CN115436058B - A method, device, equipment and storage medium for bearing fault feature extraction - Google Patents

Abstract

The invention discloses a method, a device, equipment and a computer storage medium for extracting bearing fault characteristics, which comprise the steps of collecting bearing vibration signals and carrying out windowing interception on the bearing vibration signals; calculating fractal dimension values of the intercepted signal fragments by using a Katz method; tracking the impact position generated by bearing faults in the bearing vibration signals according to the fractal dimension values, and extracting an envelope curve of the impact position; and further removing interference on the envelope line according to a peak value searching algorithm to obtain the bearing vibration fault characteristics. According to the method, the vibration sequence of the bearing is converted into the short-time fractal dimension sequence through the fractal dimension, other interference signals are restrained in the conversion process, the signals of bearing faults are more prominent, the interference signals are restrained, finally, all the signals are processed according to the peak value searching algorithm, the influence of the interference signals on the bearing fault signals is further restrained, and the bearing fault signals are more highlighted.

Description

A method, device, equipment and storage medium for bearing fault feature extraction

Technical field

The present invention relates to the technical field of rolling bearing fault diagnosis, and in particular to a method, device, equipment and computer storage medium for extracting bearing fault features.

Background technique

Rolling bearings are the most common bearings. They are often used to move heavy loads in harsh environments and are among the most prone to failure of mechanical components. 44% of large induction motor failures are related to bearing failure. Bearing failure can lead to productivity and financial losses or even catastrophic consequences. Therefore, bearing fault detection is of great significance. Most of the many methods that have been developed are based on sound, temperature, wear and vibration. Among them, vibration signature has been found to be a powerful tool for bearing fault detection. When the rolling elements pass through the defective area, small collisions occur, resulting in a series of mechanical pulses. These impulses cause the system to resonate at high speed at its natural mechanical frequency. Then the resonance frequency is modulated, and the fault characteristic frequency (Fault Characteristic Frequency: FCF) and the main information of the detection are named from the bearing fault pulse frequency. One of the most commonly used detection techniques involves amplitude demodulating the vibration signal to envelope the signal. Bearing health can then be determined based on the spectrum. Hilbert transform is often used for this purpose. However, the Hilbert transform is susceptible to noise and interference. In order to obtain reliable detection results, preprocessing and filtration are often required. A typical example is high resonance frequency technology (HRFT). The main steps of this method include filtering the vibration signal near the resonance frequency, envelope the filtered signal, and Fourier transform of the demodulated signal. The design of bandpass filters is usually based on unknown resonant frequencies. Therefore, appropriate center frequency and bandwidth are very important, and choosing an appropriate solution is a challenging task.

The most widely used envelope methods in the literature for bearing condition monitoring include Hilbert transform, EO and MMA. Although they have their own advantages, they all have their own shortcomings, which are summarized as follows: (1) Hilbert transform is susceptible to vibration interference and noise, so pre-filtering is usually required, which involves a challenging filtering (2) The EO method is limited by single component and narrow band, and its effectiveness is usually affected by multiple interferences. (3) For MMA technology, the SE structure requires a large amount of calculation and is resistant to multiple interferences. The low processing efficiency limits their application in bearing fault feature extraction.

From the above, it can be seen that how to reduce the impact of interference signals and improve fault characteristics is a problem that needs to be solved.

Contents of the invention

The purpose of the present invention is to provide a bearing fault feature extraction method, device, equipment and computer storage medium, which solves the problem of difficulty in extracting bearing fault features in the prior art.

In order to solve the above technical problems, the present invention provides a method for extracting bearing fault features, including:

Collect bearing vibration signals, and perform window interception on the bearing vibration signals;

Use Katz method to calculate the fractal dimension value of the intercepted bearing vibration signal;

Track the impact location caused by the bearing failure in the bearing vibration signal according to the fractal dimension value, and extract the envelope of the impact location;

Calculating a mean and a standard deviation for each envelope and calculating a peak value of the envelope signal based on said mean and standard deviation;

The peak value of the envelope signal whose difference between the peak value and the standard deviation of the envelope signal is less than 1 is set to 1, and the remaining envelope signals are bearing fault characteristics.

Preferably, the calculation of the fractal dimension value of the intercepted bearing vibration signal using the Katz method includes:

According to the bearing vibration signal s={s ₁ , s ₂ ,..., s _N }, the Euclidean distance between continuous signals is calculated as:

Among them, N is the total number of sampling points in the bearing vibration signal, s _i is any point in the vibration signal, and its s _i coordinate is (x _i , y _i ), i=1,2,...,N;

According to the Euclidean distance between the continuous signals, the plane range d of the waveform and the total length L of the curve are calculated. The calculation formula is:

The total length L of the curve and the planar extent d of the waveform are normalized by the average distance a between the consecutive signals, and utilize the mathematical definition formula of KatzFD Calculate the fractal dimension:

Among them, k=N-1 is the number of steps in the curve, and a=L/k.

