Gear fault feature extraction method based on improved spectral kurtosis analysis

Abstract

The invention aims to provide a gear fault feature extraction method based on improved spectral kurtosis analysis, and belongs to the field of gear fault diagnosis. The method comprises three parts of impact component analysis, frequency band weight setting and fault feature extraction, and comprises the following steps: firstly, carrying out short-time Fourier transform on a vibration signal in a given window function, and calculating respective spectral kurtosis for different frequency bands; carrying out time domain average processing on the vibration signal to obtain a gear rotation frequency related signal, calculating a power spectrum of the gear rotation frequency related signal and an energy ratio of different frequency bands in the power spectrum, setting a weighted value of the spectrum kurtosis of the corresponding frequency band according to the power spectrum, and summarizing to form a weighted spectrum kurtosis graph; and finally, performing envelope spectrum analysis by adopting Hilbert transformation, extracting fault features, and determining the position where the gear fault occurs. According to the method, the influence of interference signals on fault impact frequency band selection is reduced, so that the accuracy of fault feature extraction is improved, and the method is suitable for feature extraction and fault position judgment in an early fault state.

Description

Technical Field

[0001] The present invention relates to a fault diagnosis method, in particular to a gear fault diagnosis method. Background Art

[0002] Gearboxes are the most commonly used transmission components in machinery and equipment. They can achieve different proportions of power output by selecting different sizes of gears for combination. Gears are important components in gearboxes. Due to the influence of the working environment, friction and collision of components, and human error, gears often break, crack, and other faults, resulting in reduced work efficiency and increased accident risks.

[0003] In actual engineering practice, in order to extract gear fault features and confirm the location of the fault, the spectral kurtosis analysis method is often used to locate the effective frequency band, and the fault features are extracted through envelope analysis to achieve fault location. However, in the case of early weak faults, due to the influence of fault noise and complex transmission paths, it is difficult to effectively extract the early fault features of the gear, so it is necessary to enhance the accuracy of the extraction of early weak gear fault features. Summary of the invention

[0004] The purpose of the present invention is to provide a gear fault feature extraction method based on improved spectral kurtosis analysis, which can solve the problem that impact features caused by early gear faults are difficult to capture.

[0005] The object of the present invention is achieved in that:

[0006] The present invention provides a gear fault feature extraction method based on improved spectral kurtosis analysis, which is characterized by comprising the following steps:

[0007] (1) Obtaining the vibration signal of the gear box: Installing a sensor on the gear box to collect the vibration signals generated by the gear box under different conditions;

[0008] (2) Improved spectral kurtosis analysis: including impulse component analysis, setting frequency band weights, and calculating weighted spectral kurtosis of each frequency band;

[0009] (3) Fault feature extraction: including extraction of impulse frequency band and Hilbert envelope demodulation.

[0010] The present invention may also include:

[0011] 1. In step (2), the impact component analysis uses short-time Fourier transform and spectral kurtosis analysis to detect the impact components contained in different frequency bands:

[0012] Short-time Fourier processing of vibration signals:

[0013] The continuous vibration signal x(t) generated by the gear box is discretely sampled to obtain the signal x(n). The mathematical expression of the short-time Fourier transform of x(n) is as follows:

[0014]

[0015] In the formula, x(n) represents the input discrete vibration signal; w(n-kP) represents the window function centered on kP; k represents the number of analysis; P represents the time step; f represents the frequency; X(kP,f) represents the short-time Fourier transform value of the vibration signal;

[0016] Calculate the spectral kurtosis of each frequency band of the vibration signal:

[0017] The spectral kurtosis of the vibration signal of each layer and each frequency band obtained by short-time Fourier transform is calculated:

[0018] Calculate the 2nth order instantaneous moment S of X(kP,f) 2nX (f), used to estimate the energy intensity of the short-time Fourier transform signal X(kP,f):

[0019]

[0020] Where <> k represents k times time average;

