Synchronous measurement and analysis method and system for torsional vibration of steam turbine shaft and blade vibration

Abstract

The application discloses a synchronous measurement and analysis method and system for torsional vibration of a steam turbine shaft system and blade vibration, and belongs to the technical field of monitoring of steam turbine generator units. The method comprises noise suppression processing on multi-mode synchronous signals and correlation analysis, so as to obtain coupling modal characteristics representing the comprehensive state of the steam turbine shaft system-blade, and to evaluate and analyze the coupling modal characteristics based on a pre-constructed coupling analysis model. In the application, an adaptive filter is used to filter the blade vibration signal, and cross modulation spectrum of the torsional vibration demodulation timing information and the vibration modal characteristic component is determined based on a mutual information function, so as to accurately determine the cross phase relationship at the sampling point, obtain the coupling modal characteristics representing the comprehensive state of the steam turbine shaft system-blade, and more comprehensively and accurately reflect the overall operation state of the steam turbine.

Description

Technical Field

[0001] This invention belongs to the field of steam turbine generator set monitoring technology, specifically relating to a method and system for synchronous measurement and analysis of torsional vibration of turbine shaft system and blade vibration. Background Technology

[0002] As the core power equipment of the power system, the safe and stable operation of the steam turbine is directly related to the reliability and economy of the entire power grid. During the operation of the steam turbine, there are a variety of complex dynamic phenomena inside it. Among them, shaft torsional vibration and blade vibration are two key factors affecting the life and safety of the unit. Shaft torsional vibration refers to the torsional vibration generated by the rotor system of the steam turbine generator under the action of periodic torque. Continuous or resonant shaft torsional vibration can generate huge alternating stress in a short period of time, leading to fatigue of the shaft material. In severe cases, it can cause major accidents such as shaft breakage and coupling damage. Steam turbine blades will generate complex vibrations under the excitation of high-speed airflow, mainly divided into high-frequency flutter, low-frequency tuning vibration and resonance. When the excitation frequency is close to or equal to the natural frequency of the blade, resonance will be triggered, generating huge dynamic stress. Long-term operation under resonance conditions will lead to fatigue fracture of the blade. The fractured fragments may further damage adjacent blades, nozzles or impellers, causing more serious chain damage.

[0003] Currently, although individual monitoring technologies for shaft torsional vibration and blade vibration are relatively mature, in practical engineering applications, they are usually analyzed separately. However, considering that shaft torsional vibration and blade vibration are not isolated phenomena, on the one hand, strong blade vibration can change the distribution of steam force on the rotor circumference, forming periodic torque excitation, thereby triggering or aggravating shaft torsional vibration; on the other hand, the torsional dynamic characteristics of the shaft (such as instantaneous speed changes) can change the characteristics of the airflow excitation force acting on the blades. This separate monitoring method lacks synchronous correlation analysis, making it difficult to reveal the coupling mechanism. In addition, when the unit malfunctions, traditional independent monitoring systems give alarms for abnormal shaft torsional vibration and suspected increased blade vibration, but cannot determine the causal relationship between the two, thus limiting the ability to diagnose and trace the source of the fault. To address the above problems, we propose a method and system for synchronous measurement and analysis of turbine shaft torsional vibration and blade vibration. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a method and system for synchronous measurement and analysis of turbine shaft torsional vibration and blade vibration. This solves the problem that traditional separate monitoring methods lack synchronous correlation analysis of shaft torsional vibration and blade vibration, thus making it impossible to determine the causal relationship between shaft torsional vibration and blade vibration, which limits the ability to diagnose and trace faults.

[0005] This invention is implemented as follows: a method for synchronous measurement and analysis of torsional vibration of turbine shaft system and blade vibration, the method comprising: Based on distributed sensors, shaft rotation signals, key phase pulse signals, and blade vibration signals are collected respectively, and the shaft rotation signals, key phase pulse signals, and blade vibration signals are unified with a time reference to obtain a multi-modal synchronization signal; A multi-mode synchronization signal with a unified time base is obtained, and noise suppression processing is performed on the multi-mode synchronization signal to obtain a noise-suppressed multi-mode synchronization signal. After applying noise suppression, the multi-mode synchronization signal is subjected to correlation analysis. Based on the correlation analysis results, the coupled modal characteristics characterizing the integrated state of the turbine shaft system and blades are obtained. The coupled modal characteristics are obtained, and the coupled modal characteristics are evaluated and analyzed based on the pre-constructed coupled analysis model to output the integrated state value of the turbine shaft system-blade system. Preferably, the integrated state value of the turbine shaft system and blades is obtained, and it is determined whether the integrated state value exceeds the preset state warning threshold. If the integrated state value exceeds the preset state warning threshold, a state warning command is triggered, and the integrated state value of the turbine shaft system and blades is visualized based on the state warning command.

[0006] Preferably, the noise suppression processing for the multimodal synchronization signal includes: The multi-modal synchronization signal is acquired and processed in segments based on the turbine shaft-blade rotation cycle to obtain multi-segmented signal windows. The trend term in the multimodal synchronization signal is extracted by polynomial fitting. The trend term is removed from the multimodal synchronization signal within the segmented signal window, and the multimodal synchronization signal after trend term removal is output. The multimodal synchronization signal after removing the trend term is obtained. A bandpass filter is set based on the theoretical frequency range of shaft torsional vibration and blade vibration. The multimodal synchronization signal is then processed by bandpass filtering based on the bandpass filter. Edge detection is performed on the multimodal synchronization signal after bandpass filtering to determine the continuity of the signal. If the signal is incomplete or falsely triggered, interpolation is performed based on the multimodal synchronization signals corresponding to the previous and next sampling points to obtain the interpolated multimodal synchronization signal.

[0007] Preferably, the correlation analysis of the multimodal synchronization signal includes: Acquire the noise-suppressed multimode synchronization signal, identify the shaft rotation signal and key phase pulse signal in the multimode synchronization signal, and establish the phase relationship between the shaft rotation signals based on the key phase pulse signal; The phase relationship between the shaft rotation signals is mapped to the sampling point. Based on the sampling point, the torsional vibration amplitude and frequency at the sampling point are demodulated to generate torsional vibration demodulation timing information containing the torsional vibration amplitude and frequency. The blade vibration signal in the multimodal synchronization signal is acquired. Based on the torsional vibration demodulation timing information, the torsional vibration demodulation timing information is extended by higher-order harmonics to obtain the extended reference signal. The blade vibration signal is used as the input and the extended reference signal is used as the reference input. The blade vibration signal is adaptively filtered based on the adaptive filter to output the adaptive residual signal. The adaptive residual signal is loaded, and the adaptive residual signal is sparsified based on the Morlet wavelet dictionary. The vibration mode feature components corresponding to the blade vibration signal are extracted from the adaptive residual signal by the matching pursuit method. The torsional vibration demodulation timing information and vibration mode characteristic components are obtained. The cross-modulation spectrum of the torsional vibration demodulation timing information and vibration mode characteristic components is determined based on the mutual information function. The cross-phase relationship at the sampling point is determined by combining the torsional vibration demodulation timing information and vibration mode characteristic components cross-modulation spectrum.