Preferably, tracking the impact location caused by a bearing failure in the bearing vibration signal according to the fractal dimension value and extracting the envelope of the impact location includes:

Determine the width of the moving window according to the fractal dimension value;

Use a moving window to intercept the bearing vibration signal from beginning to end to obtain multiple signal segments;

Calculate the fractal dimension value of each bearing vibration segment;

The short-time fractal dimension sequence is obtained by splicing the fractal dimension values of all signal segments.

Preferably, using the moving window to intercept the bearing vibration sequence from beginning to end to obtain multiple signal segments includes:

The matrix formed by the multiple bearing vibration segments is:

Among them, sw _m is each signal segment, N _win is the window length, N _win =int(αf _s ), int(.) is the integer part of the value, α is the preset constant, and f _s is the sampling frequency of the waveform.

Preferably, the splicing of the fractal dimension values of all signal segments to obtain the short-term fractal dimension signal includes:

Calculate the fractal dimension value of each signal segment to obtain the short-term fractal dimension vector:

STFD=[STFD(1),STFD(2),...,STFD(m),...,];

Convert it into the short-time fractal dimension sequence: STFD=STFD ^original -MIN+1;

Among them, the vector STFD ^original is the calculated signal STFD sequence, MIN is the minimum component with the same length as STFD ^original , 1 is the unit vector with the same length as STFD ^original , and STFD is the obtained STFD sequence.

Preferably, setting the peak value of the envelope signal whose difference between the peak value and the standard deviation of the envelope signal is less than 1 to 1, and the remaining envelope signals as bearing fault characteristics includes:

The formula that defines the peak search algorithm:

Among them, μ is the average value of signal x(t), σ is the standard deviation of signal x(t), E{.} is the mathematical expectation operator, x(t)=s(t)+i(t)=A _mp e ^-βt sinω _r t+A _i cosω _i t; where A _mp is the vibration amplitude of the pulse, ω _r is the vibration frequency of the pulse, A _i is the vibration amplitude of the interference, and ω _i is the vibration frequency of the interference;

Calculate the standard deviation σ and peak K of each signal;

Determine whether the amplitude is greater than or equal to 1+σ. If the amplitude is greater than or equal to 1+σ, the amplitude remains unchanged. If the amplitude is less than 1+σ, the amplitude is set to 1;

Determine whether the peak difference between the two iterations meets the preset value. If so, stop calculating the peak value. If not, continue the iteration;

Until all signal calculations are completed, the iteration is stopped and the bearing vibration fault characteristics are obtained.

Preferably, the calculation steps of the peak search algorithm are:

S71: Initialization, let i=1,

S72: Calculate the standard deviation σ ⁱ and peak value K ⁱ of the signal segment x(i);

S73: Determine whether the amplitude in the signal segment is greater than or equal to 1+σ ⁱ ;

S74: If it is greater than or equal to 1+σ ⁱ , the amplitude remains unchanged; if it is less than 1+σ ⁱ , the amplitude is set to 1;

S75: Determine i≥N, if true, stop iteration;

S76: If not, determine whether |K ⁱ -K ^i-1 |≤ε is satisfied. If so, stop calculating K ⁱ -K ^i-1 , where ε is the minimum threshold constant;

S77: If not satisfied, set i=i+1 and return to step S72.

The invention also provides a device for extracting bearing fault characteristics, including:

A signal acquisition module is used to collect bearing vibration signals and perform window interception on the bearing vibration signals;

The fractal dimension calculation module is used to calculate the fractal dimension value of the intercepted bearing vibration signal using the Katz method;

An envelope extraction module is used to track the impact location caused by the bearing failure in the bearing vibration signal according to the fractal dimension value, and extract the envelope of the impact location;

a peak calculation module for calculating the average value and standard deviation of each envelope line, and calculating the peak value of the envelope signal based on the average value and standard deviation;

The interference suppression module is configured to set the peak value of the envelope signal whose difference between the peak value and the standard deviation of the envelope signal is less than 1 to 1, and the remaining envelope signals are bearing fault characteristics.

The invention also provides a device for extracting bearing fault characteristics, including:

A memory is used to store a computer program; a processor is used to implement the steps of the above method for extracting bearing fault characteristics when executing the computer program.

The present invention also provides a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the steps of the above method for extracting bearing fault characteristics are implemented.

The invention provides a method for extracting bearing fault characteristics. First, the bearing vibration signal is collected when the bearing is running, and windowed interception is performed. The Katz method is used to calculate the fractal dimension value of the intercepted bearing vibration signal. The fractal dimension value is used to classify the bearing. The vibration signal is converted into a short-time fractal dimension sequence, and the envelope of the impact position caused by the fault is extracted through the fractal dimension value to distinguish the fault signal from the interference signal, which can better suppress the impact of the interference signal and highlight the fault signal; finally, according to The peak search algorithm sets the amplitude value less than the standard deviation to 1 to highlight the fault signal. Finally, the bearing vibration fault characteristics are obtained. This invention converts the vibration sequence of the bearing into a short-time fractal dimension sequence through fractal dimension, and suppresses other interference signals during the conversion process, making the signal of bearing failure more prominent, suppressing the interference signal, and finally based on the peak search algorithm All signals are processed to further suppress the impact of interference signals on bearing fault signals, and the bearing fault signals are more highlighted. There is no need to use filters to filter the signals, which cannot suppress multiple interference signals and requires a huge amount of calculation. The practical scope is wider. And the present invention can realize online bearing status monitoring.

