A microphone array-based adaptive bearing fault diagnosis method

Abstract

The application discloses a microphone array-based adaptive bearing fault diagnosis method, which comprises the following steps: 1, collecting bearing fault sound array signals by using a microphone array, and selecting a bearing position as an enhanced target position; 2, establishing an adaptive robust beamformer, inputting the sound array signals into the beamformer, and obtaining an enhanced signal; 3, setting a time-frequency domain filter, solving a filter length by using a PSO algorithm, and performing time-frequency domain filtering on the enhanced signal; 4, calculating a characteristic frequency corresponding to a theoretical fault type; and 5, decomposing the filtered signal by using a wavelet transform, selecting a decomposed signal to perform envelope power spectrum analysis, and determining a fault type according to a peak frequency of the envelope power spectrum.

Description

Technical Field

[0001] This invention belongs to the field of bearing fault diagnosis technology, and relates to an adaptive bearing fault diagnosis method based on a microphone array. Background Technology

[0002] Rolling bearings are ubiquitous in industrial production, often operating in harsh environments, making their failure a major cause of industrial machinery accidents. Therefore, the monitoring and diagnosis of rolling bearings has always been a key research focus. Over the past few decades, significant progress has been made in using vibration analysis for bearing fault diagnosis. With the continuous development of signal processing, various vibration-based analysis methods have been developed, including spectral kurtosis, envelope spectrum, empirical mode decomposition, and variational mode decomposition. Although vibration analysis is a mature and effective measurement method, its limitations as a contact measurement have led to increasing attention on non-contact acoustic fault diagnosis techniques in recent years. Research has demonstrated the feasibility of acoustic-based fault diagnosis for rotating machinery. Therefore, acoustic fault diagnosis can be considered an advancement over vibration-based fault diagnosis. However, the complex acoustic environment of industrial settings leads to low signal-to-noise ratios, limiting the application of acoustic-based fault diagnosis techniques.

[0003] Current research on acoustic-based fault diagnosis mainly categorizes into two methods: single-microphone methods and array microphone methods. While single-microphone methods have proven effective in some cases, they still suffer from the inability to separate signals with similar spectral distributions and are susceptible to environmental noise. In recent years, with the application of microphone arrays, research on acoustic-based fault diagnosis has flourished. Microphone arrays can simultaneously acquire acoustic spatial and time-frequency domain information, allowing for spatial processing of beamformer output signals to extract fault features. However, in low signal-to-noise ratio environments, beamformers may lose their noise suppression effect, leading to difficulties in signal processing. Although beamforming-enhanced signals can effectively attenuate spatial noise interference, it cannot filter out noise at the enhanced points; and time-frequency domain filtering methods for extracting fault features are often applied to single-channel signals, failing to effectively suppress interference from other spatial noise. Summary of the Invention

[0004] To address the aforementioned technical problems, the purpose of this invention is to provide an adaptive bearing fault diagnosis method based on a microphone array.

[0005] This invention provides an adaptive bearing fault diagnosis method based on a microphone array, characterized by comprising:

[0006] Step 1: Use a microphone array to collect the sound array signal of the bearing failure, and select the bearing location as the target location for enhancement;

[0007] Step 2: Establish an adaptive robust beamformer, input the sound array signal into the beamformer to obtain the enhanced signal;

[0008] Step 3: Construct a time-frequency domain filter, solve for the filter length using the PSO algorithm, and perform time-frequency domain filtering on the enhanced signal;

[0009] Step 4: Calculate the characteristic frequencies corresponding to the theoretical fault types;

[0010] Step 5: Decompose the filtered signal using wavelet transform, select the decomposed signal for envelope power spectrum analysis, and determine the fault type based on the peak frequency of the envelope power spectrum.

[0011] The adaptive bearing fault diagnosis method based on a microphone array of the present invention has the following beneficial effects:

[0012] (1) A novel adaptive robust beamformer was designed, which can achieve efficient suppression of non-target direction noise signals in low signal-to-noise ratio environments.

[0013] (2) Establish a composite index for judging faults, and then design an adaptive time-frequency domain filter to effectively reduce the homogeneous noise of bearing direction sound.

