Rotor blade dynamic field fast reconstruction method based on deep neural network model

Abstract

The application discloses a kind of rotor blade dynamic field fast reconstruction methods based on deep neural network model, first acquisition includes the BTT signal of blade vibration information, then design deep neural network model, through training data set to deep neural network model is trained;Next using the deep neural network model of training to measured BTT signal is handled, reconstructs blade vibration displacement signal;Afterwards to the blade vibration displacement signal of reconstruction is carried out spectrum analysis, identifies out vibration frequency, amplitude, phase parameter;Then establish the finite element model of blade, by modal analysis determines the natural frequency and modal shape of blade;Finally utilize identified vibration parameter and modal analysis result, based on modal superposition theory fast calculation blade in vibration state under displacement field, strain field and stress field.The application can understand blade vibration displacement and stress condition in real time, from support real-time decision, improve the safety, reliability, durability of rotating machinery.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of mechanical blades, and particularly relates to a rotor blade dynamic field rapid reconstruction method based on a deep neural network model. BACKGROUND

[0002] Blades are an important component in rotating machinery, especially in turbines and compressors. The shape and design of the blades directly affect the efficiency and performance of the machine. However, blades work at high speed and in complex environments, and are prone to vibration due to factors such as fluid excitation, imbalance, mechanical looseness, and structural defects. Vibration problems not only affect the normal operation of the equipment, but also can cause fatigue failure and shorten the service life of the blades, and even cause serious safety accidents. Therefore, vibration monitoring of blades is of great significance for the reliability, safety, performance optimization, and life extension of rotating machinery. Through online monitoring and other means, the vibration characteristics of the blades under different rotational speeds and working conditions can be deeply understood, thereby supporting the identification and prevention of blade failure, extending the service life of the equipment, improving the operation reliability, and ensuring the safety of the equipment and personnel.

[0003] Dynamic strain measurement is an important part of blade vibration monitoring research. Traditional dynamic strain measurement methods mainly rely on strain gauge method, which involves pasting strain gauges on the surface of rotating blades, combining slip ring power supply and telemetry technology to achieve strain measurement. This contact measurement method has many limitations. First, the installation process of the strain gauge is very strict, often requiring modification of the rotor system, increasing the complexity of the structure. Second, the presence of strain gauges and their connecting leads increases the additional mass of the blades, especially for small structures. In addition, strain gauges are prone to failure in high temperature and high pressure environments, resulting in inaccurate measurement results. Strain gauge method also has the problems of difficult installation, easy to fall off at high speed, many additional devices, large influence on flow field, long measurement period, and high measurement cost. Especially when multiple points need to be measured, strain gauges need to be repeatedly arranged, further increasing the measurement difficulty and cost. These shortcomings make it difficult for the strain gauge method to meet the requirements of real-time monitoring of blade dynamic strain field.

[0004] BTT is a non-contact measurement technology that installs sensors on the circumference of rotating equipment to record the time when the blade tip passes through the sensor position. By analyzing the time deviation of the blade tip actually passing through the sensor from the ideal non-vibration state passing through the sensor, the displacement of the blade tip at that moment is calculated. Then the complete displacement curve of the blade tip during rotation is reconstructed using signal processing methods, and the vibration frequency, amplitude and phase of the blade are obtained through frequency spectrum analysis. Based on the finite element modal analysis of the blade and the vibration displacement of the tip, the displacement field, strain field and stress field of the entire blade can be obtained using modal superposition theory. By continuously monitoring the vibration and strain of the blade, abnormalities can be detected in time to provide early warning information and avoid catastrophic failures.

[0005] Since the engine speed is relatively fixed, the number of sensors installed on the casing directly affects the sampling frequency of the blade tip timing device. In order to make the sampling frequency meet the Nyquist criterion, the number of sensors to be installed will be large, which is difficult to achieve in actual engineering use. Therefore, identifying key information such as complex blade vibration frequency and amplitude through undersampling signals is a problem that needs to be focused on in blade tip timing vibration measurement. The existing BTT signal reconstruction method is often based on the assumption of constant speed and requires prior information, so it is sensitive to noise, has low reconstruction efficiency, and the reconstruction accuracy is limited by the sampling frequency and sparsity assumption. SUMMARY

[0006] In order to overcome the shortcomings of the prior art, the present application provides a rotor blade dynamic field fast reconstruction method based on a deep neural network model. First, the BTT signal containing blade vibration information is collected, then a deep neural network model is designed, the deep neural network model is trained through a training data set, and the mapping relationship from the BTT signal to the complete vibration signal is learned. Next, the trained deep neural network model is used to process the measured BTT signal, and the blade vibration displacement signal is reconstructed. Then, the reconstructed blade vibration displacement signal is analyzed by spectrum, and the vibration frequency, amplitude and phase parameters are identified. Then, the finite element model of the blade is established, the natural frequency and modal shape of the blade are determined through modal analysis, and the modal response of the blade is analyzed in combination with the vibration parameters. Finally, the identified vibration parameters and modal analysis results are used to quickly calculate the displacement field, strain field and stress field of the blade in the vibration state based on the modal superposition theory. The present application can realize real-time understanding of the vibration displacement and stress condition of the blade, support real-time decision-making, and improve the safety, reliability and durability of the rotating machinery.

