Fault gene-based engine degradation trend extraction method

Abstract

The engine degradation trend extraction method based on the fault gene solves the problems that an existing rotating machine degradation trend extraction method is cross in fitting, low in prediction accuracy, low in speed and incapable of meeting the prediction lead, and the engine degradation trend extraction method based on the fault gene is provided. Based on a fault gene technology, root causes and characteristics of potential faults are analyzed, a trend prediction method and a fault model are deeply combined, and a deeper reference basis is provided for trend prediction; according to the invention, through two groups of signal processing methods, feature information in fault signals is comprehensively captured, and the identification and analysis capability of different fault modes is improved. According to the method, a set of trend model identification criteria is provided and comprises a standardized process of data acquisition and processing, an index system of model evaluation and verification and a strategy of model updating and optimization, so that the system can identify and match the trend model more effectively, and the efficiency and reliability of trend model establishment are comprehensively improved.

Description

Technical Field

[0001] The invention belongs to the field of rotating machinery health management, and in particular relates to a method for extracting engine degradation trends based on fault genes. Background Art

[0002] Affected by the uncertainty of various factors such as the operating environment and workload, rotating machinery and equipment will gradually experience functional degradation and structural wear during use, which will in turn affect their operating efficiency, safety and life. By analyzing and mining the monitoring data of rotating machinery, the key features and trends in the degradation process of mechanical equipment can be extracted, and degradation anomalies can be detected at an early stage, and corresponding repair and maintenance measures can be taken. Therefore, the establishment of a method for extracting the degradation trend of rotating machinery is an important part of achieving the health management of rotating machinery. However, in the implementation of related projects, it was found that the current extraction of degradation trends has the following problems:

[0003] 1. When extracting the degradation trend of rotating machinery, it is necessary to select a suitable trend model. Different rotating machinery systems may have different degradation patterns. Currently, the selection of a suitable model mostly relies on the accumulation of knowledge and experience, which may lead to inadequate consideration of the complexity and applicability of the model.

[0004] 2. When analyzing the degradation trend of rotating machinery, it is necessary to combine the working principle, operating mechanism, structural characteristics and other influencing factors of the rotating machinery system, and associate the trend with the long-term operating status and performance of the equipment. At present, the analysis and interpretation of degradation trends are relatively weak, and the long-term trend model may not be well fitted.

[0005] 3. When predicting the degradation trend of rotating machinery, if the data selected is too large or too small, it will be unfavorable to predict the degradation trend. If it is too large, the data will be huge and the training time will be long, and if it is too small, the prediction accuracy will be inaccurate. At present, there is a lack of data quality evaluation methods, which may lead to the inability to guarantee the accuracy and speed of prediction.

[0006] 4. The prediction of the degradation trend of rotating machinery needs to consider not only the prediction accuracy but also the prediction lead time. Assuming the sampling rate f s If the frequency is 20kHz, the sampling interval T = 0.05ms. If vibration prediction is performed based on the original data, even if the prediction step is 100 steps, the prediction lead time is only 5ms, which is too small to meet the warning requirements and has little significance in practical applications.

[0007] The patent application with application number CN115114770-A discloses a baseline adaptive auxiliary power unit performance trend analysis method, which constructs a characteristic formula to evaluate the similarity between different degradation feature samples and the standard degradation pattern library model, and performs degradation feature extraction on the exhaust gas temperature (EGT) data corresponding to the auxiliary power unit (APU) throughout the life cycle, thereby obtaining the APU performance deviation, health state degradation and failure time. However, the fault features considered in this invention are not comprehensive enough, and it is impossible to simultaneously consider the impact of multiple mode failures on the health state, which may lead to low accuracy of the long-term degradation trend model.

