Intelligent detection method and system for the status of pneumatic valves on test bench

Abstract

This invention relates to a method and system for detecting the condition of engine ground test benches, specifically to a method and system for intelligent detection of the condition of pneumatic valves on test benches. To address the shortcomings of existing valve internal leakage detection technologies, such as sensitivity to temperature interference, high failure rate for minor leaks, limited correlation analysis, and high deployment costs, this invention provides an intelligent detection method for the condition of pneumatic valves on test benches. This method includes data acquisition and preprocessing; graded assessment of cylinder health status based on preprocessed data; detection of valve internal sealing status based on preprocessed data; and output of cylinder health status and valve internal sealing status to obtain intelligent detection results for the pneumatic valve condition. Furthermore, a test bench intelligent detection system for implementing the above method is also provided.

Description

Technical Field

[0001] This invention relates to a method and system for detecting the status of an engine ground test stand, specifically to a method and system for intelligent detection of the status of pneumatic valves on the test stand. Background Technology

[0002] The reliability of valves on rocket engine test benches directly affects engine test safety and accuracy. Current technologies for valve internal leakage detection often rely on single methods, such as pressure threshold alarms or leakage decay model fitting, which have the following drawbacks:

[0003] (1) Sensitive to temperature interference: Traditional threshold method cannot distinguish between temperature fluctuation (±5℃) and real leakage signal, with a false alarm rate of over 30%;

[0004] (2) High rate of missed detection for minute leaks: The leakage attenuation model is not sensitive enough to detect minute leaks (leakage rate < 0.1 L / min) (missed detection rate > 20%).

[0005] (3) Single correlation analysis: Only single sensor data or simple time domain feature analysis is used, without making full use of the correlation information of multi-dimensional pressure signals.

[0006] (4) High deployment cost: It requires a large number of professional sensors (such as acoustic array sensors), costing hundreds of thousands of yuan, and is difficult to deploy on a large scale on the test bench. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of existing valve internal leakage detection technologies, such as sensitivity to temperature interference, high failure rate of small leaks, limited correlation analysis, and high deployment costs, and to provide a method and system for intelligent detection of the status of pneumatic valves on a test bench.

[0008] To achieve the above objectives, the technical solution provided by this invention is as follows:

[0009] A method for intelligent detection of the status of pneumatic valves on a test bench, characterized by the following steps:

[0010] Data acquisition and preprocessing; the data includes pressure signals from the chambers on both sides of the piston in the cylinder, pressure signals from the valve inlet and outlet, and ambient temperature signals;

[0011] Based on the pre-processed pressure signals and ambient temperature signals of the piston chambers on both sides of the cylinder, the health status of the cylinder is graded and evaluated.

[0012] Based on the pre-processed valve inlet and outlet pressure signals, the internal sealing status of the valve is detected.

[0013] The health status of the output cylinder and the internal sealing status of the valve are used to obtain the intelligent detection results of the pneumatic valve status.

[0014] Furthermore, the data acquisition and preprocessing includes the following steps:

[0015] Synchronously acquire pressure signals from both sides of the piston chamber in the cylinder. Valve inlet and outlet pressure signals and ambient temperature signal ;

[0016] Pressure signals in the piston chambers on both sides of the cylinder Including cylinder cavity I pressure and cylinder cavity II pressure The valve inlet and outlet pressure signals Including valve inlet pressure and valve outlet pressure ;

[0017] The pressure signals collected from the piston chambers on both sides of the cylinder Valve inlet and outlet pressure signals After denoising and centering, the zero-mean signal of cylinder pressure is obtained. Zero-mean signal of valve inlet and outlet pressure .

[0018] Furthermore, the graded assessment of the cylinder's health status includes the following steps:

[0019] Calculate the zero-mean signal of cylinder pressure higher-order moments;

[0020] Establish cylinder pressure zero-mean signal A linear regression model of higher-order moments and ambient temperature;

[0021] Based on the ambient temperature signal, a linear regression model is used to analyze the zero-mean cylinder pressure signal. Compensation is performed on higher-order moments;

[0022] Based on the zero-mean signal of the compensated cylinder pressure The higher-order moments are used to classify and evaluate the health status of the cylinder.

[0023] Furthermore, the cylinder pressure zero-mean signal Higher-order moments include third-order moments and fourth moment ,

[0024] ;

[0025] ;

[0026] Where u is the length of the scroll window, and Δt is the step size. Let be the zero-mean cylinder pressure at the i-th second, where i takes the values ​​t, t+1, ... ; These correspond to cylinder cavity I and cylinder cavity II, respectively; The mean of the zero-mean cylinder pressure. The standard deviation of the zero mean cylinder pressure;

[0027] The cylinder pressure zero-mean signal The linear regression model of the higher-order moments and ambient temperature is expressed as:

[0028] ;

[0029] in, Zero-mean signal of cylinder pressure The higher-order moments are given by , where j is the index of the higher-order moment, taking a value of 1 or 2, corresponding to the third and fourth-order moments respectively. , These are the linear parameters of the linear regression model. This is random noise related to ambient temperature;

