A method and device for determining the vibration state of a steam turbine blade

Abstract

This application discloses a method and apparatus for determining the vibration state of a steam turbine blade, relating to the field of steam turbine condition monitoring technology. Its main purpose is to provide timely and accurate early warning of the vibration state of steam turbine blades, thereby reducing economic losses and safety risks. The main technical solution is as follows: Initial cracks are pre-fabricated on the steam turbine blade; these initial cracks are used to monitor the propagation under abnormal vibration stress. Acoustic emission signals and visual image sequences of the initial cracks are collected during steam turbine operation. A set of dynamic active features corresponding to the acoustic emission signals and a set of static morphological features corresponding to the visual image sequences are extracted. The dynamic active features and static morphological features are then fused to obtain a target multimodal feature vector. Based on a pre-established mapping relationship and the target multimodal feature vector, the target vibration state of the steam turbine blade is determined. The mapping relationship characterizes the correspondence between different multimodal feature vectors and different vibration states of the steam turbine blade.

Description

Technical Field

[0001] This application relates to the field of turbine condition monitoring technology, and in particular to a method and apparatus for determining the vibration state of turbine blades. Background Technology

[0002] As a core piece of equipment in the modern energy and power industry, steam turbines operate under extremely complex environments of high temperature, high pressure, and high speed for extended periods. Their rotor blades must withstand enormous centrifugal stress, aerodynamic vibration, and steam corrosion, making them critical components highly susceptible to fatigue damage. If cracks appear on the blades and are not detected in time, they will rapidly propagate under alternating stress, potentially leading to blade breakage, interstage collisions, or even complete turbine failure.

[0003] Currently, existing technologies for crack detection in in-service steam turbine blades mainly rely on two methods: offline non-destructive testing (NDT) during periodic overhauls, such as penetrant testing and ultrasonic testing; and vibration signal analysis during operation, which indirectly determines the blade condition by monitoring vibration signals from the casing or bearing housing. However, offline NDT can only be performed when the turbine is shut down, making it impossible to monitor crack initiation and early propagation in real time. Vibration signal analysis, on the other hand, is insensitive to small early cracks and specific local vibration modes such as higher-order vibration modes. More importantly, both methods can only determine the presence of cracks or roughly estimate their size, failing to capture key behavioral characteristics such as crack propagation activity and direction. This results in a significant lag in early warning of turbine blade vibration conditions, easily missing the optimal intervention opportunity and leading to serious economic losses and safety risks. Summary of the Invention

[0004] In view of the above problems, this application provides a method and device for determining the vibration state of steam turbine blades. The main purpose is to provide timely and accurate early warning of the vibration state of steam turbine blades, thereby reducing economic losses and safety risks.

[0005] To solve the above-mentioned technical problems, this application proposes the following solution: In a first aspect, this application provides a method for determining the vibration state of a steam turbine blade, the method comprising: Initial cracks are pre-formed on the turbine blades, and these initial cracks are designed to allow for monitorable propagation under abnormal vibration stress. Acoustic emission signals and visual image sequences of the initial crack were collected during turbine operation; Extract a set of dynamic activity features corresponding to the acoustic emission signal, extract a set of static morphology features corresponding to the visual image sequence, and fuse the dynamic activity features and the static morphology features to obtain a target multimodal feature vector; Based on the pre-established mapping relationship and the target multimodal feature vector, the target vibration state of the turbine blade is determined. The mapping relationship is used to characterize the correspondence between different multimodal feature vectors and different vibration states of the turbine blade.

[0006] Secondly, this application provides a device for determining the vibration state of a steam turbine blade, the device comprising: Prefabrication unit for prefabricating initial cracks on turbine blades, the initial cracks being designed to allow for monitorable propagation under abnormal vibration stress; The acquisition unit is used to acquire the acoustic emission signals and visual image sequences of the initial crack obtained by the prefabrication unit during turbine operation; The processing unit is used to extract a set of dynamic active features corresponding to the acoustic emission signal obtained by the acquisition unit, extract a set of static morphological features corresponding to the visual image sequence, and fuse the dynamic active features and the static morphological features to obtain a target multimodal feature vector. The first determining unit is used to determine the target vibration state of the turbine blade based on the pre-established mapping relationship and the target multimodal feature vector obtained by the processing unit. The mapping relationship is used to characterize the correspondence between different multimodal feature vectors and different vibration states of the turbine blade.

[0007] To achieve the above objectives, according to a third aspect of this application, a storage medium is provided, the storage medium including a stored program, wherein, when the program is executed, the device where the storage medium is located is controlled to perform the turbine blade vibration state determination method of the first aspect described above.

[0008] To achieve the above objectives, according to a fourth aspect of this application, a processor is provided, the processor being configured to run a program, wherein the program, when running, executes the turbine blade vibration state determination method of the first aspect described above.

[0009] By employing the above technical solution, this application provides a method and apparatus for determining the vibration state of a steam turbine blade. By pre-fabricating initial cracks sensitive to abnormal vibration stress on the steam turbine blade, it provides a monitoring target for capturing crack responses related to vibration state. By simultaneously acquiring acoustic emission signals and visual image sequences of the initial cracks, it achieves real-time data acquisition during operation, overcoming the time limitations of offline detection. Furthermore, it combines the dynamic real-time nature of acoustic emission signals with the intuitiveness of visual images, compensating for the shortcomings of long transmission paths and large attenuation of single vibration signals, ensuring the comprehensiveness and timeliness of data acquisition. By extracting and fusing dynamic and static morphological features, the constructed target multimodal feature vector can accurately capture crack responses. The study identifies key behavioral characteristics such as crack propagation activity and direction, and achieves comprehensive characterization of dynamic behavior and static morphology. This overcomes the limitations of existing technologies that can only determine the presence or rough estimation of crack size. Ultimately, through a pre-established mapping relationship, the multimodal feature vectors are directly correlated with the blade vibration state, enabling precise determination of the target vibration state. This represents a shift from passively detecting cracks to actively predicting vibration risks, improving the timeliness and accuracy of early warnings. This effectively avoids missing the optimal intervention opportunity due to delayed early warnings, significantly reducing safety risks such as interstage collisions and overall turbine damage caused by blade fracture, and minimizing huge economic losses caused by unplanned shutdowns or improper maintenance. This provides reliable support for the health status assessment and remaining life prediction of turbine blades.

[0010] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0011] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart of a method for determining the vibration state of a steam turbine blade provided in an embodiment of this application is shown; Figure 2 A flowchart of another method for determining the vibration state of a steam turbine blade provided in an embodiment of this application is shown; Figure 3 This paper shows a block diagram of a device for determining the vibration state of a steam turbine blade according to an embodiment of this application. Figure 4This paper shows a block diagram of another turbine blade vibration state determination device provided in an embodiment of this application. Detailed Implementation

[0012] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the present application to those skilled in the art.

