Wind tunnel health monitoring data mining method, medium and system

Abstract

The invention provides a wind tunnel health monitoring data mining method, medium and system, and belongs to the technical field of wind tunnel health monitoring data mining and processing, and the method comprises the steps: collecting a plurality of sensing signals to construct a monitoring data stream, and carrying out the filtering and noise reduction processing to obtain preprocessed data and an abnormal data set; and starting a GPU and CPU processing stream to receive the preprocessed data and the abnormal data set, performing component decomposition to obtain a steady-state data matrix, a dynamic data matrix, an abnormal feature matrix, a parameter data vector and an abnormal parameter vector, establishing a data mining feature equation set, and calculating a wind tunnel operation state evaluation result by using the wind tunnel operation state evaluation equation set. And generating a wind tunnel operation characteristic matrix and a wind tunnel state early warning index, and outputting an early warning signal according to a comparison result of the early warning index and a preset health threshold value. The technical problem that potential fault symptoms cannot be better found due to the fact that wind tunnel health detection data cannot be rapidly and efficiently mined and analyzed from the two aspects of steady state and dynamic state at present is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind tunnel health monitoring data mining and processing, and in particular, relates to a wind tunnel health monitoring data mining method, medium and system. Background Art

[0002] With the rapid development of aerospace technology, wind tunnel tests play an irreplaceable role in aircraft development, aerodynamic shape optimization and other fields. As a large-scale precision experimental equipment, the operating status of the wind tunnel directly affects the accuracy and reliability of the test data. Traditional wind tunnel monitoring methods mainly rely on the threshold judgment of a single sensor data, which cannot fully reflect the health status of the wind tunnel system. At present, wind tunnel health monitoring mainly adopts the following technical means: The first type is the fault diagnosis method based on expert experience. This method establishes an expert knowledge base and combines historical operation data to formulate fault judgment rules. However, due to the complexity of the wind tunnel system, it is difficult to cope with the changing operating conditions and cannot detect potential faults in time by relying solely on empirical rules.

[0003] The second type is the single parameter monitoring method. By placing temperature, pressure, vibration and other sensors at key locations, the equipment operating parameters are monitored in real time. This method is simple to operate, but there is an information island phenomenon, and the correlation between parameters cannot be explored, resulting in one-sided monitoring results.

[0004] The third type is the statistical analysis method. Statistical methods are used to analyze historical data and establish a parameter change law model. However, this type of method has high requirements on data quality and is difficult to adapt to the dynamic change characteristics of wind tunnel test conditions.

[0005] Since wind tunnel monitoring data requires extremely high real-time performance and has a huge amount of data, the above three methods all have the technical problem of being unable to quickly and efficiently mine and analyze wind tunnel health detection data from both steady-state and dynamic aspects, resulting in the inability to better discover potential fault signs. Summary of the invention

[0006] In view of this, the present invention provides a method, medium and system for mining wind tunnel health monitoring data, which can solve the technical problem that the prior art cannot quickly and efficiently mine and analyze wind tunnel health detection data from both steady-state and dynamic aspects, resulting in the inability to better discover potential fault signs.

[0007] The present invention is achieved in that: A first aspect of the present invention provides a method for mining wind tunnel health monitoring data, comprising collecting multiple sensor signals to construct a monitoring data stream, filtering and denoising the monitoring data stream to obtain preprocessed data and an abnormal data set, starting a GPU and a CPU processing stream to receive the preprocessed data and the abnormal data set, performing component decomposition to obtain a steady-state data matrix, a dynamic data matrix, an abnormal feature matrix, a parameter data vector and an abnormal parameter vector, establishing a data mining characteristic equation group based on the steady-state data matrix, the dynamic data matrix and the abnormal feature matrix, calculating a wind tunnel operation status evaluation result using a wind tunnel operation status evaluation equation group, generating a wind tunnel operation characteristic matrix and a wind tunnel state warning indicator, and outputting a warning signal according to a comparison result between the wind tunnel state warning indicator and a preset health threshold.

[0008] On the basis of the above technical solution, the wind tunnel health monitoring data mining method of the present invention can also be improved as follows: Among them, the multiple sensor signals include temperature sensor signals, vibration sensor signals, pressure sensor signals, sound sensor signals, current sensor signals, speed sensor signals, power sensor signals, flow sensor signals and noise sensor signals.

[0009] Furthermore, the CUDA parallel processing stream of the GPU receives the pre-processed data; and the input stream of the CPU receives the abnormal data set.

[0010] Furthermore, the preprocessed data is decomposed into components to obtain the steady-state data matrix, the dynamic data matrix and the parameter data vector; the abnormal data set is decomposed into components to obtain the abnormal feature matrix and the abnormal parameter vector.

[0011] Furthermore, the wind tunnel operation state evaluation equation group includes a state characteristic equation, a dynamic correlation equation, a steady-state mapping equation and an abnormal influence equation.

[0012] Furthermore, the input of the state characteristic equation includes the steady-state data matrix, the dynamic data matrix and the temperature sensor signal and the pressure sensor signal in the parameter data vector, and the output is the operating state characteristic vector in the wind tunnel operating state evaluation result; the input of the dynamic correlation equation includes the dynamic data matrix, the abnormal characteristic matrix and the speed sensor signal and the flow sensor signal in the parameter data vector, and the output is the operating parameter dynamic characteristic vector in the wind tunnel operating state evaluation result.