[0008] Preferably, the method for determining the cross-modulation spectrum of torsional vibration demodulation timing information and vibration mode characteristic components based on the mutual information function includes: Load torsional vibration demodulation timing information and vibration mode characteristic components, perform Fourier transform on the torsional vibration demodulation timing information and vibration mode characteristic components, and obtain the Fourier transformed torsional vibration demodulation signal and vibration mode signal; The torsional vibration demodulated signal and vibration mode signal after Fourier transform are obtained. The mutual information of the torsional vibration demodulated signal and vibration mode signal is calculated by sliding window to obtain the signal mutual information of the torsional vibration demodulated signal and vibration mode signal. The mutual information function is constructed based on the signal mutual information. The mutual information function is expressed as:

[0009]

[0010] in, Indicates torsional vibration demodulation signal and vibration mode signals Mutual information function, Indicates torsional vibration demodulation signal and vibration mode signals The joint density function, The peak value in the mutual information curve; Mutual information curves are generated based on mutual information functions, and phase compensation is performed on the torsional vibration demodulation signal using the peak value in the mutual information curves as a reference to obtain the torsional vibration modulation function. The torsional vibration modulation function is expressed as:

[0011] in, These are the torsional vibration modulation weights and the time spectrum of the vibration mode signal, respectively. The torsional vibration modulation function is obtained, and the vibration mode signal is coherently weighted based on the torsional vibration modulation function. The coherently weighted vibration mode signal and the torsional vibration modulation function are mapped to the mutual information curve. The high-order harmonic sidebands in the mutual information curve are identified, and the high-order harmonic sidebands are reconstructed using a median filter to obtain the coupled mode characteristics that characterize the integrated state of the turbine shaft system and blades.

[0012] Preferably, the coupling analysis model is based on a BiLSTM model and further includes an input layer and an output layer. The input layer is connected to the BiLSTM model, and the BiLSTM model is connected to the output layer. A parallel feature extraction layer is set between the input and output layers and the BiLSTM model. The parallel feature extraction layer includes a feature encoding block, a parallel deep feature extraction network, and a graph neural network. The input layer is used to acquire coupled modal features and pass them to the feature extraction layer. The feature encoding block identifies torsional vibration assessment features, blade vibration assessment features, and torsional vibration-blade correlation assessment features in the coupled modal features. The deep feature extraction network is used to acquire torsional vibration assessment features and torsional vibration-blade correlation assessment features. Based on the deep feature extraction network, torsional vibration anomaly indicators are extracted from the torsional vibration assessment features and torsional vibration-blade correlation assessment features. The torsional vibration anomaly indicators are weighted based on an expert rule base of rotor dynamics and fatigue theory. Based on the weighted torsional vibration anomaly indicators, a multi-head torsional vibration anomaly is constructed using a cross-attention mechanism. The evaluation matrix is ​​processed by a physical data residual network to handle the residual components between the abnormal evaluation matrix and the standard evaluation matrix. Uncertainty analysis is performed on the residual decomposition based on a Bayesian hierarchical network to obtain a quantified shaft torsional vibration evaluation value. The graph neural network is used to obtain blade vibration evaluation features and torsional vibration-blade correlation evaluation features. Based on the blade vibration evaluation features and torsional vibration-blade correlation evaluation features, a graph neural network heterogeneous graph is constructed, and the influence weights of torsional vibration-blade are explicitly learned using a cross-attention mechanism. The influence weights are assigned to the graph nodes in the graph neural network heterogeneous graph. The blade vibration evaluation value is determined based on a multi-output regression network combined with the influence weights of the graph nodes. The BiLSTM model is used to obtain shaft torsional vibration evaluation values ​​and blade vibration evaluation values, construct a torsional vibration-blade digital twin network, and update the weight parameters of the torsional vibration-blade digital twin network based on the shaft torsional vibration evaluation values ​​and blade vibration evaluation values. Global pooling is used to automatically learn the turbine shaft-blade comprehensive state classification evaluation and output the turbine shaft-blade comprehensive state value.

[0013] Preferably, the evaluation and analysis of coupled modal characteristics based on the pre-constructed coupling analysis model includes: The coupled modal features are acquired and passed to the feature extraction layer. The feature encoding block identifies torsional vibration evaluation features, blade vibration evaluation features, and torsional vibration-blade correlation evaluation features in the coupled modal features. A deep feature extraction network is used to obtain torsional vibration assessment features and torsional vibration-blade correlation assessment features. Based on the deep feature extraction network, torsional vibration anomaly indicators are extracted from the torsional vibration assessment features and torsional vibration-blade correlation assessment features. The torsional vibration anomaly index is weighted based on an expert rule base of rotor dynamics and fatigue theory, and a multi-head torsional vibration anomaly evaluation matrix is ​​constructed based on the weighted torsional vibration anomaly index using a cross-attention mechanism. The residual components between the anomaly evaluation matrix and the standard evaluation matrix are processed by physical data residual network. Uncertainty analysis is performed on the residual decomposition based on Bayesian hierarchical network to obtain quantified shaft torsional vibration evaluation values.

[0014] Graph neural networks are used to obtain blade vibration assessment features and torsional vibration-blade correlation assessment features. Based on the blade vibration assessment features and torsional vibration-blade correlation assessment features, a graph neural heterogeneous graph is constructed. The cross-attention mechanism is used to explicitly learn the influence weight of torsional vibration-blade, and the influence weight is assigned to the graph nodes in the graph neural heterogeneous graph. The blade vibration assessment value is determined based on the multi-output regression network combined with the influence weight of the graph nodes. The formula for calculating the shaft system torsional vibration assessment value is as follows:

[0015] in, This is the value for shaft torsional vibration assessment. This is the weight vector for torsional vibration anomaly indicators. These are, respectively, the torsional vibration anomaly index vector, the residual correction term output by the residual network, and the uncertainty quantification term of the Bayesian hierarchical network;

[0016] in, These are respectively random uncertainty error, model uncertainty, and prior deviation. These are the anomaly assessment matrix and the standard assessment matrix, respectively. These are the variance penalty strength and the regularization weight, respectively. The formula for calculating the blade vibration assessment value is as follows:

[0017] in, This indicates the blade vibration assessment value. These are the original features of the graph nodes and the original features of the connections between the graph nodes, respectively. The weights of the edges connecting graph nodes and their associated nodes; Obtain the shaft torsional vibration assessment value and the blade vibration assessment value, construct a torsional vibration-blade digital twin network, and update the weight parameters of the torsional vibration-blade digital twin network based on the shaft torsional vibration assessment value and the blade vibration assessment value; A global pooling-based automatic learning method is used to classify and evaluate the integrated state of the turbine shaft system and blades, and output the integrated state value of the turbine shaft system and blades.

