A robust reconstruction and state degradation assessment method for a turbine top cover dynamic system
By constructing an equivalent pressure input and time delay correction, a weak-form dynamic matrix is established. Combined with benchmark regularization and self-stiffness residual projection, the shortcomings of existing roof condition monitoring methods are addressed, enabling robust reconstruction and condition degradation assessment of the roof dynamic system, and improving the accuracy and reliability of the assessment.
Patent Information
- Application Number
- CN202611102503.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-23
- Publication Date
- 2026-11-10
AI Technical Summary
Existing methods for monitoring the condition of turbine top covers mainly rely on vibration amplitude, spectral characteristics, and trend thresholds for judgment. It is difficult to directly establish the equivalent dynamic relationship between pressure pulsation input and the three-dimensional displacement response of the top cover, resulting in inaccurate assessment of the condition degradation of the top cover.
By collecting operating data of the roof cover, performing time synchronization and filtering, constructing an equivalent pressure input and performing time delay correction, establishing a weak form dynamic matrix equation, reconstructing a healthy benchmark model using the regularized least squares method, and evaluating the three-dimensional relative stiffness changes of the roof cover by combining benchmark regularization and self-stiffness residual projection.
It achieves stable reconstruction of the top cover dynamic model under strong noise conditions, reduces parameter drift, improves the reliability of state identification results, and can quantitatively assess the direction and degree of top cover state degradation, providing a basis for operational anomaly identification and maintenance decisions.
Smart Images

Figure CN123069946A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of turbine structure condition monitoring and dynamic identification technology, and relates to a robust reconstruction and condition degradation assessment method for the turbine top cover dynamic system. Background Technology
[0002] The turbine top cover, located above the turbine's flow passage, is a crucial structural component that bears the force of water pressure, the guide vane mechanism, and the transmission of unit vibration. During unit operation, the top cover is constantly affected by factors such as hydraulic pressure pulsations, dynamic and static interference of the runner, changes in guide vane opening, connection constraints, and unit load fluctuations. Its dynamic response reflects changes in the top cover's structural stiffness, constraint state, and pressure-structure coupling relationship. To ensure the safe and stable operation of the unit, existing projects typically monitor the operating status of the top cover and its adjacent areas using pressure pulsation sensors, displacement sensors, and vibration sensors. This data, combined with operating condition data such as unit load, head, guide vane opening, and rotational speed, is used to analyze the top cover's vibration level, pressure pulsation characteristics, and abnormal states.
[0003] Existing methods for analyzing the condition of hydroelectric power plant roofs mainly include vibration amplitude monitoring, spectrum analysis, trend comparison, threshold alarm, modal testing, and finite element simulation. Vibration amplitude and spectrum analysis can be used to determine whether the roof response exceeds limits and whether the main frequency components are abnormal; pressure pulsation analysis can be used to identify hydraulic excitation characteristics and their correlation with structural response; finite element modal analysis and dynamic response calculation can be used to evaluate the roof structure's stiffness, boundary constraints, and dynamic characteristics under pressure loads. In recent years, system identification, condition feature extraction, and data-driven prediction methods based on monitoring data have also been increasingly used for the condition assessment of hydroelectric equipment, extracting information on structural condition changes from operational data and assisting in determining whether the equipment's operating condition has changed.
[0004] However, existing technologies often focus on the individual analysis of pressure pulsations or vibration responses, typically using amplitude, frequency, trend, or simulation comparisons as the basis for judgment. This makes it difficult to directly establish the equivalent dynamic relationship between pressure pulsation input and the three-dimensional displacement response of the roof. For roof degradation issues, simply judging an increase in vibration amplitude or a change in spectral composition is insufficient to further explain whether the equivalent stiffness of the roof has decreased, in which direction the degradation mainly occurs, and to what extent. Furthermore, field monitoring data often suffers from measurement noise, time delays between pressure input and structural response, variations in operating conditions, and strong correlations between candidate dynamic terms. These problems prevent conventional identification methods from stably and accurately reconstructing the roof's dynamic model. Directly using conventional methods to reconstruct dynamic parameters easily leads to derivative noise amplification, parameter drift, and misjudgments of state changes, severely restricting the engineering application of model-based state degradation assessment methods. Summary of the Invention
[0005] The technical problem to be solved by this invention is to provide a robust reconstruction and state degradation assessment method for the dynamic system of a turbine top cover, in order to solve the problem that existing turbine top cover state monitoring methods mainly rely on vibration amplitude, spectral characteristics and trend thresholds for judgment, and it is difficult to quantitatively reveal the dynamic state changes from the pressure pulsation input and the three-dimensional displacement response of the top cover.
[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a robust reconstruction and state degradation assessment method for a turbine top cover dynamic system, comprising the following steps: S1: Collect displacement response, pressure pulsation signals and operating parameters during the operation of the turbine top cover. Perform time synchronization, outlier removal, trend term removal, filtering and steady-state segment screening on the collected data to form a monitoring dataset for dynamic reconstruction. S2: Construct the original pressure pulsation measurement data into an equivalent pressure input corresponding to the three-dimensional displacement of the top cover; S3: Perform time delay correction on the equivalent pressure input, and determine the optimal time delay after correction based on the criterion of minimizing the weak form dynamic residual; S4: After completing the pressure input time delay correction, a smooth test function is introduced to construct weak form integral constraints within multiple time windows and establish weak form matrix equations for each direction of the top cover. S5: Select stable operating data of the top cover under healthy or baseline conditions, construct weak form matrix equations, use regularized least squares method to solve for healthy baseline dynamic coefficients, and establish a healthy baseline equivalent dynamic model. S6: For the current operating state data, the same data processing method as the health baseline state is used to construct the weak form matrix equation of the current state, and the current state dynamic coefficients are reconstructed by baseline regularization with reference to the health baseline dynamic coefficients to obtain the low-dimensional equivalent dynamic model of the current state. S7: Calculate the relative stiffness change estimate of the benchmark regularized reconstruction channel based on the self-stiffness correlation coefficient in the health benchmark model and the current state model; S8: Calculate the weak-form residual of the current state relative to the health baseline model, obtain the change in self-stiffness coefficient through residual projection, and calculate the relative stiffness change estimate of the residual projection channel. S9: Calculate the pressure-displacement compliance of the health baseline state and the current state in multiple target frequency bands, and obtain the relative stiffness change estimate of the compliance channel from the compliance ratio; S10: The relative stiffness changes obtained from the benchmark regularization reconstruction channel, the self-stiffness residual projection channel, and the pressure-displacement compliance channel are fused to obtain the three-dimensional relative stiffness change vector of the top cover, and the overall degradation degree and main degradation direction are determined.