Description of the drawings

In order to explain the embodiments of the present invention or the technical solutions of the prior art more clearly, the following will briefly introduce the drawings needed to describe the embodiments or the prior art. Obviously, the drawings in the following description are only For some embodiments of the present invention, those of ordinary skill in the art can also obtain other drawings based on these drawings without exerting creative efforts.

Figure 1 is a flow chart of a first specific embodiment of a bearing fault extraction method provided by the present invention;

Figure 2 is a flow chart of a second specific embodiment of the bearing fault extraction method provided by the present invention;

Figure 3 is a comparison diagram of using STFD transformation to extract bearing fault features: (a) pulse signal; (b), (c) and (d) STFD obtained using Katz, Sevcik and Higuchi methods respectively;

Figure 4 (a) Curve u of (different) STFD sequences generated under different window lengths; (b) Curve chart of the evolution of the correlation coefficient between the simulated signal envelope and the generated STFD sequence under different window lengths;

Figure 5 is a graph of interference suppression by STFD transformation of signal mixture (impulse and random interference): (a) periodic interference; (b) simulated pulse; (c) mixture of (a) and (b); (d) STFD representation of signal mixture;

Figure 6 is a graph showing the STFD and STFD-KPSA transformation results of simulated fault orientation signals with multiple cyclic interferences: (a) signal interference mixture; (b) STFD sequence without KPSA; (c) spectrum of b, (d )STFD-KPSA sequence; (e) spectrum of d;

Figure 7 is the detectability diagram of the STFD-KPSA method;

Figure 8 is a graph showing the application of the proposed method to simulated signals, (a) synthetic signal; (b) Hilbert envelope spectrum of a; (c) STFD sequence; (d) STFD-KPSA results; (e )c spectrum; (f)d spectrum;

Figure 9 is a graph showing the performance of STFD-KPSA in detecting side positioning outer ring faults: (a) original vibration signal; (b) Hilbert envelope spectrum of the original signal; (c) STFD-KPSA of the original signal Results; (d) Spectrum of STFD-KPSA results;

Figure 10 is a graph showing the performance of STFD-KPSA in detecting bottom positioning outer ring faults: (a) original vibration signal; (b) Hilbert envelope spectrum of the original signal; (c) STFD-KPSA result of the original signal; (d) Spectrum of STFD-KPSA results;

Figure 11 is a graph showing the performance of STFD-KPSA in detecting inner ring faults: (a) original vibration signal; (b) Hilbert envelope spectrum of the original signal; (c) STFD-KPSA of the original signal; (d) Spectrum of STFD-KPSA results;

Figure 12 is a structural block diagram of a device for extracting bearing fault features provided by an embodiment of the present invention.

Detailed ways

The core of the present invention is to provide a bearing fault extraction method that uses fractal dimensions to convert into short-time fractal dimension sequences to effectively suppress the impact of interference signals, and uses a peak search algorithm to further suppress interference and improve fault characteristics.

In order to enable those skilled in the art to better understand the solution of the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

Please refer to Figure 1, which is a flow chart of a first specific embodiment of the method for extracting bearing fault features provided by the present invention; the specific operation steps are as follows:

Step S101: Collect bearing vibration signals, and perform window interception on the bearing vibration signals;

Vibration sensors are used to collect signals of bearing vibration faults, and the collected bearing vibration signals are intercepted by adding windows.

Step S102: Calculate the fractal dimension value of the intercepted bearing vibration signal using the Katz method;

Step S103: Track the impact location caused by the bearing failure in the bearing vibration signal according to the fractal dimension value, and extract the envelope of the impact location;

Step S104: Calculate the average value and standard deviation of each envelope, and calculate the peak value of the envelope signal based on the average value and standard deviation;

Step S105: Set the peak value of the envelope signal whose difference between the peak value and the standard deviation of the envelope signal is less than 1 to 1, and the remaining envelope signals are bearing fault characteristics.

In this embodiment, a vibration sensor is first used to collect the bearing vibration signal, and then the Katz algorithm is used to calculate the distance between the signals and the total length of the curve, and then calculate the fractal dimension value between the signals, and then convert the bearing signal based on the fractal dimension value. Convert it into a short-time fractal dimension sequence; use the conversion of fractal dimension to suppress the interference signal, making the fault signal more prominent. According to the peak search algorithm, the interference signal is removed, the fault signal is retained, and a more complete fault signal is finally obtained. There is no need to use a filter for filtering, which simplifies the steps. The present invention can also monitor the bearing fault status online in real time.

Based on the above embodiments, this embodiment explains in detail the method for extracting bearing fault features. Please refer to Figure 2. Figure 2 is a flow chart of the second specific embodiment of the method for extracting bearing fault features provided by the present invention; specific operations Proceed as follows:

Step S201: Use a vibration sensor to collect bearing vibration signals;

Step S202: Calculate the Euclidean distance between the signals and the total length of the curve based on the bearing vibration signal;

Please refer to Figure 3. Figure 3 is a comparison chart of the present invention using STFD transformation to extract bearing fault features, in which: (a) pulse signal; (b), (c) and (d) respectively use Katz, Sevcik and Higuchi methods. From the comparison, it can be seen that the Katz method has the best effect, and the peaks of the image are evenly distinguished.

