Method and system for damage monitoring of an aeroengine turbine blade

Abstract

The application relates to the technical field of aero-engines, and provides a damage monitoring method and system for a turbine blade of an aero-engine. The method comprises the following steps: embedding a sensor array on the surface of the turbine blade; activating a self-calibration unit in a low-load stage of the engine, generating a health initial reference through benchmark correction; arranging a non-contact observation group at a tail nozzle, reading an observation data set; inputting the reference and the data set into a twin model, performing fusion and correction, and generating a damage distribution and grade; activating a group dynamics consistency model to establish a deviation index; compensating the damage distribution and grade according to the deviation index, and outputting a monitoring result. The application solves the technical problem that the turbine blade is difficult to be detected and evaluated in time and accurately when cracks, fatigue, ablation and other damages occur in the blade under the action of a complex working environment, and the technical effect that the early damage early warning capability and the fault diagnosis reliability are improved through a multi-modal data joint correction and group dynamics compensation mechanism is achieved.

Description

Technical Field

[0001] This application relates to the field of aero-engine technology, specifically to a method and system for damage monitoring of aero-engine turbine blades. Background Technology

[0002] As the core power unit of aircraft, aero-engines subject their turbine blades to prolonged exposure to high temperatures, high pressures, high-speed rotation, and the coupled effects of complex aerodynamic and thermal loads, making them highly susceptible to damage such as cracks, fatigue, ablation, and material degradation. With the continuous improvement of engine thrust and efficiency, the extreme operating conditions endured by the blades have become increasingly stringent. Traditional monitoring methods relying on manual inspections or single sensors not only struggle to detect early, minute damage in a timely manner but also suffer from unstable detection results, high false alarm rates, and delayed maintenance cycles. Furthermore, complex flow fields and intense background noise can easily mask blade damage characteristics, leading to decreased monitoring accuracy. Summary of the Invention

[0003] This application provides a damage monitoring method and system for aero-engine turbine blades, aiming to solve the technical problem of timely and accurate detection and assessment of damage such as cracks, fatigue, and ablation caused by turbine blades under complex working environments. It achieves high-precision real-time monitoring of the spatial distribution and damage level of blade damage through multimodal data joint correction and group dynamics compensation mechanism, thereby improving the early damage warning capability and the reliability of fault diagnosis.

[0004] The first aspect disclosed in this application provides a method for damage monitoring of aero-engine turbine blades. The method includes: embedding a distributed sensor array on the surface of the turbine blade, the sensor array including fiber optic grating sensors and thin-film piezoelectric sensors; activating a self-calibration excitation unit when the engine is under low load, using the self-calibration excitation unit to perform reference correction on the output response signal of the sensor array, generating an initial reference for blade health; deploying a non-contact observation group at the exhaust nozzle, reading the observation dataset of the non-contact observation group, the observation dataset including acoustic emission signals, vibration mode signals, and vortex evolution signals; sending the initial reference for blade health and the observation dataset to a digital twin model, performing multi-modal data fusion processing, and using uncertainty propagation and deviation correction operators to jointly correct the data, generating a spatial distribution and damage level of the blade damage; activating a swarm dynamics consistency model, using the swarm dynamics consistency model to establish a swarm deviation index for the turbine blade; and generating damage monitoring results after compensating for the spatial distribution and damage level of the blade damage according to the swarm deviation index.

[0005] Another aspect of this application discloses a damage monitoring system for aero-engine turbine blades. The system includes: a sensor deployment module: embedding a sensor array on the turbine blade surface, the sensor array including fiber optic grating sensors and thin-film piezoelectric sensors; a reference calibration module: activating a self-calibration excitation unit when the engine is under low load, using the self-calibration excitation unit to perform reference calibration on the output response signal of the sensor array, generating an initial reference for blade health; a data reading module: deploying a non-contact observation group at the exhaust nozzle, reading the observation dataset of the non-contact observation group, the observation dataset including acoustic emission signals, vibration mode signals, and vortex evolution signals; a fusion correction module: sending the initial reference for blade health and the observation dataset to a digital twin model, performing multi-modal data fusion processing, and using uncertainty propagation and deviation correction operators to jointly correct the data, generating the spatial distribution and damage level of the blade damage; an index establishment module: activating a swarm dynamics consistency model, using the swarm dynamics consistency model to establish a swarm deviation index for the turbine blade; and a compensation module: compensating for the spatial distribution and damage level of the blade damage according to the swarm deviation index, generating damage monitoring results.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: The aforementioned damage monitoring method for aero-engine turbine blades first embeds distributed sensors, such as fiber optic gratings and thin-film piezoelectric sensors, on the blade surface to acquire real-time strain and stress characteristics. Under low engine load, a self-calibrating excitation unit performs signal reference correction to establish the blade's initial healthy state. Subsequently, a non-contact observation device is deployed at the exhaust nozzle to collect various data, including acoustic emission, vibration modes, and vortex evolution. Then, the healthy reference and observation data are input into a digital twin model for multimodal fusion and uncertainty correction to obtain the spatial distribution and severity of blade damage. Finally, a population dynamics consistency model is introduced to establish deviation indicators for the blade population and compensate for the damage assessment results, thereby generating more accurate and reliable damage monitoring conclusions.

[0007] The above description is merely 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

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a flowchart illustrating a method for monitoring damage to turbine blades in an aero-engine, as shown in one embodiment.

[0010] Figure 2 This is a diagram of the architecture of a damage monitoring system for aero-engine turbine blades in one embodiment.