[0008] The patent application with application number CN114139767-A discloses a method, system, computer device and storage medium for predicting the health trend of equipment. The method uses the empirical mode decomposition method to extract features from the vibration data of the measured equipment within a preset time period to obtain the energy entropy matrix of the data. Based on the energy entropy matrix, the health index is determined, the health trend of the measured equipment is predicted, and finally the remaining service life of the equipment is obtained. However, the patent does not provide a set of standardized processes, and the efficiency of establishing long-term degradation trends is not high.

[0009] The patent application with application number CN202310198906.8 discloses a generalized degradation trend tracking method for predicting the remaining service life of rotating machinery. Starting from the vibration data of rotating machinery, this method provides a feature learning method for generalized degradation trend tracking for predicting the remaining service life of rotating machinery. However, this method is not deeply integrated with the fault physical model, resulting in low accuracy in establishing the long-term trend model.

[0010] In summary, the current method for extracting degradation trends of rotating machinery is difficult to select a suitable trend model in the implementation of related projects. There are problems such as poor fitting of long-term trend models, lack of guarantee of prediction accuracy and speed, and inability of prediction lead time to meet early warning needs. Summary of the invention

[0011] In view of the problems existing in the prior art, the present invention proposes a method for extracting engine degradation trends based on fault genes.

[0012] The technical solution of the present invention is:

[0013] The method for extracting engine degradation trend based on fault genes is special in that it includes the following steps:

[0014] Step 1: Based on the target model engine parameters, read the <fault mode> in the engine fault gene library;

[0015] Step 2: According to the <failure mode> in the fault gene of the engine model, read the <long-term trend model> of all potential faults, and list all possible trend models as a trend model list;

[0016] Step 3: Extract the long-term vibration trend data extracted by the airborne sensor, use the envelope analysis method to obtain the envelope curve, and then process the signal through two sets of standard algorithms respectively, and save the processed signal characteristics;

[0017] Step 4: Based on the engine trend data extracted in step 3, model identification is performed one by one in comparison with the trend model list in step 2 to generate candidate models;

[0018] Step 5: Select three most similar degradation trend models from the candidate models based on the degree of feature similarity;

[0019] Step 6: Call the threshold in the <fault manifestation> segment of the fault gene library as the threshold for setting the vibration characteristic amplitude, and judge the three most similar degradation trend models listed in step 5:

[0020] If the vibration characteristic amplitude of the listed model does not exceed the threshold, the three extracted "degradation trend models" are added to the <unknown fault mode> segment of the engine fault gene of this model, and go to step 4;

[0021] If the vibration characteristic amplitude of the fault characteristic indicated by <fault mode> in the fault gene of the engine exceeds the threshold, it is determined that the degradation trend of the engine model is extracted successfully, and the trend analysis result is output.

[0022] Furthermore, the <long-term trend model> of the potential fault includes three standard trends: linear, power and polynomial.

[0023] Furthermore, the step 3 includes the following sub-steps:

[0024] Step 3.1: Use wavelet transform method to reduce noise of the extracted airborne data;

[0025] Step 3.2: Use envelope analysis to obtain the envelope curve, then use EMD decomposition to process the signal, and save the processed signal characteristics;

[0026] Step 3.3: Use envelope analysis to obtain the envelope curve, then use VMD decomposition to process the signal and save the processed signal characteristics.

[0027] Furthermore, the step 3.1 includes the following sub-steps:

[0028] Step 3.1.1: Preprocess the raw data of <long-term trend model>;

[0029] Step 3.1.2: Use wavelet transform to perform multi-scale decomposition;

[0030] Step 3.1.3: De-noise the wavelet coefficients at each scale decomposed in step 3.1.2 by setting a threshold;

[0031] Step 3.1.4: Use inverse wavelet transform to reconstruct the signal at the denoising backend into a denoised signal.

