A signal weak feature extraction method and system

Abstract

A signal weak feature extraction method and system mainly includes the following steps: obtaining original vibration signals of a non-circular gear housing in an electromechanical device under normal and fault states and the like, decomposing the original vibration signals according to an improved empirical wavelet transform, and extracting a frequency band instantaneous envelope line of the decomposed signal components combined with an improved sequential statistical filter; then, combining preset constraint criteria such as a multi-scale correlation criterion, an entropy criterion and a fault feature frequency matching criterion, a weak feature signal modal component in the original vibration signal is screened out. The method can effectively improve the extraction effect of the non-circular gear signal weak feature, reduce the influence of noise and interference signals, improve the screening accuracy of the weak feature signal component, and help improve the accuracy of fault diagnosis and analysis.

Description

Technical Field

[0001] This invention relates to the field of signal extraction technology for electromechanical equipment, and in particular to a method and system for extracting weak features of non-circular gear signals. Background Technology

[0002] In the current field of weak feature extraction technology for non-circular gear signals, most methods improve the extraction effect by combining multiple signal processing techniques. However, these methods still have some significant problems and challenges when processing non-circular gear signals.

[0003] First, existing technologies have shortcomings in signal decomposition. Due to the complexity of non-circular gear signals, signal decomposition is often incomplete, leading to incomplete extraction of weak feature signal components. In such cases, the extracted weak features may not accurately reflect the true state of the non-circular gear, thus affecting subsequent fault diagnosis and analysis.

[0004] Secondly, existing methods lack effective constraint criteria when extracting weak feature signal components. This results in low accuracy and may contain a large amount of noise and interference signals. The presence of these noise and interference signals not only reduces the accuracy of weak feature signals but may also mislead fault diagnosis results.

[0005] Finally, the accuracy of identifying the fault characteristic frequencies of non-circular gears is also insufficient. This is because in non-circular gear signals, fault characteristic frequencies are often intertwined with other frequency components, making accurate identification difficult. This situation severely impacts the effectiveness of fault diagnosis. Summary of the Invention

[0006] In response, this disclosure provides a method and system for extracting weak signal features, applicable to the extraction of weak features from non-circular gear signals, which helps improve the accuracy of fault diagnosis and analysis.

[0007] The weak signal feature extraction method disclosed herein mainly includes the following steps:

[0008] S1, acquire the original vibration signals of the non-circular gear housing in the electromechanical equipment under normal working conditions and fault conditions, wherein the fault conditions include one or more of the following: non-circular gear wear state, gear cracking state, tooth surface detachment state, and gear scuffing state.

[0009] S2, the original vibration signal is decomposed using an improved empirical wavelet transform, and the decomposed signal components are combined with an improved sequential statistical filter to extract the instantaneous envelope of the frequency band; wherein, the improved empirical wavelet transform is used to optimize the wavelet basis function and its decomposition, and the improved sequential statistical filter is used to optimize the weight allocation and multiple filtering fusion.

[0010] S3. Based on preset constraint criteria, weak characteristic signal modal components in the original vibration signal are screened out. The preset constraint criteria include one or more of the following: multi-scale correlation criterion, entropy criterion, and fault characteristic frequency matching criterion.

[0011] The signal weak feature extraction system applying the above method mainly includes:

[0012] The data acquisition module is used to acquire the original vibration signal of the non-circular gear housing in the electromechanical equipment. The original vibration signal includes vibration signals under normal and fault conditions. The fault conditions include one or more of the following: gear wear, gear cracking, tooth surface detachment, and gear scuffing.

[0013] The signal processing module is used to decompose the original vibration signal according to the improved empirical wavelet transform, and to extract the instantaneous envelope of the frequency band by combining the decomposed signal components with the improved sequential statistical filter. The improved empirical wavelet transform is used to optimize the wavelet basis function and decomposition algorithm, and the improved sequential statistical filter is used to optimize the weight allocation and multiple filtering fusion.

[0014] The modal component screening module is used to screen out weak characteristic signal modal components in the original vibration signal by combining preset constraint criteria. The preset constraint criteria include one or more of the following: multi-scale correlation criterion, entropy criterion, and fault characteristic frequency matching criterion.