[0013] Currently, existing technologies for crack detection in in-service steam turbine blades mainly rely on two methods: offline non-destructive testing (NDT) during periodic overhauls, such as penetrant testing and ultrasonic testing; and vibration signal analysis during operation, which indirectly determines the blade condition by monitoring vibration signals from the casing or bearing housing. However, offline NDT can only be performed when the turbine is shut down, making it impossible to monitor crack initiation and early propagation in real time. Vibration signal analysis, on the other hand, is insensitive to small early cracks and specific local vibration modes such as higher-order vibration modes. More importantly, both methods can only determine the presence of cracks or roughly estimate their size, failing to capture key behavioral characteristics such as crack propagation activity and direction. This results in a significant lag in early warning of turbine blade vibration conditions, easily missing the optimal intervention opportunity and leading to serious economic losses and safety risks.

[0014] Research has shown that by introducing initial cracks sensitive to abnormal vibration stress in the stress concentration areas of turbine blades, and simultaneously collecting and fusing dynamic features characterizing structural response activity and static features characterizing structural morphological changes during operation, a multimodal comprehensive feature can be formed. Based on a pre-defined mapping rule between these multimodal features and blade vibration state, the vibration state of the turbine blade can be determined. This allows for the precise capture of key behavioral characteristics such as crack propagation activity and direction, enabling accurate identification of the turbine blade vibration state. This shifts the focus from passively detecting cracks to proactively predicting vibration risks, providing reliable support for blade health assessment and remaining life prediction, and significantly reducing safety risks and economic losses caused by delayed early warnings.

[0015] Based on the above considerations, this application provides a method for determining the vibration state of steam turbine blades. This method enables timely and accurate early warning of the vibration state of steam turbine blades, reducing economic losses and safety risks. The specific execution steps are as follows: Figure 1 As shown, it includes: 101. Pre-fabricate initial cracks on turbine blades.

[0016] The initial crack is used to detect the monitored propagation under abnormal vibration stress.

[0017] In this step, the dynamic characteristics of the turbine blades can be analyzed in advance using finite element method (FEM) simulations to accurately locate areas of concentrated vibration stress, such as the blade root and the periphery of the tie rod holes. These areas are critical locations where fatigue cracks are most likely to develop due to alternating stress during blade operation. Pre-fabricating initial cracks in these areas maximizes the crack's sensitivity to vibration stress. After identifying the stress concentration areas, one or a group of micron-level initial cracks can be created using laser etching or micro-electrical discharge machining (EDM). The initial size (such as width and depth) and orientation of the initial cracks can be optimized through multiple rounds of FEM simulations according to preset conditions. These preset conditions must meet two requirements: first, maintaining structural stability under normal operating conditions such as rated turbine operation and conventional load fluctuations, preventing unexpected propagation and avoiding impact on blade operation; and second, being highly sensitive to abnormal vibration stresses, allowing the cracks to exhibit monitorable propagation behavior when the blades are subjected to such abnormal vibration stresses. These abnormal vibration stresses correspond to resonance modes that easily lead to blade fatigue damage, such as first-order bending, second-order bending, and torsional vibration modes.

[0018] 102. Acquire acoustic emission signals and visual image sequences of the initial crack during turbine operation.

[0019] In this step, acoustic emission signals can be acquired by deploying an acoustic emission sensor network. Specifically, this network can be deployed by installing multiple high-frequency acoustic emission sensors in a ring array or uniform distribution on the impeller disk or nearby stationary components. The sensor layout needs to be optimized through signal reachability testing to ensure that the monitoring range of all sensors completely covers the area where the initial crack is located, with no signal blind spots. The sensors are fixed to the component surface using a high-temperature coupling agent and encapsulated in a high-temperature resistant protective sleeve to withstand the high-temperature environment of the turbine operation. Visual image sequences can be acquired by deploying a visual imaging system. For the deployment of this system, an observation window can be opened in the turbine cylinder corresponding to the location of the initial crack. This window is sealed with a high-temperature, high-pressure resistant transparent quartz material to ensure that it does not affect the turbine's sealing performance and operational safety. A high-resolution, high-temperature resistant endoscopic camera or industrial camera is integrated through the observation window. The camera's viewing axis is adjusted to precisely align with the area where the initial crack is located. The lens is equipped with an anti-steam fogging coating to prevent the image clarity from being affected by steam inside the cylinder.

[0020] The turbine is controlled to operate under all operating conditions sequentially, including startup, shutdown, variable load (e.g., load fluctuations covering 30%-110% of rated load), and rated operation. A synchronization trigger module ensures time synchronization between the acoustic emission acquisition system and the visual imaging system (synchronization error not exceeding 10ms). For acoustic emission signal acquisition, a continuous acquisition mode can be used to continuously record acoustic emission waveform data and extract key parameters such as event count, signal energy, amplitude, and duration in real time. Binary data storage format is used to ensure data integrity, while the activity of acoustic emission signals is monitored in real time (e.g., using event rate as the core indicator). For visual image sequence acquisition, digital images of the initial crack area can be automatically acquired at a first preset time interval (e.g., 30 minutes / time). The acquired images can include timestamps, operating parameters (load, speed, temperature), and other relevant information to facilitate subsequent data matching and analysis.

[0021] It should be noted that in order to determine whether the visual image sequence can capture the key node morphology of crack propagation, it can also be combined with the activity of acoustic emission signals. That is, a preset activity threshold is set (this threshold can be pre-calibrated through bench experiments and is 3 times the average event rate under normal working conditions). When the activity of acoustic emission signals exceeds the preset threshold, images are acquired at a smaller second preset time interval (such as 5 minutes / time).

[0022] This step aims to acquire multimodal data related to crack propagation under all operating conditions, while ensuring the synchronization and integrity of the data.

[0023] 103. Extract a set of dynamic activity features corresponding to the acoustic emission signal, extract a set of static morphology features corresponding to the visual image sequence, and fuse the dynamic activity features and static morphology features to obtain the target multimodal feature vector.

[0024] In this step, after acquiring the acoustic emission signals and visual image sequences, both can be preprocessed. Specifically, the continuously acquired acoustic emission signals are filtered to remove environmental noise (such as steam flow noise, equipment operating background noise, etc.), and invalid signals are eliminated using a thresholding method (the threshold can be set to twice the peak value of the sensor's inherent noise) to filter out valid acoustic emission events related to crack propagation. Simultaneously, the acquired visual image sequences are processed by grayscale conversion, denoising, and image enhancement to eliminate image interference caused by factors such as steam atomization and changes in illumination. Image registration technology is used to align images acquired at different time points with the initial crack image to ensure the comparability of crack morphology changes.

[0025] Time-domain and frequency-domain analysis of the preprocessed acoustic emission signal allows for the extraction of at least one of the core features, such as event rate, signal energy, and average amplitude, as dynamic activity features. The event rate is the number of acoustic emission events exceeding a threshold per unit time, reflecting the frequency of crack propagation. Signal energy is the integral of the energy released by a single acoustic emission event, reflecting the severity of crack propagation. The average amplitude is the average signal amplitude of a single acoustic emission event, directly related to the stress level driving crack propagation. Similarly, digital image processing techniques are used to analyze the crack morphology of the preprocessed image, extracting at least one of the following as static morphological features: crack length, propagation direction angle, and morphological complexity. Crack length is the macroscopic propagation length of the crack; propagation direction angle is the angle between the main crack extension direction and the reference axis; and morphological complexity characterizes complex features such as crack bifurcation and bends, and is the average of the sum of squares of the deviations of the local curvatures of the crack edge from the average curvature.

