A variable-scale adaptive denoising method for acoustic differential intelligent bearing and acoustic intelligent bearing

Abstract

The present invention proposes a variable-scale adaptive denoising method for acoustic differential intelligent bearings and an acoustic intelligent bearing. The method is to collect the expected sound signal on the end face of the outer ring of the bearing and collect the reference sound signal on the axle box cover; perform adaptive variational modal decomposition on the expected sound signal to determine the optimal decomposition layer number and bandwidth penalty coefficient; calculate the optimal modal center frequency and use it to design a Wiener filter group, use the Wiener filter group to decompose the expected sound signal and the reference sound signal to obtain multiple decomposition sub-signal pairs; use the sub-signal of the reference sound signal to filter out the in-band noise of the sub-signal of the expected sound signal, output all denoised modes as the final mode, and sum all denoised modes to obtain a reconstructed denoised signal. The method has strong adaptability and can effectively remove noise from the collected sound signal of the bearing, and is particularly suitable for improving the signal-to-noise ratio of the acoustic signal of the track side of the train bearing.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical component monitoring and fault diagnosis, and in particular to a variable-scale adaptive denoising method for an acoustic differential intelligent bearing and an acoustic intelligent bearing. Background Art

[0002] Bearings are a key part of rotating equipment, supporting their weight and high-speed rotation. Due to long-term exposure to heavy loads, alternating stresses, and complex environments, bearing structures are prone to deformation and wear, which can lead to equipment failure and shutdown, and even cause economic losses and casualties. Therefore, monitoring and diagnosing the health status of bearings is of great significance to maintaining the normal operation of equipment, and has great economic and social benefits. At present, equipment intelligence is an important way to achieve industrial modernization, which has attracted great attention from various countries. Bearing units with self-sensing, self-decision-making and self-regulation functions - smart bearings have gradually become a research hotspot.

[0003] At present, most intelligent bearings use integrated speed, temperature and vibration sensors to sense the operating status of bearings, ignoring the use of sound as an information source. In fact, acoustic signals contain rich information about bearing status, and have unique advantages of early warning and high sensitivity compared to temperature and vibration signals. They have been widely used in train trackside acoustic detection. However, due to the complex wheel-rail structure in the trackside system and interference from factors such as background noise, the noise in the acoustic signal collected by the microphone often exhibits multi-scale characteristics in the full frequency band. In addition, many problems such as sound source aliasing, signal distortion, and low signal-to-noise ratio further limit the practical application of the trackside system.

[0004] Aiming at the intelligent diagnosis and real-time processing requirements of intelligent bearing status information, it is an important requirement for bearing intelligence to realize self-diagnosis and self-learning of bearing status through adaptive processing of bearing status signals. Among them, the least mean square (LMS) algorithm is a commonly used adaptive filtering technology. Its principle is to minimize the error between the expected signal and the reference signal, realize the mining and self-learning of key data in the expected signal, and realize the enhancement of weak signals under complex noise. Adaptive filtering can use the results of the filter parameters obtained at the previous moment to automatically adjust the filter parameters at the current moment to adapt to the unknown or time-varying statistical characteristics of the signal and noise, thereby achieving optimal filtering. However, when faced with non-stationary, multi-scale noise and low signal-to-noise ratio, the LMS algorithm cannot completely filter out.

[0005] In particular, variational mode decomposition (VMD) can adaptively perceive and mine sensitive modes in signals, decompose broadband noise into multiple modal scales, and reduce the non-stationarity of noise. However, VMD relies heavily on predefined parameters. Improper parameter setting will lead to the loss of key fault information and the introduction of redundant noise, which is not conducive to the accurate perception and intelligent diagnosis of weak faults. Summary of the invention

[0006] In order to overcome the defects existing in the above-mentioned prior art, the purpose of the present invention is to provide a variable-scale adaptive denoising method for an acoustic differential intelligent bearing and an acoustic intelligent bearing.