[0021] Calculate the fourth-order moment cumulant used to measure the impulse component of the signal in this frequency band:

[0022]

[0023] Among them, C 4X (f) represents the fourth-order moment accumulation of the signal in this frequency band; S 4X (f) represents the fourth-order instantaneous moment of the signal in this frequency band; S 2X (f) represents the second-order instantaneous moment of the signal in this frequency band;

[0024] Calculate the fourth-order normalized cumulant of the signal in this frequency band to determine the spectral kurtosis value:

[0025]

[0026] Wherein, K(f) represents the spectral kurtosis value of the signal in this frequency band.

[0027] 2. Setting the frequency band weight in step (2) includes using time domain averaging to process the vibration signal and improving the spectral kurtosis analysis method by setting the spectral kurtosis weight for the signal frequency band, thereby determining the fault frequency band:

[0028] Time domain averaging:

[0029] The signal is intercepted and averaged:

[0030] Set the gear shaft rotation frequency to f a , the sampling frequency is Fs, then the sampling time interval is Based on this, the truncated period length l of the vibration signal is determined as:

[0031]

[0032] Where round() represents the rounding function;

[0033] The vibration signal x(n) is decomposed according to the obtained truncated period length to obtain a new vibration signal sequence x1(l), x2(l), x3(l)······, x N (l), add the vibration signal sequence according to the corresponding points and take the average value to obtain the frequency-related signal y(l):

[0034]

[0035] Among them, x n (l) represents the intercepted sub-signal; N represents the number of vibration signal segments;

[0036] Calculate the energy weight of each frequency band:

[0037] The power spectrum of the vibration signal y(l) is calculated, the weight of each frequency band is set, and the spectral kurtosis of each frequency band is weighted to improve the spectral kurtosis analysis results:

[0038] Perform Fourier transform on the vibration signal y(l):

[0039]

[0040] Wherein, Y(f) represents the Fourier transform value of the vibration signal at a certain frequency; j represents the imaginary unit;

[0041] Calculate its power spectrum:

[0042] P(f)=|Y(f) 2

[0043] Wherein, P(f) represents the power spectrum value of the vibration signal at a certain frequency;

[0044] Divide the power spectrum according to the number of frequency bands divided by the short-time Fourier transform, and calculate the energy mean of each divided area As the trust weight of the spectral kurtosis of this frequency band;

[0045] The trust weights calculated for each frequency band under different window lengths are summarized to form a weight matrix Z(m1,m2):

[0046]

[0047] in, Indicates the calculated frequency band energy mean; m1 indicates the layer where the calculated frequency band is located; m2 indicates the position of the calculated frequency band in this layer.

[0048] 3. Step (2) of calculating the weighted spectral kurtosis of each frequency band includes:

[0049] Multiply the spectral kurtosis of each frequency band in each layer by the corresponding frequency band weight value to obtain their respective weighted spectral kurtosis:

[0050]

[0051] in represents the weighted spectral kurtosis of the calculated frequency band; Represents the original spectral kurtosis of the calculated frequency band; Indicates the energy weight of the calculated frequency band.

[0052] 4. Extracting the impact frequency band in step (3) includes:

[0053] Summarize the weighted spectral kurtosis of all frequency bands in each layer for comparison, find the frequency band with the maximum weighted spectral kurtosis, and calculate its corresponding center frequency f i With bandwidth BW:

[0054]

[0055] BW=2 k

[0056] In the formula, f s represents the sampling frequency; k represents the layer where the maximum weighted spectral kurtosis is located; i represents the sub-frequency band where the maximum weighted spectral kurtosis of the kth layer is located.

[0057] 5. In step (3), Hilbert envelope demodulation includes:

[0058] The selected impact frequency band is band-pass filtered, and the obtained signal is subjected to Hilbert envelope analysis. The frequency with the maximum amplitude in the envelope spectrum is defined as the fault characteristic parameter, and the fault location is determined by detecting its correlation with the rotation frequency of the driving gear and the driven gear.