[0013] Furthermore, the input of the steady-state mapping equation includes the power sensor signal and the current sensor signal in the steady-state data matrix and the parameter data vector, and the output is the steady-state characteristic vector of the operating parameters in the wind tunnel operating status evaluation result; the input of the abnormal influence equation includes the vibration sensor signal and the noise sensor signal in the abnormal characteristic matrix and the abnormal parameter vector, and the output is the abnormal influence characteristic vector in the wind tunnel operating status evaluation result.

[0014] Furthermore, the wind tunnel operation characteristic data is calculated by the wind tunnel operation status evaluation equation group to obtain the wind tunnel operation status evaluation result; the wind tunnel operation status evaluation result is used to generate the wind tunnel operation characteristic matrix and the wind tunnel status early warning index.

[0015] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned method for mining wind tunnel health monitoring data.

[0016] A third aspect of the present invention provides a wind tunnel health monitoring data mining system, which includes the above-mentioned computer-readable storage medium.

[0017] Compared with the prior art, the wind tunnel health monitoring data mining method, medium and system provided by the present invention have the following beneficial effects: First, the parallel processing architecture of GPU and CPU enables efficient processing of massive monitoring data. CUDA parallel computing technology is used to process pre-processed data, while the CPU is used to process abnormal data sets, which improves data processing efficiency and lays the foundation for real-time monitoring.

[0018] Secondly, an innovative component decomposition method is proposed to decompose the monitoring data into a steady-state data matrix, a dynamic data matrix and a parameter data vector, thus achieving a multi-dimensional characterization of the wind tunnel operation status. This decomposition method can effectively extract data features and provide a reliable basis for subsequent status evaluation.

[0019] Third, a complete set of wind tunnel operation status evaluation equations was constructed, including state characteristic equations, dynamic correlation equations, steady-state mapping equations, and abnormal influence equations. These equations describe the wind tunnel operation characteristics from different perspectives and achieve a comprehensive evaluation of the system status.

[0020] Fourth, an early warning mechanism based on multi-source data was established, which realized the timely discovery and early warning of abnormal conditions through the wind tunnel operation characteristic matrix and status early warning indicators. The early warning mechanism took into account the characteristics of various sensor signals and their correlation, and improved the accuracy of early warning.

[0021] In summary, the present invention solves the technical problem that the prior art cannot quickly and efficiently mine and analyze wind tunnel health detection data from both steady-state and dynamic aspects, resulting in an inability to better discover potential fault signs. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flow chart of the method of the present invention; Figure 2 It is the characteristic distribution diagram of the steady-state data matrix in Example 2; Figure 3 This is a diagram showing how the early warning indicator H changes over time in Example 2. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0024] like Figure 1 As shown, it is a flow chart of a method for mining wind tunnel health monitoring data provided by the present invention. The method comprises the following steps: S10, collecting temperature sensor signals, vibration sensor signals, pressure sensor signals, sound sensor signals, current sensor signals, speed sensor signals, power sensor signals, flow sensor signals and noise sensor signals during the operation of the wind tunnel, and constructing a monitoring data stream; S20, filtering and denoising the monitoring data stream to obtain preprocessed data, and excluding abnormal band data in the preprocessed data during the filtering and denoising process, and recording them as abnormal data sets; S30, starting the CUDA parallel processing flow of the GPU to receive the pre-processed data; starting the input flow of the CPU to receive the abnormal data set; S40, performing component decomposition on the preprocessed data to obtain a steady-state data matrix, a dynamic data matrix and a parameter data vector; performing component decomposition on the abnormal data set to obtain an abnormal feature matrix and an abnormal parameter vector; S50, establishing a data mining characteristic equation group based on the steady-state data matrix, the dynamic data matrix and the abnormal characteristic matrix to obtain wind tunnel operation characteristic data; S60, using the wind tunnel operation characteristic data, calculating a wind tunnel operation status evaluation result based on a wind tunnel operation status evaluation equation group; S70, generating a wind tunnel operation characteristic matrix and a wind tunnel state warning index according to the wind tunnel operation state evaluation result; S80: Outputting a warning signal according to a comparison result between the wind tunnel status warning indicator and a preset health threshold.

[0025] The wind tunnel operation state evaluation equation group includes state characteristic equation, dynamic correlation equation, steady-state mapping equation and abnormal influence equation: The state characteristic equation is used to establish the correlation between wind tunnel operation parameters, the input includes the steady-state data matrix, the dynamic data matrix and the temperature sensor signal and the pressure sensor signal in the parameter data vector, and the output is the operation state characteristic vector in the wind tunnel operation state evaluation result; The dynamic correlation equation is used to analyze the dynamic change law of the wind tunnel operation parameters, and the input includes the dynamic data matrix, the abnormal feature matrix and the speed sensor signal and the flow sensor signal in the parameter data vector, and the output is the dynamic feature vector of the operation parameters in the wind tunnel operation status evaluation result; The steady-state mapping equation is used to establish a steady-state characteristic mapping of the wind tunnel operating parameters, the input includes the steady-state data matrix and the power sensor signal and the current sensor signal in the parameter data vector, and the output is the steady-state characteristic vector of the operating parameters in the wind tunnel operating status evaluation result; The abnormal influence equation is used to evaluate the influence of abnormal data on the wind tunnel operation state. The input includes the vibration sensor signal and the noise sensor signal in the abnormal feature matrix and the abnormal parameter vector. The output is the abnormal influence feature vector in the wind tunnel operation state evaluation result.

