Dynamic stress field measurement method and device for damping blade, medium and equipment

Abstract

The invention discloses a dynamic stress field measurement method and device for a damping blade, a medium and equipment, and relates to the technical field of online measurement of a dynamic stress field of a blade. The method comprises the steps of collecting blade tip displacement data of a whole circle of damping blades of the rotating equipment through a blade tip timing sensor array; the blade tip timing sensor array comprises a plurality of blade tip timing sensors, and the blade tip timing sensors are arranged in the circumferential direction of the blade at intervals; according to the blade tip displacement data of the whole circle of damping blades, the vibration frequency and response of each blade are determined by extracting modal information and phase change of adjacent blades; the response comprises a displacement amplitude and a vibration phase; and inputting the vibration frequency and the response of each blade into a pre-trained dynamic stress field prediction neural network model to obtain a dynamic stress field of the whole circle of damping blades. According to the method, full-field acquisition of the dynamic stress of the whole circle of blades can be realized, and the operation monitoring precision of the blades is improved.

Description

Technical Field

[0001] This invention relates to the field of online measurement technology of dynamic stress field of blades, and particularly to a method, device, medium and equipment for measuring the dynamic stress field of a damped blade. Background Technology

[0002] As a key load-bearing component in turbine machinery, the dynamic stress state of blades directly affects the safety and service life of the entire machine. Due to their unique structure and multi-blade coupled vibration behavior, the dynamic stress distribution of coiled damping blades is more complex and more difficult to obtain in real time, thus affecting the health monitoring and fault early warning of the equipment.

[0003] Currently, blade dynamic stress measurement mainly relies on contact sensing devices such as strain gauges to collect strain data. Although these devices can reflect local dynamic stress, they cannot achieve full-field acquisition of dynamic stress across the entire blade, making it difficult to meet the requirements for high-precision operational monitoring. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, device, medium, and equipment for measuring the dynamic stress field of damped blades to address the above-mentioned technical problems. This method can achieve full-field acquisition of dynamic stress of the entire blade and improve the accuracy of blade operation monitoring.

[0005] The present invention adopts the following technical solution: This invention provides a method for measuring the dynamic stress field of a damping blade, comprising: The blade tip displacement data of the entire circumference of the damping blades of the rotating equipment is collected by a blade tip timing sensor array; the blade tip timing sensor array includes multiple blade tip timing sensors, which are arranged at intervals along the circumference of the blade. Based on the tip displacement data of the entire damping blade, the vibration frequency and response of each blade are determined by extracting modal information and phase changes of adjacent blades; the response includes displacement amplitude and vibration phase. The vibration frequency and response of each blade are input into a pre-trained dynamic stress field prediction neural network model to obtain the dynamic stress field of the entire damped blade.

[0006] Optionally, the method for determining the installation locations of multiple blade tip timing sensors specifically includes: The installation position of each blade tip timing sensor is determined based on the spacing between the blade sheaths and the impeller structure.

[0007] Optionally, based on the tip displacement data of the entire damping blade, the vibration frequency and response of each blade are determined by extracting modal information and phase changes of adjacent blades, including: The tip displacement data of the entire damping blade is input into the vibration identification model. The vibration identification model extracts modal information and phase changes of adjacent blades to determine the vibration frequency and response of each blade. The vibration identification model includes a downsampling layer, multiple stacked convolutional modules, and a fully connected layer connected in series. Each convolutional module includes a convolutional layer, batch normalization, and activation function connected in series.

[0008] Optionally, the loss function of the vibration recognition model during training. for: ; in, and These represent the actual nodal diameter vibration frequency and the full-circle damped blade nodal diameter vibration frequency predicted by the vibration identification model, respectively. For the number of leaves, and The vibration identification model predicts the first Displacement amplitude and vibration phase of each blade, and Each is the real first Displacement amplitude and vibration phase of each blade, and The weight of the loss term; It is the Euclidean norm.