[0059] The advantages of the present invention are as follows: the present invention uses the time domain averaging method to extract relevant signals, calculates its power spectrum and divides it into frequency bands, calculates the percentage of the energy mean of each frequency band relative to the total energy and sets the trust weight of the corresponding spectral kurtosis accordingly, and performs weighted processing on the spectral kurtosis obtained based on the short-time Fourier transform, thereby reducing the influence of interference signals on the selection of fault impact frequency bands, thereby improving the accuracy of fault feature extraction, and is suitable for feature extraction and fault location judgment in early fault states. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a flow chart of the present invention;

[0061] Figure 2 The original signal waveform of the driving gear under the early crack fault state at 1224rpm and 6N·m;

[0062] Figure 3 The original signal waveform of the active gear in the early stage of broken tooth fault state under the working condition of 1224rpm and 6N·m;

[0063] Figure 4 The envelope spectrum of the vibration signal of the active gear in a healthy state under the working condition of 1224rpm and 6N·m is obtained based on the improved spectral kurtosis method;

[0064] Figure 5 The envelope spectrum of the vibration signal of the active gear under the condition of 1224rpm and 6N·m in the early crack fault state is obtained based on the improved spectral kurtosis method;

[0065] Figure 6 The envelope spectrum of the vibration signal of the active gear under the condition of early tooth breakage fault at 1224rpm and 6N·m based on the improved spectral kurtosis method;

[0066] Figure 7 The spectral kurtosis comparison diagram of the vibration signal under the early crack fault state of the active gear based on the improved spectral kurtosis method and the traditional spectral kurtosis method;

[0067] Figure 8 This is a comparison diagram of the spectral kurtosis of the vibration signal under the early tooth breakage fault state of the active gear based on the improved spectral kurtosis method and the traditional spectral kurtosis method;

[0068] Fig. 9 This is a comparison diagram of the envelope spectra obtained by the improved spectral kurtosis method and the traditional spectral kurtosis method for the vibration signal of the active gear under the early crack fault state. DETAILED DESCRIPTION

[0069] The present invention is described in more detail below with reference to the accompanying drawings:

[0070] Combination Figure 1-9 , the present invention comprises the following steps:

[0071] 1. Obtain the gear box vibration signal

[0072] Since the gear vibration signal has high sensitivity and can reflect the changes caused by faults, a sensor with satisfactory accuracy can be installed on the gear box to collect the vibration signals generated by the gear box in different states for later analysis.

[0073] 2. Improved spectral kurtosis analysis

[0074] 2.1 Analysis of impact components

[0075] In the early fault state, the impact component contained in the gear vibration signal is small and it is difficult to distinguish the content in different frequency bands. To address this problem, this patent uses short-time Fourier transform and spectral kurtosis analysis to detect the impact components contained in different frequency bands.

[0076] 2.1.1 Short-time Fourier processing of vibration signals

[0077] Since the essence of spectral kurtosis is the local characteristics at a certain frequency in the process, it is necessary to convert the signal from the time domain to the frequency domain and divide the frequency band during the calculation process. The frequency domain information obtained by fast Fourier transform is difficult to reflect the actual characteristics of the vibration signal in the frequency domain. In contrast, short-time Fourier transform can analyze the signal in the window of different time periods by setting the window function, which plays the role of time-frequency analysis and frequency band division. Therefore, this patent adopts the method of short-time Fourier transform under the window length obtained by two-layer and three-layer to obtain the frequency-amplitude characteristics of the gear vibration signal in different frequency bands.

[0078] The continuous vibration signal x(t) generated by the gear box is discretely sampled to obtain the signal x(n). The mathematical expression of the short-time Fourier transform of x(n) is as follows:

[0079]

[0080] Wherein, x(n) represents the input discrete vibration signal; w(n-kP) represents the window function centered on kP; k represents the number of analyses; P represents the time step; f represents the frequency; X(kP,f) represents the short-time Fourier transform value of the vibration signal.