[0018] On the other hand, the present invention also provides a synchronous measurement and analysis system for torsional vibration of turbine shaft system and blade vibration, the synchronous measurement and analysis system for torsional vibration of turbine shaft system and blade vibration includes: The signal acquisition module collects shaft rotation signals, key phase pulse signals, and blade vibration signals respectively based on distributed sensors, and unifies the shaft rotation signals, key phase pulse signals, and blade vibration signals with a unified time reference to obtain a multi-modal synchronization signal; The noise suppression module is used to acquire a multi-mode synchronization signal with a unified time base, perform noise suppression processing on the multi-mode synchronization signal, and obtain a noise-suppressed multi-mode synchronization signal. The modal coupling module loads the noise-suppressed multi-modal synchronization signal, performs correlation analysis on the multi-modal synchronization signal, and obtains the coupled modal characteristics characterizing the integrated state of the turbine shaft system and blades based on the correlation analysis results. The state assessment module is used to acquire coupled modal characteristics, evaluate and analyze the coupled modal characteristics based on a pre-built coupled analysis model, and output the turbine shaft-blade integrated state value. The status warning module is used to acquire the comprehensive status value of the turbine shaft system and blades, determine whether the comprehensive status value exceeds the preset status warning threshold, and if the comprehensive status value exceeds the preset status warning threshold, trigger a status warning command and visualize the comprehensive status value of the turbine shaft system and blades based on the status warning command.

[0019] Preferably, the noise suppression module includes: The signal segmentation unit is used to acquire multi-mode synchronization signals and process the multi-mode synchronization signals into segments based on the turbine shaft-blade rotation cycle to obtain multi-group segmented signal windows; The trend term removal unit uses polynomial fitting to extract the trend term in the multimodal synchronization signal, removes the trend term from the multimodal synchronization signal within the segmented signal window, and outputs the multimodal synchronization signal after trend term removal. The bandpass filter unit is used to acquire the multimodal synchronization signal after removing the trend term. The bandpass filter is set based on the theoretical frequency range of shaft torsional vibration and blade vibration, and the multimodal synchronization signal is bandpass filtered based on the bandpass filter. The interpolation processing unit is used to perform edge detection on the multimodal synchronization signal after bandpass filtering to determine the continuity of the signal. If the signal is incomplete or falsely triggered, interpolation processing is performed based on the multimodal synchronization signals corresponding to the previous and next sampling points to obtain the interpolated multimodal synchronization signal.

[0020] Compared with the prior art, the embodiments of this application have the following main advantages: In this embodiment of the invention, distributed sensors simultaneously acquire shaft rotation signals, key phase pulse signals, and blade vibration signals. These signals are unified with a time reference to form a multi-mode synchronous signal. An adaptive filter is used to filter the blade vibration signal, extracting vibration mode feature components. Furthermore, based on the mutual information function, the cross-modulation spectrum of torsional vibration demodulation timing information and vibration mode feature components is determined, thereby accurately determining the cross-phase relationship at the sampling point. Coupled mode features characterizing the integrated state of the turbine shaft-blade system are obtained. Based on a pre-constructed coupled analysis model, the coupled mode features are evaluated and analyzed, and the integrated state value of the turbine shaft-blade system is output. The evaluation and analysis not only considers the evaluation results of shaft torsional vibration and blade vibration individually, but also fully considers the correlation between the two, which can more comprehensively and accurately reflect the overall operating state of the turbine. This makes the evaluation results not only accurate, but also understandable to engineers, greatly enhancing users' trust in the intelligent diagnostic system.

[0021] In this embodiment of the invention, when processing noise suppression for multimodal synchronization signals, the integrity and quality of the input data are fundamentally guaranteed by actively identifying and repairing these defects. Furthermore, the noise suppression process effectively removes noise and interference components from the multimodal synchronization signals, improving signal quality and signal-to-noise ratio. Ultimately, the multimodal synchronization signals processed by noise suppression can more clearly reflect the true state of turbine shaft torsional vibration and blade vibration, thereby providing a more accurate basis for fault diagnosis and condition assessment, and significantly improving the accuracy and reliability of fault diagnosis.

[0022] In this embodiment of the invention, correlation analysis of multimodal synchronization signals not only considers the frequency, amplitude, and phase relationships between torsional vibration demodulation timing information and vibration mode characteristic components, but also delves into the complex coupling relationships between torsional vibration demodulation timing information and vibration mode characteristic components through higher-order harmonic spread and cross-modulation spectrum analysis. By accurately extracting the characteristic information of torsional vibration and blade vibration and quantifying their coupling relationships, the accuracy of fault diagnosis can be significantly improved. Finally, based on the coupled mode characteristics obtained from deep correlation analysis, the operating status of the steam turbine can be more accurately assessed. This method can provide more comprehensive condition assessment results, providing a more reliable basis for maintenance personnel to formulate reasonable maintenance strategies, thereby extending the service life of the steam turbine, improving operating efficiency, and reducing maintenance costs.

[0023] In this embodiment of the invention, when determining the cross-modulation spectrum of torsional vibration demodulation timing information and vibration mode characteristic components based on the mutual information function, the mutual information function is used to accurately quantify the nonlinear coupling relationship between the torsional vibration demodulation timing information and vibration mode characteristic components. Furthermore, a sliding window analysis is used to capture transient coupling dynamics. By mapping the coherently weighted vibration mode signal and the torsional vibration modulation function onto the mutual information curve, higher-order harmonic sidebands in the mutual information curve can be identified. Finally, a median filter is used to reconstruct the higher-order harmonic sidebands, obtaining coupled mode characteristics that characterize the integrated state of the turbine shaft system and blades. The mode reconstruction method effectively removes noise and interference, ensuring the accuracy and reliability of the coupled mode characteristics.

[0024] In this embodiment of the invention, when evaluating and analyzing coupled modal features based on a pre-built coupled analysis model, the torsional vibration evaluation features, blade vibration evaluation features, and torsional vibration-blade correlation evaluation features in the coupled modal features can be identified through the feature extraction layer and feature encoding block. The deep feature extraction network can automatically extract torsional vibration anomaly indicators from the torsional vibration evaluation features and the torsional vibration-blade correlation evaluation features. The deep learning method can automatically learn complex patterns in the signal, improving the accuracy and efficiency of feature extraction. Furthermore, by processing the residual components between the anomaly judgment matrix and the standard judgment matrix through the physical data residual network, the abnormal components in the signal can be further analyzed, thereby improving the sensitivity of the system and ensuring that even small abnormal changes can be detected. By constructing a torsional vibration-blade digital twin network and updating the network weight parameters based on the shaft torsional vibration evaluation value and the blade vibration evaluation value, real-time dynamic modeling of the turbine's operating status can be achieved. This digital twin network can provide more accurate operating status evaluation, providing real-time decision support for operation and maintenance personnel. Finally, the global pooling method can automatically learn key features in the signal, improving the system's adaptability and flexibility, and ensuring the accuracy and reliability of the evaluation results. Attached Figure Description

[0025] Figure 1 A flowchart illustrating the method for synchronous measurement and analysis of torsional vibration of turbine shaft system and blade vibration is shown. Figure 2 A schematic diagram of the implementation process of the noise suppression method for multimodal synchronization signals is shown. Figure 3 This diagram illustrates the implementation process of a correlation analysis method for multimodal synchronization signals. Figure 4 A schematic diagram of the implementation process of the cross-modulation spectrum method for determining torsional vibration demodulation timing information and vibration mode characteristic components based on mutual information function is shown. Figure 5A schematic diagram of the implementation process of a method for evaluating and analyzing coupled modal characteristics based on a pre-built coupled analysis model is shown. Figure 6 A schematic diagram of the structure of a synchronous measurement and analysis system for torsional vibration of turbine shaft system and blade vibration is shown. Detailed Implementation

