A spectrum self-adaptive weighting wind tunnel model vibration monitoring method
By using a spectrum adaptive weighting method, the problems of model dependence, insufficient sensitivity, and early warning delay in vibration monitoring during wind tunnel tests are solved. This enables real-time and accurate vibration status monitoring and early warning for different aircraft models, providing an adaptive and intelligent early warning mechanism.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-22
- Publication Date
- 2026-03-24
AI Technical Summary
Existing vibration monitoring methods in wind tunnel tests suffer from problems such as strong model dependence, insufficient sensitivity, lack of adaptability, and early warning delay, making it difficult to achieve real-time, accurate monitoring and early warning for different aircraft models.
By employing a spectrum adaptive weighting method, through data acquisition and preprocessing, spectrum standardization, constructing an objective function to improve the Fisher discriminant ratio, solving for the optimal weight vector and adaptive threshold warning, the method automatically learns and weights the spectrum features most sensitive to vibration state changes, constructs a health index (HI), and achieves early and accurate warning of vibration state.
It enables adaptive monitoring of different aircraft models, improves the sensitivity and accuracy of early warning of vibration conditions, reduces the need for human intervention, and provides clear fault tracing analysis capabilities and timely early warning mechanisms.
Smart Images
Figure CN121388634B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of structural health monitoring and fault diagnosis technology, and in particular to a method for monitoring and early warning of vibration status for different aircraft models in a wind tunnel testing environment. Background Technology
[0002] Wind tunnel testing is a crucial step in the development of aircraft, used to verify their aerodynamic performance and structural stability. During testing, aircraft models may experience severe vibrations such as flutter and buffeting under the influence of high-speed airflow. These vibrations can range from affecting the accuracy of test data to causing damage to the model or even destroying the wind tunnel equipment. Therefore, real-time and accurate monitoring of the model's vibration status and timely early warning are essential.
[0003] Existing vibration monitoring methods are mostly based on single time-domain indicators (such as root mean square (RMS) or kurtosis) or frequency-domain indicators (such as characteristic frequency amplitude). These methods have significant limitations:
[0004] 1. Model Dependency: Different aircraft models have vastly different natural frequencies, mode shapes, and vibration response levels. Thresholds set for a specific model cannot be directly applied to other models, requiring extensive prior knowledge and debugging work.
[0005] 2. Insufficient sensitivity: Indicators such as RMS are not sensitive to early and weak abnormal vibrations; while kurtosis, although sensitive to shock, is easily affected by noise and its trend is not monotonic, making it difficult to use for condition assessment.
[0006] 3. Lack of adaptability: Traditional methods cannot automatically identify the characteristic frequencies of the current model's vibration state, requiring manual intervention in spectrum analysis, which cannot meet the experimental needs of rapid model switching.
[0007] 4. Warning delay: Alarm methods based on absolute thresholds are often triggered only after the vibration has developed to a certain extent, resulting in a short warning time. Summary of the Invention
[0008] The purpose of this invention is to provide a vibration state monitoring method and system that is highly versatile, highly sensitive, and adaptable to different aircraft models, addressing the aforementioned shortcomings. This method aims to automatically learn and weight the spectral features most sensitive to changes in vibration state, construct a stable and sensitive health index (HI), and achieve early and accurate warning of model vibration state in wind tunnel tests.
[0009] This invention is achieved through the following scheme:
[0010] A method for vibration monitoring using a spectrum-adaptive weighted wind tunnel model includes the following steps:
[0011] S1, Data Acquisition and Preprocessing: Vibration signals are acquired through the acquisition unit within the wind tunnel test model, and the vibration signals are processed to obtain a... 1D spectrum matrix ,in For frequency points, Number of time segments;
[0012] S2, Spectrum Standardization: To eliminate the influence of differences in absolute amplitude between different aircraft models, the spectrum matrix is standardized.