[0007] The displacement response described in S1 is the equivalent displacement response of the top cover: ; in, u x ( t ), u y ( t ), u z ( t ) respectively indicate the top cover at x direction, y direction and z The equivalent displacement response in the direction, the pressure pulsation measurement point data is denoted as: ; in, m This represents the number of pressure pulsation measurement points. p 1( t ), p 2( t ),…, p m ( t The parameters are pressure pulsation signals at different locations; the operating parameters include unit load, head, guide vane opening and speed.
[0008] The equivalent pressure input described in S2 is: ; in, Construct operators for pressure input; The equivalent pressure input after construction is denoted as: ; in, p e,x ( t ), p e,y ( t ), p e,z ( t ) respectively with the top cover x , y , z The pressure input corresponding to the directional equivalent displacement response.
[0009] S3 describes the first j In each direction, the prior engineering delay is assumed to be... The allowed search scope is The candidate delay range is: ; In candidate delay Next, thej The time delay correction pressure input in each direction is: ; The optimal time delay after correction is determined based on the criterion of minimizing the weak-form dynamic residual: ; in, For the first j Each direction in candidate latency The residuals are calculated from the weak form dynamics matrix. This is the prior time delay constraint coefficient for the project.
[0010] The top cover described in S4 j The low-dimensional equivalent dynamic relationship in each direction is as follows: ; in, For the first j Equivalent acceleration response in each direction, As a candidate function library, These are the dynamic coefficients to be identified; The candidate function libraries are: ; in, Used to characterize damping-related terms, Used to characterize nonlinear stiffness terms, Used to characterize the pressure input term after time delay correction. Used to characterize pressure-displacement coupling terms; Introducing a smoothness test function During the time window Internally constructed weak formal integral constraints: ; in, For the first q One test function; No. j The weak form matrix equations for each direction are: ; in, This is the weakly form inertial term obtained by integrating the second derivative of the displacement response and the test function. This is the weakly form candidate matrix obtained by integrating the candidate function library and the test function. This is the residual term.
[0011] The first mentioned in S5 j Weak-form matrix equations of health status in each direction: ; The superscript 0 indicates the baseline health status. For the first time in a healthy state j Weak-form candidate matrices in each direction, To correspond to the weak form of the inertia term, The health baseline kinetic coefficients to be solved are... For residual terms; The health baseline kinetic coefficients are solved using the regularized least squares method: ; in, The regularization coefficient is used. The equivalent kinetic model of the health benchmark is: .
[0012] The first mentioned in S6 j The weak form matrix equations of the current state in each direction are: ; Among them, superscript c Indicates the current state. For the current state, the first j Weak-form candidate matrices in each direction, Let be the corresponding weak form inertia term. These are the dynamic coefficients to be solved under the current state. For residual terms; Based on health baseline kinetic coefficients For reference, the current state dynamic coefficients are reconstructed using benchmark regularization: ; in, As a health baseline constraint coefficient, The coefficient constraint weight matrix is defined; the first constraint weight is set for the damping term coefficient, pressure input term coefficient, pressure-displacement coupling term coefficient, and inter-direction coupling term coefficient; for the... j The first constraint weight is greater than the second constraint weight, which is set as the correlation coefficient of self-stiffness in each direction.
[0013] The equivalent self-stiffness in the j-th direction described in S7 and the current state are respectively: ; in, The first under the health baseline state j Equivalent self-stiffness in each direction, For the current state, the first j Equivalent self-stiffness in each direction, and These are the self-stiffness correlation coefficients in the healthy baseline model and the current state model, respectively; the relative stiffness change estimate of the baseline regularized reconstruction channel is: ; in, The relative stiffness change obtained by reconstructing the channel using the baseline regularization. To prevent positive numbers with excessively small denominators.
[0014] The weak-form residual of the current state relative to the health baseline model, as described in S8, is: ; in, This indicates dynamic differences in the current state data that cannot be explained by the health baseline model. For the current state, the first j Weakly formal inertial terms in each direction, This is the weak-form candidate matrix for the current state. Let be the kinetic coefficients for health baseline; let The weak form candidate matrix for the current state Correlation coefficient between medium and self-stiffness The corresponding column vector, then the change in the self-stiffness coefficient is: ; in, Here is the projection regularization coefficient; the relative stiffness change estimate for the residual projection channel is: ; in, This represents the relative stiffness change obtained from the stiffness residual projection channel.
[0015] The first mentioned in S9 j The direction, in the first b Within a target frequency band, the pressure-displacement compliance is defined as: ; in, For the first b A target frequency band filtering operator Represents the root mean square value. The equivalent pressure input is after time delay correction; the relative stiffness change of the compliance channel is estimated as follows: ; in, This is an estimate of the relative stiffness change obtained from the pressure-displacement compliance channel. This indicates taking the median value of the results for multiple target frequency bands.