[0007] The technical scheme adopted by the present application to solve its technical problems is as follows:

[0008] Step 1: Collect the BTT signal containing blade vibration information, and perform denoising and normalization preprocessing;

[0009] Step 2: Design a deep neural network model, train the deep neural network model through a training data set, and learn the mapping relationship from the BTT signal to the complete vibration signal;

[0010] Step 3: Use the trained deep neural network model to process the measured BTT signal, and reconstruct the blade vibration displacement signal;

[0011] Step 4: Perform spectrum analysis on the reconstructed blade vibration displacement signal, and identify the vibration frequency, amplitude and phase parameters;

[0012] Step 5: Establish a finite element model of the blade, determine the natural frequency and modal shape of the blade through modal analysis, and analyze the modal response of the blade in combination with the vibration parameters;

[0013] Step 6: Using the identified vibration parameters and modal analysis results, quickly calculate the displacement field, strain field and stress field of the blade in the vibration state based on the modal superposition theory.

[0014] Further, the step 1 is specifically:

[0015] Step 1-1: Sensor installation;

[0016] Install laser sensors on the outer fixed parts of the machine according to the specified angle distribution in the circumferential direction, and install encoders at the shaft ends;

[0017] Step 1-2: Blade position calibration;

[0018] Before starting operation, calibrate the order of each blade passing through the sensor position, and record the time of different blades passing through different BTT sensors;

[0019] Step 1-3: Data acquisition;

[0020] When the machine starts to run, each blade passing through the sensor will produce a pulse signal, and the time stamp of the pulse signal is recorded into the computer through the data acquisition system;

[0021] Step 1-4: Time synchronization and data processing;

[0022] Calculate the vibration deflection angle based on the arrival time and rotational speed:

[0023]

[0024] Where, T i k is the time when the kth blade passes through the ith BTT sensor, Ω n is the average rotational speed of the nth revolution, is the theoretical time of the kth blade passing through the ith BTT sensor under vibration conditions obtained from the encoder, is the vibration angle of the kth blade passing through the ith BTT sensor;

[0025] Step 1-5: Assign to the blade

[0026] All i=1,…,I; k=1,…,K signals collected by multiple sensors are correctly assigned to each blade, and the tip displacement sequence of the kth blade in the nth revolution is:

[0027]

[0028] where x i is the displacement, R is the distance from the blade tip to the center of rotation, and Θ is the vibration deflection angle. The tip displacement sequence x={x1,…,x I of each cycle is concatenated to obtain the BTT signal of the blade within a certain time range;

[0029] Steps 1-6: The displacement time sequence of each blade is zeroed, normalized, de-energized, and filtered to eliminate the static displacement part and noise of the blade.

[0030] Further, the step 2 is specifically:

[0031] Step 2-1: Select CNN+LSTM as the deep neural network model, use CNN to extract signal spatial features, and use LSTM to capture the time dynamics and long-time span correlation in the signal;

[0032] Step 2-2: Based on the simulated vibration displacement signal, a training set and a test set are established. The signal after BTT measurement matrix linear mapping is preprocessed as the input of the deep neural network model, and the complete simulated vibration displacement signal is as the output of the deep neural network model;

[0033] Step 2-3: Train the designed deep neural network model, and learn the nonlinear mapping relationship between the input and the output by adjusting the model parameters;

[0034] Step 2-4: Evaluate the reconstruction ability and accuracy of the model through cross-validation and independent test set, and adjust the hyperparameters to optimize the model performance.

[0035] Further, the step of establishing a training set and a test set based on the simulated vibration displacement signal comprises:

[0036] Step 2-2-1: Collect and compress the tip vibration signal of the simulated signal , where N represents the number of multi-modal resonance, EO i is the corresponding frequency order, A i and are the amplitude and phase, respectively;

[0037] Step 2-2-2: The original tip vibration data set is represented as X={x (1) ,...,x (i) ,...,x (l)}; In this data set, represents the i-th vibration signal, with n sampling points; The compressed vibration signal is represented as Y={y (1) ,...,y (i)..., y (l)}, wherein represents the ith measurement signal, There are m sampling points; the mapping from the original signal X to the compressed vibration signal Y is a linear mapping, that is, Y = φX, wherein is a linear mapping matrix determined by the installation angle of the BTT sensor; the acquisition process of the ith compressed signal is represented as:

[0038] y (i) = φ m×n x (i) (3)

[0039] Step 2-2-3: Dimensional processing is performed on the compressed measurement signal y(i) to obtain a proxy signal consistent with the length of the complete signal:

[0040] z (i) = φ T y (i) (4)

[0041] The proxy signal is subjected to Z-score normalization:

[0042]

[0043] Further, the sample input of the training set is set in the pre-set deep neural network model, the loss value is calculated by the pre-set loss function, the model parameters are adjusted by using the loss value, and a trained deep neural network model is obtained.