[0026] At least one of the three dynamic activity features extracted above is fused with at least one of the three static morphology features at the feature level to construct a unified target multimodal feature vector. This vector comprehensively characterizes the dynamic propagation activity and static morphological changes of the crack within a specific time period, providing comprehensive and multidimensional data support for subsequent vibration state judgment.

[0027] This step is the core of achieving deep fusion of multimodal data. By accurately extracting features and constructing a unified vector, it provides data support for subsequent vibration state judgment.

[0028] 104. Based on the pre-established mapping relationship and the target multimodal feature vector, determine the target vibration state of the turbine blade.

[0029] The mapping relationship is used to characterize the correspondence between different multimodal feature vectors and different vibration states of turbine blades.

[0030] In this step, a multimodal eigenvector-vibration state mapping database can be constructed in advance using finite element simulation, bench tests, and historical operating data. This database stores the correspondence between different multimodal eigenvectors and different typical vibration states of the turbine blades. These typical vibration states include, but are not limited to, first-order bending mode shapes, second-order bending mode shapes, torsional mode shapes, and composite mode shapes. By observing the vibration states, it is possible to intuitively determine whether the vibration stress of the turbine blades is abnormal. In other words, the vibration states provide a clear indication of whether an early warning is needed and how to intervene.

[0031] By matching the target multimodal feature vector calculated in real time with the feature vector combination in the mapping database, the target vibration state of the turbine blade can be determined. For example, when the crack propagation direction angle in the target feature vector is close to 90° and the event rate continues to increase (e.g., the increase exceeds 50%) within three consecutive sampling time windows, the blade is determined to be in a state of strong bending vibration. When the propagation direction angle is close to 0° and the morphological complexity increases significantly (e.g., more than twice the average value under normal operating conditions), the blade is determined to be in a state of torsional vibration. When multiple features in the target multimodal feature vector highly match the feature combination of a typical vibration state in the database (matching degree exceeds 85%), the typical vibration state is directly output as the target vibration state.

[0032] This step achieves a qualitative judgment of the vibration state through correlation analysis between multimodal characteristics and vibration state.

[0033] Based on the above Figure 1 As can be seen from the implementation method, the turbine blade vibration state determination method provided in this application pre-fabricates initial cracks sensitive to abnormal vibration stress on the turbine blade, providing a monitoring object for capturing crack responses related to vibration state. By simultaneously acquiring acoustic emission signals and visual image sequences of the initial cracks, real-time data acquisition during operation is achieved, breaking through the time limitations of offline detection. On the other hand, the combination of the dynamic real-time nature of acoustic emission signals and the intuitiveness of visual images compensates for the shortcomings of long transmission paths and large attenuation of single vibration signals, ensuring the comprehensiveness and timeliness of data acquisition. By extracting dynamic active features and static morphological features and fusing them, the constructed target multimodal feature vector can not only accurately capture cracks. The study also explored key behavioral characteristics such as expansion activity and expansion direction, achieving a comprehensive characterization of dynamic behavior and static morphology. This overcomes the limitations of existing technologies that can only determine the presence or rough estimation of crack size. Ultimately, through a pre-established mapping relationship, the multimodal feature vectors are directly correlated with the blade vibration state, enabling precise determination of the target vibration state. This represents a shift from passively detecting cracks to actively predicting vibration risks, improving the timeliness and accuracy of early warnings. Consequently, it effectively avoids missing the optimal intervention opportunity due to delayed early warnings, significantly reducing safety risks such as interstage collisions and overall turbine damage caused by blade fractures. It also reduces the huge economic losses caused by unplanned shutdowns or improper maintenance, providing reliable support for the health status assessment and remaining life prediction of turbine blades.

[0034] Furthermore, the preferred embodiments of this application are based on the above... Figure 1 Based on this, a detailed explanation of the process for determining the vibration state of steam turbine blades is provided, and the specific steps are as follows: Figure 2 As shown, it includes: 201. Determine the stress concentration region of the turbine blade under the target vibration mode through finite element analysis.

[0035] Among them, the target vibration mode is used to characterize one or more specific resonance modes that lead to fatigue damage or fracture of turbine blades.

[0036] In this step, 3D scanning technology is used to obtain the actual 3D contour data of the turbine blades, or based on the blade design drawings (including dimensional tolerances, chamfers, tie rod holes, and other details), a 1:1 full-size geometric model is constructed using CAD software (such as SolidWorks or UG). This ensures that the model accurately reproduces the actual structure of the blade, especially retaining key structural details that are prone to stress concentration, such as the root transition fillets, tie rod holes, and tenon joint surfaces, to avoid stress analysis deviations caused by model simplification. Based on the actual material used in the blades (such as 1Cr13 martensitic stainless steel, GH4169 high-temperature alloy, etc.), the room temperature and high-temperature mechanical property parameters of the material are retrieved, including elastic modulus, Poisson's ratio, yield strength, tensile strength, fatigue limit, and coefficient of thermal expansion. The high-temperature parameters must match the in-cylinder operating temperature during normal turbine operation (typically 300-500℃) to ensure that the simulation environment is consistent with actual operating conditions. Based on the statistical data of turbine blade failures and dynamic theory, the target vibration mode is identified as a specific resonance mode that is prone to causing fatigue damage or fracture of the blade. Specifically, it includes the first-order bending mode, the second-order bending mode, the torsional mode, and the bending-torsional composite mode. These modes are easily excited by aerodynamic excitation during blade operation, leading to local stress superposition, which is the main cause of crack initiation and propagation.

[0037] Professional finite element analysis software such as ANSYS and ABAQUS were used to mesh the blade geometric model. For areas suspected of stress concentration, such as the blade root, the periphery of the tie rod holes, and the leading / trailing edge transition region, a structured mesh was used for refinement, with the mesh element size controlled between 0.1-0.5 mm to ensure the accuracy of stress calculations. Unstructured meshes were used in other areas, with element sizes not exceeding 2 mm, ensuring both computational efficiency and overall analysis accuracy. After mesh generation, a quality check was performed to ensure that the element distortion rate was ≤5%, avoiding any impact on the calculation results due to mesh quality issues. Boundary conditions: Simulating the actual installation state of the blade, fixed constraints were applied to the mating surfaces of the blade tenon and the wheel disk, restricting their translational and rotational degrees of freedom to ensure that the constraint method was consistent with the actual stress state. Load application: Centrifugal loads, aerodynamic loads, and temperature loads were applied, while also considering temperature loads (applying a temperature field based on the actual temperature distribution inside the cylinder) to comprehensively simulate the complex stress environment of the blade during operation. Based on this, modal analysis is used to solve for the natural frequency of the blade and the mode shape of the corresponding target vibration mode, confirm the excitation conditions of the target vibration mode, perform harmonic response analysis based on the modal superposition method, calculate the steady-state stress distribution of the blade under the target vibration mode, and obtain stress cloud diagrams and stress data files.