[0015] Compared with the prior art, the beneficial effects of this disclosure are: (1) the original vibration signal is decomposed by the improved empirical wavelet transform, which improves the extraction effect of weak signal features; (2) the frequency band instantaneous envelope is extracted by combining the improved sequential statistical filter, which effectively reduces the influence of noise and interference signals; (3) the preset constraint criteria are introduced, which improves the screening accuracy of weak feature signal components; (4) it is applicable to the extraction of weak features of non-circular gear signals, which helps to improve the accuracy of fault diagnosis and analysis; (5) it is simple to implement, easy to operate, and has a wide range of application prospects. Attached Figure Description

[0016] The above and other objects, features and advantages of this disclosure will become more apparent from the more detailed description of exemplary embodiments of this disclosure taken in conjunction with the accompanying drawings, in which the same reference numerals generally represent the same components.

[0017] Figure 1 Here is a flowchart of the weak signal feature extraction process according to this disclosure;

[0018] Figure 2 This is a diagram of the internal structure of a computer device for an exemplary weak signal feature extraction system according to this disclosure. Detailed Implementation

[0019] Preferred embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While preferred embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure will be thorough and complete, and will fully convey the scope of the present disclosure to those skilled in the art.

[0020] This disclosure provides a method and system for extracting weak signal features, the process of which is shown in the appendix. Figure 1 As shown, the main steps include:

[0021] 1. Step S101:

[0022] Acquire the original vibration signal of the non-circular gear housing in the electromechanical equipment. The original vibration signal includes the vibration signal under normal working conditions and fault conditions of the non-circular gear. Fault conditions include one or more of the following: gear wear, gear cracking, tooth surface detachment, and gear scuffing.

[0023] The original vibration signal of the non-circular gear housing in electromechanical equipment can be obtained by installing suitable vibration sensors, such as accelerometers, velocity sensors, or displacement sensors, on the housing or key parts of the non-circular gear transmission system. These sensors can be connected to data acquisition devices (such as data acquisition instruments, dynamic signal analyzers, or wireless sensor nodes in an industrial IoT environment), and appropriate sampling frequencies, ranges, and filtering parameters can be set to ensure that vibration signals within the frequency range of interest are captured while avoiding noise interference. Alternatively, the data acquisition system can be activated to record vibration signals in real time. Long-term data acquisition under different operating conditions (such as no-load, load changes, and different speeds) is necessary for a comprehensive analysis of the gear system's vibration behavior.

[0024] It should be noted that these sensors should be installed as close as possible to the vibration source, or in a location where vibration can be easily transmitted, to ensure that the most direct and authentic vibration information is captured.

[0025] It should be noted that collecting vibration signals under normal operating conditions serves as a benchmark for establishing a standard model of equipment health status, facilitating comparative analysis of abnormal conditions and timely detection of minute changes. As gears wear, vibration characteristics change, manifesting as an increase in specific frequency components, the appearance of new frequency peaks, or a shift in existing frequency peaks. Analyzing vibration signals under wear conditions can predict maintenance cycles and component replacement times. Gear cracking leads to intensified vibration, especially when the crack enters the meshing stage with rotation, generating a significant impact effect. The vibration signal reflects a sudden increase in amplitude and new frequency components, aiding in the early detection of serious mechanical failures. Material loss from the gear surface causes a decrease in meshing quality, resulting in irregular vibrations with altered amplitude and frequency characteristics. Analyzing vibration signals under these conditions allows for rapid location of problem areas and assessment of equipment safety. Under extreme conditions, gears may experience metal-to-metal adhesion (scuffing). The vibration signal characteristics at this point differ significantly from normal conditions, potentially showing an increase in high-frequency components or aggravated low-frequency fluctuations. These signal characteristics can help determine whether scuffing has occurred.

[0026] It should also be noted that analyzing vibration signals under different fault conditions can more accurately predict the remaining lifespan of the equipment, allow for advance planning of maintenance, and reduce losses from sudden downtime.

[0027] 2. Step S102:

[0028] The original vibration signal is decomposed using an improved empirical wavelet transform, and the decomposed signal components are combined with an improved sequential statistical filter to extract the instantaneous envelope of the frequency band. The improved empirical wavelet transform includes optimizing the wavelet basis function and the decomposition algorithm, while the improved sequential statistical filter includes optimizing weight allocation and multi-filter fusion. Specifically:

[0029] (1) The original vibration signal is decomposed based on the improved empirical wavelet transform, including:

[0030] 1) Improve the empirical wavelet transform and propose an optimized wavelet transform basis function:

[0031] Let X represent the different states of the original vibration signal of a non-circular gear housing in electromechanical equipment. status (t), the signal center frequency is and the corresponding scale parameter δ status (k);

[0032] The optimized wavelet basis functions are:

[0033]

[0034] Where k represents the scale index or frequency index, status represents the five different states in the vibration signal, c represents the center marker, and t represents the time variable;

[0035] 2) By using a preset objective function, the orthogonality of the optimized wavelet basis functions is enhanced to improve the decomposition algorithm.

