A gear fault detection method for a gearbox

Abstract

The present invention relates to the field of gear vibration signal analysis and fault diagnosis, and specifically to a gear fault detection method for a gearbox. The method comprises: first collecting a gear vibration signal, screening all initial extreme value points, and using the product of the amplitude deviation degree of the initial extreme value point and the local accumulated energy as the amplitude mutation degree; then extracting the instantaneous frequency and instantaneous phase of each initial extreme value point, fusing the frequency consistency and phase consistency through the geometric mean method to obtain a time-frequency coupling index; determining the validity of the initial extreme value point by multiplying the time-frequency coupling index and the amplitude mutation degree; screening the true extreme value points based on the validity; extracting the inherent rotational component of the vibration signal based on the true extreme value points, jointly constructing a feature vector, and matching it with a fault feature library to achieve gear fault diagnosis. This method suppresses the influence of noise interference and pseudo-extreme value points, and improves the accuracy of fault detection.

Description

Technical Field

[0001] The present invention relates to the field of gearbox vibration signal analysis and fault diagnosis, and more particularly to a gearbox gear fault detection method. Background Art

[0002] As a key component in a vehicle's powertrain, the gears within the transmission are prone to structural faults such as wear, cracks, and missing teeth over long periods of operation. Failure to promptly detect and identify these faults can lead to decreased transmission efficiency and even serious equipment damage or accidents. Therefore, monitoring and diagnosing transmission gear status are crucial.

[0003] Traditional gear fault detection methods mostly rely on contact measurement methods such as accelerometers or strain gauges. Although these methods have high measurement accuracy, they have defects such as complex layout, difficult maintenance, and sensitivity to the working environment. They are difficult to adapt to complex working conditions such as high vibration, high temperature, and closed structures.

[0004] To this end, recent research has shifted toward non-contact condition detection methods. By collecting vibration signals from the gearbox surface and performing signal processing and feature extraction, these methods can intelligently identify and classify gear faults. Typical existing approaches include using the ITD (Intrinsic Time Scale Decomposition) algorithm to adaptively decompose the gearbox gear vibration signals. This algorithm then extracts statistical or entropy-based features of each inherent rotational component to reflect the gear health.

[0005] The core principle of the ITD algorithm is to perform modal decomposition based on local extreme points of the signal. By continuously screening and extracting local extreme points, a series of PRCs (intrinsic rotation components) are constructed. Each PRC represents the characteristics of the signal at a specific time scale. For example, the existing Chinese patent document with publication number CN111709383B discloses a gear fault detection method, device, and storage medium for a transmission. This detection method uses ITD and sample entropy to perform gear fault diagnosis. By performing ITD decomposition on vibration signals under different states, sample entropy features are extracted and compared with a reference state to enable identification under multiple fault modes.

[0006] However, in the complex operating environment of a transmission, vibration signals are often contaminated by a variety of noise sources. These include background mechanical noise caused by engine and bearing structural vibration, external disturbances such as road bumps, and sensor electrical noise. This noise can introduce numerous pseudo-extreme points into the signal. Since the ITD method relies on extreme points for modal extraction, these pseudo-extreme points can be easily misidentified as valid signal features, leading to cross-confusion of features from different fault states in the signal. This in turn leads to modal aliasing and redundant components, compromising the accuracy of signal feature extraction and resulting in inaccurate fault detection results. Summary of the Invention

[0007] In order to solve the problem of pseudo extreme points generated by noise interference, which leads to signal modal aliasing and inaccurate feature extraction, thereby reducing the accuracy of fault detection, the present invention proposes a gear fault detection method for a gearbox, comprising:

[0008] Collecting a gear vibration signal of a gearbox, wherein the gear vibration signal is composed of multiple sampling points;

[0009] Screen all initial extreme points of the vibration signal and use the amplitude deviation of each initial extreme point multiplied by the cumulative energy of the local area of ​​the initial extreme point as the amplitude mutation degree of the initial extreme point; the cumulative energy of the local area is the square sum of the amplitudes of all sampling points in the local area;

[0010] Extract the instantaneous frequency and instantaneous phase of each initial extreme point, fuse the frequency consistency and phase consistency of each initial extreme point through the geometric mean method, and obtain the time-frequency coupling index of the initial extreme point; wherein the frequency consistency is determined based on the difference between the instantaneous frequency of the initial extreme point and the reference frequency; the phase consistency is determined based on the difference between the instantaneous phase of the initial extreme point and the reference phase of the gear;

[0011] The product of the time-frequency coupling index and the amplitude mutation degree is used as the validity of the initial extreme point. The true extreme point of the vibration signal is determined based on the comparison result between the validity and the preset threshold.

[0012] Based on the true extreme points of the vibration signal, the inherent time scale decomposition algorithm is used to extract all the inherent rotational components of the vibration signal. All the inherent rotational components are combined to form a feature vector. Gear fault diagnosis is performed based on the matching results between the feature vector and the pre-established fault feature library.

[0013] This technical solution collects the vibration signals of the gearbox gears and screens the initial extreme points. It quantifies the degree of amplitude mutation by combining the amplitude deviation and the local accumulated energy, thereby accurately capturing the drastic changes in the signal and being able to preliminarily distinguish between true extreme points and pseudo extreme points. By extracting the instantaneous frequency and instantaneous phase of each initial extreme point, combining frequency consistency and phase consistency, and calculating the time-frequency coupling index, the true extreme points and pseudo extreme points are further distinguished from the perspective of time-frequency characteristics and gear dynamics. Subsequently, by combining the amplitude mutation degree and the time-frequency coupling index, the true extreme points in the vibration signal are determined through a double screening mechanism, thereby eliminating the pseudo extreme points caused by noise interference, providing a more accurate data basis for subsequent signal decomposition and feature extraction, and then obtaining accurate signal decomposition results, thereby improving the accuracy and reliability of subsequent fault detection based on the signal decomposition results.