[0011] Explanation of reference numerals in the attached diagram: Sensor deployment module 11, reference calibration module 12, data reading module 13, fusion calibration module 14, index establishment module 15, compensation module 16. Detailed Implementation

[0012] This application provides a method and system for monitoring the damage of aero-engine turbine blades, which solves the technical problem of timely and accurate detection and assessment of damage such as cracks, fatigue, and ablation caused by turbine blades under complex working environments. It achieves the technical effect of high-precision real-time monitoring of the spatial distribution and damage level of blade damage through multimodal data joint correction and group dynamics compensation mechanism, thereby improving the early damage warning capability and the reliability of fault diagnosis.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0014] It should be noted that the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such process, method, product, or device.

[0015] Example 1, as Figure 1 As shown, this application provides a method for damage monitoring of aero-engine turbine blades, the method comprising: A sensor array is embedded on the surface of the turbine blade, the sensor array including fiber optic grating sensors and thin-film piezoelectric sensors.

[0016] In this embodiment, a distributed sensor array is pre-designed and embedded on the surface of the aero-engine turbine blade. This sensor array consists of fiber optic grating sensors and thin-film piezoelectric sensors. The fiber optic grating sensors are distributed along the blade surface and can perform high-sensitivity real-time monitoring of the blade's stress state and temperature changes by detecting changes in the wavelength reflected by the grating. This is suitable for capturing subtle strains generated by the blade under high-temperature, high-speed airflow. The thin-film piezoelectric sensors are attached to or embedded in the blade surface in the form of a thin film. They can output charge signals when the blade is subjected to periodic aerodynamic loads or impact loads, thereby reflecting the dynamic response characteristics of the blade. These two types of sensors complement each other through a reasonable spatial arrangement, enabling the system to simultaneously acquire the blade's health parameters in different physical dimensions. This ensures the engine's stability and long-term reliability under extreme operating conditions and provides sufficient data support for subsequent benchmark calibration, digital twin modeling, and damage identification.

[0017] When the engine is under low load, the self-calibration excitation unit is activated, and the output response signal of the sensor array is used to perform reference correction to generate an initial reference for blade health.

[0018] In one embodiment, when the aero-engine operates at low load, a self-calibrating excitation unit is actively triggered. This unit can apply specific excitation signals, such as small-amplitude mechanical pulses or controllable frequency vibration waves, to the turbine blades without affecting normal engine operation, thereby stimulating the response of the embedded sensor array. Subsequently, the raw response signals output by the fiber optic grating sensor and the thin-film piezoelectric sensor under excitation are acquired, and the optical and charge signals are synchronously sampled and time-aligned respectively. These signals acquired in this way are highly correlated with the inherent characteristics of the blades in a non-destructive state, effectively eliminating uncertainties caused by environmental temperature fluctuations, material stress relaxation, or sensor drift. Then, the acquired raw response signals are benchmarked using a built-in calibration algorithm, including noise filtering, baseline drift correction, and normalization, to remove environmental interference and system bias. Noise filtering can be achieved through wavelet thresholding or bandpass filtering; baseline drift correction can be achieved through polynomial fitting minus baseline or high-pass filtering; and normalization can be achieved through min-max normalization or Z-score normalization. Then, based on the preprocessed signal, key feature parameters are extracted according to the preset key field names, such as the reflection wavelength drift of the fiber optic grating sensor, the voltage amplitude and frequency response characteristics of the thin-film piezoelectric sensor, etc., and the initial reference data of the blade in a healthy state is established. This initial reference of blade health will serve as the standard basis for subsequent damage comparison analysis under operating conditions, thereby ensuring that monitoring accuracy and stability can still be maintained in complex operating environments.

[0019] A non-contact observation group is deployed at the tail nozzle, and the observation data set of the non-contact observation group is read. The observation data set includes acoustic emission signals, vibration mode signals, and vortex evolution signals.

[0020] In one embodiment, a non-contact observation group is pre-deployed in the exhaust nozzle area of ​​an aero-engine. This non-contact observation group preferably consists of various non-contact sensors, such as a laser Doppler vibrator, an array of high-sensitivity microphones, and a high-speed optical measurement device. The microphone array can collect acoustic emission signals generated by the turbine blades during engine operation, reflecting the transient acoustic characteristics released by the blades under stress or crack propagation. The laser Doppler vibrator can perform high-precision measurements of the overall and local vibration modes of the blades, extracting their frequency response and modal deformation information to form vibration modal signals, thereby identifying modal changes caused by damage. The high-speed optical measurement device can acquire vortex evolution signals in the exhaust nozzle airflow, revealing the impact of blade damage on the wake structure and energy distribution. By reading the collected acoustic emission signals, vibration modal signals, and vortex evolution signals, an observation dataset can be constructed. This dataset ensures the safety and reliability of the measurement process, providing rich and accurate external observation data for subsequent digital twin modeling and multimodal data fusion.

[0021] The initial reference for leaf health and the observation dataset are sent to the digital twin model to perform multimodal data fusion processing. Uncertainty propagation and bias correction operators are used to jointly correct the data, generating the spatial distribution and damage level of leaf damage.