[0081] 2.1.2 Calculation of spectral kurtosis of vibration signal in each frequency band

[0082] The fault impact component of the vibration signal taken in the early fault state is not obvious. In order to measure the impact characteristics of each frequency band, this patent calculates the spectral kurtosis of the vibration signal of each layer and each sub-frequency band obtained by short-time Fourier transform. The specific steps are as follows:

[0083] Calculate the spectral kurtosis of the vibration signal of each layer and each frequency band obtained in step 2.1.1. The specific steps are as follows:

[0084] First calculate the 2nth order instantaneous moment S of X(kP,f) 2nX (f), which is used to estimate the energy intensity of the short-time Fourier transform signal X(kP,f), as shown in the following formula:

[0085]

[0086] Where <> krepresents k times time average.

[0087] After that, the fourth-order moment cumulant used to measure the impact component of the signal in this frequency band is calculated as shown in the following formula:

[0088]

[0089] Among them, C 4X (f) represents the fourth-order moment accumulation of the signal in this frequency band; S 4X (f) represents the fourth-order instantaneous moment of the signal in this frequency band; S 2X (f) represents the second-order instantaneous moment of the signal in this frequency band.

[0090] Finally, the fourth-order normalized cumulant of the frequency band signal is calculated to determine the spectral kurtosis value, as shown in the following formula:

[0091]

[0092] Wherein, K(f) represents the spectral kurtosis value of the signal in this frequency band.

[0093] 2.2 Setting frequency band weights

[0094] In the early fault state, the fault characteristics of the signal are small, and the interference signals generated by environmental noise, transmission path and other factors sometimes have a certain impact on the high-frequency part, making the frequency band with smaller fault impact components of this type of fault have a larger spectral kurtosis, resulting in the fault impact characteristics caused by gear faults in the final selected vibration signal frequency band not being obvious. To address this problem, this patent uses time domain averaging to process vibration signals and sets spectral kurtosis weights for signal frequency bands to improve the spectral kurtosis analysis method, thereby determining the fault frequency band.

[0095] The details are as follows:

[0096] 2.2.1 Time Domain Averaging Processing

[0097] The signal generated by gear vibration is unstable, and in addition to the fault component, the vibration signal also contains interference components caused by working noise, vibration path effect, etc. In order to improve the signal-to-noise ratio of the vibration signal, this patent uses the time domain averaging method to process the vibration signal. By intercepting the signal multiple times according to a certain length and adding the average value, it plays a role in reducing noise and highlighting the characteristics of vibration-related signals. The specific steps are as follows:

[0098] Set the gear shaft rotation frequency to f a , the sampling frequency is Fs, then the sampling time interval is Based on this, the truncated period length l of the vibration signal can be determined as:

[0099]

[0100] Where round() represents the rounding function.

[0101] The vibration signal x(n) is decomposed according to the obtained truncated period length to obtain a new vibration signal sequence x1(l), x2(l), x3(l)······, x N (l), add the vibration signal sequence according to the corresponding points and take the average value to obtain the frequency-related signal y(l), as shown in the following formula:

[0102]

[0103] Among them, x n (l) represents the intercepted sub-signal; N represents the number of vibration signal segments.

[0104] 2.2.2 Calculate the energy weight of each frequency band

[0105] Since the spectral kurtosis analysis method has poor resistance to interference components, the obtained frequency band may not have high fault characteristics. In order to reduce the influence of interference components, this patent calculates the power spectrum of the vibration signal y(l) obtained in step 2.2.1, sets the weight of each frequency band and performs weighted processing on the spectral kurtosis of each frequency band, thereby improving the spectral kurtosis analysis results. The details are as follows:

[0106] Perform Fourier transform on the vibration signal y(l):

[0107]

[0108] Wherein, Y(f) represents the Fourier transform value of the vibration signal at a certain frequency; j represents the imaginary unit.