[0014] Preferably, the method for screening all initial extreme value points of the vibration signal is as follows: divide each sampling point into a local area, and calculate the mean of the amplitudes of all sampling points in the local area as the local mean of the sampling point, and form a local mean curve with the serial number of the sampling point as the horizontal coordinate and the local mean of the sampling point as the vertical coordinate; on the local mean curve, if a local mean is greater than the local mean adjacent to its left and also greater than the local mean adjacent to its right, the sampling point corresponding to the local mean is taken as a maximum point; if a local mean is less than the local mean adjacent to its left and also less than the local mean adjacent to its right, the sampling point corresponding to the local mean is taken as a minimum point; all maximum points and all minimum points constitute all initial extreme value points of the vibration signal.

[0015] This technical solution dynamically divides the local area of ​​each sampling point and constructs a local mean curve based on the local mean to screen the initial extreme points of the vibration signal. It effectively avoids the problem of insufficient local feature capture brought about by the traditional fixed window averaging method, can more accurately identify the true extreme points in the signal, and reduce the impact of noise interference on extreme point judgment.

[0016] Preferably, the method for dividing the local area of ​​each sampling point is: setting a basic window with each sampling point as the center, and determining the dominant frequency in the basic window by the zero-crossing detection method, and calculating the coefficient of variation of the amplitudes of all sampling points in the basic window; taking the ratio of the dominant frequency to the maximum frequency in the basic window as the frequency adaptive factor of the sampling point, multiplying the amplitude variation coefficient by the frequency adaptive factor, and calculating the time-frequency complexity of the basic window; using the time-frequency complexity, adjusting the size of the basic window, and determining the range of the adjusted basic window as the local area of ​​the sampling point; wherein the size of the adjusted range of the basic window is negatively correlated with the time-frequency complexity.

[0017] This technical solution more accurately reflects the frequency and amplitude variation characteristics of a signal by determining the dominant frequency within a basic window at each sampling point and calculating the coefficient of amplitude variation. A frequency adaptation factor is introduced to dynamically adjust the range of the local area based on the ratio of the dominant frequency to the maximum frequency, adapting to the characteristics of signals in different frequency bands.

[0018] Preferably, the coefficient of variation of the amplitudes of all sampling points within the basic window is determined based on the ratio of the standard deviation to the mean of the amplitudes of all sampling points.

[0019] Preferably, the degree of deviation of the amplitude of each initial extreme point is determined by the following method: obtaining the mean and standard deviation of the amplitudes of all sampling points within the local range of each initial extreme point; calculating the absolute value of the difference between the amplitude of the initial extreme point and the mean, and taking the ratio of the absolute value to the standard deviation as the degree of deviation of the amplitude of the initial extreme point.

[0020] Preferably, the instantaneous frequency and instantaneous phase of each initial extreme point are extracted based on the following method: the vibration signal is decomposed into multiple natural mode components through empirical mode decomposition, and each natural mode component is subjected to Hilbert transform to obtain the instantaneous frequency and instantaneous phase corresponding to each initial extreme point of the vibration signal.

[0021] Preferably, frequency consistency is determined based on the following:

[0022] Obtain the reference frequency and multiple harmonic frequencies of the reference frequency, and calculate the The initial extreme point and the reference frequency The consistency of the harmonic frequencies:

[0023] , in the formula, For the The initial extreme point and the The consistency of the harmonic frequencies, is the tolerance of the reference frequency, is the natural exponential function, For the The instantaneous frequency of the initial extreme point, is the reference frequency, For the harmonic frequency; The maximum value of the consistency between the initial extreme point and all harmonic frequencies of the reference frequency is taken as the The frequency consistency of the initial extreme points.

[0024] This technical solution uses the form of a Gaussian function to accurately quantify the degree of matching between the extreme points of the signal generated during the gear meshing process and the reference frequency and harmonic frequencies based on the difference between the instantaneous frequency of each initial extreme point and the reference frequency and its harmonic frequencies. This can effectively identify vibration characteristics that are highly correlated with the reference frequency and its harmonic frequencies of the gear, and help to accurately screen out true extreme points that are highly consistent with the reference frequency.

[0025] Preferably, the phase consistency of each initial extreme point satisfies the following relationship:

[0026] , in the formula, For the The phase consistency of the initial extreme points, For the The instantaneous phase of the initial extreme point, is the reference phase, is the natural exponential function, is the tolerance of the reference phase.

[0027] This technical solution quantifies phase consistency by analyzing the difference between the instantaneous phase of each initial extreme point and the reference phase in the form of a Gaussian function. It can effectively identify the phase change characteristics in the gear vibration signal and reduce the false recognition rate caused by noise or atypical vibration modes, thereby providing another dimension of accurate judgment indicators for extracting the true extreme points.

[0028] Preferably, the true extreme points of the vibration signal are determined according to the following method: initial extreme points whose validity is greater than a preset threshold are taken as true extreme points, and initial extreme points whose validity is not greater than the preset threshold are removed as pseudo extreme points.

[0029] Preferably, the step of diagnosing gear faults based on the matching results of the feature vector and the pre-established fault feature library includes: calculating the Euclidean distance between the feature vector and each feature vector in the fault feature library, and taking the fault type corresponding to the feature vector with the smallest Euclidean distance as the diagnosis result.