[0013] S3, construct an objective function based on the improved Fisher discriminant ratio, and define a frequency weight vector;
[0014] S4. Solve for the optimal weight vector by using an iterative optimization algorithm to find the maximum value of the objective function.
[0015] S5, Health Indicators Calculation and state assessment, optimal weights With real-time acquired and standardized spectrum Multiply them to get real-time health indicator values. ;
[0016] S6, adaptive threshold warning, based on historical normal conditions. Value, calculator mean and standard deviation Set dynamic warning thresholds and gradient thresholds; when When a value exceeds the warning threshold or its gradient exceeds the gradient threshold, the system triggers warning signals of different levels.
[0017] In S1, specifically, vibration signals are collected by accelerometers installed inside the wind tunnel test model. After anti-aliasing filtering and analog-to-digital conversion, a discrete-time series signal is obtained. The signal is then segmented. A fast Fourier transform is performed on each segment to calculate its amplitude spectrum, resulting in a... 1D spectrum matrix ,in For frequency points, The number of time segments, where the number of frequency points is... It should conform to the effective frequency range under the Nyquist sampling theorem.
[0018] Specifically, in S2, the mean value of each frequency point under the normal state sample is calculated. and standard deviation The standardized formula is:
[0019]
[0020] in, The value is a very small integer to avoid division by zero errors; this step unifies the spectra of different models to a similar numerical distribution range.
[0021] In S3, specifically, a set of weights is found. This makes the weighted index It can best distinguish between the two states of "stable" and "dangerous"; objective function Designed as follows:
[0022]
[0023] The first Fisher criterion: maximizes inter-class divergence and minimizes intra-class divergence;
[0024] The inter-class scatter matrix, and These are the mean vectors of the normalized spectra for the stable and dangerous states, respectively.
[0025] The scatter matrix is the intra-class scatter matrix.
[0026] The second term, bimodal entropy regularization, is used to constrain the weight distribution.
[0027] It is the steady-state entropy, where This encourages a relatively dispersed weight distribution to avoid overfitting individual features of the current model;
[0028] It is the entropy of the dangerous state, and its physical meaning is The sum of sensitivity to weight changes under dangerous conditions; this term encourages weights to concentrate on frequencies where the amplitude changes significantly under dangerous conditions. It is the regularization coefficient. It is a balance parameter used for adjustment. and The weights between them.
[0029] In S4, the specific steps include:
[0030] S41, Solve for the initial Fisher direction This is equivalent to the generalized eigenvalue problem. ;
[0031] S42, using gradient ascent iteration Optimize until the objective function converges; the final optimal weight vector is obtained. It intuitively reflects the contribution of each frequency component to the change in vibration state; the larger the absolute value of the weight, the more critical the frequency.
[0032] Gradient components: The former is the analytic term, and the latter is the entropy gradient.
[0033] The Fisher gradient is:
[0034] The entropy gradient is: in .
[0035] In S5, this The value integrates all frequency information and is sensitive to abnormal states. Through analysis... By analyzing the time series trend of the values and their gradient, the vibration state can be assessed.
[0036] In S5, specifically, according to The criteria are based on historical normal conditions. Value, calculator mean and standard deviation Set dynamic early warning thresholds: Simultaneously, calculation Real-time changing gradient, and setting gradient threshold.
[0037] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:
[0038] 1. Strong model adaptation capability: Through spectrum normalization and innovative dual-modal entropy regularization design, the method can automatically adapt to the dynamic characteristics of different aircraft without having to reset complex thresholds for each model, making it highly versatile.
[0039] 2. Early Warning and High Sensitivity: The objective function based on improved Fisher discriminant analysis can automatically focus on the "critical frequency band" most sensitive to state changes, enabling the constructed... It is exceptionally sensitive to early-stage minor faults, overcoming the shortcomings of insufficient sensitivity in traditional time-domain indicators.
[0040] 3. Clear physical meaning: optimal weight vector It provides a clear physical explanation, allowing engineers to directly understand which frequency components are affected by changes in state transitions by observing the distribution of weights, which helps in fault tracing and analysis.