[0030] The zero-mean signal of the compensated cylinder pressure higher order moments Represented as:

[0031] ;

[0032] The specific steps for grading and assessing the health status of cylinders are as follows:

[0033] calculate Z-score standardized value ;

[0034] ;

[0035] in, This represents the average of the higher-order moments of the pressure signal corresponding to the cylinder under normal conditions. The standard deviation of the higher-order moment of the pressure signal corresponding to the cylinder under normal conditions;

[0036] according to The health status of the cylinder shall be graded and assessed according to the following grading rules:

[0037] like and If so, the cylinder is normal;

[0038] like or If so, the cylinder will degenerate;

[0039] like or If so, then the cylinder is faulty.

[0040] Furthermore, based on the preprocessed valve inlet and outlet pressure signals, the specific operation for detecting the internal sealing status of the valve includes the following steps:

[0041] Zero-mean signal of valve inlet and outlet pressure Perform principal component-pressure correlation coefficient analysis;

[0042] Zero-mean signal of valve inlet and outlet pressure Perform independent component-pressure mutual information analysis;

[0043] Zero-mean signal of valve inlet and outlet pressure Perform direct correlation coefficient analysis;

[0044] The internal leakage features, including principal component-pressure correlation coefficient, independent component-pressure mutual information, and direct correlation coefficient, are fused into a comprehensive feature vector F, which is then used to classify the internal sealing state of the valve based on the SVM model.

[0045] Furthermore, the principal component-pressure correlation coefficient analysis specifically involves:

[0046] Establish zero-mean signal for valve inlet and outlet pressure covariance matrix Zero-mean signal for inlet and outlet pressure Including zero mean valve inlet pressure and valve outlet pressure zero mean ;

[0047] Covariance matrix extracted by eigenvalue decomposition The two principal components with a cumulative variance of ≥95% are the first principal component PC1(t) and the second principal component PC2(t).

[0048] Calculate the first principal component PC1(t) and the zero mean value of the valve outlet pressure. correlation coefficient And the second principal component PC2(t) and the zero mean of the valve inlet pressure correlation coefficient ;

[0049] like and Marked as normal association, if and Mark them as abnormal associations and record the correlation coefficients. The square value and correlation coefficient The square value of .

[0050] Furthermore, the independent component-pressure mutual information analysis specifically includes:

[0051] Zero-mean signal of valve inlet and outlet pressure PCA whitening is performed to obtain the whitened signal Z(t). The FastICA algorithm is then used to optimize the whitened signal Z(t) through maximum likelihood estimation to obtain the separation matrix W. Independent sources are then separated from the separation matrix W. ,in, For leakage source signal, The signal is a noise source.

[0052] Calculate the mutual information between the leak source signal and the pressure difference. ,in, The difference between the zero mean pressure signals at the valve inlet and outlet;

[0053] Calculate the average value of the noise source signal and the zero-mean signal of the valve inlet and outlet pressures. correlation coefficient ;

[0054] like If the value is greater than 0.3 bits, it is marked as a leak source; otherwise, it is marked as no leak.

[0055] like If the value is >0.6, it is marked as abnormal noise interference; otherwise, it is marked as normal noise interference.

[0056] Furthermore, the direct correlation coefficient analysis specifically includes:

[0057] Calculate the autocorrelation function between the zero-mean difference ΔP(t) of the valve inlet and outlet pressure signals and the zero-mean difference ΔP(t+τ) of the valve inlet and outlet pressure signals at its own lag time τ. ;

[0058] ;

[0059] in, for and covariance, for variance For time delay, ; for The variance;

[0060] Fitting autocorrelation function Exponential decay model:

[0061] ;

[0062] Obtain the attenuation coefficient ;

[0063] like If the decay is too slow, it is marked as decaying normally; otherwise, it is marked as decaying normally.

[0064] calculate ,like If it is not internally leaking, mark it as internally leaking; otherwise, mark it as not internally leaking.

[0065] Furthermore, the comprehensive feature vector F is expressed as:

[0066] ;

[0067] In the SVM model, the decision function for classifying the internal sealing state of the valve is:

[0068] ;

[0069] in, For the classification results, The total number of samples, For Lagrange multipliers, For sample labels, Let g be the feature vector of the g-th support vector. For the index of the support vectors; For kernel function, For bias;

[0070] The SVM model is trained using historical samples and outputs the internal sealing state of the valve based on the comprehensive feature vector F, including two states: sealed and unsealed.