[0016] The first mentioned in S10 jThe final relative stiffness change index in each direction is: ; in, No. j The final relative stiffness change index in each direction, These are the fusion weights for the three channels, satisfying the following: The three-dimensional relative stiffness variation vector of the top cover is: The overall degradation level index is: The main direction of degradation is: ,in, D Indicates the degree of degradation of the top cover. Indicates the main direction of degradation.
[0017] The main beneficial effects of this invention are as follows: 1. It can establish an equivalent dynamic relationship between pressure pulsation input and roof displacement response. By constructing an equivalent pressure input, correcting for pressure input time delay, and applying weak formal dynamic constraints, the originally dispersed pressure pulsation data and roof displacement data are unified into a low-dimensional equivalent dynamic model, overcoming the problem that traditional amplitude, spectrum, and trend analysis cannot explain the source of state changes.
[0018] 2. Higher stability of dynamic reconstruction under strong noise conditions. By constructing weak-form dynamic equations through multi-scale time windows and smooth test functions, the second derivative of the noisy displacement signal is avoided, reducing the impact of noise amplification on the identification of dynamic coefficients and improving the robustness of model reconstruction under operational monitoring data conditions.
[0019] 3. It can reduce input response mismatch and parameter compensation. By correcting the equivalent time delay between the pressure input and the top cover response through engineering prior time delay and weak form residual correction, the parameter compensation caused by time mismatch between the pressure term, stiffness term, damping term and pressure-displacement coupling term is reduced.
[0020] 4. It can improve the comparability and reliability of the current state identification results. By establishing a health benchmark model and reconstructing the current state under the constraints of the health benchmark, the healthy state and the current state are compared under the same model structure and the same coefficient meaning, avoiding parameter drift caused by free fitting of the current state.
[0021] 5. Capable of quantitatively assessing the direction and extent of roof degradation. Through benchmark regularization reconstruction, self-stiffness residual projection, and pressure-displacement compliance channel fusion, the roof condition is obtained. x , y , zThe three-dimensional relative stiffness variation index can determine whether the top cover has deteriorated, the main direction of degradation, and the degree of degradation, providing a quantitative basis for identifying operational anomalies, tracking status trends, and making maintenance decisions. Attached Figure Description
[0022] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0023] Figure 1 This is a schematic diagram of the overall process of the robust reconstruction and state degradation assessment method for the turbine top cover dynamic system of the present invention; Figure 2 This is a schematic diagram of the arrangement of measuring points for the displacement response and pressure pulsation of the top cover in this invention; Figure 3 This is a schematic diagram of the pressure input delay correction of the present invention; Figure 4 This is a schematic diagram of the multi-scale weak form dynamic constraint construction of the present invention; Figure 5 This is a schematic diagram of the regularized reconstruction of the health benchmark model and the current state benchmark of the present invention; Figure 6 This is a schematic diagram illustrating the extraction and fusion evaluation of state degradation indicators in this invention; Figure 7 This is a schematic diagram showing the results of an embodiment of the state degradation recognition method of the present invention; Figure 8 This is a schematic diagram illustrating the robustness verification results of state degradation identification under strong noise conditions according to the present invention.
[0024] In the diagram: 1. Turbine top cover; 2. Top cover displacement measuring point; 3. Pressure pulsation measuring point in the bladeless zone; 4. Pressure pulsation measuring point in the volute; 5. Pressure pulsation measuring point in the tailrace pipe. Detailed Implementation
[0025] Example 1: S1: Multi-source operation monitoring data acquisition, collecting displacement response, pressure pulsation signals, and operating parameters during the operation of turbine top cover 1, forming a monitoring dataset for dynamic reconstruction. The equivalent displacement response of the top cover is denoted as: ; in, u x ( t ), u y ( t ), u z ( tThe values () represent the equivalent displacement responses of the top cover in the three directions. These equivalent displacement responses can be obtained from a single three-dimensional displacement measurement point, or from multiple displacement measurement points through weighted fusion, principal component extraction, or modal coordinate transformation. Pressure pulsation measurement point data are denoted as: ; in, m This represents the number of pressure pulsation measurement points. p 1( t ), p 2( t ),…, p m ( t The pressure pulsation signals are located at different positions. These pressure pulsation measurement points include pressure pulsation measurement point 3 in the bladeless zone, pressure pulsation measurement point 4 in the volute, pressure pulsation measurement point 5 in the tailrace, or other locations that can reflect hydraulic excitation characteristics. Operating parameters include unit load, head, guide vane opening, and rotational speed.
[0026] In this embodiment, according to Figure 2 The measurement point arrangement shown acquires the three-dimensional displacement response and pressure pulsation signal of the top cover, wherein top cover displacement measurement point 2 is used to obtain... x , y , z The three-dimensional equivalent displacement response, pressure pulsation measurement points 3 (bladeless zone), 4 (volute), and 5 (tailrace pipe) were used to obtain pressure pulsation signals in different hydraulic regions. The sampling frequency for both the displacement response and pressure pulsation signals was 200 Hz, the length of a single stable operating segment was 25 s, and each segment contained 5000 sampling points. Six stable operating condition segments were selected for joint calculation, with load factors of 0.60, 0.70, 0.80, 0.90, 1.00, and 0.75, corresponding to pressure amplitude factors of 0.86, 0.92, 0.98, 1.04, 1.10, and 0.95. To simulate strong noise monitoring conditions, measurement noise with a signal-to-noise ratio of 8 dB was added to the top cover displacement response, and measurement noise with a signal-to-noise ratio of 20 dB was added to the pressure pulsation signal. The results of 12 repeated samplings were then averaged. After acquisition, the data underwent time synchronization, outlier removal, trend term removal, filtering, and steady-state segment selection. Specifically, the top cover displacement response was smoothed using a third-order Savitzky-Golay process with a 0.08 s time window to reduce the impact of strong noise on subsequent weak-form integral calculations. The parameters mentioned above are consistent with the sampling frequency, segment length, noise level, average number of repetitions, and smoothing window settings in the code.