[0036] The preset objective function includes:

[0037]

[0038] Among them, h n It is the set of coefficients for the bandpass filter, used to generate the optimized wavelet basis functions, b n It is the frequency band boundary, used to determine the frequency range of different wavelet basis functions; X n It is the decomposition component of signal X in the nth frequency band; signal X is a non-circular gear vibration signal under any state, including but not limited to normal working condition, gear wear condition, gear cracking condition, tooth surface detachment condition, and gear scuffing condition, <ψ m ,ψ n >is ψ m and ψ n The inner product of these terms measures the degree of their insufficient orthogonality; ∥·∥2 represents the Euclidean norm, and the second term measures the reconstruction error of the signal decomposition, L(b n ) is the boundary optimization penalty term, which is a criterion matched to the signal frequency characteristics; α, β, γ are weighting coefficients used to balance the importance of orthogonality, reconstruction error, and boundary optimization; ψ m ψ represents the wavelet basis function corresponding to the m-th frequency band. n This represents the wavelet basis function corresponding to the nth frequency band.

[0039] The weighting coefficients α, β, and γ were optimized using experimental data and cross-validation.

[0040] Step 1: Define the objective function

[0041] First, the objective function, which includes three metrics (orthogonality, reconstruction error, and boundary optimization), can be defined as follows:

[0042] F(α,β,γ)=α·+β·R+γ·B

[0043] Where P represents the orthogonality index, R represents the reconstruction error index, B represents the boundary optimization index, and α, β, and γ are the corresponding weight coefficients.

[0044] Step 2: Prepare the experimental dataset

[0045] Collect and organize the relevant signal datasets, dividing them into training and validation sets.

[0046] Step 3: Set up the search space

[0047] Determine the search range for α, β, and γ;

[0048] In this embodiment, the search range for α, β, and γ is [0, 1].

[0049] Step 4: Initialize the weight matrix

[0050] Create a grid search matrix containing all possible combinations of α, β, γ;

[0051] In this embodiment, it is assumed that there are three linear intervals all in the range [0,1]. However, to demonstrate the generality of this method, the ranges of α, β, and γ are chosen to be [a1,b1], [a2,b2], and [a3,b3], respectively, and they are all uniformly divided into 'n1', 'n2', and 'n3' equally spaced points. Then, let:

[0052] A={α∣α∈[a1,b1],α=a1+i*(b1-a1) / n1,i=0,1,...,n1-1}

[0053] B={β∣β∈[a2,b2],β=a2+j*(b2-a2) / n2,j=0,1,...,n2-1}

[0054] C={γ∣γ∈[a3,b3],v=a3+k*(b3-a3) / n3,k=0,1,...,n3-1}

[0055] Therefore, the grid search space formed by all the weight combinations can be represented as the Cartesian product of the points within these intervals:

[0056] S = A × B × C

[0057] This means that each element in set 'S' is a set of weight values ​​'(α,β,γ)', covering all possible combinations.

[0058] In an alternative embodiment, the specific solution can be implemented through actual programming, in which case the set is converted into a list or array.

[0059] Step 5: Calculate performance indicators

[0060] For each weight combination (α,β,γ), perform the following operations:

[0061] Perform signal processing on the training data using given weights;

[0062] Calculate the orthogonality index P, the reconstruction error index R, and the boundary optimization index B;

[0063] Substitute these three indicators into the objective function F to calculate the total score;

[0064] The model's performance is evaluated on a validation set, for example, by calculating diagnostic accuracy or other relevant evaluation metrics.

[0065] Step 6: Cross-validation

[0066] K-fold cross-validation is used to reduce the risk of overfitting. For each set of weights, the average performance is calculated across all folds.

[0067] Step 7: Select the optimal weights

[0068] Compare the objective function values ​​and diagnostic performance on the validation set under different weight combinations, and select the weight combination with the best overall performance.

[0069] As a preferred option, step ⑧ is also included: verifying the optimal solution.

[0070] The model performance was validated on independent test sets using the optimal α, β, and γ values ​​to confirm its generalization ability.