[0026] The specific implementation methods of the above steps are described in detail below: The specific implementation method of step S10 is: first, through a series of sensors installed at key positions of the wind tunnel, such as temperature sensors, vibration sensors, pressure sensors, sound sensors, current sensors, speed sensors, power sensors, flow sensors and noise sensors, various monitoring data during the operation of the wind tunnel are collected in real time. These monitoring data include temperature sensor signals, vibration sensor signals, pressure sensor signals, sound sensor signals, current sensor signals, speed sensor signals, power sensor signals, flow sensor signals and noise sensor signals, which constitute a comprehensive monitoring data stream. The mathematical models of these sensor signals, as described in the claims, include various typical periodic changes, attenuation, accumulation and other characteristics. By collecting these signal data, the operation status of the wind tunnel can be fully reflected.

[0027] The specific implementation method of step S20 is: filtering and denoising the monitoring data stream collected in step S10. First, a bandpass filter is used to filter the original sensor signal data to remove high-frequency noise and low-frequency drift to obtain preprocessed data. In this process, it is also necessary to set a reasonable threshold to screen and eliminate abnormal band data in the preprocessed data to form an abnormal data set. The abnormal band in the preprocessed data may indicate an abnormal condition in the wind tunnel and needs to be extracted separately for analysis. The threshold of filtering and denoising can be adjusted according to the data characteristics of different sensors, and can generally be set to 1.5-2 times the normal value. Through this step, stable preprocessed data and abnormal data sets that need to be analyzed in detail can be effectively extracted from the original monitoring data.

[0028] The specific implementation method of step S30 is: first start the CUDA parallel processing flow of the GPU to receive the pre-processed data after filtering and noise reduction; at the same time, start the input flow of the CPU to receive the screened abnormal data set. CUDA parallel processing can greatly improve the analysis and processing speed of a large amount of pre-processed data, while the serial processing of the CPU is more suitable for analyzing abnormal data sets one by one. Such an asynchronous parallel architecture can make full use of hardware resources and improve the overall computing efficiency of the system.

[0029] The specific implementation method of step S40 is: decompose the pre-processed data and the abnormal data set into characteristic components respectively. For the pre-processed data, the wavelet decomposition algorithm is used to decompose it into a steady-state data matrix, a dynamic data matrix and a parameter data vector. Among them, the steady-state data matrix reflects the basic operating characteristics of the system, the dynamic data matrix reflects the instantaneous change characteristics of the system, and the parameter data vector contains characteristic quantities such as the baseline value and amplitude of each monitoring parameter. For the abnormal data set, the same wavelet decomposition algorithm is used to decompose it into an abnormal characteristic matrix and an abnormal parameter vector, reflecting the characteristic quantities of the system under abnormal conditions. These characteristic components provide a basis for subsequent data mining and state evaluation.

[0030] The specific implementation method of step S50 is: based on the steady-state data matrix, dynamic data matrix and abnormal characteristic matrix obtained in the previous step, a set of data mining characteristic equations is established. This set of equations includes state characteristic equations, dynamic correlation equations, steady-state mapping equations and abnormal influence equations. The state characteristic equation describes the instantaneous change characteristics and cumulative effects of the system state; the dynamic correlation equation describes the dynamic coupling relationship between system parameters, and introduces the second-order derivative term to reflect the acceleration change; the steady-state mapping equation describes the steady-state characteristics of the system parameters and their relationship with the temperature deviation; the abnormal influence equation describes the influence of abnormal vibration and noise on the system state, and introduces the second-order derivative term of abnormal energy to reflect the mutation characteristics. This set of characteristic equations links various physical mechanisms contained in the monitoring data and provides a mathematical model basis for subsequent state evaluation.

[0031] The specific implementation of step S60 is: using the data mining characteristic equation group established in step S50, combined with the steady-state data matrix, dynamic data matrix and abnormal characteristic matrix obtained in step S40, the state evaluation result of the wind tunnel operation is calculated. The state evaluation result includes the operating state characteristic vector, the operating parameter dynamic characteristic vector, the operating parameter steady-state characteristic vector and the abnormal impact characteristic vector. These characteristic vectors fully describe the current operating state of the wind tunnel and provide a basis for subsequent state warning. The various undetermined coefficients in the state evaluation equation group can be obtained by training and optimizing the complete data set collected under standard conditions using the particle swarm optimization algorithm.

[0032] The specific implementation of step S70 is: according to the wind tunnel operation status evaluation result calculated in step S60, two key outputs are generated: one is the wind tunnel operation characteristic matrix, which comprehensively reflects the steady-state characteristics, dynamic characteristics and abnormal characteristics of various operation parameters of the wind tunnel; the other is the wind tunnel state warning index, which is a comprehensive warning index obtained by weighted synthesis of the operation state characteristic vector, dynamic characteristic vector, steady-state characteristic vector and abnormal impact characteristic vector. These two outputs provide important basis for subsequent state warning. The threshold of the wind tunnel state warning index can be initially set to 1.2-1.5 times the normal value based on engineering experience, and can be further optimized and adjusted according to actual application.

[0033] The specific implementation of step S80 is: compare the wind tunnel status warning index generated in step S70 with the pre-set health threshold. If the warning index exceeds the health threshold, the corresponding warning signal is output. The setting of the health threshold can refer to the following principles: for the operating state characteristic vector and the dynamic characteristic vector, it can be set to 1.2-1.5 times the normal value; for the steady-state characteristic vector, it can be set to 1.5-2 times the normal value; for the abnormal impact characteristic vector, it can be set to 2-3 times the normal value. These thresholds can be dynamically adjusted and optimized according to the actual operating conditions. Once the warning signal is output, it means that the wind tunnel has an abnormal condition and corresponding maintenance measures need to be taken in time.