[0014] (3) Combining spatial domain filtering and time-frequency domain filtering can effectively reduce the impact of noise on the acoustic characteristics of faults. Attached Figure Description

[0015] Figure 1 This is a flowchart of an adaptive bearing fault diagnosis method based on a microphone array according to the present invention.

[0016] Figure 2 The envelope power spectrum obtained after processing the array signal by the method of the present invention;

[0017] Figure 3 The envelope power spectrum is obtained by directly performing wavelet decomposition on the vibration signal. Detailed Implementation

[0018] like Figure 1 As shown, an adaptive bearing fault diagnosis method based on a microphone array according to the present invention includes:

[0019] Step 1: Use a microphone array to collect the sound array signal of the bearing failure, and select the bearing location as the target location for enhancement.

[0020] Step 2: Establish an adaptive robust beamformer. Input the sound array signal into the beamformer to obtain the enhanced signal, specifically:

[0021] Step 2.1: Establish a traditional MVDR beamformer, specifically as follows:

[0022] The output of the MVDR beamformer is expressed as:

[0023] y(k)=w H X(k)

[0024] Where w is the weight vector of the beamformer, w H Denotes the Hermitian transpose of w; k is the number of microphones in the microphone array; the energy output of the beamformer is:

[0025] E{|y(k) 2 |}=w H Rw

[0026] Where R is the autocorrelation matrix of the array signal.

[0027] To prevent the desired signal source from being canceled out during beamformer output energy minimization, a linear constraint is imposed throughout the optimization process:

[0028] w H ad=1

[0029] Therefore, the problem with traditional MVDR beamformers is transformed into:

[0030]

[0031] Among them, a d To enhance the guidance matrix of the target direction relative to the reference point.

[0032] Step 2.2: Adjust the linear constraints of the traditional MVDR beamformer to add robustness, specifically:

[0033] The problem of traditional MVDR beamformers is transformed into:

[0034]

[0035] in, Let n be the sample covariance matrix of the array signal, and n be the length of each channel in the array signal.

[0036]

[0037] Where η is the set forgetting factor; n is the signal length. To increase the robustness of the beamformer, an inaccurate steering vector z is set to minimize the weighted power output of the array in the case of uncertainty in the steering matrix.

[0038]

[0039] Where ε represents the possible range covering the imprecise steering vector z, centered on the assumed steering vector, i.e.:

[0040] ε={Gu+a d |||u||≤1}

[0041] Where u is a set constant;

[0042] Assumption If the matrix is ​​positive definite, the optimization problem can be further transformed into the following form:

[0043]

[0044] Let G = δI, then we get the following:

[0045]

[0046] However, the nonlinear constraints in the equation are nonconvex. Under the optimal solution, a rotation is performed to select the optimal solution without affecting the optimization of the cost function, such that:

[0047]

[0048] The optimization problem can be transformed into the following convex forms:

[0049]

[0050] Construct the following Lagrange function:

[0051]

[0052] Where t(.) is the step function, thus ensuring that the constraint conditions are met, we obtain the following expression for the solution:

[0053]

[0054] Where λ is a Lagrange multiplier.

[0055] Step 2.3: By searching for the minimum weight vector, an adaptive robust beamformer is obtained. The sound array signal is input into this beamformer to obtain the enhanced signal. Specifically:

[0056] Step 2.3.1: Initialize parameters w 0 =a d α = 0.1, η = 0.97;

[0057] Step 2.3.2: Calculate the sample covariance matrix of the array signal according to the following formula:

[0058]

[0059] Step 2.3.3: Calculate the optimal step size for weight vector update:

[0060]

[0061] Where 0 < α < 1;

[0062] Step 2.3.4: Update the unconstrained weight vector:

[0063]

[0064] Step 2.3.5: Let If χ < δ||w n+1 The Lagrange multipliers are calculated according to the following formula:

[0065]

[0066] A = μ(k) 2 ((Re{π(k) H a d}) 2 -δ 2 ||π(k)|| 2 )

[0067]

[0068] Otherwise λ = 0, Return to step 2.3;

[0069] Step 2.3.5: Substitute the Lagrange multipliers into the solution expression to calculate the weight vector of the adaptive robust beamformer. Input the sound array signal into the adaptive robust beamformer to obtain the enhanced signal Y = [y1, y2…y]. n ].

