: Defining states, actions and reward functions; wherein the states are historical false alarm data corresponding to multiple fan operating conditions, fan operating parameters, multi-path interference degree indexes and signal path separation accuracy; the actions are adjustment operations of the fault triggering threshold; Obtaining the current state, selecting the current action according to the current state, adjusting the fault triggering threshold based on the current action, and updating the current state to obtain the updated state; Based on the updated state, calculate the current reward value; Based on the current state, the current action, the updated state and the current reward value, calculate the current Q value; Compare the current Q value with the target Q value, and update the threshold adjustment strategy based on the comparison result until the current Q value and the target Q value, to obtain the threshold adjustment strategy under multiple fan operating conditions.

7. A fan gear box rupture early warning device based on acoustic emission positioning technology, characterized in that, The device includes: The construction module is configured to obtain internal structure parameters of a fan gearbox, and construct a sound wave propagation model based on the internal structure parameters of the fan gearbox; The simulation module is configured to perform sound wave propagation simulation at multiple acoustic emission source positions by using the sound wave propagation model to obtain sound wave propagation simulation data; The separation module is configured to obtain real-time acoustic emission signals, and perform path separation on the real-time acoustic emission signals by using a graph convolution network model to obtain classification probability data of multi-path signal paths; The detection module is configured to perform fault signal detection by using a reinforcement learning agent model based on the sound wave propagation simulation data and the classification probability data of the multi-path signal paths to obtain fan gearbox rupture early warning information.

8. A computer device, comprising: It includes: A memory and a processor are communicatively connected, and the memory stores computer instructions; the processor executes the computer instructions to perform the fan gearbox rupture early warning method based on the acoustic emission positioning technology according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions for making a computer execute the fan gearbox rupture early warning method based on the acoustic emission positioning technology according to any one of claims 1 to 6.

10. A computer program product, characterised in that, It includes computer instructions for making a computer execute the fan gearbox rupture early warning method based on the acoustic emission positioning technology according to any one of claims 1 to 6.

Images