[0007] In order to achieve the above-mentioned object of the present invention, the present invention provides a variable-scale adaptive denoising method for acoustic differential intelligent bearings, collecting the expected sound signal x1(n) on the end face of the outer ring of the bearing and collecting the reference sound signal x2(n) on the axle box cover;

[0008] Perform adaptive variational mode decomposition on the desired sound signal x1(n) to determine the optimal decomposition layer number k opt and bandwidth penalty coefficient α opt ;

[0009] According to the optimal decomposition level k opt and bandwidth penalty coefficient α opt The optimal modal center frequency is calculated and used to design a Wiener filter bank, and the Wiener filter bank is used to decompose the desired sound signal x1(n) and the reference sound signal x2(n) to obtain multiple decomposition sub-signal pairs;

[0010] The in-band noise of the sub-signal of the desired sound signal x1(n) is filtered out using the sub-signal of the reference sound signal x2(n), all denoised modes are output as the final mode, and all denoised modes are summed to obtain the reconstructed denoised signal.

[0011] This method has strong adaptability and can effectively remove noise from the collected bearing sound signals. It is particularly suitable for improving the signal-to-noise ratio of the train bearing trackside acoustic signal.

[0012] The preferred solution of this method: the steps of performing adaptive variational modal decomposition on the desired sound signal x1(n) are:

[0013] Initialize VMD decomposition parameters, including noise relaxation setting tau, decomposition DC component setting DC, modal center frequency initialization method init and convergence criterion tol, set the decomposition layer vector K = [k1, k2, …, k p ]∈R p (p represents the dimension of the decomposition layer vector K), bandwidth penalty coefficient vector α = [α1, α2, …, α q ]∈R q(q represents the dimension of the bandwidth penalty coefficient α), initialize the kurtosis matrix Q = zeros(p,q)∈R p×q ;

[0014] Select parameter combination (k i ,α j )(1≤i≤p,1≤j≤q), perform VMD decomposition on the desired sound signal x1(n) to obtain the modal {u k The frequency domain expression of {u k (ω)} and the corresponding center frequency {ω k}:

[0015] [{u k (ω)},{ω k}]=VMD(x1,k i ,α j ,tau,DC,init,tol) (9), Where x1 is the expected sound signal, {u k (ω)}={u1(ω),u2(ω),…,u ki (ω)} is the frequency domain expression of each mode of VMD decomposition output of x1, {ω k}={ω1,ω2,…,ω ki} is the center frequency of each mode;

[0016] For {u k (ω)} is inversely transformed to obtain the time domain expression of each mode {u k (n)}, in the time domain, u k (n) Add and reconstruct the denoised signal based on VMD decomposition

[0017]

[0018] Where IFFT(·) represents inverse Fourier transform;

[0019] Calculate the reconstructed signal The kurtosis of and write it into the kurtosis matrix;

[0020]

[0021] Where N is the data reconstruction signal Length;

[0022] Traverse all combinations of decomposition parameters and bandwidth penalty parameters (k i , α j )(1≤i≤p,1≤j≤q), calculate the kurtosis of the reconstructed signal under all combinations of decomposition parameters and bandwidth penalty parameters, and obtain the kurtosis matrix Q;

[0023] Search for the maximum value of the element in the kurtosis matrix Q, and search for the decomposition parameter (k opt ,α opt ) as the optimal decomposition layer number and bandwidth penalty parameter.

[0024] The preferred solution of this method: the optimal modal center frequency {ω opt The calculation steps of} are:

[0025] With the optimal decomposition level k opt and bandwidth penalty parameter α opt The optimal modal center frequency {ω opt};

[0026] [{u opt (ω)},{ω opt}]=VMD(x1,k opt ,α opt ,tau,DC,init,tol) (12), Where x1 is the expected sound signal, {u opt (ω)}={u1(ω),u2(ω),…,u kopt (ω)} is the optimal decomposition level k opt and bandwidth penalty parameter α opt The frequency domain expression of the optimal modal shape is determined, {ω opt}={ω1,ω2,…,ω i ,…,ω kopt} is the center frequency corresponding to the optimal mode.