[0022] In one embodiment, after obtaining the initial reference and observation datasets for blade health, these datasets are simultaneously input into a digital twin model. This digital twin model is a virtual blade entity constructed based on high-fidelity blade geometry and operating parameters. Within the digital twin model, the received data is fused into an uncertainty propagation chain through mutual information processing and joint fitting to accurately reflect unstable factors in the blade's state evolution. Based on this, a deviation correction operator is used for joint correction according to the uncertainty propagation results, reducing the accumulation of deviations caused by differences in sensor sensitivity or installation position offsets. After the above processing, the digital twin model outputs the results from the processing layer, forming the spatial distribution of blade damage and damage level assessment results. This achieves an accurate mapping of the actual blade health status in the virtual environment, providing a reliable basis for subsequent early warning and maintenance decisions.

[0023] Furthermore, this application provides a method for sending the initial reference and observation dataset of leaf health to a digital twin model, including: The strain detection signal from the initial reference of blade health is read, and the acoustic emission signal from the observation dataset is read; the strain detection signal and the acoustic emission signal are processed together to establish a first uncertainty component; a second uncertainty component is established by jointly fitting the spatiotemporal distribution stability of the vortex evolution signal and the mode frequency shift of the vibration mode signal; the uncertainty propagation link of the damage state is constructed using the first uncertainty component and the second uncertainty component; and uncertainty propagation is completed according to the uncertainty propagation link.

[0024] Preferably, strain detection signals are first read from an initial reference of blade health, and acoustic emission signals are simultaneously extracted from the observation dataset. These two types of signals represent the characteristics of the blade's structural and acoustic responses, respectively. Subsequently, the extracted strain detection and acoustic emission signals undergo mutual information processing. Specifically, the strain detection and acoustic emission signals are divided into equal-length time windows, and the probability distribution characteristics of the signals are statistically analyzed within each time window, such as the frequency distribution of amplitude occurrences. Based on this probability distribution, the joint probability distribution function of the strain detection and acoustic emission signals is calculated, thereby obtaining their joint entropy and individual marginal entropy. Then, the mutual information between the strain detection and acoustic emission signals is calculated using the formula: I(X;Y)=H(X)+H(Y)−H(X,Y), where H(X) and H(Y) are the entropies of the strain detection and acoustic emission signals, respectively, H(X,Y) is the joint entropy, and I(X;Y) is the mutual information. A higher mutual information value indicates a stronger correlation between the strain detection and acoustic emission signals in terms of damage characteristics. After obtaining the mutual information, it is used as an input parameter. Combined with the changing trend of mutual information under different time windows, a first uncertainty component is established to characterize the coupling uncertainty between blade strain response and acoustic emission characteristics. This first uncertainty component can dynamically reflect the influence of local crack initiation and propagation processes on the correlation between the two types of signals, providing basic data for the subsequent construction of uncertainty propagation links. Furthermore, spatiotemporal feature analysis is performed on the vortex evolution signal. The data on the instantaneous position of the vortex core, the velocity vector of the flow field, and the vortex intensity as a function of time and space are normalized. By weighting the standard deviation of each data point, the spatiotemporal distribution stability parameters of vortex evolution under different operating conditions are obtained. Then, the modal frequencies are extracted from the vibration modal signals, focusing on identifying the modal frequency shift phenomenon caused by blade damage or material degradation, and expressing it as an offset relative to the reference modal frequency. Then, the spatiotemporal distribution stability parameters of the vortex evolution signal and the frequency shift of the vibration mode are constructed as input variables in the same eigenvector space. A joint fitting method is used for coupled modeling. During fitting, least squares fitting, multiple regression analysis, or principal component analysis combined with nonlinear regression can be used to establish the correspondence between vortex flow field disturbance and structural modal changes. Then, by evaluating the residual distribution and fitting confidence interval of the joint fitting results, a dynamic uncertainty quantification index is obtained, which is defined as the second uncertainty component. This second uncertainty component can reflect the coupling uncertainty between vortex flow field instability and structural vibration frequency changes in the damage monitoring process and serves as an important input for subsequent uncertainty propagation links. After obtaining the first and second uncertainty components, these two components are combined to construct an uncertainty propagation link under blade damage conditions. This link connects multiple components in a time-step manner, thereby dynamically describing the evolution of the damage signal with time and operating conditions.Finally, based on this uncertainty propagation link, accumulation and propagation calculations are performed on the signal components at different time steps to complete the overall uncertainty propagation and provide a quantitative basis for subsequent joint data correction and damage distribution generation.

[0025] Furthermore, this application provides a method for completing uncertainty propagation based on the aforementioned uncertainty propagation link, including: The uncertainty propagation link is discretized into N time steps; the uncertainty components in the uncertainty propagation link are weighted and accumulated according to the time relationship of the N time steps to establish a weighted accumulation result; an attenuation coefficient is set based on the relaxation characteristics of the blade material and the stability of historical operating conditions, and the attenuation processing of the weighted accumulation result is performed using the attenuation coefficient to complete the uncertainty propagation.