[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0027] Traditional separate monitoring methods lack synchronous correlation analysis between shaft torsional vibration and blade vibration, thus failing to determine the causal relationship between the two, limiting fault diagnosis and tracing capabilities. To address this issue, we propose a method and system for synchronous measurement and analysis of turbine shaft torsional vibration and blade vibration. In short, the method first collects shaft rotation signals, key phase pulse signals, and blade vibration signals using distributed sensors. Noise suppression processing is applied to the multi-modal synchronization signals, and correlation analysis is performed on the multi-modal synchronization signals. Using the correlation analysis results as a reference, coupled modal characteristics characterizing the integrated state of the turbine shaft-blade system are obtained. The coupled modal characteristics are then evaluated and analyzed based on a pre-constructed coupling analysis model. In this embodiment of the invention, distributed sensors simultaneously acquire shaft rotation signals, key phase pulse signals, and blade vibration signals. These signals are unified with a time reference to form a multi-mode synchronous signal. An adaptive filter is used to filter the blade vibration signal, extracting vibration mode feature components. Furthermore, based on the mutual information function, the cross-modulation spectrum of torsional vibration demodulation timing information and vibration mode feature components is determined, thereby accurately determining the cross-phase relationship at the sampling point. Coupled mode features characterizing the integrated state of the turbine shaft-blade system are obtained. Based on a pre-constructed coupled analysis model, the coupled mode features are evaluated and analyzed, and the integrated state value of the turbine shaft-blade system is output. The evaluation and analysis not only considers the evaluation results of shaft torsional vibration and blade vibration individually, but also fully considers the correlation between the two, which can more comprehensively and accurately reflect the overall operating state of the turbine. This makes the evaluation results not only accurate, but also understandable to engineers, greatly enhancing users' trust in the intelligent diagnostic system.

[0028] This invention provides a method for synchronous measurement and analysis of torsional vibration of turbine shaft system and blade vibration. Figure 1 A flowchart illustrating a method for synchronously measuring and analyzing torsional vibration of a steam turbine shaft system and blade vibration is shown. This method includes: S10: Based on distributed sensors, shaft rotation signals, key phase pulse signals, and blade vibration signals are collected respectively, and the shaft rotation signals, key phase pulse signals, and blade vibration signals are unified with a time reference to obtain a multi-mode synchronization signal; S20: Obtain a multi-mode synchronization signal with a unified time base, perform noise suppression processing on the multi-mode synchronization signal, and obtain a noise-suppressed multi-mode synchronization signal; S30, load the noise-suppressed multi-mode synchronization signal, perform correlation analysis on the multi-mode synchronization signal, and use the correlation analysis results as a reference to obtain the coupled mode characteristics that characterize the integrated state of the turbine shaft system-blade; S40: Obtain coupled modal characteristics, evaluate and analyze the coupled modal characteristics based on the pre-built coupled analysis model, and output the turbine shaft-blade integrated state value.

[0029] S50: Obtain the integrated status value of the turbine shaft system and blades, and determine whether the integrated status value exceeds the preset status warning threshold. S60: If the overall status value exceeds the preset status warning threshold, a status warning command is triggered, and the overall status value of the turbine shaft system and blades is visualized based on the status warning command. In this way, the abstract evaluation values ​​are transformed into an intuitive graphical interface, which helps maintenance personnel to quickly grasp the overall status and make efficient decisions.

[0030] In this embodiment of the invention, distributed sensors simultaneously acquire shaft rotation signals, key phase pulse signals, and blade vibration signals. These signals are unified with a time reference to form a multi-mode synchronous signal. An adaptive filter is used to filter the blade vibration signal, extracting vibration mode feature components. Furthermore, based on the mutual information function, the cross-modulation spectrum of torsional vibration demodulation timing information and vibration mode feature components is determined, thereby accurately determining the cross-phase relationship at the sampling point. Coupled mode features characterizing the integrated state of the turbine shaft-blade system are obtained. Based on a pre-constructed coupled analysis model, the coupled mode features are evaluated and analyzed, and the integrated state value of the turbine shaft-blade system is output. The evaluation and analysis not only considers the evaluation results of shaft torsional vibration and blade vibration individually, but also fully considers the correlation between the two, which can more comprehensively and accurately reflect the overall operating state of the turbine. This makes the evaluation results not only accurate, but also understandable to engineers, greatly enhancing users' trust in the intelligent diagnostic system.

[0031] This invention provides a method for noise suppression processing of multimodal synchronization signals. Figure 2This diagram illustrates the implementation flow of a noise suppression method for multimodal synchronization signals. The method specifically includes: S101: Acquire multimodal synchronization signals. These signals are then segmented based on the turbine shaft-blade rotation cycle to obtain multiple segmented signal windows. By binding the signals to the mechanical rotation cycle, hard alignment of signals from different modes is achieved in the time domain. This ensures that all signals are anchored within the same one or more complete mechanical cycles, regardless of any minor asynchrony in the original sampling. This establishes a time reference for subsequent cross-modal phase and correlation analysis. S102 uses polynomial fitting to extract the trend term in the multimodal synchronization signal, removes the trend term from the multimodal synchronization signal within the segmented signal window, and outputs the multimodal synchronization signal after removing the trend term. Removing the trend term from the multimodal synchronization signal within the segmented signal window can effectively remove these slowly varying interferences that are unrelated to the vibration of the machine itself, thus ensuring that the weak periodic oscillation characteristics submerged under the huge DC offset can be clearly revealed after removing the trend term, making the subsequent frequency domain analysis more sensitive and accurate. S103: Obtain the multimodal synchronization signal after removing the trend term. Set a bandpass filter based on the theoretical frequency range of shaft torsional vibration and blade vibration. Perform bandpass filtering on the multimodal synchronization signal based on the bandpass filter. By retaining the effective signal bandwidth and eliminating out-of-band noise, this step can maximize the signal-to-noise ratio. A high signal-to-noise ratio is a key prerequisite for ensuring the success of subsequent feature extraction. S104 performs edge detection on the multimodal synchronization signal after bandpass filtering to determine the continuity of the signal. If the signal is incomplete or falsely triggered, interpolation is performed based on the multimodal synchronization signals corresponding to the previous and next sampling points to obtain the interpolated multimodal synchronization signal.

[0032] In this embodiment of the invention, when processing noise suppression for multimodal synchronization signals, the integrity and quality of the input data are fundamentally guaranteed by actively identifying and repairing these defects. Furthermore, the noise suppression process effectively removes noise and interference components from the multimodal synchronization signals, improving signal quality and signal-to-noise ratio. Ultimately, the multimodal synchronization signals processed by noise suppression can more clearly reflect the true state of turbine shaft torsional vibration and blade vibration, thereby providing a more accurate basis for fault diagnosis and condition assessment, and significantly improving the accuracy and reliability of fault diagnosis.