[0034] The calculation process or equation involved in the present invention is described in detail below.

[0035] 1. The sensor signals are specifically expressed as follows: ; ; ; ; ; ; ; ; ; Where: is the time variable; is the reference value of each signal; is the amplitude of each signal; is the frequency of each signal; is the phase angle; is the temperature attenuation coefficient; is the harmonic component of each signal; is the speed drift term; is a voltage signal; is the power factor; is the noise random term; Each parameter is obtained through sensor calibration experiment.

[0036] 2. The steady-state data matrix The specific expressions are as follows: ; Where: is the steady-state component of each signal; is the sampling time point;

[0037] The steady-state components are obtained by wavelet decomposition.

[0038] 3. The dynamic data matrix The specific expressions are as follows: ; Where: is the dynamic component of each signal, obtained by wavelet decomposition.

[0039] 4. The abnormal feature matrix The specific expressions are as follows: ; Where: It is an abnormal vibration signal; is an abnormal noise signal; is an abnormal energy signal. ; is the integration time window.

[0040] 5. The state characteristic equation is specifically expressed as follows: ; Where: is the undetermined weight coefficient, which is determined by the particle swarm optimization algorithm; is the error correction term;

[0041] This equation takes into account the instantaneous change characteristics and cumulative effects of the system state.

[0042] 6. The dynamic correlation equation is specifically expressed as follows: ; Where: is the weight coefficient to be determined; is the characteristic value; is the error correction term; The equation introduces the second-order derivative term to describe the acceleration change of the system dynamic characteristics.

[0043] 7. The steady-state mapping equation is specifically expressed as follows: ; Where: is the weight coefficient to be determined; is an imaginary unit; is the error correction term; This equation takes into account the quadratic integration effect of the temperature deviation.

[0044] 8. The abnormal influence equation is specifically expressed as follows: ; Where: is the weight coefficient to be determined; is the error correction term;

[0045] The second-order derivative of abnormal energy introduced into this equation reflects the mutation characteristics of the abnormal state.

[0046] All the unknown coefficients are obtained through the following steps: collecting a complete data set under standard wind tunnel conditions and classifying and labeling it; decomposing the signal using wavelet transform; optimizing the coefficients using particle swarm optimization algorithm; and verifying the accuracy of the model using cross-validation method.

[0047] The reasons for choosing the above functional relationships are: using trigonometric functions to describe periodic change characteristics; using exponential functions to describe attenuation characteristics; introducing integral terms to reflect cumulative effects; introducing derivative terms to reflect rate of change and acceleration; using complex exponentials to describe phase relationships; using logarithmic functions to compress data ranges; and quadratic terms to reflect nonlinear characteristics.

[0048] The derivation process and parameter sources of each equation are explained in detail below.

[0049] 1. Original signal modeling derivation process: Through the analysis of the wind tunnel operation mechanism, the composition of each sensor signal includes the following characteristics: (1) Periodic changes: The sine function is used to describe the basic periodic characteristics; (2) Attenuation characteristics: described by exponential function; (3) Harmonic component: using Fourier series expansion; (4) Random perturbation: introducing noise terms; For example, temperature signal The derivation steps are: Step 1: Determine the base temperature , obtained through no-load test calibration; Step 2: Analyze the periodic variation characteristics and obtain item; Step 3: Consider the temperature attenuation law and introduce item; Step 4: Determine by least squares fitting , , , Parameter value.

[0050] 2. Steady-state data matrix Derivation process: Step 1: Perform wavelet transform on the original signal; Step 2: Select the wavelet basis function and determine the number of decomposition layers; Step 3: Extract low-frequency components as steady-state features; Step 4: Construct the steady-state data matrix.

[0051] 3. State characteristic equation Derivation process: Step 1: Based on the first law of thermodynamics, establish the energy balance equation; Step 2: Introduce the time partial derivative to describe the rate of state change; Step 3: Consider the integrated effects of temperature and pressure; Step 4: Add error correction term.

[0052] Optimization effect: can reflect the instantaneous changes of system status; takes into account the energy accumulation effect; has strong anti-interference ability.

[0053] 4. Dynamic correlation equation Derivation process: Step 1: Based on the principles of fluid mechanics, establish the dynamic characteristic equation; Step 2: Consider the quadratic effect of the speed; Step 3: Introduce the cubic term of flow rate to reflect the turbulent characteristics; Step 4: Add the second-order derivative term to describe the change in acceleration.

[0054] Parameter determination: Obtained through genetic algorithm optimization; , Optimized by particle swarm algorithm; Determined by simulated annealing algorithm.

[0055] 5. Steady-state mapping equation Derivation process: Step 1: Establish power factor mapping relationship; Step 2: Introduce complex exponential to describe phase relationship; Step 3: Add the quadratic integral term of temperature deviation; Step 4: Optimize the error correction term.

[0056] 6. Abnormal influence equation Derivation process: Step 1: Analyze the coupling mechanism of vibration and noise; Step 2: Establish an abnormal energy calculation model; Step 3: Introduce logarithmic compression and square root terms; Step 4: Add the second-order derivative to reflect the mutation characteristics.

[0057] Parameter optimization: (1) , Obtained through support vector regression; (2) Determined by cross validation; (3) Determine the range through error analysis.