[0026] Optionally, the constructed damage state uncertainty propagation link is first discretized in time, dividing the entire monitoring cycle into N equally spaced time steps. In each time step, the corresponding uncertainty component value is obtained to ensure dynamic reflection of uncertainty changes over time. Then, based on the temporal relationship between the N time steps, the uncertainty components obtained from each time step are weighted and accumulated. The weighting factor can be set according to signal strength, characteristic sensitivity, or historical monitoring importance, thus highlighting the uncertainty contribution at critical moments during the accumulation process, resulting in a weighted accumulation result reflecting the evolution of uncertainty throughout the entire cycle. Next, based on the relaxation characteristics of the blade material, the stress relaxation time constant is obtained through querying. Based on the historical operating condition stability, the standard deviation of parameter fluctuation amplitude is calculated, and the reciprocal of the sum of the standard deviation and 1 is taken to obtain the operating condition stability factor. The stress relaxation time constant and the operating condition stability factor are then weighted and superimposed to obtain an attenuation coefficient. This attenuation coefficient reflects the decreasing trend of uncertainty caused by the gradual stabilization of material properties or stress relaxation. Finally, this attenuation coefficient is used to exponentially attenuate the weighted accumulation result, thereby reducing the influence of long-term data on the current monitoring results. Through the above process, uncertainty can be effectively propagated and corrected in the time domain, so that the final uncertainty propagation result retains the overall influence of historical information and avoids excessive interference of old data information on the current damage judgment, thereby achieving reasonable convergence and reliable expression of the uncertainty of blade damage state in the time dimension.

[0027] Furthermore, this application provides a method for sending the initial reference and observation dataset of leaf health to the digital twin model, and also includes: Obtain the propagation results of uncertainty propagation, and configure uncertainty factors based on the propagation results; map the strain signal, thin film piezoelectric signal in the initial reference of blade health, and the acoustic emission signal, vibration mode signal, and vortex evolution signal in the observation dataset to the same reference space; based on the uncertainty factors, establish cross-channel deviation constraints using the minimum cross-entropy criterion, perform joint optimization of temporal differences and spatial distribution differences, and establish joint optimization results; complete the joint correction of the deviation correction operator based on the joint optimization results.

[0028] Preferably, after uncertainty propagation is complete, an uncertainty factor is configured based on the propagation result obtained from attenuation processing. Specifically, the propagation uncertainty at the current time step is divided by the maximum uncertainty to measure the reliability and weight allocation of various signals during multimodal fusion. Subsequently, the strain signal and thin-film piezoelectric signal from the initial blade health reference, as well as the acoustic emission signal, vibration mode signal, and vortex evolution signal from the observation dataset, are normalized, and the processing results are mapped to the same reference space. This reference space can be a unified time-series coordinate system to achieve alignment and comparability of cross-channel data. Then, the set uncertainty factor is introduced to assign weights to different signal channels, making the influence of high-reliability signals greater and the influence of low-reliability signals weaker during fusion. Using the same method described above, the joint probability distribution and marginal probability distribution of each channel are constructed, and the minimum cross-entropy criterion is introduced as the optimization objective function. By calculating the cross-entropy values ​​between channels, cross-channel deviation constraints are established to ensure the consistency of statistical correlation between signals from different channels. Under this cross-channel deviation constraint, optimizations are performed on both temporal and spatial distribution differences. In the temporal dimension, dynamic time warping, correlation peak alignment, or interpolation resampling corrects the time step offset of different channel signals. In the spatial dimension, blade geometric coordinate mapping and interpolation fitting methods are used to correct the distribution positions of different channel signals, reducing spatial deviations caused by different installation locations or measurement methods. By weighted fusion of the temporal and spatial optimization results, a globally consistent joint optimization result is obtained. This joint optimization result reflects the optimal matching state of each channel signal in both time and space dimensions. After obtaining the joint optimization result, it is used as the joint correction result of this deviation correction operator and transmitted from the digital twin model to the blade damage analysis model. This blade damage analysis model, built using a deep neural network, has been pre-trained through forward propagation, loss calculation, backpropagation, and parameter optimization. It can map the corresponding spatial distribution and damage level of blade damage based on the input feature data, ensuring the accuracy of damage identification and the reliability of the assessment results.

[0029] Furthermore, this application provides a method for establishing cross-channel deviation constraints using the minimum cross-entropy criterion, and performing joint optimization of temporal and spatial distribution differences, including: The time steps of the sensor array and non-contact observation group are mapped to the standardized position of the rotation period according to the blade rotation period; after normalizing the local stress amplitude and vibration mode amplitude, the angle is corrected by using the mapping of the standardized position to establish a dynamic weighting factor; and weighted compensation is performed for joint optimization based on the dynamic weighting factor.

[0030] Optionally, firstly, using the turbine blade rotation period as a reference, the time-series signals from the embedded sensor array and non-contact observation group are uniformly mapped to a standardized rotation period coordinate system. This maps each sampling time point to a standardized position of the blade rotation angle, such as a point within the 0–360° range, thus eliminating alignment errors caused by inconsistent sampling frequencies of different signal sources. Subsequently, the local stress amplitude and vibration mode amplitude in each channel signal are normalized to eliminate differences in different measurement dimensions and amplitude ranges. Then, based on the standardized position, the correspondence between the normalized amplitude and the blade rotation angle is nonlinearly fitted. The fitting process also uses methods such as least squares to establish a weighting factor that dynamically changes with the rotation period. This dynamic weighting factor can reflect the importance of the signal to the damage monitoring results at a certain moment or angle. Subsequently, during the cross-channel joint optimization process, this dynamic weighting factor is introduced to perform weighted compensation on different signals. That is, in the calculation of temporal optimization and spatial distribution optimization, high-weight signals are given greater influence, and low-weight signals are given less influence, thereby ensuring that the optimization results are closer to the actual physical state, so as to improve the fusion accuracy of cross-channel signals in the time and space dimensions and provide a reliable basis for the joint correction of the deviation correction operator.