[0071] Meanwhile, this invention also provides an intelligent detection system for the status of pneumatic valves on a test bench, used to implement the aforementioned intelligent detection method for the status of pneumatic valves on a test bench. Its key features include: a sensor module, a data acquisition module, a preprocessing module, a cylinder health assessment module, a valve sealing status module, and an output module; the sensor module includes a pressure sensor group and a temperature sensor. The pressure sensor group includes pressure sensors respectively disposed in the cavities on both sides of the piston in the cylinder, and pressure sensors disposed at the valve inlet and outlet. The temperature sensor is used to acquire the ambient temperature signal of the cylinder; the input end of the data acquisition module is connected to the output end of the sensor module, used to synchronously acquire the pressure signals of the cavities on both sides of the piston in the cylinder, the valve inlet and outlet pressure signals, and the ambient temperature signal. The preprocessing module receives pressure signals from the cylinder piston's two sides, valve inlet and outlet pressure signals, and ambient temperature signals. Its input is connected to the output of the data acquisition module, and it performs preprocessing on these signals. The cylinder health assessment module receives these signals and assesses the cylinder's health status based on the preprocessed pressure signals from the piston's two sides and the ambient temperature signal. The valve sealing status module receives these signals and detects the valve's internal sealing status based on the preprocessed valve inlet and outlet pressure signals. The output module outputs the cylinder's health status and the valve's internal sealing status, providing an intelligent detection result for the pneumatic valve's status.

[0072] The beneficial effects of this invention are:

[0073] 1. The intelligent detection method for pneumatic valve status on the test bench of this invention is based on pressure characteristics and multimodal correlation analysis. Specifically, it utilizes a temperature compensation model to separate temperature interference and achieve graded assessment of cylinder health status. The obtained cylinder health status assessment has high accuracy, with a judgment error of <5%. At the same time, through multimodal correlation analysis, the sensitivity and reliability of valve internal leakage detection are improved.

[0074] 2. The multimodal correlation analysis of the intelligent detection method for pneumatic valve status on the test bench of this invention captures the fluctuation pattern changes caused by leakage through principal component-pressure correlation coefficient analysis (sensitivity improved by 40%), identifies nonlinear correlations through independent component-pressure mutual information analysis (false alarm rate reduced by 30%), and detects abnormal correlations of the pressure signal itself through direct correlation coefficient analysis (missed detection rate reduced by 25%).

[0075] 3. The intelligent detection method for the status of pneumatic valves on the test bench of the present invention has high engineering applicability. It mainly achieves analysis and judgment through algorithms, without the need for additional hardware, and is compatible with existing pressure sensors.

[0076] 4. The intelligent detection system for the status of pneumatic valves on the test bench of the present invention has a simple structure. By setting pressure sensors and temperature sensors at specific positions on the pneumatic valves and their cylinders, the corresponding pressure signals and temperature signals are obtained, thereby realizing intelligent detection. Attached Figure Description

[0077] Figure 1 This is in Embodiment 1 of the present invention Temperature compensation model diagram;

[0078] Figure 2 This is in Embodiment 1 of the present invention Temperature compensation model diagram;

[0079] Figure 3 This is in Embodiment 1 of the present invention Temperature compensation model diagram;

[0080] Figure 4 This is in Embodiment 1 of the present invention Temperature compensation model diagram. Detailed Implementation

[0081] Example 1

[0082] This embodiment provides an intelligent detection method for the status of pneumatic valves on a test bench. The pneumatic valves include pneumatic shut-off valves and pneumatic ball valves, both of which are equipped with cylinders. The method in this invention mainly assesses the health status of the cylinder by the correlation between the pressure signals of the chambers on both sides of the piston in the cylinder and the ambient temperature signal. At the same time, it detects the internal sealing status of the valve by multimodal correlation analysis of the valve inlet and outlet pressures (principal component-pressure correlation coefficient, independent component-pressure mutual information, and direct correlation coefficient analysis). Finally, the health status of the cylinder and the internal sealing status of the valve are output to obtain the intelligent detection result of the pneumatic valve status.

[0083] The intelligent detection method for the status of pneumatic valves on the test bench of this invention specifically includes the following steps:

[0084] Step 1, Data Acquisition and Preprocessing; the data includes pressure signals from the chambers on both sides of the piston in the cylinder, pressure signals from the valve inlet and outlet, and ambient temperature signals;

[0085] Optionally, step 1 includes the following sub-steps:

[0086] Sub-step 101: Synchronously acquire pressure signals from the cylinder's piston chambers on both sides, valve inlet and outlet pressure signals, and ambient temperature signals;

[0087] Within the cylinder, a piston divides the cylinder into two chambers, denoted as Chamber I and Chamber II. The pressure signals in the chambers on either side of the piston within the cylinder... Including cylinder cavity I pressure and cylinder cavity II pressure ;

[0088] ;

[0089] Valve inlet and outlet pressure signals Including valve inlet pressure and valve outlet pressure ;

[0090] ;

[0091] The ambient temperature signal is denoted as , where t is the sampling time series point, taking values ​​of 1, 2...N, and N is the total number of time series points.

[0092] Sub-step 102: Preprocess the collected pressure signals from the cylinder piston chambers on both sides and the valve inlet and outlet pressure signals;

[0093] The collected pressure signals from the piston chambers on both sides of the cylinder and the valve inlet and outlet pressure signals are denoised using a window length method. Point, order The Savitzky-Golay filter suppresses high-frequency noise and outputs the denoised pressure signal of the piston chambers on both sides of the cylinder. and the denoised valve inlet and outlet pressure signals .