The collected vibration signal is s={s ₁ , s ₂ ,..., s _N }, where N is the total number of sampling points in the signal. An arbitrary point in the signal is represented by s _i , and its coordinates in the figure are (xi _, y _i ), i=1,2,...,N. Then the Euclidean distance between two consecutive points is

The KatzFD of the N-point signal is mathematically defined as:

Among them, L is the total length of the curve, which is obtained by the sum of the Euclidean distances of all adjacent points, that is

d is the planar extent of the waveform and can generally be thought of as the furthest of all distances between the first point of the signal (point 1) and all subsequent points. Mathematically, d can be written as

The FD value obtained from equation (2) is affected by the unit of measurement. To overcome this difficulty, Katz suggested normalizing L and d by the average distance between consecutive points (denoted by a). The average distance a is defined as a=L/k, where k is the number of steps in the curve, that is, k=N-1. In equation (2), dividing a by L and d gives

Step S203: Calculate the fractal dimension value based on the Euclidean distance and the total length of the curve;

Step S204: Use the fractal dimension value to perform window interception on the bearing vibration signal to obtain multiple bearing signal segments;

FD is used to track the impact location caused by bearing failure and extract the envelope of the impact. Typical vibration signals produced by bearing failures are periodic and include sharp rises corresponding to impacts between surfaces at the location of the defect. In order to use FD to extract features affected by periodicity, STFD was studied.

The moving window is used to truncate the original signal to obtain an STFD sequence, where each STFD value corresponds to a windowed signal part. The length of the window is selected as N _win =int(αf _s ), where int(.) represents the integer part of the value, which is the constant α pre-specified by the user, and f _s is the sampling frequency of the waveform. The constant α is set empirically. Since fault-induced pulses are often weak and masked by background noise and vibration interference, the window must be shorter to obtain satisfactory results. This is because too long a window may over-smooth the STFD series. Therefore, in bearing fault detection applications, this constant α should be smaller than the constant used for sound signal processing to ensure that the initial pulse is not missed. However, it should be noted that N _win cannot be too short. A window that is too short may lead to many hypothetical oscillations, and the impact cannot be correctly identified, as shown in Figure 4, Figure 4(a) STFD sequence generated under different window lengths (different α); Figure 4(b) Simulation under different window lengths Evolution of the correlation coefficient between the envelope signal and the resulting STFD sequence.

To obtain a point-to-point time series of STFD values, the window is moved along the input vector one sample step at a time. The window function is represented by w[n]. Then, the sequence sw _m [n] = s [n] w [mn] is a short period of the signal s [n] at time m, with length N _win . Expressing all signal segments in matrix form, the following expression can be obtained

For each signal segment w _m , the corresponding FD value represented by STFD _(m) can be calculated by equation (5). The vector STFD can be obtained by calculating the FD values of all signal segments as follows

STFD＝[STFD(1),STFD(2),...,STFD(m),...,] (7)

Without affecting the dynamic range of the STFD sequence, the following transformations are applied in subsequent subsections

STFD＝STFD ^original -MIN+1 (8)

where the vector STFD ^original is the signal STFD sequence calculated using equation (5), MIN is a vector with the same length (dimension) as STFD ^original and all its components are equal to the smallest component of STFD ^original , and 1 represents a unit that is also the same length as STFD ^original The vector, STFD represents the obtained STFD sequence. According to equation (8), the minimum value of the STFD sequence is always limited to 1.

In order to convert the original vibration signal into an STFD sequence, the FD value of each signal segment is calculated by Equation (5) and assigned to the midpoint of the signal segment. However, this will make the length of the obtained STFD sequence NN _win +1 shorter than the length N of the original signal. To maintain the signal length, the original signal is extended by copying: (a) the first half of the points at the beginning, i.e. (N _win -1)/2 points for odd N _win , (N _win /2, even), and (b ) the last (N _win -1)/2, odd N _win (N _win -2/2, even N _win ) points at the end of the original signal. Then, the length of the signal is extended to N+N _win -1. In this way, the obtained STFD sequence has the same size as the original signal.

Step S205: Calculate the fractal dimension value of each bearing signal segment;

Step S206: Splice the fractal dimension values of all bearing signal segments to obtain the short-time fractal dimension sequence;

Step S207: Remove interference from the fractal dimension sequence according to the peak search method to obtain bearing vibration fault characteristics.

Periodic interference can be expressed as And A _im and ω _im represent the amplitude and frequency of interference respectively. To evaluate the impact of interference on the demodulation of the conversion envelope, consider that a simulated pulse caused by a ^single fault is contaminated by an interference with _an amplitude larger than the _pulse t+A _i cosω _i t, where A _mp is the amplitude of the pulse, A _i >A _mp , indicating that the fault-related pulse is seriously masked by interference. Analysis of signals represented by X(t).