[0058] The existing methods for wind tunnel health monitoring data mining mainly have problems such as low processing efficiency and insufficient monitoring accuracy. Traditional monitoring methods often use a single parameter threshold or basic statistical analysis method for fault diagnosis. This method cannot effectively reflect the complex coupling relationship between multiple parameters in the wind tunnel system. At the same time, due to the huge amount of data generated during the operation of the wind tunnel and the need for real-time processing and analysis, the traditional serial calculation mode can no longer meet the needs of real-time monitoring. In addition, neural networks or deep learning methods are generally used in the existing technology for state recognition. Although these methods have strong nonlinear fitting capabilities, the training process is time-consuming and the real-time calculation complexity is high, making it difficult to meet the requirements of rapid response. The monitoring method based on pattern recognition requires the establishment of a complete fault mode library in advance, has poor recognition capabilities for newly emerging fault modes, and has a large amount of computation in the feature extraction process. More importantly, the existing technology lacks a special processing mechanism for abnormal data, and often adopts a simple elimination or smoothing method. This approach not only loses important fault feature information, but also may lead to misjudgment.

[0059] The present invention proposes an innovative data mining method for wind tunnel health monitoring, which realizes efficient data processing by establishing a GPU-CPU collaborative parallel processing architecture. This method assigns computationally intensive temperature, vibration, pressure, and sound signals to the GPU's CUDA parallel stream processing, while handing over control-related current and speed signals to the CPU for processing, giving full play to the advantages of heterogeneous computing systems. In terms of signal processing, the present invention establishes a complete mathematical model of multi-source signals, which not only considers the periodicity, attenuation characteristics, harmonic components, and random disturbances of the signal, but also realizes the precise separation of steady-state components and dynamic components through wavelet transform. In particular, in terms of abnormal data processing, an innovative abnormal energy assessment model is established to improve the accuracy of anomaly detection.

[0060] The most notable feature of the present invention is the design of four core matrix calculation equations. The state characteristic equation is based on the principle of energy balance. By introducing the time partial derivative, it can accurately describe the dynamic change characteristics of the system state. The dynamic correlation equation integrates the fluid mechanics characteristics, takes into account the nonlinear effect, and effectively reflects the complex coupling relationship between the system parameters. The steady-state mapping equation uses a complex function to describe the phase relationship, which improves the accuracy of steady-state characteristic identification. The abnormal influence equation establishes a coupling model of vibration and noise, providing a reliable basis for abnormal state identification. These equations are implemented by matrix parallel computing, which greatly improves the calculation efficiency.

[0061] In terms of parameter optimization, the present invention adopts a variety of intelligent optimization algorithms to determine the weight coefficients, introduces an adaptive mechanism to dynamically adjust the parameters, and establishes an error correction mechanism to improve the calculation accuracy. This method not only improves the warning accuracy of the system, but also significantly reduces the false alarm rate. The efficient data processing achieved by matrix calculation, the use of sparse matrix storage optimizes memory usage, the matrix block calculation improves parallel efficiency, and the pipeline processing reduces the data transmission overhead. The establishment of a multi-level warning mechanism enables the system to monitor the abnormal energy change trend in real time, dynamically evaluate the system state evolution direction, and promptly discover potential fault hazards.

[0062] The present invention has shown remarkable technical effects in practical applications. Through GPU-CPU collaborative processing and optimized matrix calculations, the data processing speed is greatly improved and the system response time is significantly shortened. The fault mode recognition method of multi-source information fusion improves the system's early warning accuracy and reduces the false alarm rate. The abnormal data processing mechanism improves the system's anti-interference ability, multiple error corrections ensure the reliability of the calculation results, and adaptive parameter adjustment improves the system's adaptability. The equation has a clear physical meaning and is convenient for fault diagnosis. The modular design and standardized interface make the system have good maintainability and scalability. These advances enable the present invention to effectively solve the real-time and accuracy problems in wind tunnel health monitoring, and it has important engineering application value.

[0063] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned method for mining wind tunnel health monitoring data.

[0064] A third aspect of the present invention provides a wind tunnel health monitoring data mining system, which includes the above-mentioned computer-readable storage medium.

[0065] Specifically, the principle of the present invention is: at the data acquisition level, a comprehensive data acquisition network is constructed by arranging multiple types of sensors such as temperature, vibration, and pressure. This multi-source data acquisition method conforms to the complexity characteristics of the wind tunnel system and can fully reflect the operating status of the system. The layout positions and sampling frequencies of various sensors are determined according to the characteristics of the wind tunnel to ensure the representativeness and integrity of the data.

[0066] At the data preprocessing level, filtering and noise reduction technology is used to process the original data, and abnormal data is identified and separated. This processing method is based on signal processing theory and effectively improves data quality through frequency domain analysis and statistical feature extraction. The separation of abnormal data is based on statistical significance tests to ensure the accurate extraction of abnormal features.

[0067] At the parallel computing level, the GPU-based CUDA architecture processes preprocessed data, while the CPU is used to process abnormal data sets. This division of labor and cooperation model fully utilizes hardware resources and improves computing efficiency. The GPU is suitable for processing large-scale matrix operations, while the CPU is more suitable for processing complex logic abnormal data analysis.

[0068] At the feature extraction level, the data is decomposed into steady-state, dynamic and parameter features through component decomposition method. This decomposition is based on time series analysis theory and can effectively capture the different operating characteristics of the system. Steady-state features reflect the basic state of the system, dynamic features reflect the changing laws of the system, and parameter features reflect the key indicators of the system.