[0109] Calculate its power spectrum:

[0110] P(f)=|Y(f) 2 (8)

[0111] Wherein, P(f) represents the power spectrum value of the vibration signal at a certain frequency.

[0112] Divide the power spectrum according to the number of frequency bands divided by the short-time Fourier transform, and calculate the energy mean of each divided area As the trust weight of the spectral kurtosis of this frequency band.

[0113] The trust weights calculated for each frequency band under different window lengths are summarized to form a weight matrix Z(m1,m2). The calculation formula is as follows:

[0114]

[0115] in, Indicates the calculated frequency band energy mean; m1 indicates the layer where the calculated frequency band is located; m2 indicates the position of the calculated frequency band in this layer.

[0116] 2.3 Calculate the weighted spectral kurtosis of each frequency band

[0117] Multiply the spectral kurtosis of each frequency band of each layer by the corresponding frequency band weight value in step 2.2.2 to obtain their respective weighted spectral kurtosis. The calculation formula is as follows:

[0118]

[0119] in represents the weighted spectral kurtosis of the calculated frequency band; Represents the original spectral kurtosis of the calculated frequency band; Indicates the energy weight of the calculated frequency band.

[0120] 3. Fault feature extraction

[0121] 3.1 Extracting the impact frequency band

[0122] Summarize the weighted spectral kurtosis of all frequency bands in each layer for comparison, find the frequency band with the maximum weighted spectral kurtosis, and calculate its corresponding center frequency f i and bandwidth BW, as shown below:

[0123]

[0124] BW=2 k (12)

[0125] In the formula, f s represents the sampling frequency; k represents the layer where the maximum weighted spectral kurtosis is located; i represents the sub-frequency band where the maximum weighted spectral kurtosis of the kth layer is located.

[0126] 3.2 Hilbert Envelope Demodulation

[0127] The selected impact frequency band is band-pass filtered, and the obtained signal is subjected to Hilbert envelope analysis. The frequency with the maximum amplitude in the envelope spectrum is defined as the fault characteristic parameter, and the fault location is determined by detecting its correlation with the rotation frequency of the driving gear and the driven gear.

[0128] Example:

[0129] Step 1: Get the gearbox vibration signal

[0130] The data in this embodiment is a gear box vibration signal x(n) obtained by discrete sampling when an early crack occurs in an active gear in a fixed-axis gear box driven by a motor and rotated by a load, and an early tooth breakage fault occurs, wherein the sampling frequency is 20480 Hz, the rotation speed is 1224 rpm, and the load is 6 N·m. Figure 2 , Figure 3 Shown is the vibration signal waveform in the early crack and early tooth break state.

[0131] Step 2: Improve spectral kurtosis analysis

[0132] Step 2.1 Impact component analysis

[0133] Step 2.1.1 Short-time Fourier processing of vibration signal

[0134] The corresponding data window length is obtained through two-layer and three-layer decomposition. The vibration signal in the window is short-time Fourier transformed according to the length of each layer of data window. The amplitude of each frequency band at different times is added and averaged, and finally the frequency amplitude information of the vibration signal in different frequency bands at different decomposition layers is obtained.

[0135] Among them, the mathematical expression of the short-time Fourier transform of the vibration signal x(n) is as follows:

[0136]

[0137] Wherein, x(n) represents the input discrete vibration signal; w(n-kP) represents the window function centered on kP, and the Hanning window is selected in this embodiment; k represents the number of analyses; P represents the time step; f represents the frequency; X(kP,f) represents the short-time Fourier transform value of the vibration signal.

[0138] Step 2.1.2 Calculate the spectral kurtosis of each frequency band

[0139] Calculate the spectral kurtosis of the vibration signal of each layer and each frequency band obtained in step 2.1.1. The specific steps are as follows:

[0140] First calculate the 2nth order instantaneous moment S of X(kP,f) 2nX (f) is used to estimate the energy intensity of the short-time Fourier transform signal X(kP,f). It is shown in the following formula:

[0141]

[0142] Where <> k represents k times time average.

