Sound signal feature extraction method based on loudness

Abstract

The invention discloses a sound signal feature extraction method based on loudness. The method comprises the steps that a signal acquisition module acquires and stores sound signals of a rolling bearing in real time, acquires original sound signals and transmits the original sound signals to a computer end; acquiring a loudness time spectrum of the sound signals by calculating the loudness of each intercepted short-time signal by adopting a time domain short-time slip interception method on the collected original sound signals; identifying and extracting a pulse component in the loudness time spectrum, analyzing the short-time signal characteristic loudness corresponding to the pulse component, designing an optimal finite impulse response filter by taking the maximum characteristic as a parameter basis, and filtering the original sound signal; and separating a low-frequency component from a high-frequency carrier in the signal by using a Hilbert transform envelope analysis method, and extracting a low-frequency characteristic frequency closely related to the fault characteristics of the rolling bearing, thereby realizing accurate diagnosis of the bearing fault.

Description

Technical Field

[0001] The invention relates to the field of rolling bearing fault diagnosis, and in particular to a method for extracting acoustic signal features based on loudness. Background Art

[0002] Rolling bearings are widely used components and are of great significance in various types of rotating machinery. They are used in different fields such as rail transportation, wind power generation, and aerospace. However, during operation, rolling bearings are exposed to complex environments and are prone to different types of failures, including deformation, wear, and cracks. Failure of rolling bearings can lead to serious safety accidents and even cause casualties. Therefore, detecting and diagnosing the status of bearings is essential to ensure the safe operation of mechanical equipment.

[0003] At present, the detection and fault diagnosis of bearing conditions using vibration signals have been widely studied. Vibration measurement is a contact-based method with advantages such as high stability and low interference. However, the installation of vibration sensors is challenging in narrow spaces and specific environments, where it becomes difficult to arrange and install these sensors. In contrast, acoustic signal measurement provides a non-contact method that overcomes these limitations. It has the advantages of minimal space requirements, no need for contact installation, and a wide monitoring range. Therefore, the application of acoustic signals in bearing fault diagnosis has received increasing attention in recent years. The main challenges faced by acoustic diagnostic technology are large acoustic signal interference, complex signal components, low signal-to-noise ratio, and significant differences between noise signals and vibration signals. It is difficult to directly apply vibration signal processing methods to noise signals and achieve satisfactory results. For example, patent CN103999517B discloses a low-power system and method for extracting and analyzing audio feature data, which can reduce the power consumption of the sound sensing function in the electronic device, but still does not fully solve the problem of how to accurately extract useful audio feature data in a high-noise environment; patent CN118190416 A discloses a rolling bearing signal feature extraction method, system and server based on sound signals, and proposes to use an acoustic intelligent perception data acquisition system installed on a rotating device to obtain acoustic data for monitoring the operating status of the equipment, effectively solving the problem of accurate sound extraction in a high-noise environment, but the system relies on specific hardware configuration, which affects its applicability in different devices or environments. In addition, when sound quality indicators are used as inputs to machine learning algorithms, the rich information they contain is usually not fully utilized. Summary of the invention

[0004] The technical problem to be solved by the present invention is the challenge of low signal-to-noise ratio and difficulty in feature extraction in acoustic signals. In order to solve the above problems, a method for extracting acoustic signal features based on loudness is provided.

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

[0006] A method for extracting acoustic signal features based on loudness, the method comprising the following steps

[0007] Step S1, the signal acquisition module collects and stores the rolling bearing acoustic signal in real time, collects the original acoustic signal and transmits it to the computer;

[0008] Step S2, using a time domain short-time sliding interception method to collect the original sound signal, and obtaining the loudness time spectrum of the sound signal by calculating the loudness of each intercepted short-time signal;

[0009] Step S3, identifying and extracting the pulse component in the loudness time spectrum, analyzing the short-time signal characteristic loudness corresponding to the pulse component, designing an optimal finite impulse response filter based on the maximum characteristic as a parameter, and filtering the original sound signal;

[0010] Step S4, using the Hilbert transform envelope analysis method, the low-frequency component in the signal is separated from the high-frequency carrier, and the low-frequency characteristic frequency closely related to the rolling bearing fault characteristics is extracted, thereby achieving accurate diagnosis of the bearing fault.