[0031] Furthermore, this application provides the generation of spatial distribution and damage level of leaf damage, and also includes: Read the engine's control parameters during low-load phases, and establish parameter compensation factors based on these parameters; use these parameter compensation factors to generate and compensate for the spatial distribution and damage level of blade damage.

[0032] Preferably, when the engine is under low load, the engine's control parameters are read. These parameters include, but are not limited to, key indicators reflecting the engine's operating status, such as fuel flow rate, compressor outlet pressure, combustion chamber temperature, rotor speed, and nozzle opening. These parameters are then normalized and weighted according to the sensitivity of each parameter to blade operating status as determined by an empirical database or experimental calibration, resulting in a parameter compensation factor. This factor characterizes the coupled influence of different control parameters on blade damage monitoring results. For example, when the combustion chamber temperature is low under low load, there may be a systematic deviation between the blade surface stress signal and the actual damage; the compensation factor can correct this deviation. Subsequently, during the generation of the blade damage spatial distribution and damage level, this parameter compensation factor is introduced into the calculation process of the blade damage analysis model to correct the initial model output. For the blade damage spatial distribution, the sum of 1 and the parameter compensation factor is multiplied by the blade damage spatial distribution for compensation correction; for the damage level, the damage level is added to the parameter compensation factor for compensation correction. In summary, the above-mentioned compensation management can eliminate the impact of abnormal fluctuations in low-load control parameters on monitoring accuracy, making the final generated damage spatial distribution closer to the actual situation, and the damage level assessment results more reliable and stable.

[0033] Activate the population dynamics consistency model and use it to establish a population deviation index for turbine blades.

[0034] In one embodiment, to improve the reliability of individual turbine blade damage monitoring results, after completing multimodal fusion and preliminary damage assessment, a swarm dynamics consistency model is activated. This model constructs a swarm dynamics reference system based on the response characteristics of multiple turbine blades under the same operating conditions. During this process, the swarm dynamics consistency model performs principal component analysis or low-rank decomposition on the acquired blade swarm response matrix and the blade response results to extract the dominant modes of the blade swarm, establishing a unified swarm consistency reference. Then, the actual response of the monitored blade is compared with the swarm consistency reference, and its residual vector is calculated. The larger the residual, the more significant the deviation from the swarm response characteristics. Finally, a swarm deviation index is defined based on the residual vector to intuitively reflect the degree of consistency between the monitored blade and the swarm health status, thus serving as a basis for further correction of blade damage distribution and severity, ensuring more robust and accurate monitoring results.

[0035] Furthermore, this application provides an activated swarm dynamics consensus model, using which a swarm dynamics consensus model is used to establish a swarm deviation index for turbine blades, including: Obtain the blade group response matrix under the same operating conditions, and synchronize the blade group response matrix and the monitored turbine blade response results to the group dynamics consistency model; use the extraction layer in the group dynamics consistency model to extract the dominant modes based on principal component analysis or low-rank decomposition, and establish a group consistency reference; use the group consistency reference to calculate the residual vector of the monitored turbine blades, and generate a group deviation index.

[0036] Preferably, sensor response data from multiple turbine blades under the same operating conditions are first collected, including strain signals, vibration modal parameters, and acoustic emission characteristics. This data is then arranged according to blade number and sampling time to construct a blade group response matrix. Simultaneously, the response results of the target monitored blade are synchronized with this group matrix in the time dimension to ensure data comparability. Subsequently, the blade group response matrix is ​​input into the feature extraction layer of the group dynamics consistency model. This feature extraction layer inherits principal component analysis and low-rank decomposition. When the dimension of the blade group response matrix is ​​high and there is a strong linear correlation between various response features, principal component analysis is used. Principal component analysis calculates the covariance of the response matrix and solves for eigenvalues ​​and eigenvectors, projecting the original high-dimensional signal onto the principal component direction with the largest variance, thereby extracting a few principal components that can represent the overall dynamic characteristics of the group. If the population response matrix exhibits significant redundancy or noise interference, a low-rank decomposition method, such as singular value decomposition (SVD), is employed. SVD decomposes the response matrix into the product of singular vectors and singular values. Larger singular values ​​and their corresponding singular vectors represent the dominant structural modes, while smaller singular values ​​correspond to noise or minor perturbations. Low-rank approximation is achieved by retaining the first few singular value components, thereby extracting the dominant population mode while reducing noise. The extracted dominant mode is used to establish a population consistency reference, which reflects the typical response pattern of a healthy population of blades under this operating condition. Subsequently, the actual response results of the monitored blades are compared with the population consistency reference, and their difference vector is calculated and defined as a residual vector to characterize the degree of deviation between the monitored blade and the population's healthy response. Finally, the residual vector is normalized or a distance metric such as Euclidean distance is used to calculate its value, converting it into a quantitative index to generate a population deviation index. The larger the index, the more significant the difference between the monitored blade and the overall population response, and the higher the potential risk of damage.

[0037] After compensating for the spatial distribution and damage level of the blade damage based on the population deviation index, damage monitoring results are generated.

[0038] In one embodiment, after obtaining the group deviation index, the index is compared with the damage identification result of a single blade. When the index value is low, it indicates that the blade's response is highly consistent with the overall group response, and the original damage distribution and level assessment results are maintained. When the index value is high, it is determined that the blade has an abnormal deviation that does not conform to the group characteristics, which may easily lead to underestimation or overestimation of the damage results. In this case, the system adjusts the initially generated damage spatial distribution according to the magnitude of the group deviation index. For example, it increases the damage weight or corrects the distribution boundary in areas with significant residuals to more realistically reflect the local damage situation. At the same time, the damage level is corrected and compensated. That is, when the deviation index exceeds a preset threshold, the damage level is appropriately increased or adjusted to a higher risk range to avoid missing potential high-risk damage. After the above compensation processing, the final damage monitoring result is output. This result combines the advantages of single-blade monitoring and group consistency correction, and can more accurately and robustly reflect the true health status of the turbine blade, providing a reliable basis for subsequent early warning and maintenance.