[0143] After that, the fourth-order moment cumulant used to measure the impact component of the signal in this frequency band is calculated as shown in the following formula:

[0144]

[0145] Among them, C 4X (f) represents the fourth-order moment accumulation of the signal in this frequency band; S 4X (f) represents the fourth-order instantaneous moment of the signal in this frequency band; S 2X (f) represents the second-order instantaneous moment of the signal in this frequency band.

[0146] Finally, the fourth-order normalized cumulant of the frequency band signal is calculated to determine the spectral kurtosis value, as shown in the following formula:

[0147]

[0148] Wherein, K(f) represents the spectral kurtosis value of the signal in this frequency band.

[0149] Step 2.2 Set the frequency band weight

[0150] Step 2.2.1 Time domain averaging

[0151] In this embodiment, the sampling frequency f s is 20480Hz, speed f a is 1224rpm, so the sampling time interval Based on the above information, the truncated period length l of the vibration signal can be calculated:

[0152]

[0153] Among them, round() represents the rounding function.

[0154] In this embodiment, the truncation period length l is 1004.

[0155] Decompose the vibration signal according to the obtained truncated period length to obtain a new vibration signal sequence x1(l),x2(l),x3(l)······,x N (l), add the vibration signal sequence according to the corresponding points and take the average value to obtain the frequency-related signal y(l), as shown in the following formula:

[0156]

[0157] where x n (l) represents the intercepted sub-signal; N represents the number of vibration signal segments.

[0158] Step 2.2.2 Calculate the energy weight of each frequency band

[0159] Perform Fourier transform on the signal y(l) obtained in step 2.2.1:

[0160]

[0161] Wherein, Y(f) represents the Fourier transform value of the vibration signal at a certain frequency; j represents the imaginary unit.

[0162] Calculate its power spectrum:

[0163] P(f)=|Y(f) 2 (20)

[0164] Wherein, P(f) represents the power spectrum value of the vibration signal at a certain frequency.

[0165] The power spectrum is divided according to the number of frequency bands divided by the short-time Fourier transform, and the energy mean of each divided area is calculated. As the trust weight of the spectral kurtosis of this frequency band.

[0166] The trust weights calculated for each frequency band under different window lengths are summarized to form a weight matrix Z(m1,m2) as shown below:

[0167]

[0168] in Indicates the calculated frequency band energy mean; m1 indicates the layer where the calculated frequency band is located; m2 indicates the position of the calculated frequency band in this layer.

[0169] Step 2.3 Calculate the weighted spectral kurtosis of each frequency band

[0170] Multiply the spectral kurtosis of each frequency band of each layer by the corresponding frequency band weight value in step 2.2.2 to obtain their respective weighted spectral kurtosis. The calculation formula is as follows:

[0171]

[0172] in represents the weighted spectral kurtosis of the calculated frequency band; Represents the original spectral kurtosis of the calculated frequency band; Indicates the energy weight of the calculated frequency band.

[0173] Step 3: Fault feature extraction

[0174] Step 3.1 Extract the impact frequency band

[0175] Summarize the weighted spectral kurtosis of all frequency bands in each layer for comparison, find the frequency band with the maximum weighted spectral kurtosis, and calculate its corresponding center frequency f i and bandwidth BW, as shown below:

[0176]

[0177] BW=2 k (twenty four)

[0178] In the formula, f s represents the sampling frequency; k represents the layer where the maximum weighted spectral kurtosis is located; i represents the sub-frequency band where the maximum weighted spectral kurtosis of the kth layer is located.