[0038] The maximum principal stress distribution data is extracted from the harmonic response analysis results. A stress concentration factor threshold is set (usually ≥1.5, stress concentration factor = local maximum principal stress / overall average stress). Regions with stress concentration factors exceeding the threshold are selected as target stress concentration areas. Typical stress concentration areas include: the rounded corner area at the blade root and tenon connection, the inner wall of the tie rod hole and the surrounding 5mm area, and the thickness abrupt change area at the leading / trailing edge of the blade. The accuracy of the finite element analysis results is verified by comparing with stress test data of similar blades (such as strain gauge tests and photoelastic test data), ensuring that the positioning deviation of the stress concentration area is ≤10%. If the deviation exceeds the allowable range, the mesh generation accuracy, load application method, or material parameters need to be adjusted, and the simulation analysis needs to be repeated until the results meet the accuracy requirements. Finally, the three-dimensional coordinate range, principal stress direction, and stress amplitude of the stress concentration area are output, providing a clear location and orientation basis for the subsequent pre-creation of initial cracks.

[0039] 202. In the stress concentration region, an initial crack with a specific orientation and initial size is prefabricated.

[0040] In this step, based on the principal stress direction of the stress concentration area output in step 201, the extension direction of the initial crack is determined to be consistent with the principal stress direction. This setting ensures that when the blade is subjected to abnormal vibration stress corresponding to the target vibration mode, the crack can extend in a controllable and monitorable manner along the principal stress direction. If there are multiple principal stress directions, a set (2-3) of initial cracks with different orientations are prefabricated, each corresponding to the principal stress direction of different target vibration modes, ensuring comprehensive coverage of dangerous vibration modes. Based on the fatigue strength calculation and finite element simulation verification of the blade, the initial crack is determined to be a micron-level safe crack, with the following specific size parameters: length: 50-200μm (adjusted according to the blade thickness, not exceeding 10% of the minimum blade thickness), ensuring that the stress intensity factor of the crack under normal operating conditions is lower than the fracture toughness threshold of the blade material, and that no unexpected extension occurs; width: ≤20μm, depth: ≤50μm (not exceeding 15% of the blade thickness), avoiding a significant decrease in the mechanical properties of the blade due to excessive crack depth, which would affect the safety of normal operation.

[0041] Initial cracks are prefabricated using laser etching or micro-electrical discharge machining (EDM). During processing, the positioning error of the laser etching or EDM equipment must not exceed ±5μm to ensure precise matching between the preset crack direction and the stress concentration direction, thus avoiding reduced sensitivity of the crack to the target vibration mode due to processing deviations. After processing, the initial crack morphology can be inspected using a high-resolution microscope to confirm that its size and orientation are consistent with the simulation design parameters before proceeding to subsequent steps.

[0042] 203. Acquire acoustic emission signals and visual image sequences of the initial crack during turbine operation.

[0043] This step combines the description of step 102 in the above method, and the same content will not be repeated here.

[0044] It should be noted that the specific execution steps for collecting the acoustic emission signal and visual image sequence of the initial crack during turbine operation are as follows: continuously collect acoustic emission signals; collect visual image sequences at a first preset time interval; when the dynamic activity characteristics of the acoustic emission signal exceed a preset activity threshold, collect visual image sequences at a second preset time interval.

[0045] The second preset time interval is shorter than the first preset time interval. After the turbine starts, the acoustic emission acquisition system starts synchronously and continuously and uninterruptedly acquires the original acoustic emission waveform data related to the initial crack according to preset parameters. During the acquisition process, the system stores the original data in binary format in real time (e.g., generating an independent data file every hour, and storing the unit operating parameters for the corresponding time period, such as speed, load, cylinder temperature, and steam pressure). Invalid interference signals are eliminated by thresholding (the threshold can be twice the peak value of the sensor's inherent noise), valid acoustic emission events are extracted, and dynamic activity characteristics such as event rate, signal energy, and average amplitude are calculated in real time to provide a basis for judgment when switching the visual acquisition mode.

[0046] During the stable operation phase of the steam turbine, the visual acquisition device is automatically triggered at a first preset time interval (30 minutes / time), acquiring 3-5 frames of initial crack area images each time. After acquisition, the best frame is selected as valid data through an image sharpness evaluation algorithm, and the remaining frames are automatically backed up. Each frame image is accompanied by a timestamp, unit operating parameters, acquisition mode identifier, and other related information to facilitate subsequent data traceability and matching. Based on this, the dynamic activity characteristics of the acoustic emission signal are compared with a preset activity threshold in real time. The preset activity threshold is used to characterize the critical point at which the crack changes from a quiescent state to an active state. Its function is to trigger enhanced monitoring. If any feature of event rate, signal energy, or average amplitude exceeds the corresponding threshold for three consecutive sampling windows, it is automatically determined as "enhanced crack activity," and a visual image sequence is immediately acquired at a second preset time interval (5 minutes / time). The acquisition process is the same as the conventional mode (3-5 frames each time, selecting the best frame) to ensure accurate capture of the dynamic changes in crack propagation. In addition, it can continuously monitor the dynamic activity characteristics of acoustic emission. When all characteristics fall below the preset activity threshold and the duration reaches 1 / 2 of the first preset time interval (e.g., 15 minutes), it will automatically resume collecting images according to the first preset time interval. However, if the characteristics continue to exceed the threshold for more than 2 hours, it will continue to collect images according to the second preset time interval and send a prompt message to the operation and maintenance platform, suggesting that it pay more attention.

[0047] 204. Extract a set of dynamic activity features corresponding to the acoustic emission signal, extract a set of static morphology features corresponding to the visual image sequence, and fuse the dynamic activity features and static morphology features to obtain the target multimodal feature vector.

[0048] This step combines the description of step 103 in the above method, and the same content will not be repeated here.

[0049] The specific steps for implementing dynamic activity features are as follows: analyze the acoustic emission signal, calculate and extract at least one feature from the event rate, signal energy and average amplitude as dynamic activity features.

[0050] Among them, the event rate characterizes the frequency of acoustic emission events per unit time, the signal energy characterizes the intensity of the energy released by the acoustic emission event, and the average amplitude characterizes the stress level that drives crack propagation.

[0051] For event rate R, signal energy E, and average amplitude A mean The calculation process is as follows: Event rate R: The number of acoustic emission events exceeding a threshold per unit time, reflecting the frequency of crack propagation. The specific formula is as follows:

[0052] in, The total number of acoustic emission events detected. This represents the sampling time window.

[0053] Signal energy E: The integral of the energy released by the acoustic emission event, reflecting the severity of crack propagation.

[0054]

[0055] in, The instantaneous amplitude of the acoustic emission signal. The duration of the event.

[0056] Average amplitude A mean The average amplitude of the acoustic emission signal is related to the stress level that drives crack propagation.

[0057]

[0058] in, The instantaneous sample value of the acoustic emission signal. This represents the number of sampling points.