[0009] Optionally, the dynamic stress field prediction neural network model includes a graph structure building module, multi-layer graph convolutional blocks, and a prediction head; the multi-layer graph convolutional blocks include... A series of concatenated graph convolutional sub-blocks, each including a graph convolutional layer, a batch normalization layer, and... Activation layer; the prediction head includes a fully connected layer; the vibration frequency and response of each blade are input into the dynamic stress field prediction model to obtain the dynamic stress field of the entire damped blade, including: In the graph structure construction module, an original adjacency matrix is ​​constructed based on the adjacency relationship between each blade. The original adjacency matrix is ​​then symmetrically normalized to obtain a normalized adjacency matrix. The vibration frequency, displacement amplitude, and vibration phase of each blade are then spliced ​​together to construct a multidimensional feature vector. The multidimensional feature vectors of all blades are then combined to form a node feature matrix. In each graph convolutional sub-block, the graph convolutional layer propagates information about the node features output from the previous layer based on the normalized adjacency matrix, and then sequentially passes the propagated node features through batch normalization layers and... The activation function is used to obtain the node features output by this layer; the node features output by the previous layer used in the first graph convolutional sub-block are the node feature matrix; the calculation formula for the graph convolutional sub-block is: in, For the first Layer node characteristics, For the first Layer node characteristics, The weight matrix is ​​a learnable matrix. for Activation function For batch normalization operations, This is the normalized adjacency matrix; In the prediction head, a fully connected layer is used to perform linear transformation and dimensionality compression on the node features output by the last layer of graph convolutional sub-blocks to obtain the dynamic stress field of the entire damping blade; the calculation formula for the fully connected layer is: in, This represents the dynamic stress field of the entire damping blade. This is the weight matrix of the fully connected layer. For bias terms, This refers to the node features output by the convolutional sub-block of the last layer of the graph.

[0010] Optionally, the graph structure building module also includes: Treat each leaf as a node and construct a graph structure based on the adjacency relationship between any two leaves; Based on the graph structure, determine the original adjacency matrix and degree matrix; the elements in the original adjacency matrix... Indicates blade and leaves Are they adjacent? Indicates blade and leaves Adjacent, Indicates blade and leaves Not adjacent; degree matrix is The matrix, The number of blades is represented by the element on the main diagonal of the degree matrix, which represents the degree of the corresponding node, while other elements are 0.

[0011] Optionally, the formula for calculating the normalized adjacency matrix is: in, This is the original adjacency matrix. For degree matrix, It is an identity matrix.

[0012] This invention provides a device for measuring the dynamic stress field of a damping blade, comprising: The acquisition module is used to acquire the tip displacement data of the entire circumference of the damping blades of the rotating equipment through the tip timing sensor array; the tip timing sensor array includes multiple tip timing sensors, which are arranged at intervals along the circumference of the blade. The first determining module is used to determine the vibration frequency and response of each blade by extracting modal information and phase changes of adjacent blades based on the tip displacement data of the entire damping blade; the response includes displacement amplitude and vibration phase. The second determining module is used to input the vibration frequency and response of each blade into a pre-trained dynamic stress field prediction neural network model to obtain the dynamic stress field of the entire damped blade.

[0013] The present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for measuring the dynamic stress field of a damping blade.

[0014] The present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the above-mentioned method for measuring the dynamic stress field of the damping blade.