[0032] Furthermore, the step 3.2 includes the following sub-steps:

[0033] Step 3.2.1: Calculate the envelope of the signal and find the average envelope;

[0034] Step 3.2.2: Calculate the intermediate signal by subtracting the envelope mean from the signal;

[0035] Step 3.2.3: Judge the intermediate signal:

[0036] If the intermediate signal satisfies the first condition of IMF, that is, the sum of the number of local maxima and local minima must be equal to the number of zero crossings or differ by at most one, then the intermediate signal is regarded as an IMF component. j , go to step 3.2.4;

[0037] Otherwise, go to step 3.3.1;

[0038] Step 3.2.4: Separate the IMF component from the original denoised signal by subtracting the IMF component from the signal;

[0039] Step 3.2.5: Judge the separated signal:

[0040] If the second condition of IMF is met, that is, the mean of the upper envelope defined by the local maximum and the lower envelope defined by the local minimum is zero, then the separated signal is taken as the original signal and go to step 3.3.1;

[0041] If the separated signal r n If is a monotonic function or a constant, go to step 3.3.6;

[0042] Otherwise, go to step 3.2.4;

[0043] Step 3.2.6: Decompose the original denoised signal into empirical mode decomposition results.

[0044] Furthermore, the step 3.3 includes the following sub-steps:

[0045] Step 3.3.1: Perform Hilbert transform on the signal to obtain the analytical signal of each mode;

[0046] Step 3.3.2: Based on the mixed-estimated center frequency of each modal analytical signal, the analytical signal of each mode is modulated and shifted to the corresponding baseband;

[0047] Step 3.3.3: Estimate the bandwidth of the signal using the H1 Gaussian smoothness method;

[0048] Step 3.3.4: Introduce the Lagrange multiplication operator λ to transform the constrained variational problem into an unconstrained variational problem and obtain the augmented Lagrange expression;

[0049] Step 3.3.5: Use the alternating direction multiplier (ADMM) algorithm combined with Parseval / Plancherel and Fourier isometric transform to optimize the components and center frequencies of each mode and search for the saddle point of the augmented Lagrange function;

[0050] Step 3.3.6: Iterate and solve the signal after alternating optimization.

[0051] Further,

[0052] In step 3.3.5, the three coefficient expressions in the augmented Lagrange expression after the alternating optimal iteration are:

[0053]

[0054] Where: γ is the noise tolerance, which meets the fidelity requirements of signal decomposition; and Corresponding to and Fourier transform of

[0055] The iteration order of step 3.3.6 is:

[0056] Step 3.3.6.1: Initialization λ 1 And the maximum number of iterations N, and n←0;

[0057] Step 3.3.6.2: Use Formula and Update the coefficient expression in the augmented Lagrange expression and ωk;

[0058] Step 3.3.6.3: Use Update

[0059] Step 3.3.6.4: Use the accuracy convergence criterion ε>0 to judge:

[0060] If not satisfied and n<N, return to step 3.3.6.2;

[0061] Otherwise, the iteration is completed and the output is and ωk.

[0062] Concept and principle of the present invention:

[0063] Since the working state of the engine is complex and changeable, and involves multiple signal types and parameters, it is difficult to identify the characteristic information of different fault modes.

[0064] like Figure 1 As shown, the present invention first analyzes the root causes and characteristics of potential faults based on fault gene technology, deeply combines the trend prediction method with the fault model, grasps the correlation between different fault modes, provides a more in-depth reference basis for trend prediction, and improves the accuracy of prediction;

[0065] Secondly, the present invention combines three standard algorithms into two groups of signal processing methods to comprehensively capture the characteristic information in the fault signal, improve the ability to identify and analyze different fault modes, and provide more effective support for fault prediction and maintenance.

[0066] Finally, the present invention proposes a set of trend model identification criteria to enable the system to more effectively identify and match trend models, improve modeling efficiency and ensure the quality of the model. These criteria include standardized processes for data collection and processing, an indicator system for model evaluation and verification, and strategies for model updating and optimization, thereby comprehensively improving the efficiency and reliability of trend model establishment.