[0069] At the state assessment level, an assessment system consisting of four core equations was constructed. The state characteristic equation is based on parameter correlation analysis, the dynamic correlation equation is based on time series pattern recognition, the steady-state mapping equation is based on characteristic mapping theory, and the abnormal impact equation is based on the anomaly detection principle. These equations complement each other and together form a complete assessment framework.

[0070] The technical solution of the present invention has strict logic and scientificity, and each processing link is interconnected to form a complete data mining system, which can effectively solve the problem of insufficient data mining in wind tunnel health monitoring.

[0071] A specific embodiment 1 of the present invention is provided below. The specific method of each step in this embodiment 1 is described in detail as follows: The specific implementation method of step S10 is: first, by installing a series of sensors at key positions of the wind tunnel, including temperature sensors, vibration sensors, pressure sensors, sound sensors, current sensors, speed sensors, power sensors, flow sensors and noise sensors, various monitoring data during the operation of the wind tunnel are collected in real time. These monitoring data include temperature sensor signals , vibration sensor signal , pressure sensor signal , sound sensor signal , current sensing signal , speed sensor signal , power sensing signal , flow sensor signal and noise sensing signals These sensing signals can be described by the following mathematical model: ; ; ; ; ; ; ; ; ; in, is the time variable; is the reference value of each signal; is the amplitude of each signal; is the frequency of each signal; is the phase angle; is the temperature attenuation coefficient; is the harmonic component of each signal; is the speed drift term; is a voltage signal; is the power factor; is the noise random term. These parameters can be obtained through sensor calibration experiments. By collecting these signal data, the operation status of the wind tunnel can be fully reflected.

[0072] The specific implementation method of step S20 is: first, a digital filter is used to filter and de-noise the monitoring data collected in step S10. Specifically, a bandpass filter is used to filter the original sensor signal data to remove high-frequency noise and low-frequency drift to obtain preprocessed data. In this process, it is also necessary to set a reasonable threshold to filter and eliminate the abnormal band data in the preprocessed data to form an abnormal data set. The abnormal band in the preprocessed data may indicate a fault condition in the wind tunnel and needs to be extracted separately for analysis. The threshold of filtering and de-noising can be adjusted according to the data characteristics of different sensors, and can generally be set to 1.5 to 2 times the normal value. Through this step, stable preprocessed data and abnormal data sets that need to be analyzed can be effectively extracted from the original monitoring data.

[0073] The specific implementation method of step S30 is: first start the CUDA parallel processing flow of the GPU to receive the pre-processed data after filtering and noise reduction; at the same time, start the input flow of the CPU to receive the screened abnormal data set. The CUDA parallel processing flow can greatly improve the analysis and processing speed of a large amount of pre-processed data, while the serial input flow of the CPU is more suitable for analyzing abnormal data sets one by one. Such an asynchronous parallel architecture can make full use of hardware resources and improve the overall computing efficiency of the system.

[0074] The specific implementation of step S40 is: decomposing the preprocessed data and the abnormal data set into characteristic components respectively. The preprocessed data is decomposed into a steady-state data matrix using a discrete wavelet transform (DWT) algorithm. , Dynamic Data Matrix And parameter data vector. Among them, the steady-state data matrix It reflects the basic operating characteristics of the system and can be expressed as: ; Dynamic Data Matrix It reflects the instantaneous change characteristics of the system and can be expressed as: ; The parameter data vector contains the baseline value and amplitude of each monitoring parameter. For the abnormal data set, the same DWT algorithm is used to decompose it into an abnormal feature matrix And the abnormal parameter vector can be expressed as: ; in, is an abnormal vibration signal, is an abnormal noise signal, is an abnormal energy signal, is the integration time window. These characteristic components provide the basis for subsequent data mining and state assessment.

[0075] The specific implementation of step S50 is: based on the steady-state data matrix obtained in the previous step , Dynamic Data Matrix and abnormal feature matrix , establish a set of data mining characteristic equations. This set of equations includes: Characteristic equation of state: ; in, is the undetermined weight coefficient, is the error correction term. This equation takes into account the instantaneous change characteristics and cumulative effects of the system state.

[0076] Dynamic correlation equation: ; in, is the undetermined weight coefficient, is the characteristic value, is the error correction term. The equation introduces the second-order derivative term to describe the acceleration change of the system dynamic characteristics.

[0077] Steady-state mapping equation: ; in, is the undetermined weight coefficient, is an imaginary unit, is the error correction term. This equation takes into account the quadratic integral effect of temperature deviation.

[0078] Abnormal influence equation: ; in, is the undetermined weight coefficient, is the error correction term. The second-order derivative of the abnormal energy introduced into this equation reflects the mutation characteristics of the abnormal state.

[0079] This set of characteristic equations links various physical mechanisms contained in the monitoring data and provides a mathematical model basis for subsequent state assessment. All the unknown coefficients can be obtained by training and optimizing the complete data set collected under standard working conditions using the particle swarm optimization algorithm.

[0080] The specific implementation of step S60 is: using the data mining characteristic equation group established in step S50, combined with the steady-state data matrix obtained in step S40 , Dynamic Data Matrix and abnormal feature matrix , calculate the status evaluation results of the wind tunnel operation. The status evaluation results include: Running state feature vector ; Running parameter dynamic feature vector ; Steady-state characteristic vector of operating parameters ; Abnormal impact feature vector ; These eigenvectors comprehensively describe the current operating status of the wind tunnel and provide a basis for subsequent status warnings.