The envelope (or instantaneous amplitude) of the analyzed signal can be calculated by the following formula:

Equation (10) shows that the main component, i.e. the interference, is envelope demodulated, but the pulse demodulation is unsuccessful. From this it can be concluded that conventional demodulation is adversely affected by strong interference. Therefore, an envelope method with the ability to suppress interference is urgently needed.

Interference suppression based on STFD transform. Please refer to Figure 5 and consider the random interference (i(t)=2cos(2πft)) as shown in Figure 5(a). Figure 5(b) and (c) are the pulse signal and the signal mixture composed of the pulse signal and interference respectively. For most practical applications, the resonant frequency is much higher than the frequency of the periodic interference, i.e., the interference frequency can be observed to be much lower than the frequency of the pulse, as shown in Figure 5(a) and (b). Let us look at the signal segment within a short window, i.e., the nth window in Figure 5(a). The Euclidean distance of the curve in the nth window, n=1,2..., is represented by L _n (defined by Equation (3)), and the plane range is d _n (defined by Equation (4)) express. If the vibration interference frequency is low enough, as long as a suitable window length is selected, L _n is almost equal to d _n in Figure 5(a), which will lead to STFD(n)≈1 (such that d _n ≈ L _n is satisfied). From this it can be concluded that interference has almost no impact on the STFD representation. For pulse signals, the difference between the planar extent (d _n ) and the Euclidean distance (L _n ) of the signal segment is larger because of the significant morphological appearance changes, as shown in Figure 5(c), according to Equation (5 ) results in STFD(n)>1. Since the resonant frequency is usually much higher than the frequency of the vibration disturbance, the STFD value generated by the pulse is much larger than the STFD value associated with the disturbance, indicating that the STFD value associated with the pulse is more significant in the STFD representation compared with the STFD value associated with the disturbance. . Therefore, the envelope of the pulse can be highlighted and interference suppressed, as shown in Figure 5(d).

Interference suppression is performed through the combination of STFD and KPSA. However, in some cases, the impact on high-frequency interference still cannot be ignored. For example, the composite signal composed of multiple interferences (cos(2π1000t), cos(2π620t) and cos(2π300t)) and simulated pulses is shown in Figure 6(a). The STFD transformation results and their spectra are shown in Figure 6(b) and (c) respectively. It can be seen that since the STFD values related to the interference have a greater impact on the STFD representation due to high frequencies, the compound effect of the interference is not successfully eliminated. Therefore, the spectrum in Figure 6(c) is mainly controlled by the interference frequency difference, which will confuse or even mislead the detection results.

In order to solve the problems caused by multiple interferences at relatively high frequencies, it is necessary to improve the proposed method in order to further suppress the effects of interference. The idea of KPSA is to extract useful information and iteratively remove unwanted signal components as much as possible until the stopping criterion is reached. The criterion was kurtosis because it has been shown to be a valid indicator of impulsivity. It is defined as:

Among them, μ and σ are the mean and standard deviation of the signal x(t) respectively, and E{.} is the mathematical expectation operator. Taking into account the uncertainty of the signal, a tolerance range for the standard deviation σ ⁱ is introduced. Therefore, in each iteration of KPSA, for example iteration i, the standard deviation σ ⁱ and kurtosis K ⁱ of the STFD transformation result (i.e. STFD ⁱ , the vector representing the STFD sequence) are calculated. The amplitude of points above 1+ ^σi remains unchanged. The amplitudes of points below 1+ ^σi are "removed", i.e. set to 1.0 (because according to equation (6), the minimum FD value is 1). In this way, a new vector STFD ^*i is constructed, representing the useful information extracted in iteration i. However, some low-energy points, that is, points with amplitudes lower than 1+ ^σi , may also be caused by faults. Therefore, a new iteration i+1 is needed to further extract the information of such low-energy pulses, as long as the difference between K ⁱ and K ^i-1 remains significant, i.e. the difference in impulsivity between two consecutive iterations is worth pursuing. In other words at each iteration, say i+1, the above process is applied to the signal remainder of iteration i, i.e. STFD ⁱ +1 = STFD ⁱ - STFD _* ⁱ +1, and the iterative process continues until the impulsiveness of the signal remainder becomes It does not matter, i.e. |K ⁱ -K ^i-1 |≤ε (“ε is a pre-specified small positive number). The final signal residue is then considered to consist of STFD components caused by noise and/or interference, and is therefore Removed. The final STFD sequence after applying KPSA (hereinafter referred to as STFD-KPSA) is correspondingly given by the following formula:

Applying the above KPSA algorithm to the STFD sequence in Figure 6(b) will give the results in Figure 6(d). As shown, the unwanted signal components have been removed, leaving only the peaks associated with the simulated fault. A more pronounced enhancement using KPSA can be observed on the spectrum shown in Figure 6(e), from which the interference-related frequency components are completely eliminated, leaving only the FCF and its harmonics.