[0044] Further, the deep neural network model determines the loss function by mean square error MSE as:

[0045]

[0046] Wherein, W and b are the weight and bias of the network, respectively.

[0047] Further, the loss function uses the back propagation algorithm to minimize the loss value and learn the related parameters using Adam as the optimizer.

[0048] Further, the step 2-4 is specifically:

[0049] The percentage root mean square difference PRD is used to evaluate the quality of signal reconstruction, which is defined as follows:

[0050]

[0051] Wherein, x and are the original vibration signal and signal reconstruction, respectively, and ||·||2 is the 2-norm.

[0052] Further, in step 3, the pre-processed signal r of the BTT measurement system is taken as input, and a nonlinear mapping relationship from the pre-processed signal to the original signal x is fitted by the trained deep neural network model, and the reconstructed signal is is represented as:

[0053]

[0054] where H represents the deep neural network model.

[0055] Further, in step 4, the frequency spectrum analysis is performed on the reconstructed signal x by fast Fourier transform (FFT) to convert the signal from the time domain to the frequency domain, and the frequency components are found from the frequency spectrum and the amplitude and phase vibration parameters are extracted.

[0056] Further, step 5 is specifically:

[0057] Step 5-1: Establish a three-dimensional geometric model of the blade, perform meshing on the geometric model and perform mesh independence test;

[0058] Step 5-2: Define material properties according to the actual blade material, including density, Young's modulus and Poisson's ratio;

[0059] Step 5-3: Select the type of modal analysis, and determine the order of solution of modal analysis, Coriolis effect and rotational speed;

[0060] Step 5-4: Obtain the motion equation and modal equation by finite element method;

[0061]

[0062] where M is the mass matrix, C is the damping matrix, K is the stiffness matrix, u(t) is the displacement vector, G is the gyroscopic effect matrix due to Coriolis force, K Ω is the stiffness matrix due to centrifugal force;

[0063] Step 5-5: Record the natural frequency of each mode by solving the eigenvalue problem, draw the Campbell diagram, extract the displacement, strain and stress modal vectors of each mode and analyze the deformation characteristics of the blade under different modes.

[0064] Further, step 6 is specifically:

[0065] Step 6-1: Compare the vibration frequency components of the reconstructed signal obtained by the deep neural network model with the Campbell diagram to determine the closest vibration mode;

[0066] Step 6-2: Calculate the participation factor of the main mode:

[0067] ​

[0068] wherein, ω i , A i and θ i are the frequency, amplitude and phase of the main vibration component of the reconstructed signal, respectively, and φ p,i is the displacement of the blade tip measurement point in the vibration mode closest to the main vibration component;

[0069] Step 6-3: Calculate the full-field displacement, strain and stress based on the modal superposition theory as follows:

[0070]

[0071] wherein, C i (t) is the time-varying coefficient of the vibration mode closest to the i-th order vibration principal component, and is the displacement vector, strain vector or stress vector of the vibration mode, and f(t) is the structure displacement field, strain field or stress field vector at time t.

[0072] The beneficial effects of the present application are as follows:

[0073] The method of the present application avoids the traditional strain gauge measurement by using non-contact BTT measurement; the complex nonlinear relationship and features between the sparse BTT signal and the complete signal can be automatically learned by using the deep neural network model, so as to improve the reconstruction accuracy and robustness, and no prior information is needed and the sampling conditions are not limited; the vibration displacement and stress condition of the blade can be understood in real time based on the reconstructed field information, which supports real-time decision making, and improves the safety, reliability and durability of the rotating machinery. BRIEF DESCRIPTION OF DRAWINGS

[0074] Figure 1 is a flowchart of the method of the present application;

[0075] Figure 2 is a structural schematic diagram of a non-contact measurement system for blade field reconstruction provided by the present application;

[0076] Figure 3 is a blade vibration signal and a rotating speed curve measured by the non-contact measurement system in the embodiment of the present application;

[0077] Figure 4 is a schematic diagram of the CNN-LSTM network structure provided in the embodiment of the present application;

[0078] Figures 5(a) to 5(b) is a complete time-domain signal of the X-direction displacement of the blade tip No. 2 and an amplitude-frequency curve reconstructed by the CNN-LSTM network in the embodiment of the present application;

[0079] Figure 6is a three-dimensional finite element model of the blade established in the embodiment of the application;

[0080] Figure 7 is a Campbell diagram of the blade in the embodiment of the application;

[0081] Figures 8(a) to 8(c) is an X-direction displacement cloud map, a Y-direction strain cloud map and a Y-direction stress cloud map of the No. 2 blade at a certain moment reconstructed by using the field reconstruction theory in the embodiment of the application. DETAILED DESCRIPTION

[0082] The application will be further described below in combination with the drawings and embodiments.