[0081] The specific implementation of step S70 is: according to the wind tunnel operation status evaluation result calculated in step S60, two key outputs are generated: The first is the wind tunnel operation characteristic matrix, which comprehensively reflects the steady-state characteristics, dynamic characteristics and abnormal characteristics of various wind tunnel operation parameters; The second is the wind tunnel status warning index, which is the characteristic vector of the running status. , dynamic feature vector , steady-state eigenvector and abnormal impact feature vector The comprehensive early warning indicator is obtained by weighted synthesis.

[0082] The threshold of the wind tunnel status warning indicator can be initially set to 1.2 to 1.5 times the normal value based on engineering experience, and can be further optimized and adjusted according to actual applications.

[0083] The specific implementation of step S80 is: compare the wind tunnel status warning index generated in step S70 with the pre-set health threshold. If the warning index exceeds the health threshold, a corresponding warning signal is output. The setting of the health threshold can refer to the following principles: For the running state feature vector and dynamic feature vector , can be set to 1.2∼1.5 times the normal value; For the steady-state eigenvector , can be set to 1.5∼2 times the normal value; For abnormal impact feature vector , can be set to 2 to 3 times the normal value.

[0084] These thresholds can be dynamically adjusted and optimized according to the actual operation situation. Once the warning signal is output, it means that the wind tunnel has an abnormal condition and corresponding maintenance measures need to be taken in time.

[0085] In order to better understand and implement the present invention, the following provides Example 2 of a specific application scenario of the present invention: An aviation research institute is conducting a test on the aerodynamic performance of a new aircraft, and needs to use the institute's large-scale advection wind tunnel for the test. In view of the complex structure and harsh working environment of the new aircraft, the researchers decided to use the wind tunnel health monitoring data mining method proposed in the present invention to conduct comprehensive monitoring and intelligent early warning of the operation status of the wind tunnel.

[0086] First, temperature sensors, vibration sensors, pressure sensors, sound sensors, current sensors, speed sensors, power sensors, flow sensors, and noise sensors were installed at key locations such as the inlet, test, and outlet sections of the wind tunnel. These sensors collected and output the following monitoring signals in real time: Temperature sensor signal : ; in, is the reference temperature, is the temperature amplitude, is the temperature change frequency, is the temperature attenuation coefficient, is the temperature attenuation coefficient.

[0087] Vibration sensor signal : ; in, is the amplitude, is the vibration frequency, is the amplitude, for Subharmonic components.

[0088] Pressure sensor signal : ; in, is the base pressure, is the pressure amplitude, is the pressure change frequency, for Subharmonic components.

[0089] Sound sensor signal : ; in, For the The amplitude of the infrasonic component, For the The frequency of the infrasonic component, For the The phase of the infrasound component.

[0090] Current sensing signal : ; in, is the reference current, is the current amplitude, is the current variation frequency, for Subharmonic components.

[0091] Speed ​​sensor signal : ; in, is the reference speed, is the speed amplitude, is the speed change frequency, is the speed drift term.

[0092] Power sensing signal : ; in, is the voltage signal, is the power factor.

[0093] Flow sensor signal : ; in, is the base flow rate, is the flow amplitude, is the flow rate change frequency, for Subharmonic components.

[0094] Noise sensor signal : ; in, For the The amplitude of the secondary noise component, For the The frequency of the secondary noise component, For the The phase of the secondary noise component, is a random noise term.

[0095] The parameters of the above-mentioned sensor signals are all determined in advance through calibration experiments.

[0096] After collecting about 10 minutes of initial monitoring data, the researchers started the data preprocessing process. First, a bandpass filter was used to filter each sensor signal to remove high-frequency noise and low-frequency drift to obtain preprocessed data. Then, the preprocessed data was tested for abnormal bands, and the threshold was set to 1.5 times the normal value. The data segments that exceeded the threshold were removed to form an abnormal data set. At this point, the preprocessed data and abnormal data sets are ready for subsequent feature decomposition and data mining analysis.

[0097] Next, the researchers started the GPU's CUDA parallel processing stream to perform eigendecomposition on the preprocessed data, and simultaneously started the CPU's serial input stream to perform eigendecomposition on the abnormal data set. Using the discrete wavelet transform (DWT) algorithm, the preprocessed data was decomposed into a steady-state data matrix. and dynamic data matrix , specifically as follows: Steady-state data matrix : ; Dynamic Data Matrix : ; For abnormal data sets, the DWT algorithm obtains the abnormal feature matrix : ; in, is an abnormal vibration signal, is an abnormal noise signal, is an abnormal energy signal. Figure 2 As shown, the characteristic distribution of the steady-state data matrix is ​​displayed through a three-dimensional surface, reflecting the correlation between the various monitoring parameters.

[0098] With these characteristic components, the researchers began to establish a set of data mining characteristic equations proposed in this invention.

[0099] First is the characteristic equation of state: ; in, is the undetermined weight coefficient, is the error correction term. is the integration time window. This equation takes into account the instantaneous change characteristics and cumulative effects of the system state.

[0100] Dynamic correlation equation: ; in, is the undetermined weight coefficient, is the characteristic value, is the error correction term. The equation introduces the second-order derivative term to describe the acceleration change of the system dynamic characteristics.

[0101] Steady-state mapping equation: ; in, is the undetermined weight coefficient, is an imaginary unit, is the error correction term. This equation takes into account the quadratic integral effect of temperature deviation.