[0094] Centralization: This refers to the denoising of the pressure signals in the piston chambers on both sides of the cylinder. and the denoised valve inlet and outlet pressure signals Centralized processing is implemented; specifically:

[0095] The average pressure signal of the piston chambers on both sides of the cylinder is first calculated using the following formula. Average pressure signal at valve inlet and outlet Then, the zero-mean value signal of the cylinder pressure is calculated. Zero-mean signal of valve inlet and outlet pressure .

[0096] ;

[0097] ;

[0098] ;

[0099] ;

[0100] ;

[0101] ;

[0102] in, The zero-average pressure in cylinder cavity I is... The zero-average pressure in cylinder cavity II; The valve inlet pressure is zero average. The valve outlet pressure is zero average.

[0103] Step 2: Based on the pre-processed pressure signals and ambient temperature signals of the piston chambers on both sides of the cylinder, the health status of the cylinder is graded and evaluated.

[0104] Optionally, step 2 includes the following sub-steps:

[0105] Sub-step 21: Calculate the zero-mean signal of cylinder pressure. higher-order moments;

[0106] Higher-order moments include third-order and fourth-order moments. With a rolling window length of u = 0.1s to 10s (covering a typical pressure fluctuation period; in this embodiment, u = 2s) and a step size Δt = 0.5s, the zero-mean value signal of the cylinder pressure within the rolling window is first calculated. Statistical characteristics; statistical characteristics include the mean of the zero mean of cylinder pressure. Standard deviation of cylinder pressure zero mean ;

[0107] ;

[0108] ;

[0109] Therefore, the zero-mean signal of cylinder pressure can be obtained. The third moment and cylinder pressure zero mean signal The fourth moment They are represented as follows:

[0110] ;

[0111] ;

[0112] in, Let be the zero-mean cylinder pressure at the i-th second, where i takes the values ​​t, t+1, ... ; These correspond to cylinder cavity I and cylinder cavity II, respectively.

[0113] Sub-step 22: Establish the cylinder pressure zero-mean signal A linear regression model of higher-order moments and ambient temperature;

[0114] Cylinder pressure zero mean signal The linear regression model of the higher-order moments and ambient temperature is expressed as:

[0115] ;

[0116] in, Zero-mean signal of cylinder pressure The higher-order moments are given by , where j is the index of the higher-order moment, taking a value of 1 or 2, corresponding to the third and fourth-order moments respectively. , These are the linear parameters of the linear regression model. This is random noise related to ambient temperature.

[0117] Estimating the linear parameters of a linear regression model using the least squares method , Required goodness of fit And significance .

[0118] Sub-step 23: Based on the ambient temperature signal, calculate the zero-mean signal of the cylinder pressure. Compensation is performed on the higher-order moments; specifically:

[0119] The zero-mean signal of the compensated cylinder pressure is calculated using the following formula. higher order moments :

[0120] ;

[0121] Compensated cylinder pressure zero-mean signal The higher-order moments are the zero-mean cylinder pressure signals calculated from sub-step 21. The deviations caused by ambient temperature are removed from the higher-order moments, while the true state information of the cylinder is retained, which means retaining the random noise related to ambient temperature.

[0122] Sub-step 24, based on the compensated cylinder pressure zero-mean signal higher order moments The health status of the cylinder is assessed in a graded manner; specifically:

[0123] Calculate the zero-mean signal of cylinder pressure after compensation higher order moments Z-score standardized value :

[0124] ;

[0125] in, This represents the average of the higher-order moments of the pressure signal corresponding to the cylinder under normal conditions. This represents the standard deviation of the higher-order moment of the pressure signal corresponding to the cylinder under normal conditions.

[0126] Based on the zero-mean signal of the compensated cylinder pressure higher order moments Z-score standardized value The health status of the cylinder shall be graded and assessed according to the following grading rules:

[0127] like and If so, the cylinder is normal;

[0128] like or If so, the cylinder will degenerate;

[0129] like or If so, then the cylinder is faulty.

[0130] Step 3: Based on the preprocessed valve inlet and outlet pressure signals, detect the valve's internal sealing status;

[0131] Optionally, step 3 includes the following sub-steps:

[0132] Sub-step 31: Analyze the zero-mean pressure signals at the valve inlet and outlet. Perform principal component-pressure correlation coefficient analysis;

[0133] Establish zero-mean signal for valve inlet and outlet pressure covariance matrix :

[0134] ;

[0135] Where n is the window length, yes The mean vector, , It is the inlet and outlet pressure signal at the m-th sampling time within the window.

[0136] Covariance matrix extracted by eigenvalue decomposition The two principal components with a cumulative variance of ≥95% are the first principal component PC1(t) and the second principal component PC2(t), where the variance of the first principal component PC1(t) is ≥80% and the variance of the second principal component PC2(t) is 15%~20%.