[0039] Furthermore, this application provides methods for generating damage monitoring results, including: Based on the damage monitoring results, a multi-level early warning threshold range is mapped to configure a damage early warning signal; the damage early warning signal is used to configure an operation and maintenance strategy, and the turbine blade maintenance management is carried out according to the operation and maintenance strategy.

[0040] Preferably, the damage levels obtained from the damage monitoring results are first mapped to pre-defined multi-level early warning threshold ranges. These multi-level early warning threshold ranges can be classified according to the engine's design life, material tolerances, and operational safety standards. For example, they can be set as normal zone, concern zone, early warning zone, and danger zone, with different zones corresponding to different damage severity ranges. When the damage level falls into a certain threshold range, a corresponding damage early warning signal is triggered. For example, when the result is in the concern zone, a prompt signal is output; when it is in the early warning zone or danger zone, an alarm signal is output with a level identifier, so that operators can obtain risk information immediately. Subsequently, based on the damage early warning signal, a pre-defined operation and maintenance strategy is automatically matched. Under low-level early warnings, the strategy may only be to increase the monitoring frequency or shorten the inspection cycle; under high-level early warnings, the strategy includes limiting engine operating conditions, adjusting task allocation, and even arranging shutdown for inspection or blade replacement. Finally, the operation and maintenance department implements corresponding maintenance management measures for the turbine blades according to the system-configured strategy, realizing closed-loop control of real-time monitoring, risk assessment, and maintenance decision-making to ensure the safety and reliability of engine operation.

[0041] Furthermore, this application, after generating the damage monitoring results, also includes: After compressing the damage monitoring results into time-series damage data, the time-series damage data is updated to the turbine blade memory bank; a stable observation factor for the turbine blade is generated based on the blade memory bank, and task allocation and damage monitoring management are performed for the corresponding turbine blade based on the stable observation factor.

[0042] Optionally, the real-time damage monitoring results are first compressed to transform the high-dimensional spatial distribution and damage level of the damage into low-dimensional time-series damage data. This compression can be achieved through methods such as key feature extraction, time step downsampling, or principal component analysis to retain the core information of the damage evolution trend while reducing storage and computational burden. Subsequently, the obtained time-series damage data is updated to the blade memory bank in chronological order. This blade memory bank serves as a dedicated data storage unit to record the damage evolution trajectory and historical operating characteristics of a single blade throughout its entire lifespan, thereby providing reliable data support for long-term monitoring and lifespan prediction. Based on the data in the blade memory bank, the time-series damage data of the target blade is extracted from the blade memory bank and analyzed. Taking damage level as an example, the difference between the damage level at the last moment and the initial damage level is calculated, and the difference is divided by the monitoring duration to obtain the damage growth rate, which reflects the average growth trend of blade damage over time. The standard deviation of the damage level is calculated to measure the fluctuation amplitude. By weighting the damage growth rate and fluctuation amplitude, a damage level stability factor is obtained. The same operation is performed for other time-series damage data, and all the calculated stability factors are weighted by importance to obtain the turbine blade stability observation factor, which reflects the long-term stability and reliability of the blade under different operating conditions. Finally, the stability observation factor is used for task allocation and damage monitoring management of the turbine blade. That is, blades with high stability can be assigned to high-load or critical flight missions; blades with poor stability or excessively rapid damage growth are assigned to low-risk missions or scheduled for maintenance or replacement. In this way, full life-cycle management based on historical memory and real-time monitoring is realized, improving the safety and economy of engine operation.

[0043] In summary, the embodiments of this application have at least the following technical effects: This embodiment first embeds a distributed sensor array on the surface of the turbine blade, including fiber optic grating sensors and thin-film piezoelectric sensors. Then, when the engine is under low load, a self-calibration excitation unit is activated to perform benchmark correction using the output response signal of the sensor array, generating an initial reference for blade health. Next, a non-contact observation group is deployed at the exhaust nozzle, and the observation dataset is read, including acoustic emission signals, vibration mode signals, and vortex evolution signals. Further, the initial blade health reference and the observation dataset are sent to a digital twin model for multimodal data fusion processing. Uncertainty propagation and deviation correction operators are used for joint data correction to generate the spatial distribution and damage level of the blade damage. Then, a swarm dynamics consistency model is activated to establish a swarm deviation index for the turbine blade. Finally, after compensating for the spatial distribution and damage level of the blade damage based on the swarm deviation index, damage monitoring results are generated. These technologies collectively address the challenge of timely and accurate detection and assessment of turbine blade damage, such as cracks, fatigue, and ablation, caused by complex operating environments. They achieve high-precision real-time monitoring of blade damage spatial distribution and damage level through multimodal data joint correction and group dynamics compensation mechanisms, thereby improving early damage warning capabilities and fault diagnosis reliability.