[0070] Step 3: Construct a time-frequency domain filter, solve for the filter length using the PSO algorithm, and perform time-frequency domain filtering on the enhanced signal, specifically as follows:

[0071] Step 3.1: Construct the filter:

[0072]

[0073] The time-frequency domain filter output is:

[0074]

[0075] Where L is the filter length and the window function is a rectangular window.

[0076] Step 3.2: Use the enhanced signal output from the beamformer as the input signal of the filter, and obtain the composite index of the signal by calculating the kurtosis, peak factor, impulse factor and envelope spectral entropy of the filter output signal;

[0077]

[0078] Where K1 is the kurtosis, K2 is the peak factor, K3 is the impulse factor, Es is the envelope spectral entropy, and σ is the standard deviation of the output signal. S is the mean. R (ω) represents the envelope spectrum of the filter output signal, which is uniformly divided into intervals according to frequency, with each interval representing a percentage of P. i (S R (ω)).

[0079] The composite exponent CI, used as the objective function for optimizing the filter length, is calculated using the following formula:

[0080]

[0081] Step 3.3: Optimize the length of the time-frequency filter using the particle swarm optimization algorithm with the composite exponent as the fitness function, and output the filtered signal obtained from the optimal filter.

[0082] Step 4: Calculate the characteristic frequencies corresponding to the theoretical fault types, specifically:

[0083]

[0084] Among them, f o f is the characteristic frequency of the outer ring fault. i f is the characteristic frequency of the outer ring fault. b f is the characteristic frequency of the outer ring fault. r Where is the shaft rotation frequency, m is the number of rolling elements, d is the diameter of the rolling elements, D is the bearing pitch diameter, and β is the contact angle.

[0085] Step 5: Decompose the filtered signal using wavelet transform, select the decomposed signal for envelope power spectrum analysis, and determine the fault type based on the peak frequency of the envelope power spectrum. Specifically:

[0086] Step 5.1: Decompose the filtered signal using wavelet transform;

[0087] Step 5.2: Plot the envelope power spectrum of the decomposed second-level signal. Based on the peak frequency of the envelope power spectrum and the characteristic frequency corresponding to the theoretical fault type, determine the fault type.

[0088] To demonstrate the effectiveness of the microphone array-based adaptive bearing fault diagnosis method proposed in this patent, a vibration signal-based fault diagnosis method is used for comparison. The experiment should follow these rules: select fault array audio signals and vibration signals with a characteristic frequency of 75Hz, and apply the same decomposition and analysis method to the vibration signal and the filtered audio signal respectively to obtain the envelope power spectrum of the signals.

[0089] The envelope power spectrum obtained after processing the array signal by the method proposed in this invention is as follows: Figure 2 As shown, there is a significant peak at the fault characteristic location; wavelet decomposition and envelope power spectrum analysis are performed directly on the vibration signal, and the envelope power spectrum is shown in the figure. Figure 3 As shown, in addition to the peak value at the characteristic frequency, a peak value also appears around 10Hz. This peak value around 10Hz is caused by the shaft center tilting downwards due to heavy load, generating an unbalanced torque during shaft rotation, which excites vibration; it is a type of interference from the same source. Therefore, the method proposed in this patent can suppress interference caused by this reason to a certain extent, making the characteristic peaks of bearing failure more obvious.