[0137] Calculate the first principal component PC1(t) and the zero mean value of the valve outlet pressure. correlation coefficient And the second principal component PC2(t) and the zero mean of the valve inlet pressure correlation coefficient :

[0138] ;

[0139] ;

[0140] in, This is the Pearson correlation coefficient.

[0141] like and Marked as normal association, if and Mark them as abnormal associations and record the correlation coefficients. The square value and correlation coefficient The square value of .

[0142] Sub-step 32: Analyze the zero-mean pressure signals at the valve inlet and outlet. Perform independent component-pressure mutual information analysis;

[0143] Zero-mean signal of valve inlet and outlet pressure PCA whitening is performed to obtain the whitened signal Z(t). The FastICA algorithm (using the nonlinear function tanh) is then used to optimize the whitened signal Z(t) through maximum likelihood estimation to obtain the separation matrix W. Independent sources are then separated from the separation matrix W. ,in, For leakage source signal, The signal is a noise source.

[0144] Calculate the mutual information between the leak source signal and the pressure difference. :

[0145] ;

[0146] in, For information entropy, This represents the difference between the zero-mean signals of the valve's inlet and outlet pressures, specifically the difference between the zero-mean valve inlet pressure and the zero-mean valve outlet pressure. .

[0147] Calculate the average value of the noise source signal and the zero-mean signal of the valve inlet and outlet pressures. correlation coefficient :

[0148] ;

[0149] like If the value is greater than 0.3 bits, it is marked as a source of leakage; otherwise, it is marked as no leakage.

[0150] like If the value is greater than 0.6, it is marked as abnormal noise interference; otherwise, it is marked as normal noise interference.

[0151] Sub-step 33: Analyze the zero-mean pressure signals at the valve inlet and outlet. Perform direct correlation coefficient analysis;

[0152] Calculate the autocorrelation function between the zero-mean difference ΔP(t) of the valve inlet and outlet pressure signals and the zero-mean difference ΔP(t+τ) of the valve inlet and outlet pressure signals at its own lag time τ. The autocorrelation function reflects the periodic fluctuations in pressure difference caused by internal leakage;

[0153] ;

[0154] in, for and covariance, for variance For time delay, ; for The variance.

[0155] Fitting autocorrelation function Exponential decay model:

[0156] ;

[0157] Obtain the attenuation coefficient ;

[0158] like If the decay is too slow, it is marked as decaying normally; otherwise, it is marked as decaying normally.

[0159] calculate autocorrelation function ,like If it is not internally leaking, mark it as internally leaking; otherwise, mark it as not internally leaking.

[0160] Sub-step 34: Fuse the internal leakage features, including principal component-pressure correlation coefficient, independent component-pressure mutual information and direct correlation coefficient, into a comprehensive feature vector F, and classify the internal sealing state of the valve based on the SVM model;

[0161] Specifically, the leakage features, including principal component-pressure correlation coefficient, independent component-pressure mutual information, and direct correlation coefficient, are fused into a comprehensive feature vector F. The comprehensive feature vector F is then expressed as:

[0162] ;

[0163] The decision function used by the SVM model to classify the internal sealing state of the valve is:

[0164] ;

[0165] in, For the classification results, The total number of samples, For Lagrange multipliers, For sample labels, Let g be the feature vector of the g-th support vector. For the index of the support vectors; It is a radial basis function kernel function; in this embodiment, the RBF kernel function is used. This is a bias. The decision function directly corresponds to the parameter values ​​of the fusion vector F. The input F is the current comprehensive feature vector, and the feature vector F of the support vectors is... g The support vectors are trained using the comprehensive feature vector F parameters of historical samples. In this embodiment, a total of 20 support vectors are set, and parameter-level similarity measurement is achieved through kernel functions to finally output the classification result.

[0166] When classifying the internal sealing state of a valve using a decision function, an SVM model (with an RBF kernel and a class weight ratio of 1:1) is first trained using the comprehensive feature vector of historical samples. The SVM model optimizes its hyperparameters through 5-fold cross-validation, with a penalty factor C1 ∈ [0.1, 100] and a kernel width parameter γ ∈ [0.01, 10]. Then, the comprehensive feature vector F of the current state is input into the trained SVM model, which ultimately outputs the classification result. This refers to the internal sealing state of the valve, which includes both sealed and unsealed states.

[0167] Step 4: Output the health status of the cylinder and the internal sealing status of the valve to obtain the intelligent detection results of the pneumatic valve status.

[0168] The intelligent detection results of pneumatic valve status include the health status of the cylinder: cylinder normal, cylinder degraded or cylinder faulty; and the internal sealing status of the valve: sealed or not sealed.