[0083] The application aims to provide a rotor blade dynamic field fast reconstruction method based on a deep neural network model, which can reconstruct the dynamic displacement, strain and stress field of a high-speed rotating blade under complex working conditions in real time and accurately under non-contact conditions by combining the BTT technology with the deep learning model, and provide reliable technical support for health monitoring and fault early warning of rotating machinery.

[0084] In view of the existing problems in sparse BTT signal reconstruction and dynamic field prediction, the application relates to a rotating blade dynamic field fast reconstruction method based on a deep neural network model, which provides reference and guidance for fault diagnosis and preventive maintenance of rotating machinery in actual engineering applications.

[0085] A rotating blade dynamic field fast reconstruction method based on a deep neural network model, comprising the following steps:

[0086] Step 1: Collect BTT signals containing blade vibration information, and perform preprocessing such as denoising and normalization;

[0087] Step 2: Design a suitable deep neural network model, train the model through a training data set, and learn the mapping relationship from the BTT signal to the complete vibration signal;

[0088] Step 3: Use the trained deep neural network model to process the measured BTT signal, and reconstruct the vibration displacement signal of the blade;

[0089] Step 4: Perform frequency spectrum analysis on the reconstructed vibration displacement signal, and identify key vibration parameters such as vibration frequency, amplitude and phase;

[0090] Step 5: Establish a finite element model of the blade, determine the natural frequency and modal shape of the blade through modal analysis, and analyze the modal response of the blade in combination with the vibration parameters;

[0091] Step 6: Use the identified vibration parameters and modal analysis results to quickly calculate the displacement field, strain field and stress field of the blade in the vibration state based on the modal superposition theory.

[0092] In one embodiment, the process of collecting BTT vibration signals and preprocessing by BTT measurement device includes the following steps:

[0093] (1) Sensor installation: Install a certain number of laser sensors on the outer fixed parts of the machine in a certain angle along the circumference, and install an encoder at the shaft end;

[0094] (2) Blade position calibration: Before starting operation, calibrate the order of each blade passing through the sensor position to accurately record the time of different blades passing through different BTT sensors;

[0095] (3) Data acquisition: When the machine starts running, each blade passing through the sensor will generate a pulse signal, and the time stamp of the pulse signal is recorded into the computer through the data acquisition system;

[0096] (4) Time synchronization and data processing: Calculate the vibration deflection angle based on the arrival time and rotational speed:

[0097]

[0098] Where T i k is the time when the kth blade passes through the ith BTT sensor, Ω n is the average rotational speed of the nth revolution, is the theoretical time of the kth blade passing through the ith BTT sensor under vibration condition obtained by the encoder, is the vibration angle of the kth blade passing through the ith BTT sensor.

[0099] (5) Assign to each blade and calculate the blade tip displacement: correctly assign all signals collected by multiple sensors to each blade, and the tip displacement sequence of the kth blade in the nth revolution is:

[0100]

[0101] Where x i is the displacement, R is the distance from the blade tip to the center of rotation, and θ is the vibration deflection angle. The tip displacement sequence x = {x1, …, x I} of each revolution is concatenated to obtain the BTT signal of the blade within a certain time range;

[0102] (6) Zeroize, normalize, de-energize and filter the displacement time sequence of each blade to eliminate the static displacement part of the blade and noise.

[0103] In one embodiment, the process of designing and training a deep neural network model includes the following steps:

[0104] (1) Select the deep neural network model of CNN+LSTM, use CNN to extract signal spatial features, and use LSTM to capture the temporal dynamics and correlation of long time span in the signal.

[0105] (2) Training and test sets are established based on simulated vibration displacement signals. The signals linearly mapped by the BTT measurement matrix are used as inputs to the deep neural network model after preprocessing, and the complete simulated vibration displacement signals are used as outputs of the deep neural network model.

[0106] (3) Train the designed neural network model and learn the nonlinear mapping relationship between input and output by adjusting the model parameters;

[0107] (4) Evaluate the model’s reconstruction capability and accuracy through cross-validation and independent test sets, and adjust hyperparameters to optimize model performance.

[0108] In one embodiment, the process of creating the training and test datasets for the deep neural network model includes:

[0109] (1) For the simulated signal of the blade tip vibration signal Data is collected and compressed, where N represents the number of multimodal resonances, and EO... i It corresponds to the frequency order, A. i and These are amplitude and phase, respectively.