[0071] It should be noted that the process may require multiple iterations to fine-tune the weights until a satisfactory performance level is achieved.

[0072] 3) The original signal is decomposed using optimized wavelet basis functions with enhanced orthogonality.

[0073] The improved empirical wavelet transform has the following significant advantages in decomposing the original vibration signal:

[0074] Enhanced Adaptability: The improved empirical wavelet transform can optimize the selection of wavelet basis functions based on the characteristics of the actual signal, making the transform more closely match the local features of the signal. This is particularly suitable for analyzing non-stationary and nonlinear vibration signals of electromechanical equipment. This adaptability can effectively extract the instantaneous features of the signal in the time and frequency domains, which is extremely beneficial for analyzing the vibration characteristics of non-circular gears under different operating conditions. The improved empirical wavelet transform provides a multi-level, multi-scale analysis framework that can decompose the original signal layer by layer and separate the energy distribution of different frequency bands. Furthermore, the improved empirical wavelet transform improves the resolution of signal details and noise suppression by optimizing the decomposition algorithm, which is beneficial for extracting key components representing fault characteristics from complex vibration signals.

[0075] Reduce computational complexity: Improvements to the traditional empirical wavelet transform may introduce more efficient computational strategies and optimization algorithms, thereby reducing computational complexity and improving the speed and efficiency of signal processing. This is particularly important for real-time online monitoring and fault diagnosis applications.

[0076] Effective noise suppression: The optimized wavelet basis function helps to enhance the suppression of noise in the signal, making the decomposed signal components more reflective of the true physical meaning, and providing a cleaner signal source for subsequent fault feature extraction.

[0077] (2) The decomposed signal components are combined with an improved sequential statistical filter to extract the instantaneous envelope of the frequency band, including:

[0078] Let the original vibration signal after improved empirical wavelet transform decomposition have several frequency band components as follows: The original vibration signal includes at least the vibration signal frequency band components under the following conditions: normal working condition of non-circular gear, gear wear condition, gear cracking condition, tooth surface detachment condition, and gear scuffing condition.

[0079] The instantaneous envelope of the frequency band is extracted by combining several frequency band components of the decomposed original vibration signal with an improved sequential statistical filter; the improved sequential statistical filter includes the following four steps:

[0080] Step 1): Adaptive threshold selection;

[0081] The local mean and standard deviation of several frequency band components of the original vibration signal after decomposition are obtained and denoted as μ(o) and δ(o).

[0082] The scaling threshold based on local standard deviation is denoted as: θ(o) = a·δ(o) + μ(o);

[0083] Where 'a' represents the dynamic adjustment coefficient, used to control the strictness of the threshold, and 'o' represents the time series index of the signal;

[0084] Step 2): Optimize weight allocation;

[0085] The instantaneous envelope e of the frequency band component of any i-th frequency band component in the vibration signal frequency band components under normal working conditions, gear wear conditions, gear cracking conditions, tooth surface detachment conditions, and gear scuffing conditions of non-circular gears. i [o] Preset weight w i ;

[0086] e i o' = w i ·e i [o]

[0087] The synthesized instantaneous envelope of the frequency band is represented as follows:

[0088]

[0089] Where N represents the number of frequency bands, and E[o] represents the instantaneous envelope of the synthesized frequency bands;

[0090] Step 3): Optimize multi-filter fusion;

[0091] Determine the size L of the neighborhood window, and presuppose that L is an odd number;

[0092] For each time point o, calculate the local maxima of the signal z[or],...,z[o],...,z[o+r] within the window, where r represents the radius of the neighborhood window;

[0093]

[0094] Set a threshold T. If z[n] is a local maximum and satisfies |z[o+1]-z[o]|+|z[o-1]-z[o]|>T, then retain the local maximum point.

[0095] Connect all the retained maxima points to form the instantaneous envelope E of the local maximum frequency band. max [o];

[0096] in:

[0097] E max [o]={z[o]:z[o]>z[or],z[n]>z[o+r], and

[0098] ∣z[o+1]-z[o]∣+∣z[o-1]-z[o]∣>T}

[0099] Step 4): Smoothing process;

[0100] After extracting the instantaneous envelope of the frequency band, the final instantaneous envelope of the frequency band is obtained using the moving average method, denoted as .