Interference suppression capability of the proposed STFD-KPSA method. In this subsection, a comprehensive evaluation of the proposed approach will be performed in terms of interference suppression. To quantify the interference level of a signal mixture x(t), including fault-induced pulses s(t) and interference i(t), the signal-to-interference ratio can be defined as

where T is the signal length, P _s is the power of the fault pulse, and _Pi is the power of the interference.

The detection performance of a method can be expressed by the number of FCF harmonics displayed by the method using a three-dimensional map. These two axes relate to the two factors to be examined. In this case, the two factors are frequency and SIR. The levels of darkness represent levels of energy, with darkest to lightest corresponding to lowest to highest energy levels. Figure 7 shows the detectability plot of the proposed method. The pulse signal is defined by equation (9), the SNR is –7dB, and the corresponding parameters are the same as listed in Table 1.

The obtained detectability plots show that: (a) the proposed method is able to handle signals with SIR as low as -33dB, (b) for signals with SIR ranging from 0 to -25dB, the proposed method can detect FCF and its cubic Harmonics, (c) Detectability then deteriorates as the SIR decreases, it can be seen that when the SIR decreases from -25dB to -33dB, only the FCF and first harmonic are observed.

Therefore, through a bearing fault feature extraction algorithm based on fractal dimension envelope demodulation proposed by this invention, the collected vibration signals are used to use the theory of fractal dimension and STFD transformation to extract bearing fault features, and KPSA is used to further extract the bearing fault features. Reduce interference and extract bearing fault characteristics.

In this embodiment, the present invention is also described in detail in combination with experimental data:

The composite signal is given by the equation: x(t)=s(t)+i(t)+n(t) (14)

where s(t) is the analog pulse signal, i(t) represents vibration interference, and n(t) represents Gaussian white noise. Use the sine function Simulate multiple disturbances produced by other electrical/mechanical components such as unbalanced shafts and gear meshes. In this case eight disturbances are added, f _i1 ... f _i8 = 1000Hz, 500Hz, 300Hz, 150Hz, 80Hz, 30Hz , 15Hz, 8Hz, A _i1 ...A _i8 = 0.5, 1, 1.5, 1, 1, 1, 1, 1.5, making SIR equal to -21dB. In addition, there is background noise in the real vibration signal; therefore, Gaussian white noise n(t) is added. The signal-to-noise ratio relative to the analog pulse signal is 7dB. The purpose of adding this type of interference is to simulate a real situation where the interference and its harmonics may occur. A mixed signal x with 16,000 sampled data points is plotted in Figure 8(a), where the affected features are heavily obscured by interference and noise. Figure 8(b) shows the Hilbert envelope spectrum of the simulated signal. As shown in the figure, the Hilbert envelope spectrum is mainly determined by the frequency difference between interferences. Since the pulse is seriously contaminated by interference and noise, the frequency information related to FCF cannot be found. The proposed method is then applied, and the results are shown in Figure 8(c) (output of STFD method) and Figure 8(d) (output of STFD-KPSA method). Spectrum analysis was performed on the STFD sequence and STFD-KPSA sequence respectively, and Figure 8(e) and (f) were obtained. It can be seen that using STFD alone cannot completely eliminate the effect of interference, as shown in Figure 8(e), where the amplitude of the second frequency line related to the interference difference is almost equivalent to the amplitude of the first FCF harmonic. This is due to the effect of multiple interferences still present in the STFD representation. Figure 8(f) provides a clearer result, where the FCF and its three harmonics are clearly identifiable, since the combined effect of the interference is further reduced using the proposed KSPA algorithm. The above analysis shows that the demodulation results of STFD-KPSA are better compared with Hilbert transform and STFD without KPSA shown in Figure 8(b) and (e) respectively. Therefore, it can be concluded that the proposed STFD-KPSA method can extract bearing fault characteristics and suppress interference and noise without using pre-filtering. Figures 9, 10 and 11 are respectively detected using the fault detection method of the present invention. Performance diagrams of side, bottom and inner ring faults, in Figure 9, Figure 10 and Figure 11 (a) original vibration signal; (b) Hilbert envelope spectrum of the original signal; (c) STFD-KPSA of the original signal Results; (d) Spectrum of STFD-KPSA results.

Please refer to Figure 12, which is a structural block diagram of a device for extracting bearing fault features provided by an embodiment of the present invention; the specific device may include:

The signal collection module 100 is used to collect bearing vibration signals and convert them into bearing vibration signal sequences;

The fractal dimension calculation module 200 is used to calculate the fractal dimension value between adjacent bearing vibration signals using the Katz method;

The envelope extraction module 300 is used to track the impact location caused by the bearing failure between each adjacent bearing vibration signal according to the fractal dimension value, and extract the envelope of the impact location;

The peak calculation module 400 is used to calculate the average value and standard deviation of each envelope line, and calculate the peak value of the envelope signal based on the average value and standard deviation;

The interference suppression module 500 is configured to set the peak value of the envelope signal whose difference between the peak value and the standard deviation of the envelope signal is less than 1 to 1, and the remaining envelope signals are bearing fault characteristics.