[0033] This invention provides a method for correlation analysis of multimodal synchronization signals. Figure 3 This diagram illustrates the implementation flow of a correlation analysis method for multimodal synchronization signals. The method specifically includes: S201: Acquire the noise-suppressed multimodal synchronization signal, identify the shaft rotation signal and key phase pulse signal within the multimodal synchronization signal, and establish the phase relationship between the shaft rotation signals using the key phase pulse signal as a reference. By identifying the shaft rotation signal and key phase pulse signal in the multimodal synchronization signal and establishing their phase relationship, the dynamic characteristics of shaft rotation can be clearly described. This clarification of the phase relationship helps to deeply understand the propagation and variation laws of shaft torsional vibration, providing crucial information for subsequent torsional vibration demodulation and correlation analysis. S202, map the phase relationship between shaft rotation signals to sampling points, demodulate the torsional vibration amplitude and frequency at the sampling points based on the sampling points, and generate torsional vibration demodulation timing information containing torsional vibration amplitude and frequency; S203: Obtain the blade vibration signal from the multi-mode synchronization signal. Using the torsional vibration demodulation timing information as a reference, perform high-order harmonic extension on the torsional vibration demodulation timing information to obtain the extended reference signal. Using the blade vibration signal as input and the extended reference signal as reference input, perform adaptive filtering on the blade vibration signal based on the adaptive filter to output the adaptive residual signal. S204 loads an adaptive residual signal and performs sparsification processing on the adaptive residual signal based on the Morlet wavelet dictionary. Then, it uses the matching pursuit method to extract the vibration mode feature components corresponding to the blade vibration signal from the adaptive residual signal. The Morlet wavelet dictionary-based sparsification decomposes the signal into a series of sparse feature components, significantly improving signal interpretability and making it easier to extract key features from complex signals. The matching pursuit method, combined with the extraction of vibration mode feature components from the adaptive residual signal, accurately identifies the modal characteristics of blade vibration. This provides high-quality feature input for subsequent coupled analysis, improving the performance and effectiveness of the entire analysis system. S205: Obtain torsional vibration demodulation timing information and vibration mode characteristic components; determine the cross-modulation spectrum of torsional vibration demodulation timing information and vibration mode characteristic components based on the mutual information function; and determine the cross-phase relationship at the sampling point by combining the torsional vibration demodulation timing information and vibration mode characteristic components.

[0034] In this embodiment of the invention, correlation analysis of multimodal synchronization signals not only considers the frequency, amplitude, and phase relationships between torsional vibration demodulation timing information and vibration mode characteristic components, but also delves into the complex coupling relationships between torsional vibration demodulation timing information and vibration mode characteristic components through higher-order harmonic spread and cross-modulation spectrum analysis. By accurately extracting the characteristic information of torsional vibration and blade vibration and quantifying their coupling relationships, the accuracy of fault diagnosis can be significantly improved. Finally, based on the coupled mode characteristics obtained from deep correlation analysis, the operating status of the steam turbine can be more accurately assessed. This method can provide more comprehensive condition assessment results, providing a more reliable basis for maintenance personnel to formulate reasonable maintenance strategies, thereby extending the service life of the steam turbine, improving operating efficiency, and reducing maintenance costs.

[0035] This invention provides a cross-modulation spectrum method for determining torsional vibration demodulation timing information and vibration mode characteristic components based on mutual information functions. Figure 4 A schematic diagram of the implementation process of the cross-modulation spectrum method for determining torsional vibration demodulation timing information and vibration mode characteristic components based on mutual information functions is shown. The cross-modulation spectrum method for determining torsional vibration demodulation timing information and vibration mode characteristic components based on mutual information functions specifically includes: S301 loads the torsional vibration demodulation timing information and vibration modal characteristic components, and performs a Fourier transform on the torsional vibration demodulation timing information and vibration modal characteristic components to obtain the Fourier transformed torsional vibration demodulation signal and vibration modal signal. The mutual information function can be calculated in either the time domain or the frequency domain. Calculating in the frequency domain allows the analysis to focus on the orthogonal components composed of Fourier basis functions, which can sometimes more effectively capture the nonlinear dependencies between specific frequencies, thus providing structured data for sliding window calculations. S302: Obtain the torsional vibration demodulated signal and vibration mode signal after Fourier transform. Perform sliding window mutual information calculation on the torsional vibration demodulated signal and vibration mode signal to obtain the signal mutual information of the torsional vibration demodulated signal and vibration mode signal. Construct a mutual information function based on the signal mutual information. Mutual information is a universal measure of the amount of information shared between two random variables. It is sensitive to any form of statistical dependence (linear, nonlinear, monotonic, non-monotonic). In steam turbines, due to factors such as material nonlinearity, geometric nonlinearity, and contact gap, the excitation of blades by torsional vibration is very likely to be nonlinear. Moreover, the mutual information function is the only powerful tool that can objectively and quantitatively reveal this complex nonlinear coupling relationship from the data. The mutual information function can be regarded as a two-dimensional correlation heat map, with the torsional vibration frequency and blade mode frequency on the horizontal and vertical axes, respectively. The color intensity of the heat map represents the magnitude of the mutual information between the frequency pairs.

[0036] The mutual information function is expressed as:

[0037]

[0038] in, Indicates torsional vibration demodulation signal and vibration mode signals Mutual information function, Indicates torsional vibration demodulation signal and vibration mode signals The joint density function, The peak value in the mutual information curve; S303, based on the mutual information function, generates a mutual information curve, and uses the peak value in the mutual information curve as a reference to perform phase compensation on the torsional vibration demodulation signal to obtain the torsional vibration modulation function. The peak value of the mutual information curve represents the strongest coupling point between torsional vibration and blade vibration, which locks the focus for subsequent in-depth analysis and avoids interference from secondary factors. The torsional vibration modulation function is expressed as:

[0039] in, These are the torsional vibration modulation weights and the time spectrum of the vibration mode signal, respectively. S304, obtain the torsional vibration modulation function, coherently weight the vibration mode signal based on the torsional vibration modulation function, map the coherently weighted vibration mode signal and torsional vibration modulation function to the mutual information curve, identify the high-order harmonic sidebands in the mutual information curve, and use the median filter to reconstruct the high-order harmonic sidebands to obtain the coupled mode characteristics characterizing the integrated state of the turbine shaft system-blade.

[0040] In this embodiment of the invention, when determining the cross-modulation spectrum of torsional vibration demodulation timing information and vibration mode characteristic components based on the mutual information function, the mutual information function is used to accurately quantify the nonlinear coupling relationship between the torsional vibration demodulation timing information and vibration mode characteristic components. Furthermore, a sliding window analysis is used to capture transient coupling dynamics. By mapping the coherently weighted vibration mode signal and the torsional vibration modulation function onto the mutual information curve, higher-order harmonic sidebands in the mutual information curve can be identified. Finally, a median filter is used to reconstruct the higher-order harmonic sidebands, obtaining coupled mode characteristics that characterize the integrated state of the turbine shaft system and blades. The mode reconstruction method effectively removes noise and interference, ensuring the accuracy and reliability of the coupled mode characteristics.