[0101] It should be noted that the improved sequential statistical filter can suppress random noise to a certain extent, especially when dealing with non-Gaussian noise. Compared with traditional filters, it has better noise immunity, thus extracting the signal envelope more accurately. In gear fault diagnosis, the improved sequential statistical filter can better highlight the weak characteristic signal modes related to the fault. The extracted envelope can clearly show information such as fault characteristic frequencies and fault development trends, which helps in further fault diagnosis and identification.

[0102] 3. Step S103:

[0103] Based on preset constraint criteria, weak characteristic signal modal components in the original vibration signal are screened out. The preset constraint criteria include: multi-scale correlation criterion, entropy criterion, and fault characteristic frequency matching criterion.

[0104] (1) Multiscale correlation criterion

[0105] The multi-scale correlation criterion includes the characteristics of the vibration signal of non-circular gears changing over time under different speeds and operating conditions, and the design of a sliding window to calculate local correlation.

[0106] Let the window length be W and the sliding step size be Δ, then for the two states, the instantaneous envelope of the frequency band and Calculate the correlation coefficient within each local window:

[0107]

[0108] Where u and v represent the working states of the non-circular gear, including normal working condition, gear wear condition, gear cracking condition, tooth surface detachment condition, and gear scuffing condition, c represents the number of windows, and i represents the i-th frequency band component. as well as These represent the contents of window c respectively. and The mean, σ Ei, With σ Ei, These represent the contents of window c respectively. and Standard deviation;

[0109] A correlation threshold is preset, and frequency band components that exceed this threshold are identified as weak characteristic signal modal components in the primary original vibration signal.

[0110] (2) The pre-defined constraint criteria also include the entropy criterion:

[0111] The entropy criterion involves calculating piecewise entropy using a sliding window approach, specifically within window c, as shown below:

[0112]

[0113] Where c represents the number of windows, p Q, Q represents the probability that intensity level Q occurs within window c. c It is the number of intensity levels within window c;

[0114] A preset entropy threshold is set, and frequency band components that exceed this threshold are identified as weak characteristic signal modal components in the primary original vibration signal.

[0115] (3) The preset constraint criteria also include the fault characteristic frequency matching criterion:

[0116] The fault characteristic frequency matching criterion includes performing spectral analysis on the instantaneous envelope of each frequency band signal component to obtain a spectrum diagram;

[0117] Based on the spectrum diagram, obtain the frequency components corresponding to the corresponding fault states in the fault characteristic frequency library. The corresponding fault states in the fault characteristic frequency library include gear wear state, gear cracking state, tooth surface detachment state, and gear scuffing state. Analyze the degree of matching between the spectrum and the characteristic frequencies.

[0118] The degree of matching between the spectrum and the characteristic frequencies is analyzed, including:

[0119]

[0120] Where S(f) represents the spectral density function of the signal, and f represents the energy distribution of the signal at different frequencies. feat It is a preset fault characteristic frequency, determined based on gear dynamics principles and fault diagnosis knowledge, and is a typical frequency related to a specific fault state, Δf. bw It refers to bandwidth. E represents the total energy of the signal from zero frequency to the Nyquist frequency. ratio It is the ratio of the frequency band energy near the characteristic frequency to the total energy, used to characterize the proportion of signal energy near the characteristic frequency to the total energy;

[0121] A threshold is preset for the ratio of frequency band energy to total energy near a characteristic frequency. Frequency band components above this threshold are identified as weak characteristic signal modal components in the primary original vibration signal.

[0122] (4) Based on preset constraint criteria, weak characteristic signal modal components in the original vibration signal are selected, including:

[0123] The correlation coefficients between different frequency bands in each state are calculated according to the multi-scale correlation criterion. The frequency band score is 1 if it is higher than the preset threshold, and 0 otherwise.

[0124] The entropy value of each frequency band between each state is calculated according to the entropy criterion. The frequency band with an entropy value higher than the preset threshold is scored as 1, otherwise it is scored as 0.

[0125] The proportion of frequency band energy near the fault characteristic frequency is calculated according to the fault characteristic frequency matching criterion. The frequency band score is 1 if the proportion is higher than the preset threshold, otherwise it is 0.

[0126] For each state, its comprehensive score across all frequency bands is calculated according to three criteria.

[0127] The weights of the multi-scale correlation criterion, entropy criterion, and fault feature frequency matching criterion are set as w, respectively. R ,w H ,w F , where w R +w H +w F =1;

[0128] Calculate the overall score for each state and each frequency band:

[0129] C Si =w R × Multiscale Correlation Criterion Score si +w H ×Entropy Criterion Score si +w F

[0130] × Fault Feature Frequency Matching Criterion Score si

[0131] Where S represents the state and i represents the frequency band;

[0132] For each frequency band, calculate its overall score C under the five states. Si Considered as "votes";

[0133] The frequency bands that meet the preset threshold number of votes contain weak characteristic signal modal components. The frequency band components obtained by wavelet transform decomposition based on the improved experience within the frequency band are then combined to obtain the final weak characteristic signal modal components.