[0030] The present invention has the following effects:

[0031] The present invention establishes a screening mechanism by combining the time domain characteristics of the vibration signal and gear dynamics to determine the true extreme points of the vibration signal, and embeds the physical laws of gear meshing, such as frequency harmonics and phase synchronization, into the signal processing process, so that the extreme point screening has both data-driven accuracy and physical-driven interpretability. In addition, the adaptive local area division and dynamic threshold strategy are combined to effectively suppress the interference of pseudo extreme points on the vibration signal decomposition of the ITD algorithm, enhance the modal independence of the inherent rotation component, thereby improving the accuracy of signal decomposition and obtaining more accurate fault detection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a schematic flow chart of the method of the present invention;

[0033] Figure 2 It is a schematic flow chart of the method of step S5 of the present invention. DETAILED DESCRIPTION

[0034] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present invention.

[0035] The present invention provides a gearbox gear fault detection method, such as Figure 1 Shown, including:

[0036] S1: Collect the gear vibration signal of the gearbox.

[0037] An ICP (Integrated Circuit Piezoelectric) sensor combines integrated circuit technology with the piezoelectric effect. It senses mechanical vibration or pressure changes through piezoelectric materials and converts these signals into electrical signals via a built-in integrated circuit for subsequent processing and analysis. Due to their compact structure, stable signal output, and high sensitivity, ICP sensors are widely used in vibration monitoring, acceleration measurement, and other fields.

[0038] Therefore, using ICP sensors, the sampling frequency , installed on a plane with good rigidity of the gearbox housing, and use this acceleration sensor to collect the vibration signal of the gearbox surface. At this moment, 500 sampling points are continuously collected to form a vibration signal According to the Nyquist sampling theorem, the highest frequency signal Need to meet: , then the theoretical maximum frequency of the gearbox vibration signal is .

[0039] S2: Dynamically divide the local area of ​​each sampling point of the vibration signal.

[0040] Gear vibration signals are often accompanied by noise, particularly high-frequency noise, which can interfere with the signal's true characteristics. Traditional algorithms often compromise the quality of overall signal decomposition. To more precisely capture the impact of noise, dynamic segmentation of each sampling point allows analysis of specific frequency components within each local region. This makes it easier to capture the true vibration characteristics of the signal and reduce noise interference with signal analysis results.

[0041] In one embodiment, the local area of ​​each sampling point is divided as follows:

[0042] Set the basic window length with each sampling point as the center (Empirical value), the basic window includes the sampling point itself, as well as the 10 sampling points on the left and 10 sampling points on the right of the sampling point.

[0043] The dominant frequency within the basic window is determined using zero-crossing detection. Zero-crossing detection is a common signal analysis method that identifies the periodic characteristics of a signal by detecting the moment when the signal passes through zero when alternating between positive and negative. In this embodiment, the zero-crossing detection method is used to determine the dominant frequency within the basic window. The dominant frequency refers to the most significant frequency component in the signal. The number and distribution of zero-crossing points can help identify the frequency characteristics of the signal, especially periodic signals. The zero-crossing detection method is used here because it is a simple and effective method that can quickly identify the periodic characteristics of a signal. This is very useful in frequency analysis and signal deconstruction, especially when analyzing the frequency of vibration signals.

[0044] Obtain the coefficient of variation of the amplitudes of all sampling points within the base window. The coefficient of variation is a statistic used to measure data fluctuations. Here, it is used to measure signal fluctuations within the base window and can reflect signal stability and noise levels. Its value is equal to the ratio of the standard deviation of the amplitudes of all sampling points to the mean. For vibration signals, a high coefficient of variation indicates greater signal fluctuations, indicating a higher likelihood of strong fluctuations or noise. A low coefficient of variation indicates a relatively stable signal.

[0045] Calculate the time-frequency complexity of the basic window:

[0046]

[0047] In this formula, The vibration signal The time-frequency complexity of the basic window of sampling points, The vibration signal The dominant frequency within the basic window of sampling points, for The maximum value of the dominant frequency of the basis window of all sampling points. represents the coefficient of variation, The vibration signal The coefficient of variation of the amplitude of all sampling points within the basic window of sampling points.

[0048] In this formula, For the The frequency adaptation factor for each sampling point is used to measure the relative magnitude of the dominant frequency within the base window at that sampling point. In signal processing, higher-frequency signals typically change faster, and their characteristics are often more subtle and susceptible to noise. Therefore, appropriately reducing the window size can help improve analysis accuracy. Conversely, lower-frequency signals typically change more slowly, and using a larger window size to capture their characteristics does not lose key information while also helping to reduce the impact of noise. The larger the value, the stronger the amplitude fluctuation of the signal, and the more complex the changes of the vibration signal within the basic window. In this case, it is necessary to narrow the window and limit the analysis range to a shorter time segment to avoid the interference of global noise or non-stationary components on feature extraction, similar to the idea of ​​magnifying local details with a microscope to capture signal details. On the contrary, The smaller it is, the more stable or regular the signal is within the basic window. In this case, it is necessary to increase the window size to integrate information from more sampling points, so that it contains more vibration cycles and enhances the reliability of the signal characteristics.

[0049] In this formula, The larger the value, the more it means that the vibration signal has high-frequency characteristics and large fluctuations within the basic window. The greater the time-frequency complexity, the smaller the window is needed to capture signal details, and vice versa.