[0102] Abnormal influence equation: ; in, is the undetermined weight coefficient, is the error correction term. The equation introduces the second-order derivative of the abnormal energy to reflect the mutation characteristics of the abnormal state.

[0103] Through calculations, the researchers obtained various status evaluation results of wind tunnel operation: Running state feature vector ; Running parameter dynamic feature vector ; Steady-state characteristic vector of operating parameters ; Abnormal impact feature vector ; The wind tunnel status warning index is obtained by weighted synthesis of these characteristic vectors. The pre-set health threshold is , so the current warning indicator exceeds the health threshold. Figure 3As shown in the figure, the change process of the warning indicator H over time is shown, and the health threshold line is marked, clearly showing the over-threshold alarm area.

[0104] The researchers then checked the wind tunnel operation characteristic matrix and found that there were large anomalies in parameters such as temperature, speed and noise. Further analysis revealed that a vibration sensor at the wind tunnel exit section had failed, leading to abnormal phenomena such as increased temperature, speed fluctuations, and increased noise.

[0105] Based on the above analysis results, the researchers immediately stopped the aerodynamic test and arranged maintenance personnel to inspect and repair the vibration sensor at the exit section. After a short shutdown, the aerodynamic test was carried out smoothly after the wind tunnel returned to normal operation.

[0106] This practice fully verified the effectiveness of the wind tunnel health monitoring data mining method proposed in this invention. Through the comprehensive analysis of multi-source monitoring data, this method can not only timely detect abnormal conditions of the wind tunnel and accurately locate the location of the fault, but also provide a strong guarantee for the safe operation of the wind tunnel by generating reliable status warning indicators. Compared with traditional single sensor signal fault diagnosis, this intelligent early warning system based on big data analysis has higher accuracy and robustness.

[0107] It should be noted that the variables involved in the present invention are explained in detail as shown in Table 1 below.

[0108] Table 1 Variable explanation table

[0109] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for mining wind tunnel health monitoring data, characterized in that: The method includes collecting multiple sensor signals to construct a monitoring data stream, filtering and denoising the monitoring data stream to obtain preprocessed data and an abnormal data set, starting a GPU and a CPU processing stream to receive the preprocessed data and the abnormal data set, performing component decomposition to obtain a steady-state data matrix, a dynamic data matrix, an abnormal feature matrix, a parameter data vector and an abnormal parameter vector, establishing a data mining characteristic equation group based on the steady-state data matrix, the dynamic data matrix and the abnormal feature matrix, calculating a wind tunnel operation status evaluation result using a wind tunnel operation status evaluation equation group, generating a wind tunnel operation characteristic matrix and a wind tunnel state warning indicator, and outputting a warning signal according to a comparison result between the wind tunnel state warning indicator and a preset health threshold.

2. The method for mining wind tunnel health monitoring data according to claim 1, characterized in that: The multiple sensor signals include temperature sensor signals, vibration sensor signals, pressure sensor signals, sound sensor signals, current sensor signals, rotation speed sensor signals, power sensor signals, flow sensor signals and noise sensor signals.

3. The method for mining wind tunnel health monitoring data according to claim 2, characterized in that: The CUDA parallel processing stream of the GPU receives the pre-processed data; and the input stream of the CPU receives the abnormal data set.

4. The method for mining wind tunnel health monitoring data according to claim 3, characterized in that: The preprocessed data is decomposed into components to obtain the steady-state data matrix, the dynamic data matrix and the parameter data vector; the abnormal data set is decomposed into components to obtain the abnormal feature matrix and the abnormal parameter vector.

5. The method for mining wind tunnel health monitoring data according to claim 4, characterized in that: The wind tunnel operation state evaluation equation group includes a state characteristic equation, a dynamic correlation equation, a steady-state mapping equation and an abnormal influence equation.

6. The method for mining wind tunnel health monitoring data according to claim 5, characterized in that: The input of the state characteristic equation includes the steady-state data matrix, the dynamic data matrix and the temperature sensor signal and the pressure sensor signal in the parameter data vector, and the output is the operating state characteristic vector in the wind tunnel operating state evaluation result; the input of the dynamic correlation equation includes the dynamic data matrix, the abnormal characteristic matrix and the speed sensor signal and the flow sensor signal in the parameter data vector, and the output is the operating parameter dynamic characteristic vector in the wind tunnel operating state evaluation result.

7. The method for mining wind tunnel health monitoring data according to claim 6, characterized in that: The input of the steady-state mapping equation includes the power sensor signal and the current sensor signal in the steady-state data matrix and the parameter data vector, and the output is the steady-state characteristic vector of the operating parameters in the wind tunnel operating status evaluation result; the input of the abnormal influence equation includes the vibration sensor signal and the noise sensor signal in the abnormal characteristic matrix and the abnormal parameter vector, and the output is the abnormal influence characteristic vector in the wind tunnel operating status evaluation result.

8. The method for mining wind tunnel health monitoring data according to claim 7, characterized in that: The wind tunnel operation characteristic data is calculated by the wind tunnel operation status evaluation equation group to obtain the wind tunnel operation status evaluation result; the wind tunnel operation status evaluation result is used to generate the wind tunnel operation characteristic matrix and the wind tunnel status early warning index.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed in a computer, the method for mining wind tunnel health monitoring data according to any one of claims 1 to 8 is used.

10. A wind tunnel health monitoring data mining system, characterized in that: A computer-readable storage medium comprising the computer-readable storage medium of claim 9.

Images