[0015] The above-mentioned at least one technical solution adopted in this invention can achieve the following beneficial effects: In this invention, a circumferentially spaced array of blade tip timing sensors ensures that the movement of each blade tip is monitored, enabling the acquisition of blade tip displacement data for the entire circumference of all blades. This overcomes the limitations of traditional contact-based measurements. Based on the blade tip displacement data for the entire circumference, modal information and phase changes of adjacent blades are extracted to determine the vibration frequency and response of each blade. Furthermore, a neural network model is integrated to map the vibration frequency and response of each blade to the dynamic stress field of the entire circumference of blades. This enables accurate prediction of the dynamic stress distribution of damped blades under complex operating conditions, thereby improving the accuracy of blade operation monitoring. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0017] Figure 1 A schematic flowchart of a method for measuring the dynamic stress field of a damping blade provided by the present invention; Figure 2 A schematic diagram of the structure of a vibration identification model provided by the present invention; Figure 3 This is a schematic diagram of the vibration response of the damped blade pitch diameter throughout its entire rotation. Figure 4This is a schematic diagram of the structure of a neural network model for predicting dynamic stress field provided by the present invention; Figure 5 A schematic diagram of another method for measuring the dynamic stress field of a damping blade provided by the present invention; Figure 6 A schematic diagram of the structure of a dynamic stress field measurement system for a damped blade provided by the present invention; Figure 7 A schematic diagram of a computer device for measuring the dynamic stress field of a damping blade, provided by the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0019] The complex pitch diameter vibration characteristics of damped ring blades make it difficult to accurately establish traditional analysis models. Finite element models require a large degree of freedom, and dynamic stress analysis methods suffer from high computational costs, poor real-time performance, and low accuracy. Therefore, there is an urgent need for a novel method that can acquire the vibration response of ring blades online and efficiently predict the dynamic stress field.

[0020] The blade tip timing measurement method is a non-contact, simple vibration monitoring technique. It uses high-response sensors deployed around the rotor to acquire the real-time transit time of the blade tip rotation and calculates the blade vibration displacement using the deviation between this time and the theoretical arrival time. This method does not interfere with blade operation and is suitable for long-term monitoring, thus offering significant advantages in rotor system vibration identification. However, traditional blade tip signal identification algorithms are only applicable to single-blade vibration identification and cannot directly reflect the pitch diameter coupling characteristics and complete response modes of blades in a coiled damping structure. Therefore, this invention proposes a dynamic stress field measurement method for damped blades. First, a non-contact sensor array is used to acquire real-time blade tip signals for the entire coil. Then, a neural network model is introduced to identify the pitch diameter vibration modes of the coiled blades online. Further, deep learning methods are integrated to map the vibration characteristics to the dynamic stress field of the entire coiled blades, enabling rapid and accurate prediction of the dynamic stress distribution of damped blades under complex operating conditions. This invention's method possesses good real-time performance and adaptability, providing a new path for intelligent monitoring of high-performance turbine machinery.

[0021] The technical solutions provided by the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0022] Figure 1This is a schematic diagram of a method for measuring the dynamic stress field of a damped blade according to the present invention, which specifically includes the following steps: S101, the blade tip displacement data of the entire circle of damping blades of the rotating equipment is collected by the blade tip timing sensor array; the blade tip timing sensor array includes multiple blade tip timing sensors, which are arranged at intervals along the circumference of the blade.

[0023] The blade tip timing sensor array includes a speed sensor and multiple blade tip timing sensors arranged in an array. The blade tip timing sensors can be fiber optic, capacitive, or eddy current displacement sensors. The array is arranged with N blade tip timing sensors to improve frequency reconstruction accuracy.

[0024] The method for determining the installation positions of multiple blade tip timing sensors specifically includes: determining the installation position of each blade tip timing sensor based on the spacing between the blade sheaths and the impeller structure.

[0025] The blade tip sensors are arranged reasonably according to the spacing between the blade sheaths and the impeller structure to ensure that each blade tip sensor can cover the blade tip movement trajectory of the corresponding area, avoid signal interference and obstruction, and meet the requirements of measurement angle and installation space.

[0026] During the operation of the rotating equipment, N tip timing sensors in the tip timing sensor array can be controlled to collect tip displacement data for the entire rotation of the damped blades, i.e. There are N sets of data in total.

[0027] S102, based on the tip displacement data of the entire damping blade, the vibration frequency and response of each blade are determined by extracting modal information and phase changes of adjacent blades; the response includes displacement amplitude and vibration phase.