[0050] Next, adjust the size of the basic window according to the following formula:

[0051]

[0052] In this formula, The vibration signal The size of the basic window after sampling points is adjusted, is the basic window length, is the minimum window length, (experience points), is the maximum window length, (experience points), Is the boundary constraint function, used to limit The value of [ ] range to avoid numerical anomalies, The vibration signal The time-frequency complexity of the basic window of sampling points, is the mean of the time-frequency complexity of all sampling points of the vibration signal within the basic window, is the natural exponential function, is the hyperbolic tangent function.

[0053] when When The time-frequency complexity within the basic window of each sampling point is small, so it is necessary to enlarge the window appropriately to integrate the information of more sampling points. Less than 0, The value range is , The value range is ,but Greater than 1, to achieve The amplification and time-frequency complexity The smaller, The closer 1, The larger the size, the greater the magnification of the window.

[0054] when When The time-frequency complexity within the basic window of each sampling point is relatively large, so it is necessary to appropriately narrow the window to capture more signal details. greater than 0, The value range is , Less than 0, then Less than 1, to achieve The reduction of time-frequency complexity The bigger, The closer , The smaller it is, the more the window will be reduced.

[0055] when When The time-frequency complexity within the basic window of the sampling points is at a general level, and it is sufficient to maintain the original basic window. 0, , , ,Keep constant.

[0056] In summary, by comparing the time-frequency complexity within a base window with the global mean, an adaptive window adjustment model is constructed using hyperbolic tangent and exponential functions: when the time-frequency complexity is lower than the mean, the base window is exponentially enlarged as the complexity decreases to incorporate more stationary signal features; when the time-frequency complexity is higher than the mean, the base window is exponentially reduced as the complexity increases to capture high-frequency details; when the time-frequency complexity equals the mean, the base window is maintained. Combined with a boundary constraint function, this scheme achieves adaptive analysis by globally enhancing the low-frequency stationary segments of the vibration signal and locally focusing on the high-frequency complex segments, effectively balancing noise suppression and feature extraction accuracy.

[0057] Finally, for the vibration signal , with the first The sampling point is the center, and the sampling point itself and the points before it are sampling point and the sampling points together constitute the local area of ​​the sampling point.

[0058] S3: Filter all initial extreme points according to the local mean of each sampling point.

[0059] For the vibration signal sampling points, the The mean of the amplitudes of all sampling points in the local area of ​​the sampling point is taken as the The local mean of all sampling points of the vibration signal is obtained using the same method.

[0060] A coordinate system is constructed with the serial number of the sampling point as the horizontal coordinate and the local mean of the sampling point as the vertical coordinate. The local mean of each sampling point corresponds to a point in the coordinate system. These points are connected in sequence to obtain the local mean curve of all sampling points of the vibration signal.

[0061] On the local mean curve, if a local mean is greater than the local mean adjacent to its left and the local mean adjacent to its right, the sampling point corresponding to the local mean is taken as the maximum point; if a local mean is smaller than the local mean adjacent to its left and the local mean adjacent to its right, the sampling point corresponding to the local mean is taken as the minimum point. All the maximum points and all the minimum points constitute all the initial extreme points of the vibration signal.

[0062] S4: Evaluate the effectiveness of each initial extreme point and screen the true extreme points based on the effectiveness.

[0063] In vibration signal analysis, due to noise interference, signal non-stationarity, and the complex operating conditions of mechanical systems, the initial extreme points obtained through local mean screening may contain a large number of false feature points. These points may be local fluctuations caused by random noise or atypical extreme values ​​caused by changes in the signal's time-frequency coupling characteristics. Therefore, it is necessary to quantitatively evaluate the effectiveness of the extreme points, eliminate unreliable initial extreme points, and retain extreme points that truly reflect the signal's physical characteristics, thereby providing an accurate foundation for subsequent vibration signal decomposition.

[0064] Specifically, if Figure 2 Shown, including:

[0065] S41: Evaluate the amplitude mutation degree of each initial extreme point.

[0066] Extreme points are local maxima or minima in a vibration signal. They can be caused by real shocks, such as transient vibrations caused by gear meshing, which create physically meaningful extreme points, as well as pseudo-shocks, such as random extreme points caused by noise or disturbances, which lack regularity. The amplitude mutation degree is analyzed for all initial extreme points, and the signal characteristics of their local regions are used to determine whether they correspond to extreme points of real shocks.

[0067] During gearbox operation, whether meshing normally or under the impact of a fault, the vibration signal experiences a sudden amplitude change at the moment of engagement, significantly deviating from the mean. Real shocks (such as normal meshing or faults) disrupt the stability of the local signal, causing the amplitude to deviate significantly from the mean. However, noise or signal fluctuations can also occasionally cause significant deviations. To further accurately distinguish, the metric of accumulated energy in a local region is introduced to analyze the signal in the local region around the initial extreme point. The waveform of a real shock (such as normal meshing or faults) typically manifests as a sudden, high-amplitude spike. At the moment of impact, the amplitude is significantly higher than the surrounding signal, forming a distinct peak. After the impact, the signal rapidly decays, and the waveform exhibits symmetry in the time domain, with energy concentration near the impact point. In contrast, waveforms generated by random noise interference typically exhibit isolated spikes or irregular fluctuations, with energy contribution primarily dependent on high-amplitude points, and no significant energy accumulation near these points.

[0068] Therefore, the amplitude deviation degree of the initial extreme point and the cumulative energy of the local area of ​​the initial extreme point can be combined to quantify the degree of amplitude mutation. The former reflects the amplitude deviation degree of the initial extreme point itself, and the latter further combines the signal characteristics of the local area of ​​the initial extreme point, which is similar to confidence assessment, to avoid misjudging the amplitude deviation caused by accidental noise as a real impact.