[0041] In this embodiment of the invention, the coupling analysis model is based on a BiLSTM model and includes an input layer and an output layer. The input layer is connected to the BiLSTM model, and the BiLSTM model is connected to the output layer. A parallel feature extraction layer is set between the input and output layers and the BiLSTM model. The parallel feature extraction layer includes a feature encoding block, a parallel deep feature extraction network, and a graph neural network. The input layer is used to acquire coupled modal features and pass them to the feature extraction layer. The feature encoding block identifies torsional vibration assessment features, blade vibration assessment features, and torsional vibration-blade correlation assessment features in the coupled modal features. The deep feature extraction network is used to acquire torsional vibration assessment features and torsional vibration-blade correlation assessment features. Based on the deep feature extraction network, torsional vibration anomaly indicators are extracted from the torsional vibration assessment features and torsional vibration-blade correlation assessment features. The torsional vibration anomaly indicators are weighted based on an expert rule base of rotor dynamics and fatigue theory. A multi-head torsional vibration model is constructed based on the weighted torsional vibration anomaly indicators using a cross-attention mechanism. The vibration anomaly assessment matrix is ​​processed by a physical data residual network to handle the residual components between the anomaly assessment matrix and the standard assessment matrix. Uncertainty analysis is performed on the residual decomposition based on a Bayesian hierarchical network to obtain a quantified shaft torsional vibration assessment value. The graph neural network is used to obtain blade vibration assessment features and torsional vibration-blade correlation assessment features. A graph neural network heterogeneous graph is constructed based on the blade vibration assessment features and torsional vibration-blade correlation assessment features. The influence weights of torsional vibration-blade are explicitly learned using a cross-attention mechanism and assigned to the graph nodes in the graph neural network heterogeneous graph. The blade vibration assessment value is determined based on a multi-output regression network combined with the influence weights of the graph nodes. The BiLSTM model is used to obtain shaft torsional vibration assessment values ​​and blade vibration assessment values, construct a torsional vibration-blade digital twin network, and update the weight parameters of the torsional vibration-blade digital twin network based on the shaft torsional vibration assessment values ​​and blade vibration assessment values. Global pooling is used to automatically learn the turbine shaft-blade comprehensive state classification assessment and output the turbine shaft-blade comprehensive state value.

[0042] This invention provides a method for evaluating and analyzing coupled modal characteristics based on a pre-constructed coupling analysis model. Figure 5 The diagram illustrates the implementation flow of a method for evaluating and analyzing coupled modal characteristics based on a pre-built coupling analysis model. Specifically, this method includes: S401, acquire coupled modal features and pass them to the feature extraction layer. The feature encoding block identifies torsional vibration evaluation features, blade vibration evaluation features and torsional vibration-blade correlation evaluation features in the coupled modal features. S402, a deep feature extraction network is used to obtain torsional vibration assessment features and torsional vibration-blade correlation assessment features, and torsional vibration anomaly indicators are extracted from the torsional vibration assessment features and torsional vibration-blade correlation assessment features based on the deep feature extraction network. S403 assigns weights to torsional vibration anomaly indicators based on an expert rule base of rotor dynamics and fatigue theory. Based on the cross-attention mechanism, a multi-head torsional vibration anomaly evaluation matrix is ​​constructed based on the weighted torsional vibration anomaly indicators. By introducing an expert rule base of rotor dynamics and fatigue theory to assign weights to the anomaly indicators, the model is infused with deep domain knowledge in the initial stage of learning, thereby ensuring that the anomaly indicators and evaluation directions that the model focuses on are in line with physical laws. S404 uses a physical data residual network to process the residual components between the anomaly evaluation matrix and the standard evaluation matrix. Based on a Bayesian hierarchical network, it performs uncertainty analysis on the residual decomposition to obtain a quantified shaft torsional vibration assessment value. The physical data residual network uses the standard evaluation matrix based on physical rules as a benchmark to calculate the difference between the model output and the physical benchmark. The goal of the network is to learn this residual and correct it, thereby achieving the complementary advantages of data and mechanism.

[0043] S405, a graph neural network is used to obtain blade vibration assessment features and torsional vibration-blade correlation assessment features, and a graph neural heterogeneous graph is constructed based on the blade vibration assessment features and torsional vibration-blade correlation assessment features. S406 employs a cross-attention mechanism to explicitly learn the influence weights of torsional vibration and blade vibration, and assigns these influence weights to graph nodes in a graph neural heterogeneous graph. Based on a multi-output regression network combined with the influence weights of graph nodes, the blade vibration assessment value is determined. The multi-head torsional vibration anomaly evaluation matrix constructed by the cross-attention mechanism presents the abstract anomaly indicators and their interrelationships in a structured matrix form. The formula for calculating the shaft system torsional vibration assessment value is as follows:

[0044] in, This is the value for shaft torsional vibration assessment. This is the weight vector for torsional vibration anomaly indicators. These are, respectively, the torsional vibration anomaly index vector, the residual correction term output by the residual network, and the uncertainty quantification term of the Bayesian hierarchical network;

[0045] in, These are respectively random uncertainty error, model uncertainty, and prior deviation. These are the anomaly assessment matrix and the standard assessment matrix, respectively. These are the variance penalty strength and the regularization weight, respectively. The formula for calculating the blade vibration assessment value is as follows:

[0046] in, This indicates the blade vibration assessment value. These are the original features of the graph nodes and the original features of the connections between the graph nodes, respectively. The weights of the edges connecting graph nodes and their associated nodes; S407: Obtain the shaft torsional vibration assessment value and the blade vibration assessment value, construct the torsional vibration-blade digital twin network, and update the weight parameters of the torsional vibration-blade digital twin network based on the shaft torsional vibration assessment value and the blade vibration assessment value. The S408 employs global pooling-based automatic learning to classify and assess the integrated condition of the turbine shaft system and blades. It outputs integrated condition values ​​for the turbine shaft system and blades, and the final output values, along with their implicit trend information, serve as a direct basis for predictive maintenance. Maintenance personnel can plan the optimal maintenance window before serious equipment failures occur based on the trends in these condition values, thereby maximizing equipment availability and reducing maintenance costs and risks.

[0047] In this embodiment of the invention, when evaluating and analyzing coupled modal features based on a pre-built coupled analysis model, the torsional vibration evaluation features, blade vibration evaluation features, and torsional vibration-blade correlation evaluation features in the coupled modal features can be identified through the feature extraction layer and feature encoding block. The deep feature extraction network can automatically extract torsional vibration anomaly indicators from the torsional vibration evaluation features and the torsional vibration-blade correlation evaluation features. The deep learning method can automatically learn complex patterns in the signal, improving the accuracy and efficiency of feature extraction. Furthermore, by processing the residual components between the anomaly judgment matrix and the standard judgment matrix through the physical data residual network, the abnormal components in the signal can be further analyzed, thereby improving the sensitivity of the system and ensuring that even small abnormal changes can be detected. By constructing a torsional vibration-blade digital twin network and updating the network weight parameters based on the shaft torsional vibration evaluation value and the blade vibration evaluation value, real-time dynamic modeling of the turbine's operating status can be achieved. This digital twin network can provide more accurate operating status evaluation, providing real-time decision support for operation and maintenance personnel. Finally, the global pooling method can automatically learn key features in the signal, improving the system's adaptability and flexibility, and ensuring the accuracy and reliability of the evaluation results.