[0028] In one embodiment, based on the tip displacement data of the entire damping blade, the vibration frequency and response of each blade are determined by extracting modal information and phase changes of adjacent blades, including: inputting the tip displacement data of the entire damping blade into a vibration identification model, extracting modal information and phase changes of adjacent blades through the vibration identification model, and determining the vibration frequency and response of each blade.

[0029] The vibration recognition model employs a one-dimensional convolutional neural network model, the structure of which is as follows: Figure 2 As shown, the vibration recognition model includes downsampling layers, multiple stacked convolutional modules, and fully connected layers connected in series; each convolutional module includes convolutional layers, batch normalization, and activation functions connected in series.

[0030] This invention uses the tip displacement data of the entire damping blade. Input into the vibration identification model, where, For the first Displacement data for each blade, This refers to the number of blades. First, the input blade tip displacement data... Downsampling is performed, i.e., 1D max pooling is used, assuming the pooling window size is... The pooled sequence for:

[0031] in, This indicates a max pooling operation. This is the size of the pooling window.

[0032] This invention uses multiple small kernel-sized convolutional modules stacked together to extract local periodic features, including modal information and phase changes between adjacent blades.

[0033] Taking the convolutional layer of the first convolutional module as an example, its input is a sequence signal of shape 1×N. This layer has P filters (convolution kernels), each with a size of 1×K, used for local sliding convolution operations in the time (leaf sequence) dimension. Specifically, the first... i Output feature map y i In position j The formula for calculation at this location is:

[0034] (1) in, For the first i The weight vector of each filter. For the corresponding bias; to ensure the output length is aligned with the input, padding is added to both ends of the sequence. One zero, For the input sequence signal ( ).

[0035] This invention introduces batch normalization between each convolutional layer and the activation function to accelerate training convergence and suppress internal covariate shift. Let the output of a certain convolutional layer be the first... i One channel, j The prenormalized value at each position is z i,j The normalization calculation process is as follows:

[0036] S1, calculate the mean and variance of this channel in the current batch: (2) in For batch size, The sequence length is given.

[0037] S2, Standardized and Linear Transformation: (3) in , For learnable scaling and offset coefficients, To prevent division by zero of small constants.

[0038] The activation function used in this invention is LeakyReLU to meet the requirements of regression tasks and training stability. Its formula is as follows: (4) The mathematical expression for a fully connected layer is as follows: (5) in, The output value of the fully connected layer; W This is the weight matrix; b This is the bias value; l Number of floors; i For the first i One neuron.

[0039] Specifically, the fully connected layer integrates the high-dimensional features of the input, with a subset of neurons used to extract global modal features and output the vibrational frequency ω. This process can be represented as:

[0040] (6) in, These are the high-dimensional features extracted by the stacked convolutional modules. This is the weight matrix. For bias terms, For activation function, This provides high-dimensional features for the input fully connected layer.

[0041] Another group of neurons maps local features and outputs the response features of each leaf (such as displacement amplitude). and vibration phase The process is as follows:

[0042] (7) in, The weight matrix is ​​the local response mapping. For bias terms, This is the activation function.

[0043] This invention uses weighted mean square error as the loss function, which can optimize the prediction accuracy of the whole-cycle modal features and the response of each blade. Optionally, the loss function of the vibration recognition model during the training process... for:

[0044] (8) in, and These represent the actual nodal diameter vibration frequency and the full-circle damped blade nodal diameter vibration frequency predicted by the vibration identification model, respectively. For the number of leaves, and The vibration identification model predicts the first Displacement amplitude and vibration phase of each blade, and Each is the real first Displacement amplitude and vibration phase of each blade, and The weight of the loss term; It is the Euclidean norm.