[0069] In short, if the amplitude deviation of an initial extreme point is greater and the cumulative energy of its local area is greater, the initial extreme point is more likely to be generated by a real shock, otherwise it is more likely to be generated by a false shock.

[0070] In one embodiment, the degree of deviation of the amplitude of each initial extreme point is determined by the following method:

[0071] Obtain the mean and standard deviation of the amplitudes of all sampling points within the local range of each initial extreme point; calculate the absolute value of the difference between the amplitude of the initial extreme point and the mean, and use the ratio of the absolute value to the standard deviation as the degree of deviation of the amplitude of the initial extreme point.

[0072] This operation essentially converts "absolute amplitude deviation" into "relative deviation relative to the local fluctuation level," that is, the standardized degree of amplitude deviation. The absolute value of the difference between the amplitude of the initial extreme point and the mean reflects the absolute deviation of the amplitude at that extreme point, and the standard deviation reflects the degree of fluctuation within the local area. The greater the fluctuation, the less significant the deviation, and vice versa.

[0073] In one embodiment, the cumulative energy of the local area of ​​each initial extreme point is equal to the square sum of the amplitudes of all sampling points in the local area. The larger the cumulative energy, the more energy concentration characteristics there are in the local area, and the more likely it is a true extreme point.

[0074] In one embodiment, the degree of amplitude mutation at each initial extreme point is equal to the product of the amplitude deviation of the initial extreme point and the cumulative energy of the local region of the initial extreme point. The amplitude deviation of the initial extreme point only reflects the amplitude characteristics of the extreme point itself, while the cumulative energy reflects the signal characteristics of the local region of the initial extreme point. The combination of the two can more comprehensively distinguish between true shocks and false shocks. The greater the amplitude deviation of an initial extreme point and the greater the cumulative energy of its local region, the more consistent the initial extreme point with the characteristics of a true shock and the more likely it is a true extreme point. Conversely, the more likely it is a false extreme point caused by accidental noise or signal fluctuations.

[0075] S42: Calculate the time-frequency coupling index of each initial extreme point.

[0076] When early wear occurs on gears, the amplitude mutation of the extreme points of the vibration signal is weak. In this case, gear wear will change the meshing stiffness, thereby causing regular changes in the relationship between the frequency components and phase of the vibration signal. Therefore, by analyzing the frequency consistency and phase consistency of each initial extreme point and calculating the time-frequency coupling index, it is possible to capture the anomalies of the characteristic frequency energy distribution and the phase synchronization changes of each frequency component at the extreme points from another dimension.

[0077] The time-frequency coupling index starts from the essential characteristics of the signal in the time-frequency domain and combines the dynamic laws of gear motion. On the basis of the degree of amplitude mutation, it further realizes the accurate identification of the extreme points of weak wear characteristics and eliminates the pseudo extreme points, providing a more reliable basis for subsequent signal decomposition and fault diagnosis.

[0078] Specifically, they include:

[0079] S421: Extract the instantaneous frequency and instantaneous phase of each initial extreme point.

[0080] EMD (Empirical Mode) decomposition and Hilbert-Transform are both classic time-frequency analysis techniques in the field of signal processing. By using EMD decomposition to decompose the vibration signal into multiple IMFs (Intrinsic Mode Components), and then performing a Hilbert transform on each IMF, the instantaneous frequency and instantaneous phase corresponding to each sampling point of the vibration signal can be obtained, and the instantaneous frequency and instantaneous phase corresponding to each initial extreme point of the vibration signal can be obtained.

[0081] For instantaneous frequency:

[0082] When the gears are meshing, the impact frequency generated by the periodic contact of the gear teeth is related to the inherent reference frequency of the gear meshing. The reference frequency is directly determined by the gear speed and the number of teeth. ,in is the speed in revolutions per minute, is the number of gear teeth, indicating the number of times the gear teeth mesh per unit time, reflecting the periodic impact frequency of a single gear meshing. The harmonic frequency of gear meshing is an integer multiple of the reference frequency and is a high-order frequency component.

[0083] The vibration signal generated by gear meshing or fault impact is periodic. When the gear teeth mesh once, the impact frequency corresponds to the reference frequency (the harmonic frequency is 1 times the reference frequency). When high-order meshing vibration or harmonics caused by faults occur, the instantaneous frequency will be an integer multiple of the reference frequency (such as 2 times, 3 times, etc.), which is related to the harmonic frequency. This is consistent with the vibration mechanism of gear transmission. Periodic impacts will form regular components related to the harmonic frequency in the time-frequency domain.

[0084] True extreme points usually correspond to gear meshing or fault shocks. Therefore, the instantaneous frequency of true extreme points is usually regular with the harmonic frequency, that is, the instantaneous frequency is related to the integer multiple of the reference frequency. The instantaneous frequency of pseudo-extreme points caused by noise may show irregular fluctuations or random jumps between multiple frequency points, but will not form an integer multiple aggregation phenomenon corresponding to the reference frequency. In other words, the instantaneous frequency is unrelated to the harmonic frequency, which is in sharp contrast to the regular frequency characteristics of true fault shocks.