[0027] The design steps of the Wiener filter bank are:

[0028] With the optimal modal center frequency {ω opt} and bandwidth penalty coefficient α opt Construct Wiener filter bank {W i}={W1,W2,…,W i ,…,W kopt}, where the Wiener filter W i is a bandpass filter, whose kernel function is defined as ω is the normalized frequency.

[0029] Using the Wiener filter bank {W i} Perform Wiener filtering on the expected sound signal x1(n) and the reference sound signal x2(n) to obtain the decomposed sub-signals {x1 (i)}、{x2 (i)};

[0030]

[0031] Then decompose the sub-signal pair (x1 (i) , x2 (i) )(1≤i≤k opt ) performs LMS filtering to remove in-band noise and outputs the final modal IMF.

[0032] The preferred solution of this method: the step of removing the in-band noise of the decomposition sub-signal pair is:

[0033] Set the length of the IIR filter to M, the current iteration step k = M, and initialize the filter coefficient to w i (k), step size parameter μ, number of iterations iter;

[0034] Update the parameter filter coefficients w iteratively according to equations (7) and (8) k (k) and the error signal e(k) until k = iter, the iteration is completed, and the optimal filter coefficient w is obtained i * ;

[0035] e(k)=x1 (i) (k)-w k T (k)u(k) (14),

[0036] w i (k+1)=w i (k)+2μu(k)e(k) (15),

[0037] In the formula, T represents the transposition operation, w i (k) is the filter coefficient of the i-th IIR filter in the current iteration, x1 (i) (k) is the input of the i-th sub-signal of the desired sound signal in the current iteration, u(k) = [x2 (i) (k),x2 (i) (k-1),…,x2 (i) (k-M+1)] T is the filter input of the i-th sub-signal of the reference sound signal in the current iteration, e(k) is the instantaneous error calculated in the current iteration, and μ is the iteration step size;

[0038] Finally, the decomposed sub-signal x1 is calculated (i) The error between the filter output and the denoised modal IMF i ;

[0039]

[0040] The present invention also proposes an acoustic intelligent bearing, comprising a bearing body, wherein the bearing body is positioned on an axis in an axle box body, a first sound collection module is provided on the outer surface of the bearing body with a sound hole on one side facing the bearing, and a second sound collection module is provided on the outer surface of the axle box body, and the output ends of the first sound collection module and the second sound collection module are both connected to the signal input end of a control module, and the control module analyzes the sound signal of the intelligent bearing according to the above-mentioned acoustic differential intelligent bearing variable-scale adaptive denoising method.

[0041] The beneficial effects of the present invention are as follows: the present invention designs a Wiener filter bank based on VMD ergodic decomposition to achieve variable-scale perception of the desired signal. Since the bandwidth of the Wiener filter is compact, the non-stationarity of the noise in the broadband is effectively suppressed, thereby ensuring a relatively ideal LMS adaptive denoising effect. Compared with the existing trackside system microphone array technology and noise reduction technology, the present invention uses fewer microphones and can eliminate in-band noise at the same time. The method has strong adaptability and can effectively improve the signal-to-noise ratio of the train bearing trackside acoustic signal.

[0042] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0044] Figure 1 It is a schematic diagram of the sound denoising process of the method;

[0045] Figure 2 It is a structural diagram of the acoustic intelligent bearing. DETAILED DESCRIPTION

[0046] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0047] In the description of the present invention, unless otherwise specified and limited, it should be noted that the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a mechanical connection or an electrical connection, or it can be the internal connection between two components. It can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to the specific circumstances.

[0048] The present invention provides an embodiment of a variable-scale adaptive denoising method for an acoustic differential intelligent bearing. A group of miniature acoustic microphone sensors with physical perception differences are arranged on the bearing body structure to construct an intelligent bearing driven by acoustic differential characteristics. Considering the propagation characteristics of acoustic signals in the intelligent bearing, a main microphone perception signal and a secondary microphone perception signal group are formed. The variational mode decomposition and adaptive filtering characteristics are further integrated to realize denoising of the sound signal of the intelligent bearing, which is applied to early weak signal fault diagnosis and intelligent status monitoring of the bearing.