[0045] The training process of the vibration recognition model includes: firstly, using a random sampling method to divide the dataset... Tr The dataset is divided into training and test sets in an 8:2 ratio. The training set is then input... Figure 2 The network is trained using a specific network structure. The parameters of each layer are iteratively updated with continuous input from the training set until a prediction accuracy of 98% is achieved, at which point training stops. After training stops, the accuracy is verified using a test set. In this invention, a test set accuracy of 96% or higher is considered sufficient for the neural network to meet the requirements.

[0046] Once the training meets the prediction accuracy requirements, the corresponding trained vibration recognition model can be obtained.

[0047] like Figure 3 As shown, Figure 3 This is a schematic diagram of the vibration response of the entire damped blade pitch diameter.

[0048] S103, the vibration frequency and response of each blade are input into the pre-trained dynamic stress field prediction neural network model to obtain the dynamic stress field of the entire damped blade.

[0049] Optionally, the dynamic stress field prediction neural network model includes a graph structure building module, multi-layer graph convolutional blocks, and a prediction head; the multi-layer graph convolutional blocks include... A series of concatenated graph convolutional sub-blocks, each including a graph convolutional layer, a batch normalization layer, and... Activation layer; the prediction head includes a fully connected layer; the vibration frequency and response of each blade are input into the dynamic stress field prediction model to obtain the dynamic stress field of the entire damped blade, including: S201, in the graph structure construction module, based on the adjacency relationship between each blade, an original adjacency matrix is ​​constructed. The original adjacency matrix is ​​then symmetrically normalized to obtain a normalized adjacency matrix. The vibration frequency, displacement amplitude, and vibration phase of each blade are then spliced ​​together to construct a multidimensional feature vector. Finally, the multidimensional feature vectors of all blades are combined to form a node feature matrix.

[0050] The graph structure construction module also includes: treating each leaf as a node and constructing a graph structure based on the adjacency relationship between any two leaves; determining the original adjacency matrix and degree matrix based on the graph structure; and the elements in the original adjacency matrix. Indicates blade and leaves Are they adjacent? Indicates blade and leaves Adjacent, Indicates blade and leaves Not adjacent; degree matrix is The matrix, The number of blades is represented by the element on the main diagonal of the degree matrix, which indicates the degree of the corresponding node; other elements are 0. The adjacency relationships between blades are defined based on the physical topology of the impeller.

[0051] Optionally, the formula for calculating the normalized adjacency matrix is: (9) in, This is the original adjacency matrix. For degree matrix, It is an identity matrix.

[0052] S202, in each graph convolutional sub-block, the graph convolutional layer propagates information about the node features output from the previous layer based on the normalized adjacency matrix, and then sequentially passes the propagated node features through the batch normalization layer and... The activation function is used to obtain the node features output by this layer; the node features output by the previous layer used in the first graph convolutional sub-block are the node feature matrix; the calculation formula for the graph convolutional sub-block is: (10) in, For the first Layer node characteristics, For the first Layer node characteristics, The weight matrix is ​​a learnable matrix. for Activation function For batch normalization operations, This is the normalized adjacency matrix.

[0053] The output of a multi-layer graph convolutional block is the output of the last layer graph convolutional sub-block.

[0054] The graph convolutional sub-blocks standardize the feature matrix to have a mean of 0 and a variance of 1; this is achieved through stacking. Layered convolutional sub-blocks gradually expand the receptive field of node features, transforming the features of each node from the vibration attributes of a single blade into spatial coupling features between the entire ring of blades.

[0055] S203, in the prediction head, a fully connected layer is used to perform linear transformation and dimensionality compression on the node features output by the last layer's graph convolutional sub-block, obtaining the dynamic stress field of the entire damping blade; the calculation formula for the fully connected layer is: (11) in, This represents the dynamic stress field of the entire damping blade. This is the weight matrix of the fully connected layer. For bias terms, This refers to the node features output by the convolutional sub-block of the last layer of the graph.