[0085] For the instantaneous phase,

[0086] In gear vibration signal analysis, instantaneous phase is a key indicator of the signal's relative position within a cycle. The fixed position of the tooth top and tooth groove meshing during gear rotation corresponds to the gear's reference phase (theoretical meshing phase). The instantaneous phase of a true extreme point (caused by normal meshing or fault impact) accurately reflects the angle (or arc) of the impact. Its difference from the reference phase is typically within a reasonable tolerance and exhibits a stable regularity across adjacent impact cycles. During uniform rotation, the phase remains consistent at the same position in each meshing cycle. Although periodic shifts may occur during a fault due to changes in the impact moment, the pattern of these shifts remains related to the gear's reference parameters. Conversely, the instantaneous phase of a pseudo-extreme point induced by noise differs significantly from the reference phase, and this difference exhibits a random distribution. The instantaneous phase exhibits no fixed pattern at different times, and the phase variation lacks correlation with parameters such as gear speed and number of teeth. The phase curve often exhibits irregular jumps or nonlinear fluctuations, failing to match the periodic phase variation of the gear mesh.

[0087] S422: Obtain the frequency consistency of each initial extreme point.

[0088] Frequency consistency is determined based on the difference between the instantaneous frequency of the initial extreme point and the reference frequency. The greater the difference, the worse the consistency, and vice versa. It is used to determine whether the instantaneous frequency of each initial extreme point is a frequency related to gear meshing.

[0089] In one embodiment, frequency consistency is determined based on:

[0090] Get the base frequency , then The harmonic frequencies are , (Experience value), 1 times, 2 times, and 3 times the harmonic frequency can better meet the needs of frequency consistency analysis and judgment, and are sufficient to cover common frequency characteristics and achieve effective evaluation of related object characteristics. When the harmonic order is too high, its energy usually decays rapidly and has little effect on the overall frequency characteristics. It can not only cover the harmonic components with relatively large energy and significant impact on system characteristics, but also will not increase the computational complexity and analysis difficulty due to the introduction of too many high-order harmonics, thus achieving a good balance between analysis effectiveness and ease of operation.

[0091] therefore, 、 and Represent the 1st, 2nd and 3rd harmonic frequencies (1 times, 2 times and 3 times the base frequency) respectively. The initial extreme point and the reference frequency The consistency of the harmonic frequencies:

[0092]

[0093] In this formula, For the The initial extreme point and the The consistency of the harmonic frequencies, is the tolerance of the reference frequency, is the natural exponential function, For the The instantaneous frequency of the initial extreme point, is the reference frequency, For the harmonic frequencies.

[0094] This formula measures the matching degree between the instantaneous frequency of each initial extreme point and the harmonic frequency through a Gaussian function. It is essentially a smooth, symmetrical, and rapidly decaying matching scoring function. is a natural exponential function. The instantaneous frequency of the initial extreme point Close to harmonic frequency hour, Approaching 0, The closer it is to 1, the higher the matching degree between the two. The more likely the initial extreme point is to be the real extreme point. The instantaneous frequency of the initial extreme point The further away from the harmonic frequency hour, The larger it is, the closer it is to 1. The closer it is to 0, the lower the matching degree between the two. The more likely the initial extreme points are to be pseudo extreme points.

[0095] In this formula, Satisfies the following relationship:

[0096]

[0097] This formula normalizes the pitch error to the full circumference of the gear and maps it to the frequency space, thereby obtaining a physically based frequency offset tolerance for estimating the standard deviation of the Gaussian matching function. Total Composite Pitch Deviation (Total Composite Pitch Deviation) is the maximum pitch error between any two teeth in one revolution of the gear. This indicator is used to reflect the maximum unevenness of the pitch distribution. It is obtained according to the general technical standards of the gear industry, for example, according to GB / T10095.1-2021 "Cylindrical Gear Precision Systems - Part 1: Definition and Permissible Values ​​of Tooth Flank Deviation on the Same Side of the Gear". The larger it is, the easier it is for the meshing frequency of the gear to fluctuate. z is the number of teeth on the gear, is the tooth pitch of the gear. The product of the number of teeth and the tooth pitch represents the theoretical circumference of the gear. It is a dimensionless ratio, which is equivalent to normalizing the maximum pitch error to the entire circumference. It conforms to the physical meaning of frequency as a parameter of whole-circle motion, ensures the accuracy of relative error calculation, and is consistent with the calculation logic of frequency fluctuation (frequency is a statistic of whole-circle motion). Multiply The pitch error mapping is transformed into frequency fluctuation error. Specifically, according to gear kinematics, the theoretical time for a gear to rotate one circle is , is the gear speed in revolutions per minute, the theoretical meshing frequency When there is a pitch error, the actual circumference becomes , causing the rotation period to fluctuate , the relative error of the circumference is ,according to is equal to the ratio of circumference to linear velocity, we can get: ; Since frequency is inversely proportional to period , then the frequency fluctuation is: Therefore, multiply by The relative error Necessary steps to convert to absolute frequency fluctuations.

[0098] Finally, the The maximum value of the consistency between the initial extreme points and all harmonic frequencies is recorded as , and as the first The frequency consistency of the initial extreme points is the same. This strategy of taking the maximum value can not only avoid the limitation of a single harmonic frequency, but also directly quantify the authenticity of the extreme point through the numerical value. The greater the consistency, the more it conforms to the inherent law of gear vibration, and the more likely it is the extreme point generated by a real impact.

[0099] S423: Obtain the phase consistency of each initial extreme point.

[0100] Phase consistency is determined based on the difference between the instantaneous phase of the initial extreme point and the reference phase. The larger the difference, the worse the consistency, and vice versa. It is used to determine whether the phase of the initial extreme point occurs near the theoretical meshing phase, so as to further verify whether the extreme point is a true extreme point.

[0101] In one embodiment, the phase consistency of each initial extreme point satisfies the following relationship:

[0102]

[0103] In this formula, For the The phase consistency of the initial extreme points, For the The instantaneous phase of the initial extreme point, is the reference phase, is the natural exponential function, is the tolerance of the reference phase, Controls whether a certain instantaneous phase is close enough to the theoretical meshing phase. The larger the value, the looser the match, and the smaller the value, the stricter the match.