[0049] In this embodiment, the acoustic differential intelligent bearing structure uses two miniature microphones to collect bearing acoustic signals, one of which is embedded in the end face of the outer ring of the bearing, with one side of the sound hole facing the bearing, recorded as the main microphone, and the other is placed on the axle box cover, with one side of the sound hole facing the axle box, recorded as the secondary microphone. The main microphone collects signals with the least intermediate transmission interface and can be as close to the fault source as possible. The received signal x1(n) has a high degree of reliability and is called the expected sound signal. In comparison, the signal collected by the secondary microphone placed on the axle box is transmitted through multiple layers of media, is far away from the fault source, and has a low signal strength. The collected signal x2(n) is mainly the noise of the entire axle box system, which is called the reference sound signal. Compared with the sound collection system placed on the trackside, the microphone integrated in the bearing body is less affected by noise and Doppler distortion, and the complexity of signal processing is also reduced. The spatial difference between the two signals is used in combination with the variable scale adaptive denoising method to quickly complete bearing detection and diagnosis.

[0050] Specifically, the variable-scale adaptive denoising method for acoustic differential intelligent bearings provided by the present invention is as follows:

[0051] The expected sound signal x1(n) is collected on the end face of the outer ring of the bearing, and the reference sound signal x2(n) is collected on the axle box cover. Figure 2 As shown, a first sound collection module with a sound hole facing the bearing on one side may be provided on the outer surface of the bearing body to collect the desired sound signal x1(n), and a second sound collection module may be provided on the outer surface of the axle box body to collect the reference sound signal x2(n).

[0052] like Figure 1 As shown, adaptive variational modal decomposition is performed on the desired sound signal x1(n) to determine the optimal decomposition layer number k opt and bandwidth penalty coefficient α opt .

[0053] Specifically, the steps of performing adaptive variational modal decomposition on the desired sound signal x1(n) are:

[0054] Initialize VMD decomposition parameters, including noise relaxation setting tau, decomposition DC component setting DC, modal center frequency initialization method init and convergence criterion tol, set the decomposition layer vector K = [k1, k2, …, k p ]∈R p , p represents the dimension of the decomposition layer vector K, and the bandwidth penalty coefficient vector α=[α1,α2,…,α q ]∈R q , q represents the dimension of the bandwidth penalty coefficient α, and the kurtosis matrix Q = zeros(p,q)∈R p×q .

[0055] Select parameter combination (k i ,α j )(1≤i≤p,1≤j≤q), perform VMD decomposition on the desired sound signal x1(n) to obtain the modal {u k The frequency domain expression of {u k (ω)} and the corresponding center frequency {ω k}:

[0056] [{u k (ω)},{ω k}]=VMD(x1,k i ,α j ,tau,DC,init,tol) (17), Where x1 is the expected sound signal, {u k (ω)}={u1(ω),u2(ω),…,u ki (ω)} is the frequency domain expression of each mode of VMD decomposition output of x1, {ω k}={ω1,ω2,…,ω ki} is the center frequency of each mode;

[0057] For {u k (ω)} is inversely transformed to obtain the time domain expression of each mode {u k (n)}, in the time domain, u k (n) Add and reconstruct the denoised signal based on VMD decomposition

[0058]

[0059] Here, IFFT(·) represents inverse Fourier transform.

[0060] Calculate the reconstructed signal The kurtosis of and write it into the kurtosis matrix;

[0061]

[0062] Where N is the data reconstruction signal Length;

[0063] Traverse all combinations of decomposition parameters and bandwidth penalty parameters (k i , α j )(1≤i≤p,1≤j≤q), calculate the kurtosis of the reconstructed signal under all combinations of decomposition parameters and bandwidth penalty parameters, and obtain the kurtosis matrix Q.