[0067] The advantages of the present invention are:

[0068] 1. In the process of trend prediction, the present invention is based on the fault gene theory, combines historical data, real-time monitoring data, structural characteristics and other information, and uses data mining methods to predict the development trend of fault modes, thereby improving the ability to accurately predict potential faults of specific models, eliminating irrelevant features in the pre-processed vibration trend data, increasing the data interval, and improving the prediction lead time.

[0069] 2. The degradation trend extraction method established by this method directly matches the trend in the standard trend library, and determines the optimal maintenance strategy through iteration with less time cost, so as to help the health management system to formulate a reasonable maintenance plan according to the actual operating status and degradation trend of the equipment, avoid excessive maintenance or insufficient maintenance, reduce maintenance costs, and improve maintenance efficiency and equipment utilization.

[0070] 3. The present invention ensures the accuracy of prediction by selecting a suitable trend model, ensures rapidity by quickly matching in a standard trend library, reduces the prediction lead time to meet the early warning requirements, and provides more effective support for fault prediction and maintenance.

[0071] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0073] Figure 1 Flowchart of the degradation trend extraction method;

[0074] Figure 2 This is the flow chart of wavelet denoising;

[0075] Figure 3 Decompose the flow chart for EMD;

[0076] Figure 4 Decompose the flow chart for VMD;

[0077] Figure 5 is the original input signal diagram;

[0078] Figure 6 It is the EMD decomposition result;

[0079] Figure 7 It is the VEM decomposition result;

[0080] Figure 8 These are three standard trend model diagrams in the standard trend library. DETAILED DESCRIPTION

[0081] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0082] This embodiment proposes a method for extracting engine degradation trends by using a fault gene method. This method models the engine degradation process based on historical data and predicts the trend of future engine performance changes, thereby achieving early warning of engine performance degradation and meeting the next generation of engine condition-based maintenance needs. The specific steps are:

[0083] Step 1: Based on the target model engine parameters, read the <fault mode> in the engine fault gene library;

[0084] Step 2: According to the <failure mode> in the fault gene of the engine model, read the <long-term trend model> of all potential faults, and list all possible trend models as a trend model list, as shown in Table 1:

[0085] Table 1 Trend model list

[0086] serial number Trend Model 1 Linear Trend 2 Power Trend 3 Polynomial Trend

[0087] Step 3: Extract the long-term vibration trend data extracted by the airborne sensor, use the envelope analysis method to obtain the envelope curve, and then process the signal through two sets of standard algorithms respectively, and save the processed signal characteristics;

[0088] Envelope analysis is a method that extracts the envelope of the signal to separate the fast oscillation part from the slow change part of the original signal, which helps to better understand and analyze the signal characteristics.

[0089] Step 3.1: If Figure 2 As shown, the wavelet transform method is used to reduce the noise of the extracted airborne data;

[0090] Step 3.1.1: Preprocess the raw data of <long-term trend model>;

[0091] The input original signal of this embodiment is as follows Figure 5 shown.

[0092] Step 3.1.2: Use wavelet transform to perform multi-scale decomposition;

[0093] Step 3.1.3: De-noise the wavelet coefficients at each scale decomposed in step 3.1.2 by setting a threshold;

[0094] Step 3.1.4: Use inverse wavelet transform to reconstruct the signal at the denoising backend into a noise-reduced signal;

[0095] Step 3.2: If Figure 3 As shown, the envelope analysis method is used to obtain the envelope curve, and then the EMD decomposition is used to process the signal, and the processed signal characteristics are saved;

[0096] EMD (Empirical Mode Decomposition) is a method to decompose a complex signal into several intrinsic mode functions (IMFs). These IMFs are functions with local characteristics that can reflect the vibration characteristics of different frequency components in the signal.