[0134] It should be noted that the show-of-hands voting method integrates information from multiple sources, including multi-scale correlation criteria, entropy criteria, and fault feature frequency matching criteria, reducing potential misjudgments from a single criterion and enhancing the system's robustness and reliability. Relying solely on a single criterion can lead to oversensitivity to certain special cases, resulting in overfitting. Through the voting mechanism, a frequency band is only considered to contain weak characteristic signal modal components when multiple criteria agree on it, which helps reduce the risk of overfitting. By combining the scores of multiple criteria, the characteristic information contained in the frequency band signal can be more comprehensively evaluated, thereby improving the accuracy of identifying weak characteristic signal modal components. Furthermore, the show-of-hands voting method is an intuitive and easy-to-understand integration method, facilitating rapid understanding and implementation by engineers and researchers, and also aiding in the interpretation and verification of results. In summary, the show-of-hands voting method helps overcome the limitations of a single criterion, improves the generalization ability and practicality of fault diagnosis systems, and helps effectively identify and extract weak characteristic signal modal components reflecting different fault states from complex vibration signals.

[0135] In an optional embodiment, the sensitivity of the recognition system can be changed by adjusting the preset threshold number of votes, depending on different application scenarios and needs, in order to adapt to different fault modes and different levels of feature signals.

[0136] In summary, the weak signal feature extraction method proposed in this disclosure acquires the original vibration signal of a non-circular gear housing in electromechanical equipment under various states such as normal and fault. The original vibration signal is decomposed according to the improved empirical wavelet transform, and the frequency band instantaneous envelope is extracted by combining the decomposed signal components with the improved sequential statistical filter. Then, the weak feature signal modal components in the original vibration signal are screened out by combining the multi-scale correlation criterion, entropy criterion and fault feature frequency matching criterion and other preset constraint criteria.

[0137] An exemplary weak signal feature extraction system applying the above method includes:

[0138] The data acquisition module is used to acquire the original vibration signals of the non-circular gear housing in the electromechanical equipment. The original vibration signals include at least the vibration signals of the non-circular gear under normal working conditions, gear wear conditions, gear cracking conditions, tooth surface detachment conditions, and gear scuffing conditions.

[0139] The signal processing module is used to decompose the original vibration signal according to the improved empirical wavelet transform, and to extract the instantaneous envelope of the frequency band by combining the decomposed signal components with the improved sequential statistical filter. The improved empirical wavelet transform includes optimizing the wavelet basis function and the decomposition algorithm, and the improved sequential statistical filter includes optimizing the weight allocation and multiple filtering fusion.

[0140] The modal component screening module is used to filter out weak characteristic signal modal components in the original vibration signal by combining preset constraint criteria. The preset constraint criteria include multi-scale correlation criteria, entropy criteria, and fault characteristic frequency matching criteria.

[0141] The above-mentioned unit modules can be embedded in the processor of the computer device in hardware form or independent of it, or they can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of the above modules.

[0142] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 2As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a method for extracting weak signal features. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0143] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0144] Acquire the original vibration signal of the non-circular gear housing in the electromechanical equipment. The original vibration signal includes at least the vibration signal of the non-circular gear under normal working conditions, gear wear conditions, gear cracking conditions, tooth surface detachment conditions, and gear scuffing conditions.

[0145] The original vibration signal is decomposed based on the improved empirical wavelet transform, and the decomposed signal components are combined with the improved sequential statistical filter to extract the instantaneous envelope of the frequency band. The improved empirical wavelet transform includes optimizing the wavelet basis function and the decomposition algorithm, and the improved sequential statistical filter includes optimizing the weight allocation and multiple filtering fusion.

[0146] Based on preset constraint criteria, weak characteristic signal modal components in the original vibration signal are screened out. The preset constraint criteria include multi-scale correlation criteria, entropy criteria, and fault characteristic frequency matching criteria.

[0147] The above technical solutions are merely exemplary embodiments of the present invention. For those skilled in the art, based on the application methods and principles disclosed in the present invention, it is easy to make various types of improvements or modifications, and not limited to the methods described in the specific embodiments of the present invention. Therefore, the methods described above are merely preferred and not restrictive.