[0027] S2: Equivalent Pressure Input Construction. Since multiple pressure pulsation measurement points are not directly equivalent to the equivalent input acting on the three-dimensional response of the top cover, this invention transforms the original pressure pulsation measurement point data into an equivalent pressure input corresponding to the three-dimensional displacement response of the top cover. Let the original pressure pulsation measurement point data be p. m ( t If the equivalent pressure input is: ; in, Operators are constructed for the pressure input, which may include one or more of the following: target frequency band filtering, normalization processing, weighted fusion of measurement points, principal component extraction, or operating condition correction. The constructed equivalent pressure input is denoted as: ; in, p e,x ( t ), p e,y ( t ), p e,z ( t ) respectively with the top cover x , y , z The pressure input corresponding to the directional equivalent displacement response.
[0028] In this embodiment, The process employs target frequency band selection and normalization. Specifically, after removing the mean and trend from the three pressure pulsation signals from the bladeless area pressure pulsation measuring point 3, the volute pressure pulsation measuring point 4, and the tailrace pipe pressure pulsation measuring point 5, the pressure components within the target analysis frequency band of 0.15 Hz to 20 Hz are extracted as effective pressure inputs for low-dimensional equivalent dynamic reconstruction. This target frequency band is used to retain pressure pulsation components related to the low-order dynamic response of the top cover, while suppressing excessively low-frequency trend terms and high-frequency interference. After processing, the original three pressure pulsation signals are respectively formed into... p e,x ( t ), p e,y ( t ), p e,z ( t This is used for subsequent pressure input delay correction and weak form dynamic constraint construction. The code also sets the pressure analysis frequency band to 0.15 Hz to 20 Hz for low-order cap dynamics reconstruction.
[0029] S3: Pressure Input Delay Correction. The pressure pulsation measuring point and the top cover displacement measuring point 2 differ in spatial location, propagation path, and acquisition link, resulting in an equivalent time delay between the pressure input and the top cover response. For the first... j One direction, Let the prior time delay of the project be... The allowed search scope is The candidate delay range is: In candidate delay Next, the j The time delay correction pressure input in each direction is: ; Furthermore, the modified optimal time delay is determined based on the criterion of minimizing the weak-form dynamic residual: ; in, For the first j Each direction in candidate latency The residuals are calculated from the weak form dynamics matrix. This is the prior time delay constraint coefficient for the project.
[0030] In this embodiment, the sampling frequency is 200 Hz, and one sampling point corresponds to 0.005 s. The sampling frequency is set according to the pressure measurement point location, signal propagation path, and data acquisition link conditions. x , y , z The prior engineering delays for the three directions are 4, 3, and 5 sampling points, corresponding to times of 0.020 s, 0.015 s, and 0.025 s, respectively. The allowed search range is Δτ = 1 sampling point; therefore, the candidate search ranges for the three directions are [3, 5], [2, 4], and [4, 6], respectively. (Engine prior delay constraint coefficients are also specified.) The value is set to 0.05. In the specific calculation, the initial estimate of the time delay is first obtained based on the correlation between the pressure input and the displacement response within the target frequency band. Then, the weak-form dynamic residuals corresponding to the candidate time delays are calculated one by one in the engineering prior time delay neighborhood. The candidate time delay with the smallest objective function is selected as the final correction time delay. Figure 3 (a) shows a segment of pressure input and displacement response. Figure 3 (b) A comparison of the prior engineering delay, the initial estimated bandwidth delay, and the corrected delay is given. Figure 3 (c) The pressure input delay correction procedure is given. The above delay and search parameters are consistent with the three-dimensional delay, local search radius, and delay prior constraint settings in the code.
[0031] S4: Multi-scale weak-form dynamic constraint construction. After completing the pressure input time delay correction, the top cover is established. j Low-dimensional equivalent dynamic relationships in each direction: ; in, For the first j Equivalent acceleration response in each direction, As a candidate function library, These are the dynamic coefficients to be identified. The candidate function library can be written as: ; Among them, the triaxial displacement term is used to characterize the linear coupling between directions. Used to characterize damping-related terms, Used to characterize nonlinear stiffness terms, Used to characterize the pressure input term after time delay correction. This is used to characterize the pressure-displacement coupling term. Since measurement noise exists in the field displacement response, directly calculating acceleration from the displacement response can easily lead to noise amplification. Therefore, a smoothed test function is introduced. The test function and its first derivative are both zero at both ends of the time window, i.e. and Within the time window Internally constructed weak formal integral constraints: ; in, For the first q A test function. This formula transfers the second derivative of the displacement response to the test function through integration by parts, thereby avoiding the direct calculation of acceleration from a strong noise displacement signal.
[0032] In this embodiment, the multi-scale time window lengths are 0.6 s, 0.9 s, 1.2 s, 1.6 s, and 2.0 s, and the adjacent window sliding step size is 0.12 s; the test functions are of order 0 to 3, totaling 4 types. For each stable operating condition segment, each time window, and each test function, the weak-form inertia term and the candidate function integral term are calculated respectively; then, the integral results corresponding to the 6 stable operating condition segments, 5 types of time windows, and 4 types of test functions are superimposed row by row to form the first... j Joint weak form matrix equations in all directions: ; in, This is the weakly form inertial term obtained by integrating the second derivative of the displacement response and the test function. This is the weakly form candidate matrix obtained by integrating the candidate function library and the test function. This is the residual term. Figure 4This embodiment illustrates the process from smooth displacement response and time-delayed pressure input to multi-scale windows, smooth test functions, weak-form integral calculations, and matrix equation construction. Thus, this embodiment completes the conversion of multi-source monitoring data into weak-form dynamic matrix equations, providing a unified data foundation for subsequent health baseline model establishment and current state model reconstruction.