[0059] For static morphological features, the specific steps are as follows: perform digital image processing on the visual image sequence, calculate and extract at least one feature from crack length, propagation direction angle and morphological complexity as static morphological features.

[0060] Among them, crack length is the macroscopic extension length of crack calculated by edge detection and pixel calibration, extension direction angle is the angle between the extension direction of the crack trunk and the preset reference axis of the blade, and morphological complexity is an indicator characterizing the degree of crack bifurcation or turning.

[0061] The calculation process for crack length L, propagation direction angle θ, and morphological complexity C is as follows: Crack length L: The macroscopic propagation length of the crack is calculated through edge detection and pixel calibration.

[0062]

[0063] in, These are the pixel coordinates of the crack edge.

[0064] Propagation direction angle θ: Calculates the angle between the crack main extension direction and a specific reference axis (such as the blade axis).

[0065]

[0066] in, , These are the coordinates of the two ends of the crack.

[0067] Morphological complexity C: An index characterizing complex features such as crack bifurcation and bends.

[0068]

[0069] in, The local curvature at the crack edge, The mean curvature.

[0070] After calculating the dynamic activity features and static morphology features mentioned above, the dynamic activity features and static morphology features are fused at the feature level to construct a unified target multimodal feature vector. The expression for this target multimodal feature vector is as follows:

[0071] This feature vector comprehensively characterizes the dynamic behavior and static morphology of a crack within a specific time period. Specifically, it includes [event rate, signal energy, average amplitude, crack length, propagation direction angle, and morphological complexity].

[0072] 205. Obtain multimodal feature vector samples of turbine blades under different typical vibration states.

[0073] In this step, a multimodal feature vector and vibration state mapping database can be constructed in advance using finite element simulation, bench tests, and historical operating data. Specifically, for finite element simulation, the mechanical response of the blade under different typical vibration states (such as first-order bending mode, second-order bending mode, torsional mode, and composite mode) can be simulated using finite element software, and the multimodal feature vector samples corresponding to each vibration state can be calculated. For bench test data acquisition, a turbine blade simulation test bench can be built, and different types and intensities of vibration loads can be artificially applied. Acoustic emission signals and visual images can be collected simultaneously, feature vector samples can be extracted, and associated with known vibration states. For historical data integration, historical fault data of similar turbine blades can be collected, and multimodal feature data corresponding to specific vibration states can be selected and added to the database. In addition, cluster analysis can be performed on all sample data to establish corresponding rules for different vibration states and combinations of eigenvector parameters. For example, the first-order bending mode corresponds to the characteristic combination of "expansion direction angle of about 45°, medium event rate, and medium signal energy", while the strong bending vibration state corresponds to the characteristic combination of "expansion direction angle close to 90°, continuous increase in event rate, and significant increase in signal energy".

[0074] For the three types of samples mentioned above, a unified preprocessing can be performed. First, the 3σ criterion can be used to remove abnormal samples whose feature parameters exceed the normal range. Then, all feature parameters can be mapped to the [0,1] interval. Next, a clear vibration state label (such as first-order bending mode, torsional mode, normal state, etc.) can be added to each group of samples to ensure that there is a clear correspondence between the sample and the vibration state.

[0075] 206. Construct a feature-vibration state mapping database based on multimodal feature vector samples.

[0076] The feature-vibration state mapping database is used to store the correspondence between different multimodal feature vectors and different vibration states of turbine blades.

[0077] In this step, all preprocessed samples are divided into a training set and a validation set in a 7:3 ratio. The training set is used to construct mapping rules, and the validation set is used to verify the accuracy of the rules. The proportion of samples for each typical vibration state in the training set and validation set is kept consistent to avoid data bias. The K-means clustering algorithm is used to cluster the multimodal feature vector samples in the training set. The number of clusters is set to the number of typical vibration states (e.g., 5 classes, including the normal state). Cluster centers are iteratively optimized to ensure that samples corresponding to the same vibration state are clustered together, and the cluster boundaries for samples of different vibration states are clear. Statistical analysis is performed on the feature parameters of each cluster (corresponding to one vibration state) to determine the value range of each feature (e.g., in the cluster of strong bending vibration modes, the expansion direction angle ∈ [85°, 95°], the event rate ∈ [4.5 times / minute, 7 times / minute]), mean, and standard deviation, forming the "feature fingerprint" of that vibration state. Based on clustering results and feature fingerprints, quantization mapping rules are established. For example, if a sample feature vector satisfies "expansion direction angle ∈ [85°, 95°] ∧ event rate ∈ [4.5, 7] ∧ signal energy ∈ [180, 250]", then it corresponds to a strong bending mode shape; if a sample feature vector satisfies "expansion direction angle ∈ [0°, 10°] ∧ topography complexity ∈ [1.8 × 10⁻⁶]", then it corresponds to a strong bending mode shape. -6 3×10 -6 If the value is ], then it corresponds to the torsional vibration mode; the mapping rule for the normal state is "expansion direction angle ∈ [0.5, 1.5] ∧ crack length expansion ≤ 5 μm / hour ∧ all features are within the normal cluster range".

[0078] The validation set samples are input into the association rules for matching tests. The recognition accuracy for each vibration state is calculated, which is the ratio of the number of correctly matched samples to the total number of validation set samples. If the recognition accuracy for a certain type of vibration state is lower than 85%, the feature value range for that class is adjusted or training set samples are supplemented. The clustering and rule building steps are repeated until the recognition accuracy for all vibration states is ≥90%.

[0079] A mapping database is constructed using a relational database (such as MySQL) or a time-series database (such as InfluxDB), specifically containing three tables: a vibration state information table, a feature mapping rule table, and a sample raw data table. The vibration state information table stores the vibration state number, name, description (e.g., "001-First-order bending mode-First-order bending resonance along the blade's length"), and hazard level. The feature mapping rule table stores the value ranges (minimum, maximum, mean, standard deviation) of each feature parameter corresponding to each vibration state, as well as the mapping rule expression. The sample raw data table stores all preprocessed sample data (feature vectors, vibration state labels, data sources) for subsequent database updates and rule optimization. Indexes are created for the vibration state number and feature parameter value ranges to quickly match the corresponding vibration state.

[0080] 207. Match the target vibration state corresponding to the target multimodal feature vector in the feature-vibration state mapping database based on the mapping relationship.

[0081] In this step, the target multimodal feature vector is compared with all vibration state mapping rules in the "Feature Mapping Rule Table" of the mapping database to filter out candidate vibration states (there may be one or more) that satisfy the condition that "all features are within the range of the feature values ​​of the vibration state". If there is no candidate state with completely matching features, the process proceeds to the fuzzy matching stage. If there is one candidate state, it is initially determined to be the correct state.

[0082] For cases with no candidate states or multiple candidate states, the cosine similarity algorithm can be used to calculate the similarity between the target feature vector and the "mean vector of feature fingerprints" of each vibration state. The specific formula is as follows:

[0083] in, For the target multimodal feature vector, Let be the mean vector of the characteristic fingerprint of a certain vibration state.