[0104] This formula measures phase consistency through Gaussian function and is suitable for determining whether the phase of a certain initial extreme point occurs near the reference phase, so as to further verify whether the extreme point may be caused by gear meshing behavior. The closer , indicating that when the instantaneous phase of the extreme point is close to the reference phase, The closer to 0, The closer it is to 1, the higher the phase synchronization between the initial extreme point and the reference phase is. On the contrary, when The more deviated The farther, The closer it is to 1, The closer it is to 0, the worse the synchronization of the extreme point is, and it may be a pseudo extreme point.

[0105] In this formula, The calculation method is:

[0106]

[0107] Among them, TCPD is the maximum pitch deviation of the gear, The pitch circle radius of the gear is the core concept of gear transmission theory. When the gear is transmitting, the linear speed on the pitch circle is equal, and the torque and speed are also defined based on the pitch circle. The unit is millimeter, , The module is in millimeters and is the standard size parameter of the gear, representing the ratio of the tooth pitch to the circumference. is the number of teeth on the gear.

[0108] In summary, the phase consistency of each initial extreme point reflects the mathematical correspondence between the angle fluctuation caused by the pitch error and the synchronization phase, which enhances the physical consistency and scene adaptability of phase judgment.

[0109] in, The derivation process is: Due to the pitch error , then the estimated gear meshing time is offset by : , is the maximum pitch deviation, is the gear meshing linear velocity. Then convert the time offset into phase error: the gear angular frequency is , is the speed of the gear, then: , and then according to the gear peripheral speed formula: , we can get: , which simply estimates the effect of the gear's own pitch error on the phase, and is therefore used to represent the allowable phase fluctuation range during gear meshing, i.e., the phase tolerance.

[0110] S424: Determine a time-frequency coupling index based on frequency consistency and phase consistency.

[0111] The frequency consistency and phase consistency of each initial extreme point are fused by the geometric mean method to obtain the time-frequency coupling index of the initial extreme point. Specifically, the following relationship is satisfied:

[0112]

[0113] In this formula, For the The time-frequency coupling index of the initial extreme point, For the The frequency consistency of the initial extreme points. For the Phase consistency of the initial extreme points. The essence of taking the square root of the product is to take the geometric mean of the two indicators. The geometric mean is a common fusion method when both indicators are important and cannot be "inflated" by a unilateral maximum value. Its characteristic is that it gives equal "multiplicative weight" to both factors. If either factor is too low, the result will be significantly suppressed, but it will not drop as quickly as the product, thereby increasing fault tolerance.

[0114] This dual constraint mechanism not only conforms to the physical laws of gear dynamics, but also has been verified through engineering practice for its noise resistance and fault sensitivity. It is an essential link to improve the accuracy of gear fault detection.

[0115] S43: Validity is determined by combining the amplitude mutation degree and time-frequency coupling index.

[0116] Specifically, for each initial extreme point, the product of its time-frequency coupling index and the amplitude mutation degree is used as the validity of the initial extreme point. The time-frequency coupling index, as an effective supplement to the amplitude mutation degree, enhances the credibility of the identification of true extreme points from multiple dimensions.

[0117] A dual criterion is formed through the multiplication mechanism to ensure that only the initial extreme points with significant amplitude mutation and time-frequency characteristics that are more consistent with the laws of gear motion will be judged as true extreme points, thereby effectively improving the recognition accuracy and suppressing noise interference.

[0118] S44: Filter true extreme points based on effectiveness.

[0119] Setting dynamic thresholds :

[0120]

[0121] In this formula, and The mean and standard deviation of the effectiveness of all initial extreme points of the vibration signal are respectively calculated. The mean is the average level and the standard deviation is the fluctuation range. The two are added together to obtain the reasonable range of effectiveness. is the working condition adjustment coefficient (e.g. 1.2 for high-speed working condition and 0.8 for low-speed working condition). The threshold is dynamically adjusted according to the overall energy distribution of the signal.

[0122] Among all the initial extreme points, if the validity of an initial extreme point is greater than or equal to , the initial extreme point is the true extreme point, otherwise it is a false extreme point and is eliminated. Finally, all the true extreme points of the vibration signal are obtained.

[0123] S5: Decompose the vibration signal based on the true extreme points and extract signal features for fault diagnosis.

[0124] The algorithm decomposes the complex signal into a series of physically meaningful Serving size, each Represents a natural vibration mode of the signal. The algorithm is based on the extraction of the true extreme points of the vibration signal Quantity, get to Represents different oscillation components from high frequency to low frequency, all The components are combined to form a feature vector .

[0125] Pre-detect the fault feature database, collect the gearbox gear vibration signals under different fault types and degrees, and follow the operations of steps S1 to S5 to obtain a set of feature vectors of multiple vibration signals corresponding to each vibration type.

[0126] The characteristic vector corresponding to the vibration signal of the gearbox gear to be detected , and each feature vector in the fault feature database is compared one by one, and the inverse of the Euclidean distance is used to measure the similarity between the two. The fault type corresponding to the feature vector with the largest similarity (the smallest Euclidean distance) is taken as the fault type of the gearbox gear to be detected, and the gearbox gear fault detection result is obtained.

[0127] While various embodiments of the present invention have been shown and described herein, it will be obvious to those skilled in the art that such embodiments are provided by way of example only.