[0033] S5: Establishment of the equivalent dynamic model for the health baseline. Selecting stable operating data of the top cover under healthy or baseline conditions, construct the equivalent pressure input, correct for pressure input time delay, and construct multi-scale weak-form dynamic constraints according to the method described in Part I, thus obtaining the... j Weak-form matrix equations of health status in each direction: ; The superscript 0 indicates the baseline health status. For the first time in a healthy state j Weak-form candidate matrices in each direction, To correspond to the weak form of the inertia term, The health baseline kinetic coefficients to be solved are... This represents the residual term. To reduce the impact of noise and candidate correlation on the health baseline model, the regularized least squares method is used to solve for the health baseline dynamic coefficients: ; in, This is the regularization coefficient.
[0034] In this embodiment, the health status data still uses six stable operating condition segments, each 25 seconds long, with a sampling frequency of 200 Hz. After time delay correction and weak-form integration calculations are performed on each operating condition segment, they are superimposed row by row to form a joint weak-form equation system. For x , y , z All three directions utilize the same candidate function library, the same time window set, and the same test function set to ensure consistency in the construction of the health benchmark model across all three directions. In this embodiment, the regularization coefficient... Take 10 -6 Under healthy conditions, the three equivalent frequencies are approximately 5.56 Hz, 7.78 Hz, and 10.37 Hz, respectively, corresponding to an equivalent self-stiffness of approximately: After solving in three directions, the equivalent dynamic model of the health baseline is obtained: This model is used to characterize the pressure input, displacement response, damping, stiffness, nonlinear terms, and pressure-structure coupling relationship of the roof in a healthy state. It serves as a dynamic reference for subsequent reconstruction of the current state model and comparison of state degradation.
[0035] S6: Current State Baseline Regularized Reconstruction. For the current operating state data, the same data processing method, equivalent pressure input construction method, pressure input delay correction method, and weak-form dynamic constraint construction method as the health baseline state are used to obtain the... j The weak form matrix equation of the current state in each direction: ; Among them, superscript c Indicates the current state. For the current state, the first j Weak-form candidate matrices in each direction, Let be the corresponding weak form inertia term. These are the dynamic coefficients to be solved under the current state. This represents the residual term. Since the current state data may be affected by measurement noise, operating condition differences, and candidate factor correlations, directly solving the current state dynamic coefficients can easily cause drift in the damping term, pressure input term, pressure-displacement coupling term, and directional coupling term. Therefore, this invention uses the healthy baseline dynamic coefficients... For reference, the current state dynamic coefficients are reconstructed using benchmark regularization: ; in, As a health baseline constraint coefficient, This is the coefficient constraint weight matrix. For coefficients that are easily affected by noise and differences in operating conditions, such as damping terms, pressure input terms, pressure-displacement coupling terms, and inter-directional coupling terms, relatively large constraint weights are set; for the... j One direction and The corresponding self-stiffness correlation coefficient is set with a relatively small constraint weight to preserve the effective offset generated by changes in structural state. In this embodiment, the current state data and the healthy baseline state use the same six stable operating conditions, the same 200 Hz sampling frequency, the same 25 s data segment length, the same target pressure frequency band, and the same weak-form window parameters. To verify the invention's ability to identify state degradation, the current state is set to be only slightly different from the healthy state. z The equivalent stiffness decreases, that is: The corresponding relative stiffness change is set as follows: In the current state The main direction of degradation is towards x Xianghe yTo prevent stiffness degradation. In this embodiment, the health baseline constraint weight for non-self-stiffness related terms is 0.08, and the health baseline constraint weight for self-stiffness related terms is 10. -6 This setting allows for the suppression of non-degenerate parameter drift while enabling... The change in self-stiffness can be preserved in the current state model.
[0036] Complete them separately x , y , z After solving for the current state coefficients in the three directions, we obtain the set of coefficients for the current state model: Compared with health benchmark models: Correspondingly, the low-dimensional equivalent dynamic model of the current state can be expressed as: ; in, This is the candidate function library after pressure input time delay correction under the current state. The health baseline model and the current state model use the same candidate function library, the same weak form construction method, and the same coefficient arrangement order; therefore, the corresponding coefficients in both models have the same dynamic meaning. Furthermore, the th... j In the candidate function library of each direction and The corresponding self-stiffness correlation coefficient is denoted as The correlation coefficients of self-stiffness between the healthy baseline state and the current state are as follows: and By assembling the above models, the dynamic characteristics of the current state can be expressed in a low-dimensional equivalent dynamic form consistent with the health benchmark model. This isomorphic and comparable model can not only output the complete dynamic coefficients of the current state, but also provide a unified basis for subsequent self-stiffness residual projection, pressure-displacement compliance calculation, and extraction of triaxial relative stiffness degradation indices. Figure 5 The relationship between the health baseline model, the regularized reconstruction of the current state baseline, and the assembly of the low-dimensional equivalent dynamic model of the current state is illustrated.