[0084] A similarity threshold of 85% is preset. If the highest similarity value is ≥85%, the target vibration state is determined to be the vibration state corresponding to the highest similarity. If the highest similarity value is <85%, it is determined to be an "unknown vibration state". The similarity data between the target feature vector and all vibration states is recorded for subsequent database optimization.

[0085] Furthermore, the initially determined target vibration state can be cross-validated by combining it with the current operating parameters of the turbine (speed, load, cylinder temperature). For example, if it is determined to be a "second-order bending mode", it is necessary to confirm whether the current speed is close to the second-order bending natural frequency of the blade. If the operating parameters contradict the excitation conditions of the vibration state, the matching process needs to be re-executed (to check for errors in the preprocessing of the target feature vector). After successful verification, the complete information of the target vibration state is output, which includes: vibration state name, description, hazard level, matching basis (the degree of fit between key features and mapping rules, such as "the extension direction angle = 88° conforms to the range of extension direction angle values ​​[85°, 95°] for a strong bending mode"), and similarity value. At the same time, the target feature vector and matching results are stored in the "sample original data table" of the mapping database to provide data support for database updates. If three consecutive matches result in "unknown vibration state", or if the matching results are inconsistent with the operating parameters, an "abnormal data alarm" will be automatically triggered, prompting maintenance personnel to check the multimodal acquisition system and feature extraction process. At the same time, the original data and matching logs during the abnormal period will be retained to facilitate troubleshooting.

[0086] Furthermore, in order to assess the dynamic load level borne by the turbine blades, a quantitative assessment of the turbine blades can also be performed. The specific steps are as follows: based on the signal energy and event rate, the equivalent vibration stress amplitude corresponding to the current crack leading edge of the initial crack is quantitatively calculated to assess the dynamic load level borne by the turbine blades.

[0087] Regarding the equivalent vibration stress amplitude σ eq The quantitative calculation is performed using the following formula:

[0088] in, , These are empirical coefficients determined based on calibration experiments.

[0089] Furthermore, in order to determine the health status of the blades, a multi-level early warning mechanism can be established. The specific steps are as follows: based on the target vibration state, determine the preset safety threshold of each feature in the target multimodal feature vector; compare the current value of each feature in the target multimodal feature vector with the preset safety threshold of each feature to determine one or more over-limit features; and make multi-level early warning decisions on the health status of the turbine blades according to the type and degree of over-limit features.

[0090] It should be noted that, based on the type and degree of exceeding the limit characteristics, the specific execution steps for multi-level early warning decision-making on the health status of turbine blades are as follows: when only one or more dynamic active characteristics exceed the first-level safety threshold in the corresponding safety threshold, a primary early warning is triggered; when the current value of a dynamic active characteristic continuously exceeds the first-level safety threshold, or the current value of any static morphology characteristic exceeds the second-level safety threshold in the corresponding safety threshold, an intermediate early warning is triggered; when the current value of a dynamic active characteristic exceeds the critical-level safety threshold in the corresponding safety threshold, and at the same time the current value of a static morphology characteristic shows an accelerating expansion trend, a high-level early warning is triggered.

[0091] In this step, a preset safety threshold is used to characterize the critical line at which the blade's health deteriorates from an acceptable range to a risk level requiring attention / intervention, triggering an early warning decision. For the three dynamic active features and three static morphological features in the target multimodal feature vector, a three-level classification can be established: "First-level safety threshold (early warning trigger line), Second-level safety threshold (risk escalation line), and Critical-level safety threshold (emergency response line)." All thresholds can be determined based on a feature-vibration state mapping database, bench test calibration, and blade material fatigue characteristics, thus ensuring the scientific validity and relevance of the thresholds.

[0092] Regarding dynamic activity characteristics, the first-level safety threshold T1 characterizes the upper limit of characteristic values ​​under normal operating conditions. Exceeding this threshold indicates that the crack begins to respond to vibration stress. Specifically, it can be determined by the characteristic mean + 2 standard deviation of normal vibration state samples in a statistical mapping database or the characteristic peak value of normal vibration conditions in bench tests. The second-level safety threshold T2 characterizes a significant increase in crack propagation activity, with vibration stress approaching 70% of the material's fatigue limit. Specifically, it can be determined by the characteristic mean of slightly abnormal vibration states in bench tests or the characteristic value corresponding to a crack propagation rate ≥ 0.5 μm / h in finite element simulations. The critical-level safety threshold T3 characterizes the rapid crack propagation stage, with vibration stress reaching 80% of the material's fatigue limit. Specifically, it can be determined by the characteristic mean of severely abnormal vibration states in bench tests or the critical characteristic value for crack instability propagation determined by blade material fatigue tests.

[0093] Regarding static morphological characteristics, the first-level safety threshold S1 characterizes the stable range of characteristics under normal operating conditions. Exceeding this range indicates that morphological changes have begun to accumulate. Specifically, it can be determined by the upper limit of the characteristic value range of normal vibration state samples in the mapping database or the characteristic stability value of the blade after pre-existing cracks in bench tests. The second-level safety threshold S2 characterizes significant morphological changes and a slight impact of crack propagation on the blade's mechanical properties. Specifically, it can be determined by the characteristic mean of moderate abnormal vibration states in bench tests or the characteristic value corresponding to a 10% decrease in blade load-bearing capacity in finite element simulations. The critical-level safety threshold S3 characterizes morphological changes approaching the dangerous limit, indicating a risk of blade fracture. Specifically, it can be determined by the crack size corresponding to the fracture toughness of the blade material, the characteristic threshold requiring shutdown for maintenance in the maintenance standards, or the upper limit of the characteristics of dangerous vibration states in the mapping database.

[0094] The system extracts the characteristic safety thresholds corresponding to the target vibration state from the mapping database. For example, if the target vibration state is a strong bending mode (high risk level), the specific safety threshold corresponding to that state is matched; if it is no abnormal vibration (low risk level), the normal working condition threshold is matched to ensure that the thresholds are accurately matched with the vibration risk level. At the same time, the current values ​​of each feature are extracted from the target multimodal feature vector and preprocessed. For dynamic active features, the average value of three consecutive sampling windows is taken as the current value to avoid misjudgment caused by accidental fluctuations in a single window. For static morphology features, the relative change is calculated by comparing with the baseline feature value of the initial crack, and the change is taken as the current value. The current value of each feature is compared with the corresponding level of safety threshold one by one to determine whether it exceeds the limit and the degree of exceeding the limit. If the current value of a feature is greater than the corresponding safety threshold, it is determined to be an over-limit feature. The same feature may exceed multiple safety thresholds at the same time. It can be recorded according to the highest level of exceeding the limit. If it only exceeds the first level of safety threshold and does not reach the second level of safety threshold, the degree of exceeding the limit is slightly exceeding the limit. If it exceeds the second level of safety threshold and does not reach the critical level of safety threshold, the degree of exceeding the limit is moderate exceeding the limit. If it exceeds the critical level of safety threshold, the degree of exceeding the limit is severe exceeding the limit.