[0056] Include The dynamic stress vector of each element completely describes the dynamic stress field distribution of the entire blade circumference, and a circumferential dynamic stress distribution diagram can be drawn according to the blade number.

[0057] The prediction head can include multiple fully connected layers. The input of the first fully connected layer is the output of the multi-layer graph convolutional block. The input of each subsequent layer is the output of the previous layer. The output of the last layer is the output of the prediction head, which is the dynamic stress field of the entire damping blade.

[0058] Neural network models for predicting dynamic stress fields, such as Figure 4 As shown, it includes four parts: graph structure building module, multi-layer graph convolutional block, prediction head, and loss function.

[0059] Optionally, the graph structure construction module in this invention is mainly used to implement node definition, edge relationship construction, and adjacency matrix normalization. Taking node definition as an example, each node carries a feature vector extracted by the previous-order neural network, including: displacement response amplitude, velocity amplitude, and modal coefficients. Let the total number of nodes be... The initial node feature matrix is ​​then represented as:

[0060] (12) in, The feature dimensions for each node.

[0061] This invention employs topological and neighborhood criteria to determine the connectivity between top nodes. Specifically, if two nodes share an element face or edge in the finite element mesh, an undirected edge is established between them; if the spatial distance between the two nodes does not exceed a preset threshold... r If the two are physically connected, then an edge is established.

[0062] In this invention, a symmetric normalization strategy is used to generate a normalized adjacency matrix, the mathematical expression of which is: (13) (14) (15) (16) In the convolutional neural network model of the dynamic stress prediction method of this invention, the multi-layer graph convolutional block module is the core computing unit, used to extract the spatial correlation features between nodes layer by layer. It is composed of concatenated sub-blocks (GCNBlockL) with the same structure, and the input of each sub-block is the node feature matrix output by the previous layer. The output is First, graph convolution is performed in the sub-block, with the following formula: .

[0063] in, This is the adjacency matrix after symmetric normalization. Features of the l-th layer nodes; The weight matrix is ​​trainable. These are features that were not activated after convolution.

[0064] In this invention, the Rectified Luminous Interval (ReLU) function is selected as the activation unit for the neural network to improve the convergence speed of the neural network. The formula is as follows:

[0065] (17) in, The output of the operation; for The activation function.

[0066] In this invention, the optimizer for the convolutional neural network is an adaptive moment estimation optimizer, and the node-level mean squared error is selected as the loss function to evaluate the predicted value. The calculation formula is as follows: (18) in, The total number of nodes. The model predicts the first Each node stress value This represents the actual stress value.

[0067] The training process of the neural network model for predicting dynamic stress fields includes: firstly, using random sampling to divide the dataset... Tr The dataset is divided into training and test sets in an 8:2 ratio. The training set is then input... Figure 4Training is performed in the corresponding convolutional neural network. The parameters of each layer of the neural network are iteratively updated with continuous input from the training set until the prediction accuracy reaches 98%, at which point training stops. After training stops, the accuracy is verified using a test set. In this invention, a test set accuracy of 96% or higher is considered sufficient for the neural network to meet the requirements.

[0068] Once the training meets the prediction accuracy requirements, the corresponding trained convolutional neural network model can be obtained.

[0069] After the above three steps, the online measurement of the dynamic stress field of the damped blade proposed in this invention is completed. For this measurement method, the new measurement signal measured by any sensor is input into the neural network model for predicting the dynamic stress field of the blade under complex pitch vibration trained in step S103 after online identification of the whole-circle blade pitch vibration in step S102, so as to quickly obtain the dynamic stress field of the whole-circle damped blade.

[0070] In a specific embodiment, such as Figure 5 As shown, the present invention also provides a method for measuring the dynamic stress field of a damping blade, which includes the following steps: S501, build a timed measurement system for the entire blade tip.

[0071] Among them, such as Figure 6 As shown, the whole-circle blade tip timing measurement system includes a test blade, a blade tip timing sensor array, a data acquisition system, and a host computer.