[0110] (2) The original blade tip vibration dataset is represented as X={x (1) ,...,x (i) ,...,x (l) In this dataset, This represents the i-th vibration signal. There are n sampling points. The compressed vibration signal is represented as Y = {y} (1) ,...,y (i) ,...,y (l)},in This represents the i-th measurement signal. There are m sampling points. The mapping from the original signal X to the compression vibration signal Y is a linear mapping, i.e., Y = φX, where... The linear mapping matrix is ​​determined by the installation angle of the BTT sensor. The acquisition process of the i-th compressed signal is represented as:

[0111] y (i) =φ m×n x (i) (14)

[0112] (3) The compressed measurement signal y(i) is processed to obtain a proxy signal consistent with the length of the complete signal:

[0113] z (i) = φ T y (i) (15)

[0114] In order to accelerate network training, the proxy signal is Z-score normalized:

[0115]

[0116] In one embodiment, the samples in the training sample set are input into a pre-set deep learning model, the loss value is calculated by a pre-set loss function, the model parameters are adjusted using the loss value, and a trained deep learning model is obtained.

[0117] In one embodiment, the loss function is determined by mean square error (MSE):

[0118]

[0119] Where W and b are the weights and biases of the network, respectively.

[0120] In one embodiment, the loss value is minimized using the backpropagation algorithm with Adam as the optimizer to learn the related parameters.

[0121] In one embodiment, the reconstruction ability and accuracy of the model are evaluated by cross-validation and independent test sets, the hyperparameters are adjusted to optimize the model performance, and the percentage root mean square difference (PRD) is used to evaluate the quality of signal reconstruction, which is defined as follows:

[0122]

[0123] Where x and are the original vibration signal and signal reconstruction, respectively, and ||·||2 is the 2-norm.

[0124] In one embodiment, the preprocessed signal r of the BTT measurement system is input, the trained deep neural network model is used to fit the nonlinear mapping relationship from the preprocessed signal to the original signal x, and the reconstructed signal is represented as:

[0125]

[0126] Where H represents the deep neural network model.

[0127] In one embodiment, the reconstructed signal​ Convert from time domain signal to frequency domain, find out the main frequency components from the spectrum and extract the amplitude and phase vibration parameters.

[0128] In one of the embodiments, the specific steps of modal analysis of the rotating blade are as follows:

[0129] (1) Establish a three-dimensional geometric model of the rotating blade, mesh the geometric model and perform mesh independence test;

[0130] (2) According to the material of the actual blade, define the material properties, including density, Young's modulus and Poisson's ratio;

[0131] (3) Select the type of modal analysis, and determine the solution order of modal analysis, Coriolis effect and rotating speed;

[0132] (4) Obtain the motion equation and modal equation by finite element method:

[0133]

[0134] Where, M is the mass matrix, C is the damping matrix, K is the stiffness matrix, u(t) is the displacement vector, G is the gyro effect matrix due to Coriolis force, K Ω is the stiffness matrix due to centrifugal force;

[0135] (5) Record the natural frequency of each mode by solving the eigenvalue problem, draw the Campbell diagram, extract the displacement, strain and stress modal vectors of each mode and analyze the deformation characteristics of the blade under different modes.

[0136] In one of the embodiments, the fast reconstruction process of displacement, strain and stress field of the rotating blade specifically includes the following steps:

[0137] (1) Compare the vibration frequency components of the reconstructed signal obtained by the deep neural network model with the Campbell diagram, and determine the vibration mode closest to it;

[0138] (2) Calculate the modal coefficient of the main mode

[0139]

[0140] Where, ω i , A i and θ i are the frequency, amplitude and phase of the main vibration component of the reconstructed signal, φ p,i is the displacement of the blade tip measuring point in the vibration mode closest to the main vibration component;

[0141] (3) Calculate the full-field displacement, strain and stress based on modal superposition theory as:

[0142]

[0143] wherein C i (t) is the time-varying coefficient of the vibration mode closest to the i-th order vibration principal component, is the displacement vector, strain vector or stress vector of the vibration mode, and f(t) is the structural displacement field, strain field or stress field vector at time t.

[0144] Embodiment:

[0145] Referring to Figure 1 , one embodiment of the present application provides a rotor blade dynamic field fast reconstruction method based on a deep neural network model, which comprises the following steps:

[0146] Step 1: Collect BTT signals containing blade vibration information, and perform pretreatment such as de-energization and filtering;

[0147] As shown in Figure 2 , first, 5 laser sensors are installed on the outer fixed part of the rotating machinery with 12 blades in a circumferential direction according to the angle distribution of {0°, 20°, 90°, 145°, 195°}, and an encoder is installed at the shaft end;

[0148] Before starting operation, the order of each blade passing through the sensor position is calibrated to accurately record the time when different blades pass through different BTT sensors;

[0149] When the blade passes through the BTT sensor, the sensor will generate a voltage pulse in the output signal, and if the blade tip deflects, the occurrence time of this pulse signal will be advanced or delayed relative to the "non-vibration" blade signal calculated from the time reference signal. At the same time, the encoder will also generate a pulse sequence in each rotation period, which not only serves as a time reference signal for the blade tip timing, but also records the time stamp of the pulse into the computer through the data acquisition system.