[0097] It includes the following sub-steps:

[0098] Step 3.2.1: Calculate the envelope of the signal and find the average envelope;

[0099]

[0100] Where x(t) is the original denoised signal, H[] is the Hilbert transform, and x0(t) is the average envelope;

[0101] In this embodiment, the three original signals after noise reduction in Table 1 are fitted with the maximum and minimum points of the original signals by spline interpolation to form upper and lower envelopes, and the average envelopes of the three signals are calculated;

[0102] Step 3.2.2: Calculate the intermediate signal by subtracting the envelope mean from the signal;

[0103] h1(t)=x(t)-x0(t)

[0104] Among them, h1(t) is the intermediate signal;

[0105] Step 3.2.3: Judge the intermediate signal:

[0106] If the intermediate signal satisfies the first condition of IMF, that is, the sum of the number of local maxima and local minima must be equal to the number of zero crossings or differ by at most one, then the intermediate signal is regarded as an IMF component. j , go to step 3.2.4;

[0107] Otherwise, go to step 3.2.1;

[0108] Step 3.2.4: Separate the IMF component from the original denoised signal by subtracting the IMF component from the signal;

[0109] Step 3.2.5: Determine the separated signal:

[0110] If the second condition of IMF is met, that is, the mean of the upper envelope defined by the local maximum and the lower envelope defined by the local minimum is zero, then the separated signal is taken as the original signal and go to step 3.3.1;

[0111] If the separated signal r n If is a monotonic function or a constant, go to step 3.3.6;

[0112] Otherwise, go to step 3.2.4;

[0113] Step 3.2.6: Decompose the original denoised signal into empirical mode decomposition results;

[0114]

[0115] Among them, r n is the signal after the nth separation, n is the number of IMF component signals;

[0116] The three signals in this embodiment all obtain 5 IMF components and the signal central trend r n , so the expressions of the three signals after decomposition are:

[0117]

[0118] The original noise reduction signal of this embodiment is processed by EMD decomposition and the result is as follows: Figure 6 shown.

[0119] Step 3.3: Use envelope analysis to obtain the envelope curve, then use VMD decomposition to process the signal, and save the processed signal characteristics;

[0120] VMD (Variational Mode Decomposition) is a signal decomposition method based on optimization theory. It decomposes the signal into multiple vibration modes by solving the optimization problem. VMD can effectively process non-stationary and nonlinear signals and has a good mathematical foundation and strong numerical stability.

[0121] VMD assumes that all components are narrowband signals concentrated near their respective center frequencies. Therefore, VMD establishes a constrained optimization problem based on the component narrowband condition to estimate the center frequency of the signal component and reconstruct the corresponding component.

[0122] VMD processes the signal in three steps. The process is as follows Figure 4 As shown, it includes the following sub-steps:

[0123] Step 3.3.1: Perform Hilbert transform on the signal to obtain the analytical signal of each mode;

[0124] Step 3.3.2: Based on the mixed-estimated center frequency of each modal analytical signal, the analytical signal of each mode is modulated and shifted to the corresponding baseband;

[0125] Step 3.3.3: Estimate the bandwidth of the signal using the H1 Gaussian smoothness method;

[0126]

[0127] Where K is the number of modes to be decomposed (a positive integer), {u k}、{ω k} respectively correspond to the kth modal component and center frequency after decomposition, δ(t) is the Dirac function, and * is the convolution operator;

[0128] Step 3.3.4: Introduce the Lagrange multiplication operator λ to transform the constrained variational problem into an unconstrained variational problem and obtain the augmented Lagrange expression:

[0129]

[0130] In the formula, α is the quadratic penalty factor, which is used to reduce the interference of Gaussian noise;

[0131] Step 3.3.5: Use the alternating direction multiplier (ADMM) algorithm combined with Parseval / Plancherel and Fourier isometric transform to optimize the components and center frequencies of each mode and search for the saddle point of the augmented Lagrange function;

[0132] u after alternately searching for optimal k ,ω k The expressions for and λ are as follows:

[0133]

[0134] Where: γ is the noise tolerance, which meets the fidelity requirements of signal decomposition; and Corresponding to and Fourier transform of