[0169] Example 2

[0170] This embodiment provides an intelligent detection system for the status of pneumatic valves on a test bench, used to implement the intelligent detection method for the status of pneumatic valves on a test bench in Embodiment 1. The intelligent detection system for the status of pneumatic valves on a test bench includes:

[0171] Sensor module: includes a pressure sensor group and a temperature sensor, with a sampling frequency ≥100Hz. The pressure sensor group includes pressure sensors (accuracy 0.1%FS, range 0~10MPa) respectively installed in the chambers on both sides of the piston in the cylinder, and pressure sensors (accuracy 0.1%FS, range 0~2MPa) installed at the valve inlet and valve outlet. The temperature sensor is used to acquire the ambient temperature signal of the cylinder (accuracy 0.5℃, range -20~80℃).

[0172] Data acquisition module: The input end connects to the output end of the sensor module and is used to synchronously acquire data from the above sensors and store it in a time-series database (supports historical data backtracking for ≥30 days).

[0173] Preprocessing module: The input end is connected to the output end of the data acquisition module, and it is used to preprocess the pressure signals of the cylinder piston on both sides, the valve inlet and outlet pressure signals, and the ambient temperature signals acquired by the data acquisition module.

[0174] Cylinder health assessment module: The input end is connected to the output end of the preprocessing module, and it is used to grade and assess the health status of the cylinder based on the pressure signals of the cylinder chambers on both sides of the piston and the ambient temperature signals after preprocessing.

[0175] Valve sealing status module: The input end is connected to the output end of the preprocessing module, and it is used to detect the internal sealing status of the valve based on the preprocessed valve inlet and outlet pressure signals.

[0176] Output module: Used to output the health status of the cylinder and the internal sealing status of the valve, obtain intelligent detection results of the pneumatic valve status, and support threshold alarm and remaining life prediction (error ≤10%).

[0177] Simulation Experiment 1: This simulation experiment uses the cylinder of a pneumatic ball valve (cylinder volume 0.002m³, working pressure 5MPa) as an example. The health status of the cylinder is assessed using the method described in Example 1, specifically including the following steps:

[0178] 1. Pressure sensors (sampling rate 100Hz) are installed in the cavities on both sides of the piston in the cylinder, and temperature sensors are installed near the cylinder to collect pressure signals and ambient temperature signals in the cavities on both sides of the piston in the cylinder, respectively.

[0179] 2. Preprocessing: A Savitzky-Golay filter (window length 51 points, order 3) was used to filter and reduce noise in the pressure signals in the cylinder on both sides of the piston (the pressure noise amplitude was reduced from 0.05MPa to 0.01MPa).

[0180] 3. Calculation of higher-order moments: With a rolling window length of u = 2s, calculate the third and fourth moments of a certain cylinder movement as follows: =0.12, =0.05, =0.13, =2.97.

[0181] 4. Temperature compensation: such as Figures 1 to 4 As shown, the fitted linear regression model is as follows:

[0182] =0.02 +0.001+ ( =0.98, p<0.1), after compensation In the calculation of the corresponding Z-score standardized value, the mean of the higher-order moments The standard deviation of higher-order moments is 0.2. It is 0.8.

[0183] =0.028 +0.001+ ( =0.97, p<0.1), after compensation In the calculation of the corresponding Z-score standardized value, the mean of the higher-order moments The standard deviation of higher-order moments is 0.3. It is 0.7.

[0184] =0.019 +0.001+ ( =0.98, p<0.1), after compensation In the calculation of the corresponding Z-score standardized value, the mean of the higher-order moments The standard deviation of higher-order moments is 0.2. It is 0.8.

[0185] =0.029 +0.001+ ( =0.98, p<0.1), after compensation In the calculation of the corresponding Z-score standardized value, the mean of the higher-order moments The standard deviation of higher-order moments is 0.4. It is 0.9.

[0186] At 25℃, the compensated higher-order moments are:

[0187] ;

[0188] ;

[0189] ;

[0190] =|(-0.651-0.3) / 0.7|=0.951;

[0191] ;

[0192] =|(-0.346-0.2) / 0.8|=0.683;

[0193] ;

[0194] =|(2.244-0.4) / 0.9|=2.049;

[0195] because If the cylinder is determined to be degraded, in actual operation, if it is determined to be degraded 5 times in a row, it needs to be repaired.

[0196] Simulation Experiment 2: This simulation experiment uses a pneumatic shut-off valve (sealing pressure 1MPa, volume) , Taking a valve as an example, the method in Example 1 is used to detect the internal sealing state of the valve, specifically including the following steps:

[0197] 1. Obtain the valve inlet pressure of the pneumatic shut-off valve. and valve outlet pressure After the valve is closed, continuously collect the valve inlet pressure and valve outlet pressure for 30 minutes. Example data: Range 0.98~1.02MPa The range is 0.98~1.02 MPa; and the collected valve inlet pressure is... and valve outlet pressure Preprocessing is required;

[0198] 2. Principal Component-Pressure Correlation Coefficient Analysis: The first principal component PC1(t) and the zero mean of the valve outlet pressure correlation coefficient =0.45, its square =0.20, the second principal component PC2(t) and the zero mean of the valve inlet pressure correlation coefficient =0.40, its square =0.16;

[0199] 3. Independent component-pressure mutual information analysis: =0.32 bits (>0.3 bits) =0.55 (<0.6), confirming the existence of a leak source;