[0041] 4. A dual early warning mechanism, combining... The dual criteria of dynamic threshold and changing gradient can monitor both steady-state shifts and sudden changes in state, providing more timely and reliable early warnings.
[0042] 5. Automation and intelligence: The entire process, from feature extraction and weight optimization to threshold setting, can be completed automatically, reducing reliance on expert experience and manual intervention, making it suitable for engineering applications. Attached Figure Description
[0043] Figure 1 This is an overall flowchart of the method of the present invention;
[0044] Figure 2 This is a complete 90-second vibration data graph from Example 2;
[0045] Figure 3 This is a graph showing the monitoring of kurtosis and Gini coefficient values using common methods in Example 2;
[0046] Figure 4 The optimal weight in Example 2 A schematic diagram;
[0047] Figure 5 For the training trains in Example 2 Monitoring map;
[0048] Figures 6-7 The monitoring results are for different train routes in Example 2. Detailed Implementation
[0049] All features disclosed in this specification, or steps in all methods or processes disclosed herein, may be combined in any way, except for mutually exclusive features and / or steps.
[0050] Any feature disclosed in this specification (including any appended claims and abstract) may be replaced by other equivalent or similar features, unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is merely one example of a series of equivalent or similar features.
[0051] In the description of this invention, it should be understood that the terms "upper," "lower," "left," "right," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a predetermined orientation, or be constructed and operated in a predetermined orientation. Therefore, they should not be construed as limitations on this invention.
[0052] Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature.
[0053] Example 1
[0054] like Figure 1 As shown, the present invention provides a technical solution:
[0055] A method for vibration monitoring using a spectrum-adaptive weighted wind tunnel model includes the following steps:
[0056] S1: Data Acquisition and Preprocessing. Vibration signals are collected using accelerometers installed in the wind tunnel test model. After anti-aliasing filtering and analog-to-digital conversion, a discrete-time series signal is obtained. The signal is then segmented (e.g., each segment is set to 1 second). A Fast Fourier Transform (FFT) is performed on each segment to calculate its amplitude spectrum, resulting in a... 1D spectrum matrix ,in For frequency points, The number of time segments, where the number of frequency points is... It should conform to the effective frequency range under the Nyquist sampling theorem.
[0057] S2: Spectrum Standardization (Model Adaptation) – To eliminate the influence of absolute amplitude differences between different aircraft models, row standardization is performed on the spectrum matrix. The mean value of each frequency point under normal state samples is calculated. and standard deviation The standardized formula is:
[0058]
[0059] in, This should be a very small integer to avoid division by zero errors. This step unifies the spectra of different models to a similar numerical distribution range.
[0060] S3: Construct an objective function based on the improved Fisher discriminant ratio and define a frequency weight vector. The core idea is to find a set of weights. This makes the weighted index It can best distinguish between the two states of "stable" and "dangerous"; objective function Designed as follows:
[0061]
[0062] The first term (Fisher's criterion): maximizes inter-class divergence and minimizes intra-class divergence.
[0063] The inter-class scatter matrix, and These are the mean vectors of the normalized spectra for the stable and dangerous states, respectively.
[0064] It is the intra-class scatter matrix.
[0065] The second term (bimodal entropy regularization term): This is the core innovation of this invention, used to constrain the weight distribution, making it both generalizable and sensitive.
[0066] It is the steady-state entropy, where This encourages a relatively dispersed weight distribution to avoid overfitting individual features of the current model and to enhance the generalization ability of the method.
[0067] It is the entropy of the dangerous state, and its physical meaning is The sum of sensitivity to weight changes under hazardous conditions. This encourages the concentration of weights on frequencies where amplitudes change significantly under hazardous conditions, thereby improving monitoring sensitivity. It is the regularization coefficient. It is a balance parameter used for adjustment. and The weights between them.