Claims

1. A gear failure detection method for a gearbox, characterized in that: include: Collecting a gear vibration signal of a gearbox, wherein the gear vibration signal is composed of multiple sampling points; All initial extreme points of the vibration signal are screened, and the amplitude mutation degree of each initial extreme point is calculated by multiplying the amplitude deviation of the initial extreme point by the cumulative energy of the local area of ​​the initial extreme point. The cumulative energy of the local area is the square sum of the amplitudes of all sampling points in the local area. The amplitude deviation degree of each initial extreme point is obtained by obtaining the mean and standard deviation of the amplitudes of all sampling points within the local area of ​​each initial extreme point. The absolute value of the difference between the amplitude of the initial extreme point and the mean is calculated, and the ratio of this absolute value to the standard deviation is used as the degree of deviation of the amplitude of the initial extreme point. The degree of deviation of the amplitude of the initial extreme point reflects the amplitude characteristics of the extreme point itself, and the cumulative energy of the local area of ​​the initial extreme point reflects the signal characteristics of the local area of ​​the initial extreme point. The combination of the two can comprehensively distinguish between real shocks and false shocks. Extract the instantaneous frequency and instantaneous phase of each initial extreme point, fuse the frequency consistency and phase consistency of each initial extreme point through the geometric mean method, and obtain the time-frequency coupling index of the initial extreme point; wherein the frequency consistency is determined based on the difference between the instantaneous frequency of the initial extreme point and the reference frequency; the phase consistency is determined based on the difference between the instantaneous phase of the initial extreme point and the reference phase of the gear; The product of the time-frequency coupling index and the amplitude mutation degree is used as the validity of the initial extreme point. The true extreme point of the vibration signal is determined based on the comparison result between the validity and the preset threshold. Based on the true extreme points of the vibration signal, the inherent time scale decomposition algorithm is used to extract all the inherent rotational components of the vibration signal. All the inherent rotational components are combined to form a feature vector. A fault feature library containing various fault types of transmission gears is pre-established. Gear fault diagnosis is performed based on the matching results between the feature vector and the pre-established fault feature library.

2. The gear fault detection method according to claim 1, characterized in that: The method for screening all initial extreme points of the vibration signal is: Divide each sampling point into a local area, and calculate the mean of the amplitudes of all sampling points in the local area as the local mean of the sampling point, and construct a local mean curve with the sequence number of the sampling point as the horizontal coordinate and the local mean of the sampling point as the vertical coordinate; On the local mean curve, if a local mean is greater than the local mean adjacent to its left and the local mean adjacent to its right, the sampling point corresponding to the local mean is taken as the maximum point; if a local mean is less than the local mean adjacent to its left and the local mean adjacent to its right, the sampling point corresponding to the local mean is taken as the minimum point; all the maximum points and all the minimum points constitute all the initial extreme points of the vibration signal.

3. The gear fault detection method according to claim 2, characterized in that: The method of dividing the local area of ​​each sampling point is: With each sampling point as the center, a basic window is set, and the dominant frequency within the basic window is determined by the zero-crossing detection method. At the same time, the coefficient of variation of the amplitude of all sampling points within the basic window is calculated; The ratio of the dominant frequency to the maximum frequency in the basic window is used as the frequency adaptation factor of the sampling point. The amplitude variation coefficient is multiplied by the frequency adaptation factor to calculate the time-frequency complexity of the basic window. The size of the basic window is adjusted using the time-frequency complexity, and the range of the adjusted basic window is determined as the local area of ​​the sampling point; wherein the size of the adjusted range of the basic window is negatively correlated with the time-frequency complexity.

4. The gear fault detection method according to claim 3, characterized in that: The coefficient of variation of the amplitudes of all sampling points within the basic window is determined based on the ratio of the standard deviation to the mean of the amplitudes of all sampling points.

5. The gear fault detection method according to claim 1, characterized in that: The instantaneous frequency and instantaneous phase of each initial extreme point are extracted based on the following method: The vibration signal is decomposed into multiple intrinsic mode components through empirical mode decomposition, and then each intrinsic mode component is subjected to Hilbert transform to obtain the instantaneous frequency and instantaneous phase corresponding to each initial extreme point of the vibration signal.

6. The gear fault detection method according to claim 5, characterized in that: Frequency consistency is determined based on the following: Get the base frequency and multiple harmonic frequencies of the base frequency; calculate the The initial extreme point and the reference frequency The consistency of the harmonic frequencies: , in the formula, For the The initial extreme point and the The consistency of the harmonic frequencies, is the tolerance of the reference frequency, is the natural exponential function, For the The instantaneous frequency of the initial extreme point, is the reference frequency, For the harmonic frequencies; The first The maximum value of the consistency between the initial extreme point and all harmonic frequencies of the reference frequency is taken as the The frequency consistency of the initial extreme points.

7. The gear fault detection method according to claim 5, characterized in that: The phase consistency of each initial extreme point satisfies the following relationship: , in the formula, For the The phase consistency of the initial extreme points, For the The instantaneous phase of the initial extreme point, is the reference phase, is the natural exponential function, is the tolerance of the reference phase.

8. The gear fault detection method according to claim 3, characterized in that: The true extreme points of the vibration signal are determined as follows: The initial extreme value points whose validity is greater than the preset threshold are regarded as true extreme value points, and the initial extreme value points whose validity is not greater than the preset threshold are regarded as pseudo extreme value points and eliminated.

9. The gear fault detection method according to claim 1, characterized in that: The steps of performing gear fault diagnosis based on the matching results of the feature vector and the pre-established fault feature library include: The Euclidean distance between the feature vector and each feature vector in the fault feature library is calculated, and the fault type corresponding to the feature vector with the smallest Euclidean distance is taken as the diagnosis result.

Images