[0044] Example 2, based on the same inventive concept as the damage monitoring method for aero-engine turbine blades in the foregoing examples, such as... Figure 2 As shown, this application provides a damage monitoring system for aero-engine turbine blades. The system includes: a sensor deployment module 11, which embeds a sensor array on the surface of the turbine blade, the sensor array including fiber optic grating sensors and thin-film piezoelectric sensors; a reference calibration module 12, which activates a self-calibration excitation unit when the engine is under low load, and uses the self-calibration excitation unit to perform reference calibration by executing the output response signal of the sensor array to generate an initial reference for blade health; a data reading module 13, which deploys a non-contact observation group at the exhaust nozzle and reads the observation dataset of the non-contact observation group, the observation dataset including acoustic emission signals, vibration mode signals, and vortex evolution signals; a fusion calibration module 14, which sends the initial reference for blade health and the observation dataset to a digital twin model, performs multi-modal data fusion processing, and uses uncertainty propagation and deviation correction operators to perform joint calibration of the data to generate the spatial distribution and damage level of the blade damage; an index establishment module 15, which activates a population dynamics consistency model and uses the population dynamics consistency model to establish a population deviation index for the turbine blade; and a compensation module 16, which compensates for the spatial distribution and damage level of the blade damage according to the population deviation index, and generates damage monitoring results.

[0045] Furthermore, the fusion correction module 14 is also used to perform the following method: The strain detection signal from the initial reference of blade health is read, and the acoustic emission signal from the observation dataset is read; the strain detection signal and the acoustic emission signal are processed together to establish a first uncertainty component; a second uncertainty component is established by jointly fitting the spatiotemporal distribution stability of the vortex evolution signal and the mode frequency shift of the vibration mode signal; the uncertainty propagation link of the damage state is constructed using the first uncertainty component and the second uncertainty component; and uncertainty propagation is completed according to the uncertainty propagation link.

[0046] Furthermore, the fusion correction module 14 is also used to perform the following method: The uncertainty propagation link is discretized into N time steps; the uncertainty components in the uncertainty propagation link are weighted and accumulated according to the time relationship of the N time steps to establish a weighted accumulation result; an attenuation coefficient is set based on the relaxation characteristics of the blade material and the stability of historical operating conditions, and the attenuation processing of the weighted accumulation result is performed using the attenuation coefficient to complete the uncertainty propagation.

[0047] Furthermore, the fusion correction module 14 is also used to perform the following method: Obtain the propagation results of uncertainty propagation, and configure uncertainty factors based on the propagation results; map the strain signal, thin film piezoelectric signal in the initial reference of blade health, and the acoustic emission signal, vibration mode signal, and vortex evolution signal in the observation dataset to the same reference space; based on the uncertainty factors, establish cross-channel deviation constraints using the minimum cross-entropy criterion, perform joint optimization of temporal differences and spatial distribution differences, and establish joint optimization results; complete the joint correction of the deviation correction operator based on the joint optimization results.

[0048] Furthermore, the fusion correction module 14 is also used to perform the following method: The time steps of the sensor array and non-contact observation group are mapped to the standardized position of the rotation period according to the blade rotation period; after normalizing the local stress amplitude and vibration mode amplitude, the angle is corrected by using the mapping of the standardized position to establish a dynamic weighting factor; and weighted compensation is performed for joint optimization based on the dynamic weighting factor.

[0049] Furthermore, the fusion correction module 14 is also used to perform the following method: Read the engine's control parameters during low-load phases, and establish parameter compensation factors based on these parameters; use these parameter compensation factors to generate and compensate for the spatial distribution and damage level of blade damage.

[0050] Furthermore, the indicator establishment module 15 is also used to perform the following method: Obtain the blade group response matrix under the same operating conditions, and synchronize the blade group response matrix and the monitored turbine blade response results to the group dynamics consistency model; use the extraction layer in the group dynamics consistency model to extract the dominant modes based on principal component analysis or low-rank decomposition, and establish a group consistency reference; use the group consistency reference to calculate the residual vector of the monitored turbine blades, and generate a group deviation index.

[0051] Furthermore, the compensation module 16 is also used to perform the following method: Based on the damage monitoring results, a multi-level early warning threshold range is mapped to configure a damage early warning signal; the damage early warning signal is used to configure an operation and maintenance strategy, and the turbine blade maintenance management is carried out according to the operation and maintenance strategy.

[0052] Furthermore, the compensation module 16 is also used to perform the following method: After compressing the damage monitoring results into time-series damage data, the time-series damage data is updated to the turbine blade memory bank; a stable observation factor for the turbine blade is generated based on the blade memory bank, and task allocation and damage monitoring management are performed for the corresponding turbine blade based on the stable observation factor.

[0053] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0054] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0055] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for damage monitoring of aero-engine turbine blades, characterized in that, The method includes: A sensor array is embedded on the surface of a turbine blade, the sensor array including fiber optic grating sensors and thin-film piezoelectric sensors; When the engine is under low load, the self-calibration excitation unit is activated, and the output response signal of the sensor array is used to perform reference correction to generate an initial reference for blade health. A non-contact observation group is deployed at the tail nozzle, and the observation data set of the non-contact observation group is read. The observation data set includes acoustic emission signals, vibration mode signals, and vortex evolution signals. The initial reference and observation dataset of leaf health are sent to the digital twin model to perform multimodal data fusion processing. Uncertainty propagation and bias correction operators are used to jointly correct the data to generate the spatial distribution and damage level of leaf damage. Activate the population dynamics consistency model and use the population dynamics consistency model to establish a population deviation index for turbine blades; After compensating for the spatial distribution and damage level of the blade damage based on the population deviation index, damage monitoring results are generated.