[0095] If only the dynamic activity feature slightly exceeds the limit (only exceeds T1, but does not reach T2), and the static morphology features do not exceed S1, and the exceeding features do not show a continuous growth trend (e.g., the feature value increase is ≤10% for two consecutive sampling windows), then it is a primary warning. At this time, a minor risk warning can be issued, i.e., an automatic pop-up window will be displayed to remind the maintenance personnel. The warning information includes: primary warning - crack vibration response start, name and current value of the exceeding feature, and target vibration status.

[0096] If one or more of the dynamic activity characteristics slightly exceed the limit (exceeding T1) and the duration is ≥30 minutes, or if a moderate exceedance occurs (exceeding T2 but not reaching T3), and one or more of the static morphology characteristics exceed S2 but not reaching S3, then a medium-level warning is issued. At this time, a risk escalation prompt can be issued, i.e., triggering an audible and visual alarm (local alarm in the data center + push notification from the remote operation and maintenance platform). The warning information is marked as medium-level warning - crack propagation risk escalation, with an attached trend chart of the over-limit characteristics, and automatically generates operating condition adjustment suggestions (such as suggesting that the load be stabilized at 80%-90% of the rated load to avoid load fluctuations exceeding ±10%), which are then pushed to the turbine control system.

[0097] If one or more of the dynamic activity characteristics exceed the critical safety threshold T3, and the static morphology characteristics show an accelerated propagation trend (e.g., crack length propagation ≥15μm within 1 hour, or morphological complexity increase ≥30% within 1 hour), then a high-level warning is issued. At this time, emergency risk management can be carried out, i.e., the highest level alarm is triggered (continuous audible and visual alarm + SMS / telephone notification to maintenance personnel), the warning information is marked "High-level warning - extremely high risk of blade fracture", the three most recent visual images (clearly showing the crack propagation situation) are uploaded simultaneously, and an emergency shutdown request is automatically sent to the turbine control system. If no manual confirmation is received within 10 minutes, the emergency shutdown process is initiated according to the preset procedure (gradually reducing load and speed to avoid secondary damage caused by sudden shutdown).

[0098] It should be noted that the determination of the warning level can also incorporate the equivalent vibration stress amplitude, that is, by adding a safety threshold corresponding to the equivalent vibration stress amplitude. The comparison between the equivalent vibration stress amplitude and the safety threshold can be used to determine whether the warning level should be upgraded, or the comparison can be used as one of the factors in determining the warning level. This embodiment does not limit this aspect.

[0099] Furthermore, as a response to the above Figure 1-2 The implementation of the method embodiment shown in this application provides a turbine blade vibration state determination device. This device is used to provide timely and accurate early warning of turbine blade vibration state, reducing economic losses and safety risks. The embodiment of this device corresponds to the aforementioned method embodiment. For ease of reading, this embodiment will not repeat the details of the aforementioned method embodiment, but it should be clear that the device in this embodiment can correspondingly implement all the contents of the aforementioned method embodiment. Specifically, as shown... Figure 3 As shown, the device includes: Prefabrication unit 301 is used to prefabricate initial cracks on turbine blades, the initial cracks being designed to allow for monitorable propagation under abnormal vibration stress; Acquisition unit 302 is used to acquire the acoustic emission signal and visual image sequence of the initial crack obtained by the prefabrication unit 301 during turbine operation; Processing unit 3003 is used to extract a set of dynamic active features corresponding to the acoustic emission signal obtained by acquisition unit 302, extract a set of static morphological features corresponding to the visual image sequence, and fuse the dynamic active features and the static morphological features to obtain a target multimodal feature vector. The first determining unit 304 is used to determine the target vibration state of the turbine blade based on the pre-established mapping relationship and the target multimodal feature vector obtained by the processing unit 303. The mapping relationship is used to characterize the correspondence between different multimodal feature vectors and different vibration states of the turbine blade.

[0100] Furthermore, such as Figure 4 As shown, the prefabrication unit 301 includes: The determination module 3011 is used to determine the stress concentration area of ​​the turbine blade under the target vibration mode through finite element analysis. The target vibration mode is used to characterize one or more specific resonance modes that lead to fatigue damage or fracture of the turbine blade. The prefabrication module 3012 is used to prefabricate the initial crack with a specific orientation and initial size in the stress concentration region obtained by the determining module 3011.

[0101] Furthermore, such as Figure 4 As shown, the acquisition unit 302 includes: The first acquisition module 3021 is used to continuously acquire the acoustic emission signal; The second acquisition module 3022 is used to acquire the visual image sequence at a first preset time interval; The second acquisition module 3022 is further configured to acquire the visual image sequence at a second preset time interval when the dynamic activity characteristics of the acoustic emission signal obtained by the first acquisition module 3021 exceed a preset activity threshold, wherein the second preset time interval is less than the first preset time interval.

[0102] Furthermore, such as Figure 4 As shown, the processing unit 303 includes: The first processing module 3031 is used to analyze the acoustic emission signal, calculate and extract at least one of the event rate, signal energy and average amplitude as the dynamic activity feature, wherein the event rate represents the frequency of acoustic emission events per unit time, the signal energy represents the intensity of energy released by the acoustic emission events, and the average amplitude represents the stress level that drives crack propagation.

[0103] Furthermore, such as Figure 4 As shown, the processing unit 303 includes: The second processing module 3032 is used to perform digital image processing on the visual image sequence, calculate and extract at least one feature among crack length, propagation direction angle and morphological complexity as the static morphological feature. The crack length is the macroscopic propagation length of the crack calculated by edge detection and pixel calibration. The propagation direction angle is the angle between the main extension direction of the crack and the preset reference axis of the blade. The morphological complexity is an index characterizing the degree of crack bifurcation or bend.

[0104] Furthermore, such as Figure 4 As shown, the device further includes: The acquisition unit 305 is used to acquire multimodal feature vector samples of turbine blades under different typical vibration states before the first determination unit 304. The construction unit 306 is used to construct a feature-vibration state mapping database based on the multimodal feature vector samples obtained by the acquisition unit 305. The feature-vibration state mapping database is used to store the correspondence between different multimodal feature vectors and different vibration states of turbine blades. The first determining unit 304 is specifically used for, Based on the mapping relationship, the target vibration state corresponding to the target multimodal feature vector is matched in the feature-vibration state mapping database.

[0105] Furthermore, such as Figure 4 As shown, the device further includes: The second determining unit 307 is used to determine a preset safety threshold for each feature in the target multimodal feature vector based on the target vibration state; The comparison unit 308 is used to compare the current value of each feature in the target multimodal feature vector with the preset safety threshold of each feature obtained by the second determining unit 307, and determine one or more out-of-limit features; The early warning unit 309 is used to make multi-level early warning decisions on the health status of the turbine blades based on the type and degree of the over-limit features obtained by the comparison unit 308.

[0106] Furthermore, such as Figure 4 As shown, the early warning unit 309 includes: The first early warning module 3091 is used to trigger a primary early warning when only one or more of the current values ​​of the dynamic activity features exceed the first level security threshold in the corresponding security thresholds; The second early warning module 3092 is used to trigger a medium-level early warning when the current value of the dynamic activity feature continuously exceeds the first level safety threshold, or when the current value of any static morphology feature exceeds the second level safety threshold in the corresponding safety threshold. The second early warning module 3093 is used to trigger an advanced early warning when the current value of the dynamic activity feature exceeds the critical level safety threshold in the corresponding safety threshold, and at the same time the current value of the static morphology feature shows an accelerated expansion trend.