[0011] In step S2, the loudness-time spectrum of the acoustic signal is obtained by using a short-time sliding interception method, and the specific process is as follows:

[0012] Step S21, creating a window function w(k) to intercept the original time series x(k) (k=1, 2, N);

[0013] Step S22, by introducing the sliding parameter p and combining the window function with the time series, intercept the time series x i (k):

[0014] x i (k) = x(k)·w(k-(i-1)p);

[0015] where x i represents the time series obtained by the i-th intercept;

[0016] Parameter p controls the speed of sliding interception;

[0017] L represents the width truncated from the original time series, satisfying 0<p<L (p is an integer);

[0018] Step S23, from x i Extract the non-zero elements and construct the slip-truncated short time series y i :

[0019] y i (j) = x i((i-1)p+j);

[0020] Satisfies 0<j<L;

[0021] L represents the width truncated from the original time series, satisfying 0<p<L (p is an integer);

[0022] This short time series can be used to obtain loudness time spectrum and extract signal pulse characteristics;

[0023] Step S24, short-term signal y i Apply the Zwicker loudness model to calculate the short-term signal loudness;

[0024] Step S25, record the loudness of the short-time signal as Ld(i), and convert Ld(i) into the time domain using the following formula:

[0025]

[0026] Among them, t i represents the time instant of the i-th sampling point; fs represents the sampling frequency, that is, the number of times the signal is sampled per unit time;

[0027] Step S26, by performing sliding interception on the entire sound signal, calculating the short-time signal loudness value at each interception, and obtaining the loudness-time spectrum of the sound signal Ld = [Ld(t1), Ld(t2) ... Ld(t i )], where Ld(t i ) represents the loudness-time spectrum at a specific time point t i The loudness value of .

[0028] In step S3, the specific process of identifying and extracting the pulse component in the loudness time spectrum is as follows:

[0029] Step S31, calculate the overall mean value E(Ld) of the loudness spectrum as the basis for judging the pulse signal:

[0030]

[0031] Where N Ld is the number of data points contained in L; t j Represents the time of the jth sampling point in the discrete time series, 1≤j≤N Ld ;

[0032] Step S32: When the impact occurs, the signal loudness increases rapidly, and the judgment variable A(i) is proposed to judge Ld(t i ) is a pulse signal point; at the same time, the threshold C is introduced h and C1, considering the rise and decay process of the pulse signal, Ld(t i) as a pulse signal is as follows:

[0033]

[0034] Application formula A(i)=Ld(t i )-E(Ld) determines whether each element in the loudness-time spectrum Ld is a pulse signal point, and records all time points t in the loudness spectrum that meet the requirements j , get the corresponding short-time signal y j .

[0035] The filter parameters are designed according to the loudness characteristics of the sound signal. The specific process is:

[0036] Compute the slip-truncated short time series y using the Zwicker loudness model i The specific loudness distribution of the bark bark is extracted to find the bark range that contributes most to the loudness, i.e. the maximum specific loudness bark bark. mspl ;

[0037] By analyzing the loudness of a large number of short-term sequences, we can get bark mspl The distribution of bark mspl The number of occurrences can reflect the characteristics of bearing failure, so the occurrence mspl The bark with the highest order is used as the basis for filter parameter design;

[0038] The frequency domain range of the filter and the conversion relationship between barks are calculated:

[0039]

[0040]

[0041] The resulting frequency range will be used as the passband frequency of the bandpass filter.