Claims

1. A method for extracting weak signal features, characterized in that, Includes the following steps: S1, acquire the original vibration signals of the non-circular gear housing in the electromechanical equipment under normal working conditions and fault conditions, wherein the fault conditions include one or more of the following: non-circular gear wear state, gear cracking state, tooth surface detachment state, and gear scuffing state. S2, the original vibration signal is decomposed using an improved empirical wavelet transform, and the decomposed signal components are combined with an improved sequential statistical filter to extract the instantaneous envelope of the frequency band; wherein, the improved empirical wavelet transform is used to optimize the wavelet basis function and its decomposition, and the improved sequential statistical filter is used to optimize the weight allocation and multiple filtering fusion. S3. Combined with preset constraint criteria, weak characteristic signal modal components in the original vibration signal are screened out. The preset constraint criteria include one or more of the following: multi-scale correlation criterion, entropy criterion, and fault characteristic frequency matching criterion. The step of combining the decomposed signal components with an improved sequential statistical filter to extract the instantaneous envelope of the frequency band specifically includes: Suppose that the original vibration signal after decomposition using the improved empirical wavelet transform has several frequency band components. The original vibration signal includes several frequency band components: vibration signal frequency band components under normal and fault conditions of non-circular gears; The decomposed original vibration signal is combined with several frequency band components using an improved sequential statistical filter to extract the instantaneous envelope of the frequency bands, including the following steps: Step 1), Adaptive threshold selection: Obtain the local mean and standard deviation of several frequency band components of the decomposed original vibration signal under different states, and denot them as follows: as well as ; Set the proportional threshold based on the local standard deviation as follows: ; in, This represents the dynamic adjustment coefficient, used to control the strictness of the threshold; 'o' represents the time series index of the signal. Step 2), optimize weight allocation: For any first component of the vibration signal frequency band under normal and fault conditions of non-circular gears Instantaneous envelope of each frequency band component Preset weights The synthesized instantaneous envelope of the frequency band is represented as follows: Where N represents the number of frequency bands, This represents the instantaneous envelope of the synthesized frequency band; Step 3), optimize multi-filter fusion: Determine the size L of the neighborhood window, and presuppose that L is an odd number; For each time point o, calculate the signal within the window. The local maxima of , where r represents the radius of the neighborhood window; Set a threshold T. If z[n] is a local maximum and satisfies... If so, then retain the maximum point; Connect all the retained maxima points to form the instantaneous envelope of the local maximum frequency band. ; in: Step 4), Smoothing: After extracting the instantaneous envelope of the frequency band, the final instantaneous envelope of the frequency band is obtained using the moving average method, denoted as . .

2. The method according to claim 1, characterized in that, The fault states include one or more of the following: non-circular gear wear state, gear cracking state, tooth surface detachment state, and gear scuffing state.

3. The method according to claim 1 or 2, characterized in that, The step of decomposing the original vibration signal based on the improved empirical wavelet transform specifically includes: The different states of the original vibration signal of the non-circular gear housing in electromechanical equipment are recorded as follows: The signal center frequency is and corresponding scale parameters , The optimized wavelet transform basis functions are: Where k represents the scale index or frequency index, status represents the five different states in the vibration signal, c represents the center marker, and t represents the time variable; The orthogonality of the optimized wavelet basis functions is enhanced by a preset objective function, wherein the preset objective function includes: in, This is the set of coefficients for the bandpass filter, used to generate the optimized wavelet basis functions; This serves as the frequency band boundary, used to determine the frequency range of different wavelet basis functions; For signal Decomposition components in the nth frequency band; signal Vibration signals of non-circular gears under any normal or fault condition; for and The inner product of these is used to measure the degree of their insufficient orthogonality; L represents the Euclidean norm, and the second term measures the reconstruction error of the signal decomposition; The boundary optimization penalty term is a criterion that matches the signal frequency characteristics; These are weighting coefficients used to balance the importance of orthogonality, reconstruction error, and boundary optimization. This represents the wavelet basis function corresponding to the m-th frequency band. This represents the wavelet basis function corresponding to the nth frequency band; The original signal is decomposed using optimized wavelet basis functions with enhanced orthogonality.