[0179] The results obtained in this step in this embodiment are as follows: in the early crack state, the spectrum kurtosis is maximum when the analysis window length is 384, the center frequency is 2400 Hz, and the bandwidth is 362.04; in the early tooth break state, the spectrum kurtosis is maximum when the analysis window length is 384, the center frequency is 4520 Hz, and the bandwidth is 362.04.

[0180] Step 3.2 Hilbert envelope analysis

[0181] The impact frequency band selected in step 3.1 is band-pass filtered, and the obtained signal is subjected to Hilbert envelope analysis, and the frequency with the maximum amplitude in the envelope spectrum is defined as the fault characteristic parameter. Table 1 shows the fault characteristic parameter f under each gear state in this embodiment. p . Figure 4 , Figure 5 , Figure 6 Shown are the envelope spectra for each gear state.

[0182] Table 1 Gear fault characteristic parameters under different conditions

[0183] Gear Status <![CDATA[Characteristic parameter f p > <![CDATA[Amplitude (m / s 2 )]]> healthy 20Hz 0.005 Early cracks 20Hz 0.010 Early tooth fracture 20Hz 0.035

[0184] from Figure 5 , Figure 6 It can be seen that the fault characteristic parameters obtained under the early crack and early broken tooth fault conditions of the active gear are both 20Hz, which is highly correlated with the rotation frequency of the active gear at a speed of 1224rpm; and relative to Figure 4 Medium health state envelope spectrum, Figure 5 , Figure 6 The amplitude of the characteristic parameter under the envelope spectrum of the medium fault state is obviously larger. Therefore, it can be determined that the driving gear has a fault, which is consistent with the actual situation.

[0185] Figure 7 , Figure 8 The figure shows the comparison of the improved spectral kurtosis and the traditional spectral kurtosis when the analysis window length is 384 when the driving gear is in the early crack state and the early tooth break state. It can be seen that the overall distribution of the spectral kurtosis in the spectral kurtosis diagram obtained by the improved spectral kurtosis analysis method is more concentrated, and Fig. 9 The fault features are more obvious in the envelope spectrum obtained by the improved spectral kurtosis analysis method shown.

[0186] In summary, this patent designs a method for extracting early gear fault features based on improved spectral kurtosis analysis to address the shortcoming that the fault frequency band cannot be accurately located when the early gear fault features are not obvious. By operating the vibration signal obtained from the experiment, it can be seen that the results obtained by this method are in line with reality, so it can be better applied to the extraction of early gear fault features.

Claims

1. A gear fault feature extraction method based on improved spectral kurtosis analysis, characterized by: The steps include: (1) Obtaining the vibration signal of the gear box: Installing a sensor on the gear box to collect the vibration signals generated by the gear box under different conditions; (2) Improved spectral kurtosis analysis: including impulse component analysis, setting frequency band weights, and calculating weighted spectral kurtosis of each frequency band; (3) Fault feature extraction: including extraction of impulse frequency band and Hilbert envelope demodulation.

2. The gear fault feature extraction method based on improved spectral kurtosis analysis according to claim 1 is characterized by: In step (2), the impact component analysis uses short-time Fourier transform and spectral kurtosis analysis to detect the impact components contained in different frequency bands: Short-time Fourier processing of vibration signals: The continuous vibration signal x(t) generated by the gear box is discretely sampled to obtain the signal x(n). The mathematical expression of the short-time Fourier transform of x(n) is as follows: In the formula, x(n) represents the input discrete vibration signal; w(n-kP) represents the window function centered on kP; k represents the number of analysis; P represents the time step; f represents the frequency; X(kP,f) represents the short-time Fourier transform value of the vibration signal; Calculate the spectral kurtosis of each frequency band of the vibration signal: The spectral kurtosis of the vibration signal of each layer and each frequency band obtained by short-time Fourier transform is calculated: Calculate the 2nth order instantaneous moment S of X(kP,f) 2nX (f), used to estimate the energy intensity of the short-time Fourier transform signal X(kP,f): Where <> k represents k times time average; Calculate the fourth-order moment cumulant used to measure the impulse component of the signal in this frequency band: Among them, C 4X (f) represents the fourth-order moment accumulation of the signal in this frequency band; S 4X (f) represents the fourth-order instantaneous moment of the signal in this frequency band; S 2X (f) represents the second-order instantaneous moment of the signal in this frequency band; Calculate the fourth-order normalized cumulant of the signal in this frequency band to determine the spectral kurtosis value: Wherein, K(f) represents the spectral kurtosis value of the signal in this frequency band.