[0072] After installing the leaf tip timing sensor array according to the measurement requirements, connect it to the data acquisition system in sequence and check the signal and measurement effect. Adjust it according to the actual situation until the measurement requirements are met.

[0073] S502 measures and obtains the tip displacement data of the entire blade circumference.

[0074] S503 uses the proposed whole-circle blade pitch diameter vibration identification method to build a neural network and train the vibration identification model to meet the requirements.

[0075] S504, obtains the vibration frequency and response of the entire damped blade pitch diameter.

[0076] S505, designing convolutional neural network structures based on data characteristics.

[0077] S506. The proposed method for predicting blade dynamic stress under pitch vibration is used to build a convolutional neural network and train the dynamic stress field prediction neural network model to meet the requirements.

[0078] S507, obtain the dynamic stress distribution of the entire damping blade.

[0079] The S508 enables online measurement of the dynamic stress field of the entire damped blade.

[0080] When applying the dynamic stress field measurement method for damped blades provided by this invention, it is not necessary to rely on... Figure 1 The steps shown are executed in sequence. The specific execution order of each step can be determined as needed, and this invention does not impose any restrictions on it.

[0081] The above describes a method for measuring the dynamic stress field of a damping blade according to one or more embodiments of the present invention. Based on the same idea, the present invention also provides a corresponding device for measuring the dynamic stress field of a damping blade, the device comprising: The acquisition module is used to acquire the tip displacement data of the entire circumference of the damping blades of the rotating equipment through the tip timing sensor array; the tip timing sensor array includes multiple tip timing sensors, which are arranged at intervals along the circumference of the blade. The first determining module is used to determine the vibration frequency and response of each blade by extracting modal information and phase changes of adjacent blades based on the tip displacement data of the entire damping blade; the response includes displacement amplitude and vibration phase. The second determining module is used to input the vibration frequency and response of each blade into a pre-trained dynamic stress field prediction neural network model to obtain the dynamic stress field of the entire damped blade.

[0082] Specific limitations regarding the dynamic stress field measurement device for damped blades can be found in the limitations of the dynamic stress field measurement method for damped blades mentioned above, and will not be repeated here. Each module in the aforementioned dynamic stress field measurement device for damped blades can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0083] The present invention also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 The provided method for measuring the dynamic stress field of damping blades.

[0084] The present invention also provides Figure 7 The schematic diagram of the computer device shown is as follows: Figure 7 As shown, at the hardware level, this computer device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then executes it to achieve the above. Figure 1 The provided method for measuring the dynamic stress field of damping blades.

[0085] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0086] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this invention.

Claims

1. A method for measuring the dynamic stress field of a damped blade, characterized in that, include: The blade tip displacement data of the entire rotation of the damping blades of the rotating equipment are collected by a blade tip timing sensor array; The blade tip timing sensor array includes multiple blade tip timing sensors, which are arranged at circumferential intervals along the blade. Based on the tip displacement data of the entire damping blade, the vibration frequency and response of each blade are determined by extracting modal information and phase changes of adjacent blades; the response includes displacement amplitude and vibration phase. The vibration frequency and response of each blade are input into a pre-trained dynamic stress field prediction neural network model to obtain the dynamic stress field of the entire damped blade.

2. The method according to claim 1, characterized in that, The method for determining the installation locations of multiple blade tip timing sensors specifically includes: The installation position of each blade tip timing sensor is determined based on the spacing between the blade sheaths and the impeller structure.

3. The method according to claim 1, characterized in that, Based on the tip displacement data of the entire damping blade, the vibration frequency and response of each blade are determined by extracting modal information and phase changes of adjacent blades, including: The tip displacement data of the entire damping blade is input into the vibration identification model. The vibration identification model extracts modal information and phase changes of adjacent blades to determine the vibration frequency and response of each blade. The vibration identification model includes a downsampling layer, multiple stacked convolutional modules, and a fully connected layer connected in series. Each convolutional module includes a convolutional layer, batch normalization, and activation function connected in series.