[0037] S7: Benchmark regularized reconstruction channel relative stiffness estimation. For the first... j In each direction, the displacement terms in the candidate function library that correspond to this direction will be selected. The corresponding coefficient is denoted as the self-stiffness correlation coefficient. Since the displacement terms in the low-dimensional equivalent dynamic equations typically appear in the form of restoring forces, the healthy baseline state and the current state can be defined. j The equivalent self-stiffnesses in each direction are as follows: ; in, The first under the health baseline state j Equivalent self-stiffness in each direction, For the current state, the first j The equivalent self-stiffness in each direction. When the healthy state and the current state use the same measuring points, the same data processing methods, and the same or similar operating conditions, the relative change of the mass-normalized equivalent self-stiffness coefficient is used to characterize the change in stiffness state of the top cover in the corresponding direction. and These are the self-stiffness correlation coefficients in the healthy baseline model and the current state model, respectively. The first type of relative stiffness change estimate can be obtained from the regularized reconstruction results of the current state baseline: ; in, The relative stiffness change obtained by reconstructing the channel using the baseline regularization. To prevent positive numbers with excessively small denominators.
[0038] In this embodiment, the three-dimensional equivalent self-stiffnesses under the healthy baseline state are respectively The current state is set to... z Reduce the equivalent stiffness by 20%, that is ,and x Xianghe y The direction remains unchanged. Therefore, the relative stiffness change of the current state is: The baseline regularization reconstruction channel can initially reflect the trend of self-stiffness changes in the current state, but since the healthy baseline constraint will suppress coefficient offset, relying solely on this channel may make the degradation amplitude too conservative.
[0039] S8: Relative stiffness estimation via self-stiffness residual projection. To avoid excessive constraints on stiffness degradation magnitude imposed by the baseline regularization process, the weak-form residuals of the current state relative to the healthy baseline model are calculated: ; in, This indicates dynamic differences in the current state data that cannot be explained by the health baseline model. For the current state, the first j Weakly formal inertial terms in each direction, This is the weak-form candidate matrix for the current state. Let be the kinetic coefficient for the health baseline. The weak form candidate matrix for the current state Correlation coefficient between medium and self-stiffness The corresponding column vectors, then the change in stiffness coefficient can be obtained from the residual projection: ; in, Let be the projection regularization coefficient. The corresponding estimate of the relative stiffness change in the residual projection is: ; in, This is the relative stiffness change obtained from the self-stiffness residual projection channel. Through this step, the dynamic residual of the current state relative to the healthy baseline model can be further transformed into triaxial self-stiffness change information. This channel can compensate for the constraint effect of baseline regularization reconstruction on the degradation amplitude, ensuring that the current state... z This allows for a more complete reflection of stiffness reduction information, while avoiding deviations in degradation magnitude caused by relying solely on a single coefficient of the current state model. This step corresponds to... Figure 6 The "self-stiffness residual projection channel" in the text.
[0040] S9: Construction of the pressure-displacement compliance channel. To further improve the stability of stiffness degradation amplitude assessment under strong noise conditions, this invention introduces a pressure-displacement compliance channel. For the first... j The direction, in the first b Within a target frequency band, the pressure-displacement compliance is defined as: ; in, For the first b A target frequency band filtering operator Represents the root mean square value. This is the equivalent pressure input after time delay correction. The compliance under the healthy baseline state and the current state are denoted as follows: and When the equivalent stiffness of the roof in a certain direction decreases, the displacement response in that direction usually increases under similar pressure input, corresponding to an increase in pressure-displacement compliance. Therefore, the relative stiffness change of the compliance channel can be estimated from the compliance ratio between the healthy state and the current state. ; in, This is an estimate of the relative stiffness change obtained from the pressure-displacement compliance channel. This indicates taking the median value of the results for multiple target frequency bands.
[0041] In this embodiment, the pressure-displacement compliance calculation employs a joint judgment of multiple target frequency bands, including the low-frequency pressure response band, the band near 4 times the rotational speed, the band near 8 times the rotational speed, and the overall target analysis band. The unit's rated rotational speed is approximately 1.852 Hz, 4 times the rotational speed is approximately 7.41 Hz, and 8 times the rotational speed is approximately 14.81 Hz. Therefore, in this embodiment, the compliance calculation frequency bands can be taken as 0.30 Hz to 3.50 Hz, approximately 6.21 Hz to 8.61 Hz, approximately 13.01 Hz to 16.61 Hz, and 0.50 Hz to 20.00 Hz. The pressure-displacement compliance is calculated for each frequency band under both healthy and current conditions, and the median value of the multi-frequency band results is taken as the compliance channel output. Through this step, the change in top cover stiffness can be supplemented by judging the relationship between pressure input and displacement response amplitude. z With reduced stiffness, under similar pressure input, z The displacement response is enhanced, and the corresponding compliance channel provides an estimate of the negative relative stiffness change; while x Xianghe y Without stiffness degradation, the flexibility change is small. This result provides an auxiliary judgment criterion independent of the model coefficients for subsequent three-channel fusion, reducing the risk of bias caused by noise, operating condition differences, or candidate correlation of single model coefficients. This step corresponds to... Figure 6 The "pressure-displacement compliance channel" in the text.
[0042] S10: Three-channel fusion and state degradation assessment. This invention fuses the relative stiffness changes obtained from the benchmark regularized reconstruction channel, the self-stiffness residual projection channel, and the pressure-displacement compliance channel to obtain the third... j The final relative stiffness degradation index in each direction: ; in, No. j The final relative stiffness change index in each direction, These are the fusion weights for the three channels, satisfying the following: The fusion weights are determined based on the consistency of the results from the three channels. In this embodiment, the average value of the baseline regularized reconstruction channel and the self-stiffness residual projection channel is calculated first: Then, the fusion weight is determined based on the difference between the pressure-displacement compliance channel and the average value: And order: ,in, Indicates will Limited to the range Inside. When the pressure-displacement compliance channel is not aligned with the average value direction and ,make This is to reduce the impact of channels with large differences on the fusion results.