[0042] Hilbert transform envelope analysis is used to diagnose rolling bearing faults. The specific process is as follows:

[0043] Step S41, for a given time series s(t), the Hilbert transform is defined as the convolution of s(t) and h(t):

[0044]

[0045] Where * is the convolution operator symbol;

[0046] H{} is the Hilbert transform operator;

[0047] Step S42, further obtaining the frequency domain representation of the Hilbert transform:

[0048]

[0049] Where S(f) is the Fourier transform of S(t), H(f) is the Fourier transform of H(t);

[0050]

[0051] Step S43, using s(t) and Construct the analysis signal z(t):

[0052]

[0053] a(t) contains the amplitude information of the signal;

[0054] Contains the phase information of the signal;

[0055] Step S44, for the signal s'(t) containing the fault component in the time series s(t), the analysis signal z'(t) is constructed using the above formula:

[0056] f a Represents the low-pass frequency component containing fault information;

[0057] Step S45: for the signal s'(t) containing the fault component, obtain Calculate the absolute value of the analytical signal z'(t) and obtain the envelope signal of the signal s'(t):

[0058]

[0059] Step S46, using the envelope signal obtained above, compare the frequency component with the highest energy proportion in the envelope signal with the fault passing frequency of the monitored bearing. If the fault frequencies are consistent, it can be determined that the bearing is faulty, thereby achieving accurate diagnosis of the bearing fault.

[0060] Beneficial effects of the present invention: The present invention can effectively extract key information from rolling bearing acoustic signals, use loudness indicators as the determining factor of filter parameters, successfully design the optimal FIR filter, enhance the fault information in the original acoustic signal, and apply envelope analysis to the filtered signal, which can effectively diagnose rolling bearing faults. The present invention combines the unique sound quality characteristics of acoustic signals and can perform targeted processing on acoustic signals. Compared with traditional signal processing methods such as spectral kurtosis, more effective and accurate diagnostic results are obtained. The method proposed by the present invention solves the challenges of low signal-to-noise ratio and difficult feature extraction in acoustic signals, and has a certain reference value for realizing the diagnosis of rolling bearing faults in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a flow chart of the method of the present invention.

[0062] Figure 2 is a flow chart of the acoustic signal feature extraction method of the present invention;

[0063] Figure 3 is a diagram of the slippage interception process of the acoustic signal of the present invention;

[0064] Figure 4 is the collected sound signal and the loudness-time spectrum of the sound signal of the present invention;

[0065] Figure 5 It is a schematic diagram of extracting a pulse signal of the present invention;

[0066] Figure 6 It is the specific loudness and maximum characteristic loudness Bark barkmspl of the short time series of the present invention;

[0067] Figure 7 It is the barkmsplBark analysis diagram of different short-term sequences of the present invention;

[0068] Figure 8 It is a schematic diagram of envelope analysis of the original signal after filtering of the present invention;

[0069] Fig. 9 It is a schematic diagram of envelope spectrum analysis of a signal after filtering using the loudness-based filter parameter determination method of the present invention;

[0070] Fig.10 It is a schematic diagram of envelope analysis of the signal after filtering using the spectral kurtosis method. DETAILED DESCRIPTION

[0071] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0072] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same technical meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0073] This embodiment provides a method for extracting acoustic signal features based on loudness. Figure 1 As shown, the steps include:

[0074] Step S1, the signal acquisition module collects and stores the acoustic signal of the rolling bearing in real time, collects the original acoustic signal x(k), and transmits it to the PC; preferably, the signal acquisition module collects the acoustic signal of the rolling bearing through the acoustic sensor AWA14425, stores the collected acoustic signal in the acquisition card NI-9234 through the data export line, develops acquisition software to communicate with the acquisition card, and stores the collected original acoustic signal x(k) in the computer, thus completing the entire sampling process.