[0064] Search for the maximum value of the element in the kurtosis matrix Q, and search for the decomposition parameter (k opt ,α opt ) as the optimal decomposition layer number and bandwidth penalty parameter.

[0065] Get the optimal decomposition level k opt and bandwidth penalty parameter α opt After that, according to the optimal decomposition layer number k opt and bandwidth penalty coefficient α opt Calculate the optimal modal center frequency {ω opt}A Wiener filter bank is designed based on this, and the Wiener filter bank is used to decompose the expected sound signal x1(n) and the reference sound signal x2(n) to obtain multiple decomposition sub-signal pairs.

[0066] Specifically, the optimal modal center frequency {ω opt The calculation steps of} are:

[0067] With the optimal decomposition level k opt and bandwidth penalty parameter α opt The optimal modal center frequency {ω opt};

[0068] [{u opt (ω)},{ω opt}]=VMD(x1,k opt ,α opt ,tau,DC,init,tol) (20), Where x1 is the expected sound signal, {u opt (ω)}={u1(ω),u2(ω),…,u kopt (ω)} is the optimal decomposition level k opt and bandwidth penalty parameter α opt The frequency domain expression of the optimal mode is determined, {ω opt}={ω1,ω2,…,ω i ,…,ω kopt} is the center frequency corresponding to the optimal mode.

[0069] Here, the optimal VMD decomposition is performed on the desired sound signal x1(n), and the previous traversal decomposition is only to search for the optimal decomposition layer number k opt and bandwidth penalty coefficient α opt , that is, to realize the variable scale perception of the desired signal, extract the optimal scale range, and the optimal modal center frequency {ω opt} and α opt According to the optimal scale range obtained by edge scale perception, a Wiener filter bank can be designed to perform Wiener filtering decomposition on the desired sound signal x1(n) and the reference sound signal x2(n).

[0070] The design steps of the Wiener filter bank are:

[0071] With the optimal modal center frequency {ω opt} and bandwidth penalty coefficient α opt Construct Wiener filter bank {W i}={W1,W2,…,W i ,…,W kopt}, where the Wiener filter W i is a bandpass filter, whose kernel function is defined as ω is the normalized frequency.

[0072] The optimal mode output of VMD corresponds to the center frequency {ω opt}={ω1,ω2,…,ω i ,…,ω kopt} is also the normalized output. The specific normalization algorithm can be found in the paper that proposed the VMD algorithm.

[0073] Using the Wiener filter bank {W i} Perform Wiener filtering on the expected sound signal x1(n) and the reference sound signal x2(n) to obtain the decomposed sub-signals {x1 (i)}、{x2 (i)};

[0074]

[0075] Then, the sub-signal of the reference sound signal x2(n) is used to filter out the in-band noise of the sub-signal of the desired sound signal x1(n), all denoised modes are output as the final mode, and all denoised modes are summed to obtain the reconstructed denoised signal.

[0076] In this embodiment, the sub-signal pair (x1 (i) , x2 (i) )(1≤i≤k opt ) performs LMS filtering to remove in-band noise and then outputs the final modal IMF.

[0077] The steps to remove the in-band noise of the decomposed sub-signal pair are:

[0078] Set the length of the IIR filter to M, the current iteration step k = M, and initialize the filter coefficient to w i (k), step size parameter μ, number of iterations iter;

[0079] Update the parameter filter coefficients w iteratively according to equations (7) and (8) k (k) and the error signal e(k) until k = iter, the iteration is completed, and the optimal filter coefficient w is obtained i * ;

[0080] e(k)=x1 (i) (k)-w k T (k)u(k) (22),

[0081] w i (k+1)=w i (k)+2μu(k)e(k) (23),

[0082] In the formula, T represents the transposition operation, w i (k) is the filter coefficient of the i-th IIR filter in the current iteration, x1 (i) (k) is the input of the i-th sub-signal of the desired sound signal in the current iteration, u(k) = [x2 (i) (k),x2 (i) (k-1),…,x2 (i) (k-M+1)] T is the filter input of the i-th sub-signal of the reference sound signal in the current iteration, e(k) is the instantaneous error calculated in the current iteration, and μ is the iteration step size;

[0083] Finally, the decomposed sub-signal x1 is calculated (i) The error between the filter output and the denoised modal IMF i ;

[0084]

[0085] Set all modal {IMF i}Sum and get the reconstructed noise reduction signal Right now Complete the denoising of the sound signal of the smart bearing.