[0048] On the other hand, embodiments of the present invention also provide a synchronous measurement and analysis system for turbine shaft torsional vibration and blade vibration. Figure 6 This diagram illustrates the structure of a synchronous measurement and analysis system for torsional vibration of a steam turbine shaft system and blade vibration. The system includes: The signal acquisition module 100 collects shaft rotation signals, key phase pulse signals, and blade vibration signals respectively based on distributed sensors, and unifies the shaft rotation signals, key phase pulse signals, and blade vibration signals with a unified time reference to obtain a multi-modal synchronization signal; The noise suppression module 200 is used to acquire a multi-mode synchronization signal with a unified time base, perform noise suppression processing on the multi-mode synchronization signal, and obtain a noise-suppressed multi-mode synchronization signal. The noise suppression module 200 includes: The signal segmentation unit 210 is used to acquire multi-mode synchronization signals and process the multi-mode synchronization signals into segments based on the turbine shaft-blade rotation cycle to obtain multi-group segmented signal windows; The trend removal unit 220 uses polynomial fitting to extract the trend term in the multimodal synchronization signal, removes the trend term from the multimodal synchronization signal within the segmented signal window, and outputs the multimodal synchronization signal after trend removal. The bandpass filter unit 230 is used to acquire the multimodal synchronization signal after removing the trend term. The bandpass filter is set based on the theoretical frequency range of shaft torsional vibration and blade vibration. The multimodal synchronization signal is bandpass filtered based on the bandpass filter. The interpolation processing unit 240 is used to perform edge detection on the multimodal synchronization signal after bandpass filtering, determine the continuity of the signal, and if the signal is incomplete or falsely triggered, perform interpolation processing based on the multimodal synchronization signals corresponding to the previous and next sampling points to obtain the interpolated multimodal synchronization signal.

[0049] The turbine shaft torsional vibration and blade vibration synchronous measurement and analysis system also includes: The modal coupling module 300 loads a noise-suppressed multi-modal synchronization signal, performs correlation analysis on the multi-modal synchronization signal, and obtains the coupled modal characteristics characterizing the integrated state of the turbine shaft system and blades based on the correlation analysis results. The state assessment module 400 is used to acquire coupled modal characteristics, evaluate and analyze the coupled modal characteristics based on a pre-built coupled analysis model, and output the turbine shaft-blade integrated state value. The status warning module 500 is used to acquire the comprehensive status value of the turbine shaft system and blades, determine whether the comprehensive status value exceeds the preset status warning threshold, and if the comprehensive status value exceeds the preset status warning threshold, trigger a status warning command and visualize the comprehensive status value of the turbine shaft system and blades based on the status warning command.

[0050] In summary, this invention provides a method and system for synchronous measurement and analysis of turbine shaft torsional vibration and blade vibration. In this embodiment, distributed sensors simultaneously collect shaft rotation signals, key phase pulse signals, and blade vibration signals, unifying these signals with a unified time reference to form a multi-mode synchronous signal. An adaptive filter is used to filter the blade vibration signal, extracting vibration mode characteristic components. Furthermore, based on the mutual information function, the cross-modulation spectrum of the torsional vibration demodulation timing information and the vibration mode characteristic components is determined, thereby accurately determining the cross-phase relationship at the sampling point and obtaining the coupled modal characteristics characterizing the integrated state of the turbine shaft-blade system. Based on a pre-constructed coupled analysis model, the coupled modal characteristics are evaluated and analyzed, outputting the integrated state value of the turbine shaft-blade system. The evaluation and analysis not only considers the individual evaluation results of shaft torsional vibration and blade vibration but also fully considers the correlation between the two, which can more comprehensively and accurately reflect the overall operating state of the turbine. This makes the evaluation results not only accurate but also understandable to engineers, greatly enhancing users' trust in the intelligent diagnostic system.

[0051] It should be noted that, for the sake of simplicity, the foregoing embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to the present invention. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0052] It should be understood that the disclosed apparatus can be implemented in other ways, given the several embodiments provided in this application. For example, the apparatus embodiments described above are merely illustrative; the division of units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or communication connections shown or discussed may be through some interfaces; the indirect coupling or communication connections between devices or units may be telecommunications or other forms.

[0053] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on these embodiments, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art can still combine, add, delete, or otherwise adjust the features of the various embodiments of the present invention according to the circumstances without conflict or creative effort, thereby obtaining different technical solutions that do not fundamentally depart from the concept of the present invention. These technical solutions also fall within the scope of protection of the present invention.

Claims

1. A method for synchronous measurement and analysis of torsional vibration and blade vibration of a steam turbine shaft system, characterized in that, The method includes: Based on distributed sensors, shaft rotation signals, key phase pulse signals, and blade vibration signals are collected respectively, and the shaft rotation signals, key phase pulse signals, and blade vibration signals are unified with a time reference to obtain a multi-modal synchronization signal; A multi-mode synchronization signal with a unified time base is obtained, and noise suppression processing is performed on the multi-mode synchronization signal to obtain a noise-suppressed multi-mode synchronization signal. After applying noise-suppressed multimodal synchronization signals, correlation analysis is performed on the multimodal synchronization signals. Based on the correlation analysis results, coupled modal characteristics characterizing the integrated state of the turbine shaft system and blades are obtained. The coupled modal characteristics are obtained, and the coupled modal characteristics are evaluated and analyzed based on the pre-constructed coupled analysis model to output the integrated state value of the turbine shaft system and blades.

2. The method for synchronous measurement and analysis of turbine shaft torsional vibration and blade vibration according to claim 1, characterized in that, The method further includes: The integrated status value of the turbine shaft system and blades is obtained. It is determined whether the integrated status value exceeds the preset status warning threshold. If the integrated status value exceeds the preset status warning threshold, a status warning command is triggered, and the integrated status value of the turbine shaft system and blades is visualized based on the status warning command.

3. The method for synchronous measurement and analysis of turbine shaft torsional vibration and blade vibration according to claim 1, characterized in that, The noise suppression processing for the multimodal synchronization signal includes: The multi-modal synchronization signal is acquired and processed in segments based on the turbine shaft-blade rotation cycle to obtain multi-segmented signal windows. The trend term in the multimodal synchronization signal is extracted by polynomial fitting. The trend term is removed from the multimodal synchronization signal within the segmented signal window, and the multimodal synchronization signal after trend term removal is output. The multimodal synchronization signal after removing the trend term is obtained. A bandpass filter is set based on the theoretical frequency range of shaft torsional vibration and blade vibration. The multimodal synchronization signal is then processed by bandpass filtering based on the bandpass filter. Edge detection is performed on the multimodal synchronization signal after bandpass filtering to determine the continuity of the signal. If the signal is incomplete or falsely triggered, interpolation is performed based on the multimodal synchronization signals corresponding to the previous and subsequent sampling points to obtain the interpolated multimodal synchronization signal.

4. The method for synchronous measurement and analysis of turbine shaft torsional vibration and blade vibration according to claim 1, characterized in that, The correlation analysis of the multimodal synchronization signal includes: The noise-suppressed multimode synchronization signal is acquired, and the shaft rotation signal and key phase pulse signal in the multimode synchronization signal are identified. The phase relationship between the shaft rotation signals is established with the key phase pulse signal as a reference. The phase relationship between the shaft rotation signals is mapped to the sampling point. Based on the sampling point, the torsional vibration amplitude and frequency at the sampling point are demodulated to generate torsional vibration demodulation timing information containing the torsional vibration amplitude and frequency. The blade vibration signal in the multimodal synchronization signal is acquired. Based on the torsional vibration demodulation timing information, the torsional vibration demodulation timing information is extended by higher-order harmonics to obtain the extended reference signal. The blade vibration signal is used as the input and the extended reference signal is used as the reference input. The blade vibration signal is adaptively filtered based on the adaptive filter to output the adaptive residual signal. The adaptive residual signal is loaded, and the adaptive residual signal is sparsified based on the Morlet wavelet dictionary. The vibration mode feature components corresponding to the blade vibration signal are extracted from the adaptive residual signal by the matching pursuit method. The torsional vibration demodulation timing information and vibration mode characteristic components are obtained. The cross-modulation spectrum of the torsional vibration demodulation timing information and vibration mode characteristic components is determined based on the mutual information function. The cross-phase relationship at the sampling point is determined by combining the torsional vibration demodulation timing information and vibration mode characteristic components cross-modulation spectrum.