[0150] Based on the time signals generated by the BTT sensor and the encoder, the vibration deflection angle of each blade at each BTT sensor is calculated

[0151]

[0152] wherein T i k is the time when the k-th blade passes through the i-th BTT sensor, Ω n is the average rotational speed of the n-th revolution, is the theoretical time when the k-th blade passes through the i-th BTT sensor under vibration conditions obtained from the encoder, is the vibration angle of the k-th blade passing through the i-th BTT sensor.

[0153] All signals collected by the BTT sensors By correctly assigning the blades to each blade according to their order of travel, the tip displacement sequence of the k-th blade in the nth revolution can be obtained as follows:

[0154]

[0155] in, R is the displacement of the k-th blade recorded by the i-th sensor in the n-th revolution, and R is the distance from the blade tip to the center of rotation. The sequence of tip displacements for each revolution is x = {x...} 1 ,…,x I The BTT signal of the blade within a certain time range is obtained by cascading the signals.

[0156] The displacement time series of each blade is subjected to depotentialization and filtering to eliminate the static displacement component and noise of the blade. Figure 3 This is the BTT signal for the second blade within 20 revolutions at a speed of 3400 rpm.

[0157] Step 2: Design a suitable deep neural network model, train the model using a training dataset, and learn the mapping relationship from BTT signal to complete vibration signal;

[0158] Design one Figure 4 The deep neural network model shown combines CNN and LSTM. The CNN consists of only three convolutional layers with 32, 64, and 1 kernels respectively, all with a kernel size of 5. No pooling layers are used to ensure the vibration signal length remains constant. The activation function for all three convolutional layers is ReLU. A final convolutional kernel ensures an initial reconstructed signal with the same size as the original signal. The LSTM has 250 cells, and the final fully connected layer has 256 neurons, ensuring the shape of the secondary reconstructed signal is the same as the original signal.

[0159] Simulated blade tip vibration signal Data is collected and compressed, where N represents the number of multimodal resonances, and EO... i It corresponds to the frequency order, A. i and These represent amplitude and phase, respectively. Simulated signals are generated for different combinations of the number, frequency order, amplitude, and phase of multimodal resonances, resulting in a dataset represented as X = {x}. (1) ,...,x (i) ,...,x (l)};

[0160] The complete simulated vibration signal is compressed using the BTT method. The acquisition process of the i-th compressed signal is represented as follows:

[0161] y (i) =φ m×n x (i) (25)

[0162] in, The linear mapping matrix is ​​determined by the BTT sensor mounting angle, and the compressed vibration signal dataset is represented as Y = {y (1) ,...,y (i) ,...,y (l)};

[0163] The compressed measurement signal y(i) is up-dimension processed to obtain a proxy signal with the same length as the complete signal.

[0164] z (i) =φ T y (i) (26)

[0165] To accelerate network training, Z-score normalization is performed on the agent signal.

[0166]

[0167] The loss function is set as follows

[0168]

[0169] Where W and b are the weights and biases of the network, respectively. The samples in the training sample set are input into a pre-set deep learning model. The loss value is minimized and the relevant parameters are learned by using the backpropagation algorithm with Adam as the optimizer.

[0170] The percentage root mean square error (PRD) is defined to evaluate the quality of signal reconstruction.

[0171]

[0172] Where x and These are the original vibration signal and the reconstructed signal, respectively. ||·||2 is the 2-norm. The reconstruction capability and accuracy of the model are evaluated through cross-validation and independent test sets. Hyperparameters are adjusted to optimize model performance.

[0173] Step 3: Use the trained deep neural network model to process the measured BTT signal and reconstruct the blade vibration displacement signal;

[0174] Will Figure 2 The signal measured by the non-contact BTT measurement system is used as input after being upgraded and normalized. The complete vibration signal shown in Figure 5(a) is reconstructed by the trained deep neural network model.

[0175] Step 4: Perform spectral analysis on the reconstructed vibration displacement signal by Fast Fourier Transform (FFT) to identify key vibration parameters such as vibration order, amplitude, etc. as shown in Figure 5(b);

[0176] Step 5: Establish a finite element model of the blade, determine the natural frequency and modal shape of the blade through modal analysis, and analyze the modal response of the blade in combination with the vibration parameters;

[0177] Establish a three-dimensional geometric model of the rotating blade, perform meshing on the geometric model and conduct mesh independence test, and establish a finite element model as shown in Figure Figure 6 ;

[0178] Define material properties according to the actual blade material, including density, Young's modulus, and Poisson's ratio;

[0179] Select the type of modal analysis, and determine the solution order, Coriolis effect, and rotational speed of the modal analysis;

[0180] Solve the modal equation to obtain the natural frequency of each mode, the displacement, strain, and stress modal vectors of the mode, and the Campbell diagram as shown in Figure Figure 7 .