4. The method according to claim 3, characterized in that, Loss function of vibration recognition model during training for: ； in, and These represent the actual nodal diameter vibration frequency and the full-circle damped blade nodal diameter vibration frequency predicted by the vibration identification model, respectively. For the number of leaves, and The vibration identification model predicts the first... Displacement amplitude and vibration phase of each blade, and Each is the real first Displacement amplitude and vibration phase of each blade, and The weight of the loss term; It is the Euclidean norm.

5. The method according to claim 1, characterized in that, The dynamic stress field prediction neural network model includes a graph structure building module, multi-layer graph convolutional blocks, and a prediction head; the multi-layer graph convolutional blocks include... A series of concatenated graph convolutional sub-blocks, each including a graph convolutional layer, a batch normalization layer, and... Activation layer; The prediction head includes a fully connected layer; the step of inputting the vibration frequency and response of each blade into the dynamic stress field prediction model to obtain the dynamic stress field of the entire damped blade includes: In the graph structure construction module, an original adjacency matrix is ​​constructed based on the adjacency relationship between each blade. The original adjacency matrix is ​​then symmetrically normalized to obtain a normalized adjacency matrix. The vibration frequency, displacement amplitude, and vibration phase of each blade are then spliced ​​together to construct a multidimensional feature vector. The multidimensional feature vectors of all blades are then combined to form a node feature matrix. In each graph convolutional sub-block, the graph convolutional layer propagates information about the node features output from the previous layer based on the normalized adjacency matrix, and then sequentially passes the propagated node features through batch normalization layers and... The activation function is used to obtain the node features output by this layer; the node features output by the previous layer used in the first graph convolutional sub-block are the node feature matrix; the calculation formula for the graph convolutional sub-block is: in, For the first Layer node characteristics, For the first Layer node characteristics, The weight matrix is ​​a learnable matrix. for Activation function For batch normalization operations, This is the normalized adjacency matrix; In the prediction head, a fully connected layer is used to perform linear transformation and dimensionality compression on the node features output by the last layer of graph convolutional sub-blocks to obtain the dynamic stress field of the entire damping blade; the calculation formula for the fully connected layer is: in, This represents the dynamic stress field of the entire damping blade. This is the weight matrix of the fully connected layer. For bias terms, This refers to the node features output by the convolutional sub-block of the last layer of the graph.

6. The method according to claim 5, characterized in that, The graph structure building module also includes: Treat each leaf as a node and construct a graph structure based on the adjacency relationship between any two leaves; Based on the graph structure, determine the original adjacency matrix and degree matrix; the elements in the original adjacency matrix... Indicates blade and leaves Are they adjacent? Indicates blade and leaves Adjacent, Indicates blade and leaves Not adjacent; degree matrix is The matrix, The number of blades is represented by the element on the main diagonal of the degree matrix, which represents the degree of the corresponding node, while other elements are 0.

7. The method according to claim 5, characterized in that, The formula for calculating the normalized adjacency matrix is: in, This is the original adjacency matrix. For degree matrix, It is an identity matrix.

8. A device for measuring the dynamic stress field of a damped blade, characterized in that, include: The acquisition module is used to acquire the tip displacement data of the damping blades of the rotating equipment throughout the entire rotation using a blade tip timing sensor array; The blade tip timing sensor array includes multiple blade tip timing sensors, which are arranged at circumferential intervals along the blade. The first determining module is used to determine the vibration frequency and response of each blade by extracting modal information and phase changes of adjacent blades based on the tip displacement data of the entire damping blade; the response includes displacement amplitude and vibration phase. The second determining module is used to input the vibration frequency and response of each blade into a pre-trained dynamic stress field prediction neural network model to obtain the dynamic stress field of the entire damped blade.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

10. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as described in any one of claims 1 to 7.

Images