[0043] In this embodiment, degradation is set to z Reduce stiffness by 20%, x Xianghe y No degradation occurs. After fusion of the three channels, the following is obtained: z The relative stiffness variation identification value is approximately 0.194, compared to the set value The value is close to 0.200, with an absolute error of approximately 0.006. x Xianghe y The recognition result is close to 0, with no obvious false alarms. Therefore, the three-dimensional relative stiffness variation vector of the roof is obtained: Furthermore, an index for the overall degree of degradation can be defined: And determine the main direction of degradation: ,in, D Indicates the degree of degradation of the top cover. Indicates the primary degradation direction. Based on a preset threshold. , can be used for the first Classify the state in each direction: ,in, This can be determined based on the unit type, historical operating data, simulation calibration results, or maintenance experience. In this embodiment, D Approximately 0.194, with the main degradation direction being... z Towards. If we take the state classification threshold... ,but x Xianghe y To satisfy This is considered normal. z To satisfy It was determined to be a significant degradation. Figure 7 (a) A comparison between the identified values and the set values of the three-dimensional relative stiffness variation is given. Figure 7 (b) gives z The relative changes in equivalent stiffness, equivalent damping, pressure coupling, and nonlinear correlation state indices, among which The change is most pronounced towards the equivalent stiffness, compared to the set value. z Consistent with stiffness degradation. To verify stability under strong noise conditions, this embodiment further varied the displacement response signal-to-noise ratio, and repeated calculations were performed under conditions of 30 dB, 20 dB, 10 dB, and 8 dB, respectively. z The relative stiffness changes. Figure 8 Robustness verification results for state degradation identification under strong noise conditions are presented. Figure 8 (a) The noisy z-direction displacement response and the smoothed z-direction displacement response were compared to illustrate that the smoothing process can reduce the influence of measurement noise while preserving the main trend of displacement response. Figure 8(b) A comparison of the baseline regularized reconstruction channel, self-stiffness residual projection channel, pressure-displacement compliance channel, and three-channel fusion results and set values is presented to verify the consistency of different channels in identifying the main degradation direction and the accuracy of the fusion results. Figure 8 (c) The z-direction relative stiffness change identification results are given for displacement response signal-to-noise ratios of 30 dB, 20 dB, 10 dB, and 8 dB, respectively, to evaluate the identification stability of the present invention under different noise levels. The results show that within the above noise ranges, the z-direction relative stiffness change identification values are stably distributed at approximately [value missing]. 0.203 to Within the range of 0.194, all values are close to the set value. The value is 0.200, and the z-axis can be correctly identified as the main degradation direction. This shows that the present invention can convert pressure pulsation input and top cover displacement response into a three-dimensional relative stiffness change index under strong noise operation monitoring data conditions, identify the main degradation direction and degree, and provide a quantitative basis for operation anomaly identification, status trend tracking and maintenance decision-making.
[0044] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The embodiments and features described in these embodiments can be arbitrarily combined without conflict. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.
Claims
1. A robust reconstruction and state degradation assessment method for a turbine top cover dynamic system, characterized in that: Includes the following steps: S1: Collect displacement response, pressure pulsation signal and operating parameters of the turbine top cover (1) during operation, and perform time synchronization, outlier removal, trend term removal, filtering and steady-state segment screening on the collected data to form a monitoring dataset for dynamic reconstruction. S2: Construct the original pressure pulsation measurement data into an equivalent pressure input corresponding to the three-dimensional displacement of the top cover; S3: Perform time delay correction on the equivalent pressure input, and determine the optimal time delay after correction based on the criterion of minimizing the weak form dynamic residual; S4: After completing the pressure input time delay correction, a smooth test function is introduced to construct weak form integral constraints within multiple time windows and establish weak form matrix equations for each direction of the top cover. S5: Select stable operating data of the top cover under healthy or baseline conditions, construct weak form matrix equations, use regularized least squares method to solve for healthy baseline dynamic coefficients, and establish a healthy baseline equivalent dynamic model. S6: For the current operating state data, the same data processing method as the health baseline state is used to construct the weak form matrix equation of the current state, and the current state dynamic coefficients are reconstructed by baseline regularization with reference to the health baseline dynamic coefficients to obtain the low-dimensional equivalent dynamic model of the current state. S7: Calculate the relative stiffness change estimate of the benchmark regularized reconstruction channel based on the self-stiffness correlation coefficient in the health benchmark model and the current state model; S8: Calculate the weak-form residual of the current state relative to the health baseline model, obtain the change in self-stiffness coefficient through residual projection, and calculate the relative stiffness change estimate of the residual projection channel. S9: Calculate the pressure-displacement compliance of the health baseline state and the current state in multiple target frequency bands, and obtain the relative stiffness change estimate of the compliance channel from the compliance ratio; S10: The relative stiffness changes obtained from the benchmark regularization reconstruction channel, the self-stiffness residual projection channel, and the pressure-displacement compliance channel are weighted and fused according to the fusion weight determined based on the consistency of the results of the three channels to obtain the three-dimensional relative stiffness change vector of the top cover, and the overall degradation degree and main degradation direction are determined accordingly.
2. The robust reconstruction and state degradation assessment method for the turbine top cover dynamic system according to claim 1, characterized in that: The displacement response described in S1 is the equivalent displacement response of the top cover: ; in, u x ( t ), u y ( t ), u z ( t ) respectively indicate the top cover at x direction, y direction and z The equivalent displacement response in the direction, the pressure pulsation measurement point data is denoted as: ; in, m This represents the number of pressure pulsation measurement points. p 1( t ), p 2( t ),…, p m ( t The parameters are pressure pulsation signals at different locations; the operating parameters include unit load, head, guide vane opening and speed.