The device for extracting bearing fault features in this embodiment is used to implement the aforementioned method for extracting bearing fault features. Therefore, the specific implementation of the device for extracting bearing fault features can be found in the embodiments of the method for extracting bearing fault features mentioned above, for example , the signal acquisition module 100, the fractal dimension calculation module 200, the envelope extraction module 300, and the interference suppression module 400 are respectively used to implement steps S101, S102, S103 and S104 in the above method of extracting bearing fault features, so its specific implementation For the method, reference may be made to the corresponding descriptions of various partial embodiments, and details will not be described again here.

Specific embodiments of the present invention also provide a device for extracting bearing fault features, including: a memory for storing a computer program; a processor for implementing the steps of the above method for extracting bearing fault features when executing the computer program .

Specific embodiments of the present invention also provide a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the steps of the above method for extracting bearing fault characteristics are implemented. .

Each embodiment in this specification is described in a progressive manner. Each embodiment focuses on its differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other. As for the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple. For relevant details, please refer to the description in the method section.

Those skilled in the art may further realize that the units and algorithm steps of each example described in connection with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of both. In order to clearly illustrate the possible functions of hardware and software, Interchangeability, in the above description, the composition and steps of each example have been generally described according to functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may implement the described functionality using different methods for each specific application, but such implementations should not be considered to be beyond the scope of the present invention.

The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly in hardware, in software modules executed by a processor, or in a combination of both. Software modules may be located in random access memory (RAM), memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or anywhere in the field of technology. any other known form of storage media.

The above has introduced in detail the bearing fault feature extraction method, device, equipment and computer-readable storage medium provided by the present invention. This article uses specific examples to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method and the core idea of the present invention. It should be noted that those skilled in the art can make several improvements and modifications to the present invention without departing from the principles of the present invention, and these improvements and modifications also fall within the scope of the claims of the present invention.

Claims (5)