[0135] Step 3.3.6: Iterate the signal after alternating optimization, and the iteration order is:

[0136] Step 3.3.6.1: Initialization λ 1 And the maximum number of iterations N, and n←0;

[0137] Step 3.3.6.2: Use equations (1) and (2) to update the coefficient expressions in the augmented Lagrange expression and ωk;

[0138] Step 3.3.6.3: Update using formula (3)

[0139] Step 3.3.6.4: Use the accuracy convergence criterion ε>0 to judge:

[0140] If not satisfied and n<N, return to step 3.3.6.2;

[0141] Otherwise, the iteration is completed and the output is and ωk.

[0142] The three original noise reduction signals of this embodiment are decomposed by VMD and the obtained results are as follows: Figure 7shown.

[0143] Step 4: Based on the engine trend data extracted in step 3.2 and step 3.3, model identification is performed one by one against the trend model list in step 2 to generate candidate models;

[0144] First, extract the envelope features of the original denoised signal decomposed from step 3.2 to step 3.3, and then compare the obtained IMF frequency domain features and each point value of the IMF time domain features with the following: Figure 8 The corresponding standard trend model features in the standard trend library shown are compared one by one by constructing the Pearson correlation coefficient:

[0145] When the similarity between two features reaches 90%, the trend model is extracted and listed as an alternative model;

[0146] Step 5: Select three most similar degradation trend models from the candidate models based on the degree of feature similarity;

[0147] The three most similar degradation trend models selected from the candidate models in this embodiment are:

[0148] Table 2: List of the three most similar degradation trend models

[0149] Sort order Similarity Trend Model 1 94.32％ Polynomial Trend 2 84.81％ Power Trend 3 88.24％ Linear Trend

[0150] Step 6: Call the threshold in the <fault manifestation> segment of the fault gene library as the threshold for setting the vibration characteristic amplitude, and judge the three most similar degradation trend models listed in step 5:

[0151] If the vibration characteristic amplitude of the listed model does not exceed the threshold, the three extracted "degradation trend models" are added to the <unknown fault mode> segment of the engine fault gene of this model, and go to step 4;

[0152] If the vibration characteristic amplitude of the fault characteristic indicated by <fault mode> in the fault gene of the engine exceeds the threshold, it is determined that the degradation trend of the engine model is extracted successfully, and the trend analysis result is output.

[0153] 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 various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should be included in the protection scope of the present invention.

Claims

1. A method for extracting engine degradation trends based on fault genes, characterized in that: The following steps are involved: Step 1: Based on the target model engine parameters, read the <fault mode> in the engine fault gene library; Step 2: According to the <failure mode> in the fault gene of the engine model, read the <long-term trend model> of all potential faults, and list all possible trend models as a trend model list; Step 3: Extract the long-term vibration trend data extracted by the airborne sensor, use the envelope analysis method to obtain the envelope curve, and then process the signal through two sets of standard algorithms respectively, and save the processed signal characteristics; Step 4: Based on the engine trend data extracted in step 3, model identification is performed one by one in comparison with the trend model list in step 2 to generate candidate models; Step 5: Select three most similar degradation trend models from the candidate models based on the degree of feature similarity; Step 6: Call the threshold in the <fault manifestation> segment of the fault gene library as the threshold for setting the vibration characteristic amplitude, and judge the three most similar degradation trend models listed in step 5: If the listed model vibration characteristic amplitude does not exceed the threshold, the three extracted "degradation trend models" are added to the <unknown fault mode> segment of the engine fault gene of this model, and go to step 4; If the vibration characteristic amplitude of the fault characteristic indicated by <fault mode> in the fault gene of the engine exceeds the threshold, it is determined that the degradation trend of the engine model is extracted successfully, and the trend analysis result is output.

2. The method for extracting engine degradation trend based on fault genes according to claim 1, characterized in that: The <long-term trend model> of the potential fault includes three standard trends: linear, power and polynomial.