3. The robust reconstruction and state degradation assessment method for the turbine top cover dynamic system according to claim 1, characterized in that: The equivalent pressure input described in S2 is: ; in, Construct operators for pressure input; The equivalent pressure input after construction is denoted as: ; in, p e,x ( t ), p e,y ( t ), p e,z ( t ) respectively with the top cover x , y , z The pressure input corresponding to the directional equivalent displacement response.
4. The robust reconstruction and state degradation assessment method for the turbine top cover dynamic system according to claim 1, characterized in that: S3 describes the first j In each direction, the prior engineering delay is assumed to be... The allowed search scope is The candidate delay range is: ; In candidate delay Next, the j The time delay correction pressure input in each direction is: ; The optimal time delay after correction is determined based on the criterion of minimizing the weak-form dynamic residual: ; in, For the first j Each direction in candidate latency The residuals are calculated from the weak form dynamics matrix. This is the prior time delay constraint coefficient for the project.
5. The robust reconstruction and state degradation assessment method for the turbine top cover dynamic system according to claim 1, characterized in that: The top cover described in S4 j The low-dimensional equivalent dynamic relationship in each direction is as follows: ; in, For the first j Equivalent acceleration response in each direction, As a candidate function library, These are the dynamic coefficients to be identified; The candidate function libraries are: ; in, Used to characterize damping-related terms, Used to characterize nonlinear stiffness terms, Used to characterize the pressure input term after time delay correction. Used to characterize pressure-displacement coupling terms; Introducing a smoothness test function During the time window Internally constructed weak formal integral constraints: ; in, For the first q One test function; No. j The weak form matrix equations for each direction are: ; in, This is the weakly form inertial term obtained by integrating the second derivative of the displacement response and the test function. This is the weakly form candidate matrix obtained by integrating the candidate function library and the test function. This is the residual term.
6. The robust reconstruction and state degradation assessment method for the turbine top cover dynamic system according to claim 1, characterized in that: The first mentioned in S5 j Weak-form matrix equations of health status in each direction: ; The superscript 0 indicates the baseline health status. For the first time in a healthy state j Weak-form candidate matrices in each direction, To correspond to the weak form of the inertia term, The health baseline kinetic coefficients to be solved are... For residual terms; The health baseline kinetic coefficients are solved using the regularized least squares method: ; in, The regularization coefficient is used. The equivalent kinetic model of the health benchmark is: 。 7. The robust reconstruction and state degradation assessment method for the turbine top cover dynamic system according to claim 1, characterized in that: The first mentioned in S6 j The weak form matrix equations of the current state in each direction are: ; Among them, superscript c Indicates the current state. For the current state, the first j Weak-form candidate matrices in each direction, For the corresponding weak form inertia term, These are the dynamic coefficients to be solved under the current state. For residual terms; Based on health baseline kinetic coefficients For reference, the current state dynamic coefficients are reconstructed using benchmark regularization: ; in, As a health baseline constraint coefficient, The coefficient constraint weight matrix is defined; the first constraint weight is set for the damping term coefficient, pressure input term coefficient, pressure-displacement coupling term coefficient, and inter-direction coupling term coefficient; for the... j The first constraint weight is greater than the second constraint weight, which is set as the correlation coefficient of self-stiffness in each direction.
8. The robust reconstruction and state degradation assessment method for the turbine top cover dynamic system according to claim 1, characterized in that: The equivalent self-stiffness in the j-th direction described in S7 and the current state are respectively: ; in, The first under the health baseline state j Equivalent self-stiffness in each direction, For the current state, the first j Equivalent self-stiffness in each direction, and These are the self-stiffness correlation coefficients in the healthy baseline model and the current state model, respectively; the relative stiffness change estimate of the baseline regularized reconstruction channel is: ; in, The relative stiffness change obtained by reconstructing the channel using the baseline regularization. To prevent positive numbers with excessively small denominators.
9. The robust reconstruction and state degradation assessment method for the turbine top cover dynamic system according to claim 1, characterized in that: The weak-form residual of the current state relative to the health baseline model, as described in S8, is: ; in, This indicates dynamic differences in the current state data that cannot be explained by the health baseline model; For the current state, the first j Weakly formal inertial terms in each direction, This is the weak-form candidate matrix for the current state. Let be the kinetic coefficients for health baseline; let The weak form candidate matrix for the current state Correlation coefficient between medium and self-stiffness The corresponding column vector, then the change in self-stiffness coefficient is: ; in, Here is the projection regularization coefficient; the relative stiffness change estimate for the residual projection channel is: ; in, This represents the relative stiffness change obtained from the stiffness residual projection channel.
10. The robust reconstruction and state degradation assessment method for the turbine top cover dynamic system according to claim 1, characterized in that: The first mentioned in S9 j The direction, in the first b Within a target frequency band, the pressure-displacement compliance is defined as: ; in, For the first b A target frequency band filtering operator Represents the root mean square value. The equivalent pressure input is after time delay correction; the relative stiffness change of the compliance channel is estimated as follows: ; in, This is an estimate of the relative stiffness change obtained from the pressure-displacement compliance channel. This indicates taking the median value of the results for multiple target frequency bands.
11. The robust reconstruction and state degradation assessment method for the turbine top cover dynamic system according to claim 1, characterized in that: The first mentioned in S10 j The final relative stiffness change index in each direction is: ; in, No. j The final relative stiffness change index in each direction, These are the fusion weights for the three channels, satisfying the following: The three-dimensional relative stiffness variation vector of the top cover is: The overall degradation level index is: The main direction of degradation is: ,in, D Indicates the degree of degradation of the top cover. Indicates the main direction of degradation.