4. The method according to claim 1 or 2, characterized in that, In step S3, when the preset constraint criterion adopts the multi-scale correlation criterion, step S3 specifically includes: Let the window length be The sliding step size is Then, for the instantaneous envelope of the frequency band in both states and Calculate the correlation coefficient within each local window: Where u and v represent the working state of the non-circular gear, including normal and fault states; c represents the number of windows; and i represents the i-th frequency band component. as well as These represent the contents of window c respectively. and The mean; and These represent the contents of window c respectively. and Standard deviation; A correlation threshold is preset, and frequency band components that exceed this threshold are identified as weak characteristic signal modal components in the primary original vibration signal.

5. The method according to claim 1 or 2, characterized in that, In step S3, when the preset constraint criterion adopts the entropy criterion, step S3 specifically includes: The piecewise entropy is calculated using a sliding window approach, specifically within window c, as shown below: Where c represents the number of windows, This represents the probability that intensity level Q occurs within window c. It is the number of intensity levels within window c; A preset entropy threshold is set, and frequency band components that exceed this threshold are identified as weak characteristic signal modal components in the primary original vibration signal.

6. The method according to claim 1 or 2, characterized in that, In step S3, when the preset constraint criterion adopts the fault characteristic frequency matching criterion, step S3 specifically includes: Spectral analysis is performed on the instantaneous envelope of each frequency band signal component to obtain the spectrum diagram; Based on the spectrum diagram, obtain the frequency components corresponding to the corresponding fault states in the fault characteristic frequency library. The fault states include one or more of the following: gear wear state, gear cracking state, tooth surface detachment state, and gear scuffing state. Analyze the degree of matching between the spectrum and the characteristic frequencies. The degree of matching between the analyzed spectrum and the characteristic frequency is expressed as follows: in, The spectral density function represents the energy distribution of the signal at different frequencies; This refers to a preset typical frequency associated with a specific fault state; For bandwidth, This represents the total energy of the signal from zero frequency to the Nyquist frequency; It is the ratio of the frequency band energy near the characteristic frequency to the total energy, used to characterize the proportion of signal energy near the characteristic frequency to the total energy; A threshold is preset for the ratio of frequency band energy to total energy near a characteristic frequency. Frequency band components above this threshold are identified as weak characteristic signal modal components in the primary original vibration signal.

7. The method according to claim 1 or 2, characterized in that, In step S3, when the preset constraint criteria adopt three criteria—multi-scale correlation criterion, entropy criterion, and fault feature frequency matching criterion—step S3 specifically includes: The correlation coefficients between different frequency bands in each state are calculated according to the multi-scale correlation criterion. The frequency band score is 1 if it is higher than the preset threshold, and 0 otherwise. The entropy value of each frequency band between each state is calculated according to the entropy criterion. The frequency band with an entropy value higher than the preset threshold is scored as 1, otherwise it is scored as 0. The proportion of frequency band energy near the fault characteristic frequency is calculated according to the fault characteristic frequency matching criterion. The frequency band score is 1 if the proportion is higher than the preset threshold, otherwise it is 0. For each state, its comprehensive score across all frequency bands is calculated according to three criteria. The weights of the multi-scale correlation criterion, entropy criterion, and fault feature frequency matching criterion are set as follows: ,in ; Calculate the overall score for each state and each frequency band: Where S represents the state. Represents frequency band; For each frequency band, calculate its overall score across the five states. Considered as "votes"; The frequency bands that meet the preset threshold number of votes contain weak characteristic signal modal components. The frequency band components obtained by wavelet transform decomposition based on the improved experience within the frequency band are the final weak characteristic signal modal components.

8. A device for extracting weak signal features using the method described in any one of claims 1-7, characterized in that, include: The data acquisition module is used to acquire the original vibration signal of the non-circular gear housing in the electromechanical equipment, the original vibration signal including vibration signals under normal and fault conditions; The fault states include one or more of the following: gear wear state, gear cracking state, tooth surface detachment state, and gear scuffing state. The signal processing module is used to decompose the original vibration signal according to the improved empirical wavelet transform, and to extract the instantaneous envelope of the frequency band by combining the decomposed signal components with the improved sequential statistical filter. The improved empirical wavelet transform is used to optimize the wavelet basis function and decomposition algorithm, and the improved sequential statistical filter is used to optimize the weight allocation and multiple filtering fusion. The modal component screening module is used to screen out weak characteristic signal modal components in the original vibration signal by combining preset constraint criteria. The preset constraint criteria include one or more of the following: multi-scale correlation criterion, entropy criterion, and fault characteristic frequency matching criterion.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

Images