[0075] Step S2, using the time domain short-time sliding interception method to collect the original sound signal x(k), and obtaining the loudness time spectrum Ld of the sound signal by calculating the loudness of each intercepted short-time signal;

[0076] Figure 2 This is a schematic diagram of the short-time sliding interception method for the original signal x(k). The specific process is:

[0077] S21, create a window function w(k) to intercept the original time series x(k) (k=1,2,N), the window function expression is:

[0078]

[0079] Where N is the length of the discrete time series;

[0080] S22, by introducing the sliding parameter p and combining the window function with the time series, intercepts the time series x i (k):

[0081] x i (k) = x(k)·w(k-(i-1)p);

[0082] where x i represents the time series obtained by the i-th intercept;

[0083] Parameter p controls the speed of sliding interception;

[0084] L represents the width truncated from the original time series;

[0085] Satisfies 0<p<L(p∈integer);

[0086] S23, from x i Extract the non-zero elements and construct the short time series y of the slip intercept i :

[0087] y i (j) = x i ((i-1)p+j);

[0088] Satisfies 0<j<L;

[0089] This short time series can be used to obtain loudness time spectrum and extract signal pulse characteristics;

[0090] S24, for short-term signal y i Apply the Zwicker loudness model to calculate short-term signals;

[0091] S25, record the loudness of the short-time signal as Ld(i), and convert Ld(i) into the time domain using the following formula:

[0092]

[0093] Among them, t i represents the time instant of the i-th sampling point; fs represents the sampling frequency, that is, the number of times the signal is sampled per unit time;

[0094] S26, by performing sliding interception on the entire sound signal, calculating the short-term signal loudness value at each interception, and obtaining the loudness time spectrum Ld(t i ), where Ld(t i ) represents the loudness-time spectrum at a specific time point t i The loudness value of .

[0095] Figure 3 is the loudness-time spectrum of the acoustic signal obtained from the original acoustic signal x(k).

[0096] Step S3, identifying and extracting the pulse component in the loudness time spectrum Ld, analyzing the short-time signal characteristic loudness corresponding to the pulse component, designing an optimal finite impulse response filter (FIR filter) based on the maximum characteristic loudness Bark as a parameter, and filtering the original sound signal x(k);

[0097] Figure 4 This is a schematic diagram of the detection and extraction of pulse signals in the loudness-time spectrum. The specific process is:

[0098] S31, calculating the overall mean value E(Ld) of the loudness spectrum as a basis for judging the pulse signal, specifically as follows:

[0099]

[0100] Where N Ld is the number of data points contained in L; t j Represents the time of the jth sampling point in the discrete time series, 1≤j≤N Ld ;

[0101] S32, when the impact occurs, the signal loudness increases rapidly, and the judgment variable A(i) is introduced to judge Ld(t i ) is a pulse signal point; at the same time, the threshold C is introducedh and C1, considering the rise and decay process of the pulse signal, Ld(t i ) as a pulse signal. The necessary and sufficient conditions are as follows:

[0102]

[0103] Where A(i) = Ld(t i )-E(Ld), determine whether each element in Ld is a pulse signal point, and record all time points t in the loudness spectrum that meet the requirements j , and obtain the corresponding short-time signal y j ;

[0104] S33, calculate y using the Zwicker loudness model i The specific loudness distribution of bark is extracted to extract the maximum characteristic loudness range that contributes most to the loudness. mspl ;

[0105] like Figure 5 As shown, the specific loudness and maximum characteristic loudness bark of the short time series are obtained. mspl ;

[0106] By analyzing the loudness of a large number of short-term sequences, we can get bark mspl The distribution of bark mspl The number of occurrences can reflect the characteristics of bearing failure;

[0107] according to Figure 6 The Bark analysis chart of the short-time series shown, select mspl The bark with the highest order is used as the basis for filter parameter design;

[0108] S34, calculate the frequency domain range of the filter and the conversion relationship between barks:

[0109]

[0110]

[0111] The obtained frequency range will be used as the passband frequency of the bandpass filter;

[0112] Preferably, the effect of the subsequent Hilbert transform is related to the characteristics of the original signal. Using a loudness-based filter parameter determination method to design a filter to process the original signal can greatly enhance the fault characteristics in the signal and effectively improve the application effect of envelope analysis in acoustic signals.

[0113] S4, using the Hilbert transform envelope analysis method, separates the low-frequency component x(t) from the high-frequency carrier y(t) in the signal, and extracts the low-frequency characteristic frequency that is closely related to the rolling bearing fault characteristics, thereby achieving accurate diagnosis of the bearing fault.