2. The damage monitoring method for aero-engine turbine blades as described in claim 1, characterized in that, Sending the initial reference for leaf health and the observation dataset to the digital twin model includes: Read the strain detection signal from the initial reference of leaf health, and read the acoustic emission signal from the observation dataset; The strain detection signal and the acoustic emission signal are processed together to establish a first uncertainty component. A second uncertainty component is established by jointly fitting the spatiotemporal distribution stability of the vortex evolution signal and the modal frequency shift of the vibration mode signal. The uncertainty propagation link of the damage state is constructed using the first uncertainty component and the second uncertainty component; Uncertainty propagation is completed according to the aforementioned uncertainty propagation link.

3. The damage monitoring method for aero-engine turbine blades as described in claim 2, characterized in that, Uncertainty propagation is completed according to the aforementioned uncertainty propagation link, including: The uncertainty propagation link is discretized into N time steps; Based on the time relationship of N time steps, perform weighted accumulation of uncertainty components in the uncertainty propagation link and establish the weighted accumulation result; A decay coefficient is set based on the relaxation characteristics of the blade material and the stability of historical operating conditions. The decay coefficient is then used to perform a decay process on the weighted cumulative result in order to complete the uncertainty propagation.

4. The damage monitoring method for aero-engine turbine blades as described in claim 1, characterized in that, Sending the initial reference and observation dataset of leaf health to the digital twin model also includes: Obtain the propagation result of uncertainty propagation, and configure the uncertainty factor based on the propagation result; The strain signal, thin film piezoelectric signal in the initial reference of leaf health, and the acoustic emission signal, vibration mode signal, and vortex evolution signal in the observation data set are mapped to the same reference space; Based on the aforementioned uncertainty factors, cross-channel deviation constraints are established using the minimum cross-entropy criterion, joint optimization of temporal differences and spatial distribution differences is performed, and joint optimization results are established. The joint correction of the deviation correction operator is completed based on the joint optimization results.

5. The damage monitoring method for aero-engine turbine blades as described in claim 4, characterized in that, Cross-channel deviation constraints are established using the minimum cross-entropy criterion, and joint optimization of temporal and spatial distribution differences is performed, including: The time steps of the sensor array and non-contact observation group are mapped to the standardized position of the rotation period according to the blade rotation period; After normalizing the local stress amplitude and vibration mode amplitude, the angle is corrected by mapping the standardized position, and a dynamic weighting factor is established. Weighted compensation is performed based on the dynamic weighting factors for joint optimization.

6. The damage monitoring method for aero-engine turbine blades as described in claim 1, characterized in that, Activate the population dynamics consistency model and use the population dynamics consistency model to establish a population deviation index for turbine blades, including: Obtain the blade group response matrix under the same operating conditions, and synchronize the blade group response matrix and the monitored turbine blade response results to the group dynamics consistency model; By utilizing the extraction layer in the population dynamics consensus model, dominant modes are extracted based on principal component analysis or low-rank decomposition to establish a population consensus reference. The residual vector of the turbine blades monitored is calculated using the aforementioned group consistency reference to generate a group deviation index.

7. The damage monitoring method for aero-engine turbine blades as described in claim 1, characterized in that, Generate damage monitoring results, including: Based on the damage monitoring results, a multi-level early warning threshold range is mapped to configure a damage early warning signal; The damage warning signal is used to configure an operation and maintenance strategy, and the turbine blades are maintained and managed according to the operation and maintenance strategy.

8. The damage monitoring method for aero-engine turbine blades as described in claim 1, characterized in that, After generating the damage monitoring results, the following are also included: After compressing the damage monitoring results into time-series damage data, the time-series damage data is updated to the turbine blade memory bank. Based on the blade memory library, a stable observation factor for the turbine blade is generated, and based on the stable observation factor, the corresponding turbine blade is assigned tasks and damage monitoring is managed.

9. The damage monitoring method for aero-engine turbine blades as described in claim 1, characterized in that, The generation of spatial distribution and damage level of leaf damage also includes: Read the engine control parameters during the low-load phase and establish a parameter compensation factor based on the control parameters; The aforementioned parameter compensation factor is used to generate and compensate for the spatial distribution and damage level of leaf damage.

10. A damage monitoring system for aero-engine turbine blades, characterized in that, The system is used to perform the damage monitoring method for aero-engine turbine blades according to any one of claims 1-9, the system comprising: Sensor deployment module: A sensor array is embedded on the surface of the turbine blade, the sensor array including fiber optic grating sensors and thin-film piezoelectric sensors; Reference calibration module: When the engine is under low load, the self-calibration excitation unit is activated, and the output response signal of the sensor array is executed by the self-calibration excitation unit to perform reference calibration and generate an initial reference for blade health. Data reading module: Deploys a non-contact observation group at the tail nozzle and reads the observation dataset of the non-contact observation group, which includes acoustic emission signals, vibration mode signals, and vortex evolution signals; Fusion correction module: Sends the initial reference of leaf health and observation dataset to the digital twin model, performs multimodal data fusion processing, uses uncertainty propagation and bias correction operators to jointly correct the data, and generates the spatial distribution and damage level of leaf damage; Indicator establishment module: Activate the population dynamics consistency model and use the population dynamics consistency model to establish the population deviation index of turbine blades; Compensation module: After compensating for the spatial distribution and damage level of the blade damage according to the population deviation index, damage monitoring results are generated.

Images