5. The method for synchronous measurement and analysis of turbine shaft torsional vibration and blade vibration according to claim 4, characterized in that, The cross-modulation spectrum for determining torsional vibration demodulation timing information and vibration mode characteristic components based on mutual information functions includes: Load torsional vibration demodulation timing information and vibration mode characteristic components, perform Fourier transform on the torsional vibration demodulation timing information and vibration mode characteristic components, and obtain the Fourier transformed torsional vibration demodulation signal and vibration mode signal; The torsional vibration demodulated signal and vibration mode signal after Fourier transform are obtained. The mutual information of the torsional vibration demodulated signal and vibration mode signal is calculated by sliding window to obtain the signal mutual information of the torsional vibration demodulated signal and vibration mode signal. The mutual information function is constructed based on the signal mutual information. Mutual information curves are generated based on mutual information functions, and phase compensation is performed on the torsional vibration demodulation signal using the peak value in the mutual information curves as a reference to obtain the torsional vibration modulation function. The torsional vibration modulation function is obtained, and the vibration mode signal is coherently weighted based on the torsional vibration modulation function. The coherently weighted vibration mode signal and the torsional vibration modulation function are mapped to the mutual information curve. The high-order harmonic sidebands in the mutual information curve are identified, and the high-order harmonic sidebands are reconstructed using a median filter to obtain the coupled mode characteristics that characterize the integrated state of the turbine shaft system and blades.

6. The method for synchronous measurement and analysis of turbine shaft torsional vibration and blade vibration according to claim 5, characterized in that, The coupling analysis model is based on the BiLSTM model and also includes an input layer and an output layer. The input layer is connected to the BiLSTM model, and the BiLSTM model is connected to the output layer. A parallel feature extraction layer is set between the input and output layers and the BiLSTM model. The parallel feature extraction layer includes a feature encoding block, a parallel deep feature extraction network, and a graph neural network.

7. The method for synchronous measurement and analysis of turbine shaft torsional vibration and blade vibration according to claim 6, characterized in that, The evaluation and analysis of coupled modal characteristics based on the pre-built coupling analysis model includes: The coupled modal features are acquired and passed to the feature extraction layer. The feature encoding block identifies torsional vibration evaluation features, blade vibration evaluation features, and torsional vibration-blade correlation evaluation features in the coupled modal features. A deep feature extraction network is used to obtain torsional vibration assessment features and torsional vibration-blade correlation assessment features. Based on the deep feature extraction network, torsional vibration anomaly indicators are extracted from the torsional vibration assessment features and torsional vibration-blade correlation assessment features. The torsional vibration anomaly index is weighted based on an expert rule base of rotor dynamics and fatigue theory, and a multi-head torsional vibration anomaly evaluation matrix is ​​constructed based on the weighted torsional vibration anomaly index using a cross-attention mechanism. The residual components between the anomaly evaluation matrix and the standard evaluation matrix are processed by physical data residual network. Uncertainty analysis is performed on the residual decomposition based on Bayesian hierarchical network to obtain quantified shaft torsional vibration evaluation values.

8. The method for synchronous measurement and analysis of turbine shaft torsional vibration and blade vibration according to claim 7, characterized in that, The evaluation and analysis of coupled modal characteristics based on the pre-built coupling analysis model also includes: Graph neural networks are used to obtain blade vibration assessment features and torsional vibration-blade correlation assessment features. Based on the blade vibration assessment features and torsional vibration-blade correlation assessment features, a graph neural heterogeneous graph is constructed. The cross-attention mechanism is used to explicitly learn the influence weight of torsional vibration-blade, and the influence weight is assigned to the graph nodes in the graph neural heterogeneous graph. The blade vibration assessment value is determined based on the multi-output regression network combined with the influence weight of the graph nodes. Obtain the shaft torsional vibration assessment value and the blade vibration assessment value, construct a torsional vibration-blade digital twin network, and update the weight parameters of the torsional vibration-blade digital twin network based on the shaft torsional vibration assessment value and the blade vibration assessment value; A global pooling-based automatic learning method is used to classify and evaluate the integrated state of the turbine shaft system and blades, and output the integrated state value of the turbine shaft system and blades.

9. A synchronous measurement and analysis system for torsional vibration and blade vibration of a steam turbine shaft system, used to implement the synchronous measurement and analysis method for torsional vibration and blade vibration of a steam turbine shaft system as described in any one of claims 1-8, characterized in that, The turbine shaft torsional vibration and blade vibration synchronous measurement and analysis system includes: The signal acquisition module collects shaft rotation signals, key phase pulse signals, and blade vibration signals respectively based on distributed sensors, and unifies the shaft rotation signals, key phase pulse signals, and blade vibration signals with a unified time reference to obtain a multi-modal synchronization signal; The noise suppression module is used to acquire a multi-mode synchronization signal with a unified time base, perform noise suppression processing on the multi-mode synchronization signal, and obtain a noise-suppressed multi-mode synchronization signal. The modal coupling module loads the noise-suppressed multi-modal synchronization signal, performs correlation analysis on the multi-modal synchronization signal, and obtains the coupled modal characteristics characterizing the integrated state of the turbine shaft system and blades based on the correlation analysis results. The state assessment module is used to acquire coupled modal characteristics, evaluate and analyze the coupled modal characteristics based on a pre-built coupled analysis model, and output the turbine shaft-blade integrated state value. The status warning module is used to acquire the comprehensive status value of the turbine shaft system and blades, determine whether the comprehensive status value exceeds the preset status warning threshold, and if the comprehensive status value exceeds the preset status warning threshold, trigger a status warning command and visualize the comprehensive status value of the turbine shaft system and blades based on the status warning command.

10. The synchronous measurement and analysis system for turbine shaft torsional vibration and blade vibration according to claim 9, characterized in that, The noise suppression module includes: The signal segmentation unit is used to acquire multi-mode synchronization signals and process the multi-mode synchronization signals into segments based on the turbine shaft-blade rotation cycle to obtain multi-group segmented signal windows; The trend term removal unit uses polynomial fitting to extract the trend term in the multimodal synchronization signal, removes the trend term from the multimodal synchronization signal within the segmented signal window, and outputs the multimodal synchronization signal after trend term removal. The bandpass filter unit is used to acquire the multimodal synchronization signal after removing the trend term. The bandpass filter is set based on the theoretical frequency range of shaft torsional vibration and blade vibration, and the multimodal synchronization signal is bandpass filtered based on the bandpass filter. The interpolation processing unit is used to perform edge detection on the multimodal synchronization signal after bandpass filtering to determine the continuity of the signal. If the signal is incomplete or falsely triggered, interpolation processing is performed based on the multimodal synchronization signals corresponding to the previous and next sampling points to obtain the interpolated multimodal synchronization signal.

Images