[0181] Step 6: Use the identified vibration parameters and modal analysis results to quickly calculate the displacement field, strain field, and stress field of the blade in the vibration state based on the modal superposition theory.

[0182] Compare the vibration frequency components of the reconstructed signal obtained from the deep neural network model with the Campbell diagram, and find that the vibration signal only contains a 6th order component, which has a frequency close to the 1st order natural frequency. From the Campbell diagram, it can be concluded that resonance can be excited at a rotational speed of 3400 revolutions per minute;

[0183] It can be calculated that at a rotational speed of 3400 revolutions per minute, the vibration of the blade is dominated by the 1st order mode, and the participation coefficient of the 1st order mode is

[0184]

[0185] where ω i , A i , and θ i are the frequency, amplitude, and phase of the main vibration component of the reconstructed signal, respectively, and φ p,i is the displacement at the blade tip measurement point in the vibration mode closest to the main vibration component;

[0186] The full-field displacement, strain, and stress calculated based on the modal superposition theory are

[0187]

[0188] wherein C i (t) is a time-varying coefficient of the vibration mode closest to the i-th order vibration principal component, is a displacement vector, a strain vector or a stress vector of the vibration mode, and f(t) is a structure displacement field, a strain field or a stress field vector at time t. The X-direction displacement cloud chart, the Y-direction strain cloud chart and the Y-direction stress cloud chart of the blade at time t=134.319s are shown in Figures 8(a), 8(b) and 8(c), respectively.

[0189] The present application utilizes a non-contact vibration measurement method, avoiding the limitations of the traditional strain measurement method. Through a visual method, the change trend of the maximum stress point of the blade resonance in the time domain and the spatial domain can be clearly shown. The method provided by the present application has high precision, and the prediction result is close to the actual result. The blade multi-modal maximum stress prediction system based on non-contact measurement provided by the present application has a simple process and is easy to implement. The above only describes the preferred embodiments of the present application, which can be applied in the vibration test of fan blades, compressor blades and turbine blades of rotary machines such as aero-engines, gas turbines and steam turbines, and is not used to limit the present application.

Claims

1. A method for fast reconstruction of rotor blade dynamic field based on deep neural network model, characterized in that, The method comprises the following steps: Step 1: collecting BTT signals containing blade vibration information, and performing denoising and normalization preprocessing; Step 2: designing a deep neural network model, training the deep neural network model through a training data set, and learning a mapping relationship from the BTT signal to the complete vibration signal; Step 2-1: selecting CNN+LSTM as the deep neural network model, using CNN to extract signal spatial features, and using LSTM to capture time dynamics and long-time span correlations in the signal; Step 2-2: establishing a training set and a test set based on simulated vibration displacement signals, the signals linearly mapped through a BTT measurement matrix being used as inputs of the deep neural network model after preprocessing, and complete simulated vibration displacement signals being used as outputs of the deep neural network model; Step 2-3: training the designed deep neural network model, and learning a nonlinear mapping relationship between the input and the output by adjusting model parameters; Step 2-4: evaluating the reconstruction capability and precision of the model through cross-validation and an independent test set, and adjusting hyperparameters to optimize the performance of the model; Step 3: using the trained deep neural network model to process the measured BTT signal, and reconstructing the blade vibration displacement signal; Step 4: performing frequency spectrum analysis on the reconstructed blade vibration displacement signal, and identifying vibration frequency, amplitude and phase parameters; Step 5: establishing a finite element model of the blade, determining the natural frequency and modal shape of the blade through modal analysis, and analyzing the modal response of the blade in combination with the vibration parameters; Step 6: using the identified vibration parameters and the modal analysis result, and quickly calculating the displacement field, strain field and stress field of the blade in a vibration state based on modal superposition theory.

2. The method of claim 1, wherein the method is characterized by: The step 1 specifically comprises: Step 1-1: sensor installation; installing laser sensors on the outer fixed part of the machine in a circumferential direction according to a specified angle, and installing an encoder at a shaft end; Step 1-2: blade position calibration; before starting operation, calibrating the order of each blade passing through the sensor position, and recording the time of different blades passing through different BTT sensors; Step 1-3: data acquisition; when the machine starts to operate, each blade passing through the sensor will generate a pulse signal, and the time stamp of the pulse signal is recorded into a computer through a data acquisition system; Step 1-4: time synchronization and data processing; calculating a vibration deflection angle based on the arrival time and the rotating speed: wherein T i k is the time of the kth blade passing the ith BTT sensor, Ω n is the average rotational speed of the nth revolution, is the theoretical time of the kth blade passing the ith BTT sensor in the vibration condition obtained by the encoder, is the vibration angle of the kth blade passing the ith BTT sensor; Step 1-5: Assigning to the blade and calculate the blade tip displacement; All data collected by multiple sensors The signal is correctly distributed to each blade, and the tip displacement sequence of the k-th blade in the nth revolution is: where x i is the displacement, R is the distance from the blade tip to the center of rotation, θ is the vibration deflection angle, and the tip displacement sequence x = {x1,..., x I N} of each cycle is concatenated to obtain the BTT signal of the blade within a certain time range; Step 1-6: zeroing, normalizing, de-energizing and filtering the displacement time sequence of each blade to eliminate the static displacement part and noise of the blade.