[0090] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An adaptive bearing fault diagnosis method based on a microphone array, characterized in that, include: Step 1: Use a microphone array to collect the sound array signal of the bearing failure, and select the bearing location as the target location for enhancement; Step 2: Establish an adaptive robust beamformer, input the sound array signal into the beamformer to obtain the enhanced signal; Step 3: Construct a time-frequency domain filter, solve for the filter length using the PSO algorithm, and perform time-frequency domain filtering on the enhanced signal; Step 4: Calculate the characteristic frequencies corresponding to the theoretical fault types; Step 5: Decompose the filtered signal using wavelet transform, select the decomposed signal for envelope power spectrum analysis, and determine the fault type based on the peak frequency of the envelope power spectrum. Step 2 specifically involves: Step 2.1: Establish a traditional MVDR beamformer; Step 2.2: Adjust the constraints of the linear form of the traditional MVDR beamformer to add robustness; Step 2.3: By searching for the minimum weight vector, an adaptive robust beamformer is obtained. The sound array signal is then input into the beamformer to obtain the enhanced signal. Step 2.2 specifically involves: The problem of traditional MVDR beamformers is transformed into: in, Let n be the sample covariance matrix of the array signal, and n be the length of each channel in the array signal. in, The forgetting factor is set; n is the signal length. To increase the robustness of the beamformer, an inaccurate steering vector z is set to minimize the weighted power output of the array, given the uncertainty in the steering matrix. in, To cover the possible range of the imprecise steering vector z, we take the assumed steering vector as the center, that is: in, A constant set; Assumption If the matrix is ​​positive definite, the optimization problem can be further transformed into the following form: set up Then we get the following: However, the nonlinear constraints in the equation are nonconvex. Under the optimal solution, a rotation is performed to select the optimal solution without affecting the optimization of the cost function, such that: The optimization problem can be transformed into the following convex forms: Construct the following Lagrange function: Where t(.) is the step function, thus ensuring that the constraint conditions are met, we obtain the following expression for the solution: in, For Lagrange multipliers; Step 3 specifically involves: Step 3.1: Construct the filter: The time-frequency domain filter output is: Where L is the filter length and the window function is a rectangular window; Step 3.2: Use the enhanced signal output from the beamformer as the input signal of the filter, and obtain the composite index of the signal by calculating the kurtosis, peak factor, impulse factor and envelope spectral entropy of the filter output signal; Where K1 is kurtosis, K2 is peak factor, K3 is impulse factor, and Es is the envelope spectral entropy; It is the standard deviation of the output signal. The mean, The envelope spectrum of the filter output signal is divided into intervals uniformly according to frequency, with each interval representing a percentage of the total frequency. ; The composite exponent CI, used as the objective function for optimizing the filter length, is calculated using the following formula: Step 3.3: Optimize the length of the time-frequency filter using the particle swarm optimization algorithm with the composite exponent as the fitness function, and output the filtered signal obtained from the optimal filter.

2. The adaptive bearing fault diagnosis method based on a microphone array as described in claim 1, characterized in that, Step 2.1 specifically involves: The output of the MVDR beamformer is expressed as: in, It is the weight vector of the beamformer. express Hermitian transpose; k is the number of microphones in the microphone array; the energy output of the beamformer is: in, This is the autocorrelation matrix of the array signal; To prevent the desired signal source from being canceled out during beamformer output energy minimization, a linear constraint is imposed throughout the optimization process: Therefore, the problem with traditional MVDR beamformers is transformed into: in, To enhance the guidance matrix of the target direction relative to the reference point.

3. The adaptive bearing fault diagnosis method based on a microphone array according to claim 1, characterized in that, Step 2.3 specifically involves: Step 2.3.1: Initialize parameters ; Step 2.3.2: Calculate the sample covariance matrix of the array signal according to the following formula: Step 2.3.3: Calculate the optimal step size for weight vector update: in, ; Step 2.3.4: Update the unconstrained weight vector: Step 2.3.5: Let ,if Calculate the Lagrange multipliers according to the following formula: otherwise , Return to step 2.3; Step 2.3.5: Substitute the Lagrange multipliers into the solution expression to calculate the weight vector of the adaptive robust beamformer. Input the sound array signal into the adaptive robust beamformer to obtain the enhanced signal. .

4. The adaptive bearing fault diagnosis method based on a microphone array according to claim 1, characterized in that, Step 4 specifically involves: in, The characteristic frequency of the outer ring fault. The characteristic frequency of the outer ring fault. The characteristic frequency of the outer ring fault. Where is the shaft rotation frequency, m is the number of rolling elements, d is the diameter of the rolling elements, and D is the bearing pitch diameter. It represents the contact angle.

5. The adaptive bearing fault diagnosis method based on a microphone array according to claim 1, characterized in that, Step 5 specifically involves: Step 5.1: Decompose the filtered signal using wavelet transform; Step 5.2: Plot the envelope power spectrum of the decomposed second-level signal. Based on the peak frequency of the envelope power spectrum and the characteristic frequency corresponding to the theoretical fault type, determine the fault type.

Images