3. The method for extracting engine degradation trend based on fault genes according to claim 1, characterized in that: The step 3 includes the following sub-steps: Step 3.1: Use wavelet transform method to reduce noise of the extracted airborne data; Step 3.2: Use envelope analysis to obtain the envelope curve, then use EMD decomposition to process the signal, and save the processed signal characteristics; Step 3.3: Use envelope analysis to obtain the envelope curve, then use VMD decomposition to process the signal and save the processed signal characteristics.

4. The method for extracting engine degradation trend based on fault genes according to claim 3, characterized in that: The step 3.1 includes the following sub-steps: Step 3.1.1: Preprocess the raw data of <long-term trend model>; Step 3.1.2: Use wavelet transform to perform multi-scale decomposition; Step 3.1.3: De-noise the wavelet coefficients at each scale decomposed in step 3.1.2 by setting a threshold; Step 3.1.4: Use inverse wavelet transform to reconstruct the signal at the denoising backend into a denoised signal.

5. The method for extracting engine degradation trend based on fault genes according to claim 4, characterized in that: The step 3.2 includes the following sub-steps: Step 3.2.1: Calculate the envelope of the signal and find the average envelope; Step 3.2.2: Calculate the intermediate signal by subtracting the envelope mean from the signal; Step 3.2.3: Judge the intermediate signal: If the intermediate signal satisfies the first condition of IMF, that is, the sum of the number of local maxima and local minima must be equal to the number of zero crossings or differ by at most one, then the intermediate signal is regarded as an IMF component. j , go to step 3.2.4; Otherwise, go to step 3.3.1; Step 3.2.4: Separate the IMF component from the original denoised signal by subtracting the IMF component from the signal; Step 3.2.5: Judge the separated signal: If the second condition of IMF is met, that is, the mean of the upper envelope defined by the local maximum and the lower envelope defined by the local minimum is zero, then the separated signal is taken as the original signal and go to step 3.3.1; If the separated signal r n If is a monotonic function or a constant, go to step 3.3.6; Otherwise, go to step 3.2.4; Step 3.2.6: Decompose the original denoised signal into empirical mode decomposition results.

6. The method for extracting engine degradation trend based on fault genes according to claim 5, characterized in that: The step 3.3 includes the following sub-steps: Step 3.3.1: Perform Hilbert transform on the signal to obtain the analytical signal of each mode; Step 3.3.2: Based on the mixed-estimated center frequency of each modal analytical signal, the analytical signal of each mode is modulated and shifted to the corresponding baseband; Step 3.3.3: Estimate the bandwidth of the signal using the H1 Gaussian smoothness method; Step 3.3.4: Introduce the Lagrange multiplication operator λ to transform the constrained variational problem into an unconstrained variational problem and obtain the augmented Lagrange expression; Step 3.3.5: Use the alternating direction multiplier (ADMM) algorithm combined with Parseval / Plancherel and Fourier isometric transform to optimize the components and center frequencies of each mode and search for the saddle point of the augmented Lagrange function; Step 3.3.6: Iterate and solve the signal after alternating optimization.

7. The method for extracting engine degradation trend based on fault genes according to claim 6, characterized in that: In step 3.3.5, the three coefficient expressions in the augmented Lagrange expression after the alternating optimal iteration are: Where: γ is the noise tolerance, which meets the fidelity requirements of signal decomposition; and Corresponding to and Fourier transform of The iteration order of step 3.3.6 is: Step 3.3.6.1: Initialization λ 1 And the maximum number of iterations N, and n←0; Step 3.3.6.2: Use Formula and Update the coefficient expression in the augmented Lagrange expression and ω k ; Step 3.3.6.3: Use Update Step 3.3.6.4: Use the accuracy convergence criterion ε>0 to judge: If not satisfied and n<N, return to step 3.3.6.2; Otherwise, the iteration is completed and the output is and ω k .

Images