[0068] S4: Find the optimal weight vector by using an iterative optimization algorithm to find the maximum value of the objective function. The steps are as follows.
[0069] Step 1: Solve for the initial Fisher orientation This is equivalent to the generalized eigenvalue problem.
[0070] Step 2: Use gradient ascent iteration Optimize until the objective function converges. The final optimal weight vector is obtained. It intuitively reflects the contribution of each frequency component to the change in vibration state; the larger the absolute value of the weight, the more critical the frequency.
[0071] Gradient components: The former is the analytic term, and the latter is the entropy gradient.
[0072] The Fisher gradient is:
[0073] The entropy gradient is: in
[0074] S5: Health Indicators ) Calculation and state evaluation, assigning optimal weights With real-time acquired and standardized spectrum Multiply them to get real-time health indicator values. .Should The value integrates all frequency information and is sensitive to abnormal states. Through analysis... By analyzing the time series trend of the values and their gradient, the vibration state can be assessed.
[0075] S6: Adaptive threshold warning, based on... The criteria are based on historical normal conditions. Value, calculator mean and standard deviation Set dynamic early warning thresholds: At the same time, calculation Real-time changing gradient, setting a gradient threshold. When When a value exceeds an absolute value threshold or its gradient exceeds a gradient threshold, the system triggers different levels of warning signals.
[0076] Example 2
[0077] Taking a wind tunnel test of a certain type of aircraft model as an example, data from three runs were collected. One dangerous run was used as training (hereinafter referred to as the first run), and the other two runs (one normal operation and one dangerous run) were used for actual testing. The sampling frequency for all three runs was set to 300Hz, and 150 points per second were obtained after FFT. The amplitude spectrum of the training train was measured. The sampling time for the training train was approximately 90 seconds. The first 70 seconds of the test were in a stable state, and the last 20 seconds were in a dangerous state. At approximately 78 seconds, a strong vibration occurred, and the train was shut down urgently.
[0078] S1: Acquire complete 90 seconds of raw vibration data, and obtain a 150*90 spectrum matrix after FFT. The complete 90 seconds of vibration data is shown below. Figure 2 As shown.
[0079] Vibration was monitored per second using common methods such as RMS value, kurtosis value, and Gini coefficient value. The monitoring results are as follows: Figure 3 As shown, it can be seen that commonly used methods are ineffective for vibration monitoring and early warning.
[0080] S2: Calculate the average value of each frequency point based on the data from the first 70 seconds. and standard deviation And the entire spectrum matrix is standardized.
[0081] S3: Construct an objective function based on the improved Fisher discriminant ratio; find a set of weights. This makes the weighted index It can best distinguish between the two states of "stable" and "dangerous". Objective function Designed as follows:
[0082] ;
[0083] S4: Set regularization parameters Entropy balance parameter Construct the objective function as described in the specification and solve for the optimal weights. Optimal weight like Figure 4 As shown.
[0084] S5: Optimal weights Multiply by the real-time spectrum data to obtain the real-time... value.
[0085] S6: Calculate the first 70 seconds (steady-state time) The mean and standard deviation, according to The method sets a threshold. The monitoring system is run, and when... When a value exceeds a threshold or its gradient increases abnormally, the system issues an audible and visual alarm to alert the operator or automatically triggers the train shutdown protection mechanism. (Training train number) Monitoring chart Figure 5 As shown.
[0086] Save the optimal weights of the accelerometer. The corresponding steps are then implemented for the next vehicle, thus enabling online monitoring of different models and different vehicle trips. It is worth noting that in step S5 of the new model, only a small amount of steady-state time (e.g., 30 seconds) needs to be collected. The threshold is set using the mean and standard deviation. Monitoring results for different train routes are as follows: Figures 6-7 As shown.