3. The method of claim 2, wherein the method further comprises: The step of establishing a training set and a test set based on simulated vibration displacement signals comprises: Step 2-2-1: Tip-vibration signal to analog signal is collected and compressed, where N represents the number of multi-modal resonances, EO i is the corresponding frequency order, A i and are the amplitude and phase, respectively; Step 2-2-2: The original tip vibration data set is represented as X = {x (1) ,...,x (i) ,...,x (l)}; in this data set, represents the ith vibration signal, with n sampling points; the compressed vibration signal is represented as Y = {y (1) ,...,y (i) ,...,y (l)}, wherein represents the ith measurement signal, with m sampling points; the mapping from the original signal X to the compressed vibration signal Y is a linear mapping, that is, Y = φX, wherein is a linear mapping matrix determined by the BTT sensor mounting angle; the acquisition process of the ith compressed signal is represented as: y (i) = φ m×n x (i) (3) Step 2-2-3: performing dimensionality processing on the compressed measurement signal y(i) to obtain an agent signal consistent with the length of the complete signal: z (i) = φ T y (i) (4) performing Z-score normalization on the agent signal:

4. The method of claim 3, wherein the method further comprises: The sample input of the training set is set into a pre-set deep neural network model, a loss value is calculated through a pre-set loss function, model parameters are adjusted using the loss value, and a trained deep neural network model is obtained; The deep neural network model determines the loss function through mean square error MSE as follows: wherein W and b are the weight and bias of the network respectively; The loss function uses a back propagation algorithm to minimize the loss value and learn related parameters with Adam as an optimizer.

5. The method of claim 4, wherein the method further comprises: The step 2-4 is specifically: The percentage root mean square difference (PRD) is used to evaluate the quality of signal reconstruction and is defined as follows: where x and are the original vibration signal and the signal reconstruction, respectively, and || · ||2is the 2-norm.

6. The method of claim 5, wherein the method further comprises: In step 3, the preprocessed signal r of the BTT measurement system is taken as input, and a nonlinear mapping relationship from the preprocessed signal to the original signal x is fitted by the trained deep neural network model, and the reconstructed signal is is represented as: Where H represents the deep neural network model.

7. The method of claim 6, wherein the method further comprises: The spectral analysis in step 4 is performed by a Fast Fourier Transform, FFT, of the reconstructed signal The conversion from the time domain signal to the frequency domain, the finding of the frequency components from the spectrum and the extraction of the amplitude and phase vibration parameters.

8. The method of claim 7, wherein the method further comprises: The step 5 is specifically: Step 5-1: Establish a three-dimensional geometric model of the blade, mesh the geometric model, and perform mesh independence testing; Step 5-2: Define material properties, including density, Young's modulus, and Poisson's ratio, according to the actual blade material; Step 5-3: Select the type of modal analysis, determine the order of solution of modal analysis, Coriolis effect, and rotational speed; Step 5-4: Obtain the motion equation and modal equation by the finite element method; where M is the mass matrix, C is the damping matrix, K is the stiffness matrix, u(t) is the displacement vector, G is the gyroscopic matrix due to Coriolis forces, K Ω is the stiffness matrix due to centrifugal forces; Step 5-5: Record the natural frequency of each mode by solving the eigenvalue problem, draw a Campbell diagram, extract the displacement, strain, and stress modal vectors of each mode, and analyze the deformation characteristics of the blade under different modes.

9. The method of claim 8, wherein the method further comprises: The step 6 is specifically: Step 6-1: Compare the vibration frequency components of the reconstructed signal obtained by the deep neural network model with the Campbell diagram to determine the closest vibration mode; Step 6-2: Calculate the modal coefficients of the main modes: where ω i , A i , and θ i are the frequency, amplitude, and phase of the dominant vibration component of the reconstructed signal, respectively, and φ p,i is the displacement at the blade tip measurement point in the vibration mode closest to the dominant vibration component. Step 6-3: Calculate the full-field displacement, strain, and stress based on the modal superposition theory: where C i (t) is a time-varying coefficient of the vibration mode closest to the i-th order vibration principal component, is a displacement vector, a strain vector, or a stress vector of the vibration mode, and f(t) is a structure displacement field, a strain field, or a stress field vector at time t.

Images