[0200] 4. Direct Correlation Coefficient Analysis:

[0201] 4.1. Calculation of Autocorrelation Function: Calculate the autocorrelation coefficient of the pressure value. ;

[0202] ;

[0203] In this embodiment, α is approximately ;

[0204] 4.2. Anomaly Detection:

[0205] Under normal operating conditions, It decays exponentially with increasing τ;

[0206] When internal leakage occurs, the periodic fluctuations caused by the leakage will lead to The decay slows down; at τ=3s, The value remains at 0.35 (under normal operating conditions for this valve). =0.12);

[0207] 5. SVM classification decision:

[0208] The comprehensive feature vector is obtained based on the above calculations:

[0209] ;

[0210] Output "Unsealed".

[0211] because (Abnormal association) (Abnormal association);

[0212] 0.32 bits > 0.3 bits (a leak source exists);

[0213] =0.55<0.6 (Noise interference is normal);

[0214] (The decay is too slow) (Internal leakage).

[0215] All key parameters point to "unsealed," consistent with the "unsealed" output of the SVM model, thus verifying the effectiveness of the detection method in this invention.

Claims

1. A method for intelligent detection of the status of pneumatic valves on a test bench, characterized in that, Includes the following steps: Data acquisition and preprocessing; the data includes pressure signals from the chambers on both sides of the piston in the cylinder, pressure signals from the valve inlet and outlet, and ambient temperature signals; preprocessing specifically involves processing the acquired pressure signals from the chambers on both sides of the piston in the cylinder. Valve inlet and outlet pressure signals After denoising and centering, the zero-mean signal of cylinder pressure is obtained. Zero-mean signal of valve inlet and outlet pressure ; Based on the pre-processed pressure signals from both sides of the piston chamber in the cylinder and the ambient temperature signal, the health status of the cylinder is graded and assessed; specifically: Calculate the zero-mean signal of cylinder pressure higher-order moments; Establish cylinder pressure zero-mean signal A linear regression model of higher-order moments and ambient temperature; Based on the ambient temperature signal, a linear regression model is used to analyze the zero-mean cylinder pressure signal. Compensation is performed on higher-order moments; Based on the zero-mean signal of the compensated cylinder pressure The higher-order moments are obtained by compensating the zero-mean signal of the cylinder pressure. The Z-score normalized value of the higher-order moment is used to classify and evaluate the health status of the cylinder. Based on the pre-processed valve inlet and outlet pressure signals, the internal sealing status of the valve is detected. The health status of the output cylinder and the internal sealing status of the valve are used to obtain the intelligent detection results of the pneumatic valve status.

2. The intelligent detection method for the status of pneumatic valves on a test bench according to claim 1, characterized in that, The data collection includes the following steps: Synchronously acquire pressure signals from both sides of the piston chamber in the cylinder. Valve inlet and outlet pressure signals and ambient temperature signal ; Pressure signals in the piston chambers on both sides of the cylinder Including cylinder cavity I pressure and cylinder cavity II pressure The valve inlet and outlet pressure signals Including valve inlet pressure and valve outlet pressure .

3. The intelligent detection method for the status of pneumatic valves on a test bench according to claim 2, characterized in that: The cylinder pressure zero-mean signal Higher-order moments include third-order moments and fourth moment , ； ； Where u is the length of the scroll window, and Δt is the step size. Let be the zero-mean cylinder pressure at the i-th second, where i takes the values ​​t, t+1, ... ; These correspond to cylinder cavity I and cylinder cavity II, respectively; The mean of the zero-mean cylinder pressure. The standard deviation of the zero mean cylinder pressure; The cylinder pressure zero-mean signal The linear regression model of the higher-order moments and ambient temperature is expressed as: ； in, Zero-mean signal of cylinder pressure The higher-order moments are given by , where j is the index of the higher-order moment, taking a value of 1 or 2, corresponding to the third and fourth-order moments respectively. , These are the linear parameters of the linear regression model. This is random noise related to ambient temperature; The zero-mean signal of the compensated cylinder pressure higher order moments Represented as: ； The specific steps for grading and assessing the health status of cylinders are as follows: calculate Z-score standardized value ; ； in, This represents the average of the higher-order moments of the pressure signal corresponding to the cylinder under normal conditions. The standard deviation of the higher-order moment of the pressure signal corresponding to the cylinder under normal conditions; according to The health status of the cylinder shall be graded and assessed according to the following grading rules: like and If so, the cylinder is normal; like or If so, the cylinder will degenerate; like or If so, then the cylinder is faulty.

4. The intelligent detection method for the status of pneumatic valves on a test bench according to claim 3, characterized in that: Based on the preprocessed valve inlet and outlet pressure signals, the specific operation for detecting the internal sealing status of the valve includes the following steps: Zero-mean signal of valve inlet and outlet pressure Perform principal component-pressure correlation coefficient analysis; Zero-mean signal of valve inlet and outlet pressure Perform independent component-pressure mutual information analysis; Zero-mean signal of valve inlet and outlet pressure Perform direct correlation coefficient analysis; The internal leakage features, including principal component-pressure correlation coefficient, independent component-pressure mutual information, and direct correlation coefficient, are fused into a comprehensive feature vector F, which is then used to classify the internal sealing state of the valve based on the SVM model.