[0087] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for vibration monitoring using a spectrum-adaptive weighted wind tunnel model, characterized by the following steps: S1, Data Acquisition and Preprocessing: Vibration signals are acquired through the acquisition unit within the wind tunnel test model, and the vibration signals are processed to obtain a... 1D spectrum matrix ,in For frequency points, Number of time segments; S2, Spectrum Standardization: To eliminate the influence of differences in absolute amplitude between different aircraft models, the spectrum matrix is standardized. S3, construct the objective function based on the improved Fisher discriminant ratio, and define the frequency weight vector; specifically in S3, find a set of weights. This makes the weighted index It can best distinguish between the two states of "stable" and "dangerous"; objective function Designed as follows: The first Fisher criterion: maximizes inter-class divergence and minimizes intra-class divergence; The inter-class scatter matrix, and These are the mean vectors of the normalized spectra for the stable and dangerous states, respectively. The scatter matrix is the intra-class scatter matrix. The second term, bimodal entropy regularization, is used to constrain the weight distribution. It is the steady-state entropy, where This encourages a relatively dispersed weight distribution to avoid overfitting individual features of the current model. It is the entropy of the dangerous state, and its physical meaning is The sum of sensitivity to weight changes under dangerous conditions; this term encourages weights to concentrate on frequencies where the amplitude changes significantly under dangerous conditions. It is the regularization coefficient. It is a balance parameter used for adjustment. and The weights between them; S4, Solve for the optimal weight vector, and use an iterative optimization algorithm to find the maximum value of the above objective function; S5, Health Indicators Calculation and state assessment, optimal weights With real-time acquired and standardized spectrum Multiply them to get real-time health indicator values. ; S6, adaptive threshold alert, based on historical normal conditions. Values, calculate their mean and standard deviation Set dynamic warning thresholds and gradient thresholds; when When a value exceeds the warning threshold or its gradient exceeds the gradient threshold, the system triggers warning signals of different levels.
2. The method for vibration monitoring using a spectrum-adaptive weighted wind tunnel model as described in claim 1, characterized in that: In S1, specifically, vibration signals are collected by accelerometers installed inside the wind tunnel test model. After anti-aliasing filtering and analog-to-digital conversion, a discrete-time series signal is obtained. The signal is then segmented. A fast Fourier transform is performed on each segment to calculate its amplitude spectrum, resulting in a... 1D spectrum matrix ,in For frequency points, The number of time segments, where the number of frequency points is... It should conform to the effective frequency range under the Nyquist sampling theorem.
3. The method for vibration monitoring using a spectrum-adaptive weighted wind tunnel model as described in claim 1, characterized in that: In S2, specifically, the mean value of each frequency point under the normal state sample is calculated. and standard deviation The standardized formula is: in, The value is a very small integer to avoid division by zero errors; this step unifies the spectra of different models to a similar numerical distribution range.
4. The method for vibration monitoring using a spectrum-adaptive weighted wind tunnel model as described in claim 1, characterized in that: In S4, the specific steps include: S41, Solve for the initial Fisher direction This is equivalent to the generalized eigenvalue problem. ; S42, using gradient ascent iteration Optimize until the objective function converges; the final optimal weight vector is obtained. It intuitively reflects the contribution of each frequency component to the change in vibration state; the larger the absolute value of the weight, the more critical the frequency. Gradient components: The former is the analytical term, and the latter is the entropy gradient. The Fisher gradient is: ; The entropy gradient is: ;in .
5. The method for vibration monitoring of a spectrum-adaptive weighted wind tunnel model as described in claim 1, characterized in that: In S5, this The value integrates all frequency information and is sensitive to abnormal states. Through analysis... By analyzing the time series trend of the values and their gradient, the vibration state can be assessed.
6. The method for vibration monitoring of a spectrum-adaptive weighted wind tunnel model as described in claim 1, characterized in that: In S5, specifically, according to The criteria are based on historical normal conditions. Value, calculator mean and standard deviation Set dynamic early warning thresholds: Simultaneously, calculation Real-time changing gradient, and setting gradient threshold.