[0114] S41, for a given time series s(t), the Hilbert transform is defined as the convolution of s(t) and h(t), as follows:

[0115]

[0116] Where * is the convolution operator symbol;

[0117] H{} is the Hilbert transform operator;

[0118] S42, further obtain the frequency domain representation of Hilbert transform:

[0119]

[0120] Where S(f) is the Fourier transform of s(t), and H(f) is the Fourier transform of h(t), as follows:

[0121]

[0122] S43, using s(t) and Construct the analysis signal z(t):

[0123]

[0124] a(t) contains the amplitude information of the signal;

[0125] Contains the phase information of the signal;

[0126] Step S44, for the signal s'(t) containing the fault component in the time series s(t), the analysis signal z'(t) is constructed using the above formula:

[0127] f a Represents the low-pass frequency component containing fault information;

[0128] Step S45: for the signal s'(t) containing the fault component, obtain Calculate the absolute value of the analytical signal z'(t) and obtain the envelope signal of the signal s'(t):

[0129]

[0130] Step S46, using the envelope signal obtained above, compare the frequency component with the highest energy proportion in the envelope signal with the fault passing frequency of the monitored bearing. If the fault frequencies are consistent, it can be determined that the bearing is faulty, thereby achieving accurate diagnosis of the bearing fault.

[0131] Preferably, in bearing fault diagnosis, the periodic collision of rolling elements with the fault area generates characteristic frequencies, which are usually in the low frequency range and are closely related to the fault characteristics of the bearing, so the Hilbert transform is suitable for extracting bearing fault information.

[0132] Figure 7 , Figure 8 , Fig. 9 Schematic diagrams of envelope spectrum analysis of the original signal, the signal after filtering using the loudness-based filter parameter determination method, and the signal after filtering using the spectral kurtosis method.

[0133] contrast Figure 7 and Figure 8 It can be found that the envelope analysis of the filtered signal obtained by this method can effectively emphasize the fault characteristic frequency f r and f o , to achieve accurate diagnosis of bearing faults; comparison Figure 7 and Fig. 9 It can be found that when the conventional vibration processing method spectral kurtosis method is used for acoustic signal processing, it is affected by the high-frequency noise in the signal, and satisfactory analysis results cannot be obtained, and the diagnostic accuracy is greatly reduced.

[0134] In summary, the loudness-based filter parameter determination method uses the loudness index as the determining factor of the filter parameters and successfully designs the optimal finite impulse response filter, which greatly enhances the fault information in the original sound signal. After performing envelope analysis on the filtered signal, the rolling bearing fault diagnosis can be effectively performed, which effectively solves the problem of low signal-to-noise ratio and difficult feature extraction in acoustic signals.

[0135] The above is only a preferred embodiment of the present invention. It should be pointed out that for those skilled in the art, several changes and improvements can be made without departing from the overall concept of the present invention, which should also be regarded as the scope of protection of the present invention.

Claims

1. A method for extracting acoustic signal features based on loudness, characterized in that: The method includes the following steps Step S1: The signal acquisition module performs real-time acquisition and storage of the acoustic signal of the rolling bearing, collects the original acoustic signal, and transmits it to the computer terminal; Step S2: The collected original acoustic signal is subjected to the time-domain short-time sliding interception method, and the loudness-time spectrum of the acoustic signal is obtained by calculating the loudness of each short-time signal intercepted; Step S3: Identify and extract the pulse components in the loudness-time spectrum, analyze the loudness of the short-time signal characteristics corresponding to the pulse components, design the optimal finite impulse response filter based on the maximum characteristics as the parameter basis, and filter the original acoustic signal; Step S4: Using the method of Hilbert transform envelope analysis, separate the low-frequency components from the high-frequency carrier in the signal, extract the low-frequency characteristic frequencies closely related to the rolling bearing fault characteristics, so as to realize the accurate diagnosis of bearing faults.