5. The intelligent detection method for the status of pneumatic valves on a test bench according to claim 4, characterized in that, The principal component-pressure correlation coefficient analysis specifically refers to: Establish zero-mean signal for valve inlet and outlet pressure covariance matrix Zero-mean signal for inlet and outlet pressure Including zero mean valve inlet pressure and valve outlet pressure zero mean ; Covariance matrix extracted by eigenvalue decomposition The two principal components with a cumulative variance of ≥95% are the first principal component PC1(t) and the second principal component PC2(t). Calculate the first principal component PC1(t) and the zero mean value of the valve outlet pressure. correlation coefficient And the second principal component PC2(t) and the zero mean of the valve inlet pressure correlation coefficient ; like and Marked as normal association, if and Mark them as abnormal associations and record the correlation coefficients. The square value and correlation coefficient The square value of .

6. The intelligent detection method for the status of pneumatic valves on a test bench according to claim 5, characterized in that, The independent component-pressure mutual information analysis specifically refers to: Zero-mean signal of valve inlet and outlet pressure PCA whitening is performed to obtain the whitened signal Z(t). The FastICA algorithm is then used to optimize the whitened signal Z(t) through maximum likelihood estimation to obtain the separation matrix W. Independent sources are then separated from the separation matrix W. ,in, For leakage source signal, The signal is a noise source. Calculate the mutual information between the leak source signal and the pressure difference. ,in, The difference between the zero mean pressure signals at the valve inlet and outlet; Calculate the average value of the noise source signal and the zero-mean signal of the valve inlet and outlet pressures. correlation coefficient ; like If the value is greater than 0.3 bits, it is marked as a leak source; otherwise, it is marked as no leak. like If the value is >0.6, it is marked as abnormal noise interference; otherwise, it is marked as normal noise interference.

7. The intelligent detection method for the status of pneumatic valves on a test bench according to claim 6, characterized in that, The direct correlation coefficient analysis specifically includes: Calculate the autocorrelation function between the zero-mean difference ΔP(t) of the valve inlet and outlet pressure signals and the zero-mean difference ΔP(t+τ) of the valve inlet and outlet pressure signals at its own lag time τ. ; ； in, for and covariance, for variance For time delay, ; for The variance; Fitting autocorrelation function Exponential decay model: ； Obtain the attenuation coefficient ; If α ≤ 0.5 s -1 If the decay is too slow, it is marked as decaying normally; otherwise, it is marked as decaying normally. calculate ,like If it is not internally leaking, mark it as internally leaking; otherwise, mark it as not internally leaking.

8. The intelligent detection method for the status of pneumatic valves on a test bench according to claim 7, characterized in that, The comprehensive feature vector F is represented as: ； In the SVM model, the decision function for classifying the internal sealing state of the valve is: ； in, For the classification result, G is the total number of samples. For Lagrange multipliers, For sample labels, Let g be the feature vector of the g-th support vector. For the index of the support vectors; For kernel function, For bias; The SVM model is trained using historical samples and outputs the internal sealing state of the valve based on the comprehensive feature vector F, including two states: sealed and unsealed.

9. A test bench pneumatic valve status intelligent detection system, used to implement the test bench pneumatic valve status intelligent detection method according to any one of claims 1-8, characterized in that: It includes a sensor module, a data acquisition module, a preprocessing module, a cylinder health assessment module, a valve sealing status module, and an output module; The sensor module includes a pressure sensor group and a temperature sensor. The pressure sensor group includes pressure sensors respectively installed in the cavities on both sides of the piston in the cylinder, and pressure sensors installed at the valve inlet and valve outlet. The temperature sensor is used to acquire the ambient temperature signal of the cylinder. The input end of the data acquisition module is connected to the output end of the sensor module, and is used to synchronously acquire the pressure signals of the cylinder on both sides of the piston, the valve inlet and outlet pressure signals, and the ambient temperature signal. The input end of the preprocessing module is connected to the output end of the data acquisition module, and is used to preprocess the pressure signals of the cylinder piston on both sides, the valve inlet and outlet pressure signals, and the ambient temperature signals acquired by the data acquisition module. The input end of the cylinder health assessment module is connected to the output end of the preprocessing module, and is used to classify and assess the health status of the cylinder based on the pressure signals of the cylinder chambers on both sides of the piston and the ambient temperature signals after preprocessing. The input end of the valve sealing status module is connected to the output end of the preprocessing module, and is used to detect the internal sealing status of the valve based on the preprocessed valve inlet and outlet pressure signals. The output module is used to output the health status of the cylinder and the internal sealing status of the valve, and to obtain and output the intelligent detection results of the pneumatic valve status.

Images