[0086] The present invention also proposes an embodiment of an acoustic intelligent bearing, such as Figure 2As shown, it includes a bearing body 2, the bearing body 2 is positioned on a shaft 3 in an axle box 1, a first sound collection module 4 with a sound hole facing the bearing on one side is provided on the outer surface of the bearing body 2, and a second sound collection module 5 is provided on the outer surface of the axle box 1. The first sound collection module 4 can be close to the fault source to the greatest extent, and there are few intermediate transmission interfaces, and the received signal reliability is high. In comparison, the second sound collection module 5 is far away from the fault source, and after being transmitted through multiple layers of media, the signal strength is small, and the collected signal is mainly the noise of the entire axle box system. Collecting two sound signals with spatial differences can realize the intelligent monitoring and diagnosis of bearings more comprehensively. Preferably, the first sound collection module 4 is arranged on the outer ring end face of the bearing body 2, and the second sound collection module 5 is arranged on the outer side of the axle box cover of the axle box 1. The output ends of the first sound collection module 4 and the second sound collection module 5 are both connected to the signal input end of a control module 6, and the control module 6 analyzes the sound signal of the intelligent bearing according to the above-mentioned acoustic differential intelligent bearing variable scale adaptive denoising method. Among them, the first sound collection module 4 and the second sound collection module 5 are preferably but not limited to using microphones.

[0087] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0088] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A variable-scale adaptive denoising method for acoustic differential intelligent bearings, characterized in that: Collect the expected sound signal x1 on the end face of the bearing outer ring, and collect the reference sound signal x2 on the axle box cover; Perform adaptive variational mode decomposition on the desired sound signal x1 to determine the optimal decomposition layer number k opt and bandwidth penalty coefficient α opt ; According to the optimal decomposition level k opt and bandwidth penalty coefficient α opt The optimal modal center frequency is calculated and used to design a Wiener filter bank, and the Wiener filter bank is used to decompose the desired sound signal x1 and the reference sound signal x2 to obtain multiple decomposition sub-signal pairs; The in-band noise of the sub-signal of the desired sound signal x1 is filtered out using the sub-signal of the reference sound signal x2, all denoised modes are output as the final mode, and all denoised modes are summed to obtain the reconstructed denoised signal.

2. The variable-scale adaptive denoising method for acoustic differential intelligent bearings according to claim 1 is characterized in that: The steps of adaptive variational modal decomposition of the desired sound signal x1 are: Initialize VMD decomposition parameters, including noise relaxation setting tau, decomposition DC component setting DC, modal center frequency initialization method init and convergence criterion tol, set the decomposition layer vector K = [k1, k2, …, k p ]∈R p , p represents the dimension of the decomposition layer vector K, and the bandwidth penalty coefficient vector α=[α1,α2,…,α q ]∈R q , q represents the dimension of the bandwidth penalty coefficient α, and the kurtosis matrix Q = zeros(p,q)∈R p×q ; Select parameter combination (k i ,α j )(1≤i≤p,1≤j≤q), perform VMD decomposition on the desired sound signal x1 to obtain the modal {u k The frequency domain expression of {u k (ω)} and the corresponding center frequency {ω k }: [{u k (ω)},{ω k }]=VMD(x1,k i ,α j ,tau,DC,init,tol)(1), where x1 is the expected sound signal, {u k (ω)}={u1(ω),u2(ω),…,u ki (ω)} is the frequency domain expression of each mode of VMD decomposition output of x1, {ω k }={ω1,ω2,…,ω ki } is the center frequency of each mode; For {u k (ω)} is inversely transformed to obtain the time domain expression of each mode {u k (n)}, in the time domain, u k (n) Add and reconstruct the denoised signal based on VMD decomposition Where IFFT(·) represents inverse Fourier transform; Calculate the reconstructed signal The kurtosis of and write it into the kurtosis matrix; Where N is the data reconstruction signal Length; Traverse all combinations of decomposition layers and bandwidth penalty coefficients (k i , α j )(1≤i≤p,1≤j≤q), calculate the kurtosis of the reconstructed signal under all combinations of decomposition layers and bandwidth penalty coefficients, and obtain the kurtosis matrix Q; Search for the maximum value of the element in the kurtosis matrix Q, and search for the decomposition parameter (k opt ,α opt ) as the optimal decomposition layer number and bandwidth penalty coefficient.