[0107] Furthermore, such as Figure 4 As shown, the device further includes: The calculation unit 310 is used to quantitatively calculate the equivalent vibration stress amplitude corresponding to the current crack leading edge of the initial crack based on the signal energy and the event rate, so as to evaluate the dynamic load level borne by the turbine blade.

[0108] Furthermore, embodiments of this application also provide a storage medium for storing a computer program, wherein the computer program, when running, controls the device where the storage medium is located to execute the above-described... Figure 1-2 The method for determining the vibration state of turbine blades described herein.

[0109] Furthermore, embodiments of this application also provide a processor for running a program, wherein the program executes the above-described... Figure 1-2 The method for determining the vibration state of turbine blades described herein.

[0110] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0111] It is understood that the relevant features in the above methods and apparatus can be referenced interchangeably. Furthermore, the terms "first," "second," etc., in the above embodiments are used to distinguish between embodiments and do not represent the superiority or inferiority of any particular embodiment.

[0112] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0113] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, this application is not directed to any particular programming language. It should be understood that the content of this application described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of this application.

[0114] In addition, the memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0115] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0116] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0117] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0118] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0119] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0120] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0121] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0122] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0123] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0124] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for determining the vibration state of a steam turbine blade, characterized in that, The method includes: Initial cracks are pre-formed on the turbine blades, and these initial cracks are designed to allow for monitorable propagation under abnormal vibration stress. Acoustic emission signals and visual image sequences of the initial crack were collected during turbine operation; Extract a set of dynamic activity features corresponding to the acoustic emission signal, extract a set of static morphology features corresponding to the visual image sequence, and fuse the dynamic activity features and the static morphology features to obtain a target multimodal feature vector; Based on the pre-established mapping relationship and the target multimodal feature vector, the target vibration state of the turbine blade is determined. The mapping relationship is used to characterize the correspondence between different multimodal feature vectors and different vibration states of the turbine blade.

2. The method according to claim 1, characterized in that, Pre-introducing initial cracks on the turbine blades includes: The stress concentration region of the turbine blade under the target vibration mode is determined by finite element analysis. The target vibration mode is used to characterize one or more specific resonance modes that lead to fatigue damage or fracture of the turbine blade. In the stress concentration region, an initial crack with a specific orientation and initial size is pre-formed.

3. The method according to claim 1, characterized in that, Acquiring acoustic emission signals and visual image sequences of the initial crack during turbine operation, including: The acoustic emission signals are continuously acquired; The visual image sequence is acquired at a first preset time interval; When the dynamic activity characteristics of the acoustic emission signal exceed a preset activity threshold, the visual image sequence is acquired at a second preset time interval, where the second preset time interval is less than the first preset time interval.

4. The method according to claim 1, characterized in that, Extract a set of dynamic activity features corresponding to the acoustic emission signal, including: The acoustic emission signal is analyzed, and at least one of the following features—event rate, signal energy, and average amplitude—is calculated and extracted as the dynamic activity feature: the event rate represents the frequency of acoustic emission events per unit time, the signal energy represents the intensity of energy released by the acoustic emission events, and the average amplitude represents the stress level driving crack propagation.

5. The method according to claim 1, characterized in that, Extract a set of static morphological features corresponding to the visual image sequence, including: The visual image sequence is subjected to digital image processing to calculate and extract at least one feature among crack length, propagation direction angle and morphological complexity as the static morphological feature. The crack length is the macroscopic propagation length of the crack calculated by edge detection and pixel calibration. The propagation direction angle is the angle between the main extension direction of the crack and the preset reference axis of the blade. The morphological complexity is an index characterizing the degree of crack bifurcation or bend.

6. The method according to claim 1, characterized in that, Before determining the target vibration state of the turbine blade based on the pre-established mapping relationship and the multimodal feature vector, the method further includes: Obtain multimodal feature vector samples of turbine blades under different typical vibration states; Based on the multimodal feature vector samples, a feature-vibration state mapping database is constructed. The feature-vibration state mapping database is used to store the correspondence between different multimodal feature vectors and different vibration states of turbine blades. The determination of the typical vibration state of the turbine blade based on the pre-established mapping relationship and the target multimodal feature vector includes: Based on the mapping relationship, the target vibration state corresponding to the target multimodal feature vector is matched in the feature-vibration state mapping database.

7. The method according to claim 1 or 6, characterized in that, The method further includes: Based on the target vibration state, a preset safety threshold for each feature in the target multimodal feature vector is determined; The current value of each feature in the target multimodal feature vector is compared with the preset safety threshold of each feature to determine one or more out-of-limit features; Based on the type and degree of the exceeded characteristics, a multi-level early warning decision is made on the health status of the turbine blades.

8. The method according to claim 7, characterized in that, Based on the type and degree of the exceeded characteristics, a multi-level early warning decision is made regarding the health status of the turbine blades, including: A primary warning is triggered when the current value of only one or more of the dynamic activity features exceeds the first-level security threshold in the corresponding security threshold. When the current value of the dynamic activity feature continuously exceeds the first level security threshold, or the current value of any static morphology feature exceeds the second level security threshold in the corresponding security threshold, a medium-level warning is triggered. When the current value of the dynamic activity feature exceeds the critical level security threshold in the corresponding security threshold, and at the same time the current value of the static morphology feature shows an accelerating expansion trend, an advanced warning is triggered.

9. The method according to claim 3, characterized in that, The method further includes: Based on the signal energy and the event rate, the equivalent vibration stress amplitude corresponding to the current crack leading edge of the initial crack is quantitatively calculated to assess the dynamic load level borne by the turbine blade.

10. A device for determining the vibration state of a steam turbine blade, characterized in that, The device includes: Prefabrication unit for prefabricating initial cracks on turbine blades, the initial cracks being designed to allow for monitorable propagation under abnormal vibration stress; The acquisition unit is used to acquire the acoustic emission signals and visual image sequences of the initial crack obtained by the prefabrication unit during turbine operation; The processing unit is used to extract a set of dynamic active features corresponding to the acoustic emission signal obtained by the acquisition unit, extract a set of static morphological features corresponding to the visual image sequence, and fuse the dynamic active features and the static morphological features to obtain a target multimodal feature vector. The first determining unit is used to determine the target vibration state of the turbine blade based on the pre-established mapping relationship and the target multimodal feature vector obtained by the processing unit. The mapping relationship is used to characterize the correspondence between different multimodal feature vectors and different vibration states of the turbine blade.

11. A storage medium, characterized in that, The storage medium includes a stored program, wherein, when the program is executed, it controls the device containing the storage medium to perform the turbine blade vibration state determination method as described in any one of claims 1 to 9.

12. A processor, characterized in that, The processor is used to run a program, wherein the program executes the turbine blade vibration state determination method as described in any one of claims 1 to 9.

Images