1. A method for extracting bearing fault features, which is characterized by including: Collect bearing vibration signals, and perform window interception on the bearing vibration signals; Use Katz method to calculate the fractal dimension value of the intercepted bearing vibration signal; Track the impact position caused by the bearing failure in the bearing vibration signal according to the fractal dimension value, and extract the envelope of the impact position; including the following steps: determining the width of the moving window according to the fractal dimension value; using the moving window Intercept the bearing vibration signal from beginning to end to obtain multiple signal segments; calculate the fractal dimension value of each bearing vibration segment; splice the fractal dimension values of all signal segments to obtain the short-time fractal dimension sequence; The use of the moving window to intercept the bearing vibration signal from beginning to end to obtain multiple signal segments includes: The matrix formed by the multiple bearing vibration segments is: Among them, sw _m is each signal segment, N _win is the window length, N _win =int(αf _s ), int(.) is the integer part of the value, α is the preset constant, and f _s is the sampling frequency of the waveform; The short-term fractal dimension sequence obtained by splicing the fractal dimension values of all signal segments includes: Calculate the fractal dimension value of each signal segment to obtain the short-term fractal dimension vector: STFD=[STFD(1),STFD(2),...,STFD(m),...,]; Convert it into the short-time fractal dimension sequence: STFD=STFD ^original -MIN+1; Among them, the vector STFD ^original is the calculated signal STFD sequence, MIN is the minimum component with the same length as STFD ^original , 1 is the unit vector with the same length as STFD ^original , and STFD is the obtained STFD sequence; Calculating a mean and a standard deviation for each envelope and calculating a peak value of the envelope signal based on said mean and standard deviation; The peak value of the envelope signal whose difference between the peak value and the standard deviation of the envelope signal is less than 1 is set to 1, and the remaining envelope signals are bearing fault characteristics; The method of setting the peak value of the envelope signal whose difference between the peak value and the standard deviation of the envelope signal is less than 1 to 1, and the remaining envelope signals as bearing fault characteristics includes: The formula that defines the peak search algorithm: Among them, μ is the average value of signal x(t), σ is the standard deviation of signal x(t), E{.} is the mathematical expectation operator, x(t)=s(t)+i(t)=A _mp e ^-βt sinω _r t+A _i cosσ _i t; where A _mp is the vibration amplitude of the pulse, ω _r is the vibration frequency of the pulse, A _i is the vibration amplitude of the interference, and ω _i is the vibration frequency of the interference; Calculate the standard deviation σ and peak K of each signal; Determine whether the amplitude is greater than or equal to 1+σ. If the amplitude is greater than or equal to 1+σ, the amplitude remains unchanged. If the amplitude is less than 1+σ, the amplitude is set to 1; Determine whether the peak difference between the two iterations meets the preset value. If so, stop calculating the peak value. If not, continue the iteration; Until all signal calculations are completed, stop iteration and obtain the bearing vibration fault characteristics; The calculation steps of the peak search algorithm are: S71: Initialization, let i=1, S72: Calculate the standard deviation σ ⁱ and peak value K ⁱ of the signal segment x(i); S73: Determine whether the amplitude in the signal segment is greater than or equal to 1+σ ⁱ ; S74: If it is greater than or equal to 1+σ ⁱ , the amplitude remains unchanged; if it is less than 1+σ ⁱ , the amplitude is set to 1; S75: Determine i≥N, if true, stop iteration; S76: If not, determine whether |K ⁱ -K ^i-1 |≤ε is satisfied. If so, stop calculating K ⁱ -K ^i-1 , where ε is the minimum threshold constant; S77: If not satisfied, set i=i+1 and return to step S72. 2. The method of claim 1, wherein calculating the fractal dimension value of the intercepted bearing vibration signal using the Katz method includes: According to the bearing vibration signal s={s ₁ , s ₂ ,..., s _N }, the Euclidean distance between continuous signals is calculated as: Among them, N is the total number of sampling points in the bearing vibration signal, s _i is any point in the vibration signal, and its s _i coordinate is (x _i , y _i ), i=1,2,...,N; According to the Euclidean distance between the continuous signals, the plane range d of the waveform and the total length L of the curve are calculated. The calculation formula is: The total length L of the curve and the planar extent d of the waveform are normalized by the average distance a between the consecutive signals, and utilize the mathematical definition formula of KatzFD Calculate the fractal dimension: Among them, k=N-1 is the number of steps in the curve, and a=L/k. 3. A device for extracting bearing fault characteristics, which is characterized by including: A signal acquisition module is used to collect bearing vibration signals and perform window interception on the bearing vibration signals; The fractal dimension calculation module is used to calculate the fractal dimension value of the intercepted bearing vibration signal using the Katz method; Extracting an envelope module, used to track the impact position caused by a bearing failure in the bearing vibration signal according to the fractal dimension value, and extract the envelope of the impact position; determine the width of the moving window according to the fractal dimension value; use the The moving window intercepts the bearing vibration signal from beginning to end to obtain multiple signal segments; calculates the fractal dimension value of each bearing vibration segment; splices the fractal dimension values of all signal segments to obtain the short-time fractal dimension sequence ; The use of the moving window to intercept the bearing vibration signal from beginning to end to obtain multiple signal segments includes: The matrix formed by the multiple bearing vibration segments is: Among them, sw _m is each signal segment, N _win is the window length, N _win =int(αf _s ), int(.) is the integer part of the value, α is the preset constant, and f _s is the sampling frequency of the waveform; The short-term fractal dimension sequence obtained by splicing the fractal dimension values of all signal segments includes: Calculate the fractal dimension value of each signal segment to obtain the short-term fractal dimension vector: STFD=[STFD(1),STFD(2),...,STFD(m),...,]; Convert it into the short-time fractal dimension sequence: STFD=STFD ^original -MIN+1; Among them, the vector STFD ^original is the calculated signal STFD sequence, MIN is the minimum component with the same length as STFD ^original , 1 is the unit vector with the same length as STFD ^original , and STFD is the obtained STFD sequence; a peak calculation module for calculating the average value and standard deviation of each envelope line, and calculating the peak value of the envelope signal based on the average value and standard deviation; The interference suppression module is used to set the peak value of the envelope signal whose difference between the peak value and the standard deviation of the envelope signal is less than 1 to 1, and the remaining envelope signal is a bearing fault characteristic; the envelope signal is If the difference between the peak value of the signal and the standard deviation is less than 1, the peak value of the envelope signal is set to 1. The remaining envelope signals are bearing fault characteristics including: The formula that defines the peak search algorithm: Among them, μ is the average value of signal x(t), σ is the standard deviation of signal x(t), E{.} is the mathematical expectation operator, x(t)=s(t)+i(t)=A _mp e ^-βt sinω _r t+A _i cosω _i t; where A _mp is the vibration amplitude of the pulse, ω _r is the vibration frequency of the pulse, A _i is the vibration amplitude of the interference, and ω _i is the vibration frequency of the interference; Calculate the standard deviation σ and peak K of each signal; Determine whether the amplitude is greater than or equal to 1+σ. If the amplitude is greater than or equal to 1+σ, the amplitude remains unchanged. If the amplitude is less than 1+σ, the amplitude is set to 1; Determine whether the peak difference between the two iterations meets the preset value. If so, stop calculating the peak value. If not, continue the iteration; Until all signal calculations are completed, stop iteration and obtain the bearing vibration fault characteristics; The calculation steps of the peak search algorithm are: S71: Initialization, let i=1, S72: Calculate the standard deviation σ ⁱ and peak value K ⁱ of the signal segment x(i); S73: Determine whether the amplitude in the signal segment is greater than or equal to 1+σ ⁱ ; S74: If it is greater than or equal to 1+σ ⁱ , the amplitude remains unchanged; if it is less than 1+σ ⁱ , the amplitude is set to 1; S75: Determine i≥N, if true, stop iteration; S76: If not, determine whether |K ⁱ -K ^i-1 |≤ε is satisfied. If so, stop calculating K ⁱ -K ^i-1 , where ε is the minimum threshold constant; S77: If not satisfied, set i=i+1 and return to step S72. 4. A device for extracting bearing fault characteristics, which is characterized by including: Memory, used to store computer programs; A processor, configured to implement the steps of a bearing fault feature extraction method according to any one of claims 1 or 2 when executing the computer program. 5. A computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the method of claim 1 or 2 is implemented. Steps of the method for extracting bearing fault features.