3. The variable-scale adaptive denoising method for acoustic differential intelligent bearings according to claim 1 is characterized in that: Optimal modal center frequency {ω opt The calculation steps of} are: With the optimal decomposition level k opt and bandwidth penalty coefficient α opt As the decomposition level, the optimal modal center frequency {ω opt }; [{u opt (ω)},{ω opt }]=VMD(x1,k opt ,a opt ,tau,DC,init,tol) (4), Where x1 is the expected sound signal, According to the optimal decomposition level k opt and bandwidth penalty coefficient α opt The frequency domain expression of the determined optimal mode, is the center frequency corresponding to the optimal mode.

4. The variable-scale adaptive denoising method for acoustic differential intelligent bearings according to claim 3 is characterized in that: The design steps of the Wiener filter bank are: With the optimal modal center frequency {ω opt } and bandwidth penalty coefficient α opt Constructing Wiener filter banks The Wiener filter W i is a bandpass filter, whose kernel function is defined as ω is the normalized frequency.

5. The variable-scale adaptive denoising method for acoustic differential intelligent bearings according to claim 1 is characterized in that: Using the Wiener filter bank {W i } Perform Wiener filtering on the expected sound signal x1 and the reference sound signal x2 to obtain the decomposed sub-signals {x1 (i) }、{x2 (i) }; Then decompose the sub-signal pair (x1 (i) , x2 (i) )(1≤i≤k opt ) performs LMS filtering to remove in-band noise and outputs the final modal IMF.

6. The variable-scale adaptive denoising method for acoustic differential intelligent bearings according to claim 1 or 5, characterized in that: The steps to remove the in-band noise of the decomposed sub-signal pair are: Set the length of the IIR filter to M, the current iteration step k = M, and initialize the filter coefficient to w i (k), step size parameter μ, number of iterations iter; Update the filter coefficients w iteratively according to equations (6) and (7) i (k) and the error signal e(k) until k = iter, the iteration is completed, and the optimal filter coefficient w is obtained i * ; e(k)=x1 (i) (k)-w i T (k)u(k)(6), w i (k+1)=w i (k)+2μu(k)e(k)(7), In the formula, T represents the transposition operation, w i (k) is the filter coefficient of the i-th IIR filter in the current iteration, x1 (i) (k) is the input of the i-th sub-signal of the desired sound signal in the current iteration, u(k) = [x2 (i) (k),x2 (i) (k-1),…,x2 (i) (k-M+1)] T is the filter input of the i-th sub-signal of the reference sound signal in the current iteration, e(k) is the instantaneous error calculated in the current iteration, and μ is the iteration step size; Finally, the decomposed sub-signal x1 is calculated (i) The error between the filter output and the denoised modal IMF i ; 7. An acoustic intelligent bearing, characterized in that: The invention comprises a bearing body, wherein the bearing body is positioned on a shaft in an axle box body, a first sound collection module is provided on the outer surface of the bearing body with a sound hole on one side facing the bearing, and a second sound collection module is provided on the outer surface of the axle box body, and the output ends of the first sound collection module and the second sound collection module are both connected to the signal input end of a control module, and the control module analyzes the sound signal of the intelligent bearing according to the variable-scale adaptive denoising method of the acoustic differential intelligent bearing according to any one of claims 1 to 6.

Images