3. The gear fault feature extraction method based on improved spectral kurtosis analysis according to claim 1 is characterized by: Setting the frequency band weight in step (2) includes processing the vibration signal by time domain averaging and improving the spectral kurtosis analysis method by setting the spectral kurtosis weight for the signal frequency band, thereby determining the fault frequency band: Time domain averaging: The signal is intercepted and averaged: Set the gear shaft rotation frequency to f a , the sampling frequency is Fs, then the sampling time interval is Based on this, the truncated period length l of the vibration signal is determined as: Where round() represents the rounding function; The vibration signal x(n) is decomposed according to the obtained truncated period length to obtain a new vibration signal sequence x1(l), x2(l), x3(l)······, x N (l), add the vibration signal sequence according to the corresponding points and take the average value to obtain the frequency-related signal y(l): Among them, x n (l) represents the intercepted sub-signal; N represents the number of vibration signal segments; Calculate the energy weight of each frequency band: The power spectrum of the vibration signal y(l) is calculated, the weight of each frequency band is set, and the spectral kurtosis of each frequency band is weighted to improve the spectral kurtosis analysis results: Perform Fourier transform on the vibration signal y(l): Wherein, Y(f) represents the Fourier transform value of the vibration signal at a certain frequency; j represents the imaginary unit; Calculate its power spectrum: P(f)=|Y(f)| 2 Wherein, P(f) represents the power spectrum value of the vibration signal at a certain frequency; Divide the power spectrum according to the number of frequency bands divided by the short-time Fourier transform, and calculate the energy mean of each divided area As the trust weight of the spectral kurtosis of this frequency band; The trust weights calculated for each frequency band under different window lengths are summarized to form a weight matrix Z(m1,m2): in, Indicates the calculated frequency band energy mean; m1 indicates the layer where the calculated frequency band is located; m2 indicates the position of the calculated frequency band in this layer.

4. The gear fault feature extraction method based on improved spectral kurtosis analysis according to claim 1 is characterized by: Step (2) of calculating the weighted spectral kurtosis of each frequency band includes: Multiply the spectral kurtosis of each frequency band in each layer by the corresponding frequency band weight value to obtain their respective weighted spectral kurtosis: in represents the weighted spectral kurtosis of the calculated frequency band; Represents the original spectral kurtosis of the calculated frequency band; Indicates the energy weight of the calculated frequency band.

5. The gear fault feature extraction method based on improved spectral kurtosis analysis according to claim 1 is characterized by: Extracting the impact frequency band in step (3) includes: Summarize the weighted spectral kurtosis of all frequency bands in each layer for comparison, find the frequency band with the maximum weighted spectral kurtosis, and calculate its corresponding center frequency f i With bandwidth BW: BW=2 k In the formula, f s represents the sampling frequency; k represents the layer where the maximum weighted spectral kurtosis is located; i represents the sub-frequency band where the maximum weighted spectral kurtosis of the kth layer is located.

6. The gear fault feature extraction method based on improved spectral kurtosis analysis according to claim 1 is characterized by: The Hilbert envelope demodulation in step (3) includes: The selected impact frequency band is band-pass filtered, and the obtained signal is subjected to Hilbert envelope analysis. The frequency with the maximum amplitude in the envelope spectrum is defined as the fault characteristic parameter, and the fault location is determined by detecting its correlation with the rotation frequency of the driving gear and the driven gear.

Images