2. The method for extracting acoustic signal features based on loudness according to claim 1, characterized in that: In the said Step S2, the loudness-time spectrum of the acoustic signal is obtained by the short-time sliding interception method. The specific process is as follows: Step S21: Create a window function w(k) to intercept the original time series x(k) (k = 1, 2, N); Step S22, by introducing the sliding parameter p and combining the window function with the time series, intercept the time series x i (k): x i (k)=x(k)·w(k-(i-1)p); where x i represents the time series obtained by the i-th intercept; The parameter p controls the speed of the sliding interception; L represents the width truncated from the original time series, satisfying 0 < p < L (p is an integer); Step S23, from x i Extract the non-zero elements and construct the slip-truncated short time series y i : y i (j)=x i ((i-1)p+j); Satisfying 0 < j < L; L represents the width truncated from the original time series, satisfying 0 < p < L (p is an integer); This short-time series can be used to obtain the loudness-time spectrum and extract the signal pulse characteristics; Step S24, short-term signal y i Apply the Zwicker loudness model to calculate the short-term signal loudness; Step S25: Record the loudness of the short-time signal as Ld(i), and convert Ld(i) into the time domain by the following formula: Among them, t i represents the time instant of the i-th sampling point; fs represents the sampling frequency, that is, the number of times the signal is sampled per unit time; Step S26, by performing sliding interception on the entire sound signal, calculating the short-time signal loudness value at each interception, and obtaining the loudness-time spectrum of the sound signal Ld = [Ld(t1), Ld(t2) ... Ld(t i )], where Ld(t i ) represents the loudness-time spectrum at a specific time point t i The loudness value of .

3. The method for extracting acoustic signal features based on loudness according to claim 1, characterized in that: In the said Step S3, the specific process of identifying and extracting the pulse components in the loudness-time spectrum is as follows: Step S31: Calculate the overall mean E(Ld) of the loudness spectrum as the basis for judging the pulse signal: Where N Ld is the number of data points contained in L; t j Represents the time of the jth sampling point in the discrete time series, 1≤j≤N Ld ; Step S32: When the impact occurs, the signal loudness increases rapidly, and the judgment variable A(i) is proposed to judge Ld(t i ) is a pulse signal point; at the same time, the threshold C is introduced h and C l , considering the rise and decay process of the pulse signal, Ld(t i ) as a pulse signal is as follows: Application formula A(i)=Ld(t i )-E(Ld) determines whether each element in the loudness-time spectrum Ld is a pulse signal point, and records all time points t in the loudness spectrum that meet the requirements j , get the corresponding short-time signal y j .

4. The method for extracting acoustic signal features based on loudness according to claim 3, characterized in that: Design the filter parameters according to the loudness characteristics of the acoustic signal. The specific process is as follows: Compute the slip-truncated short time series y using the Zwicker loudness model i The specific loudness distribution of the bark bark is extracted to find the bark range that contributes most to the loudness, i.e. the maximum specific loudness bark bark. mspl ; By analyzing the loudness of a large number of short-term sequences, we can get bark mspl The distribution of bark mspl The number of occurrences can reflect the characteristics of bearing failure, so the occurrence mspl The bark with the highest order is used as the basis for filter parameter design; Calculate the frequency range of the filter and the conversion relationship between barks: The obtained frequency range will be used as the band-pass frequency of the band-pass filter.

5. The method for extracting acoustic signal features based on loudness according to claim 1, characterized in that: Adopt Hilbert transform envelope analysis for fault diagnosis of rolling bearings. The specific process is as follows: Step S41: For a given time series s(t), the Hilbert transform is defined as the convolution of s(t) and h(t): where * is the convolution operation symbol; H{} is the Hilbert transform operator; Step S42: Further obtain the frequency-domain representation of the Hilbert transform: where S(f) is the Fourier transform of S(t), and H(f) is the Fourier transform of H(t); Step S43, using s(t) and Construct the analysis signal z(t): a(t) contains the amplitude information of the signal; Contains the phase information of the signal; Step S44: For the signal s'(t) containing the fault component in the time series s(t), construct the analytic signal z'(t) of the signal s'(t) using the above formula: f a Represents the low-pass frequency component containing fault information; Step S45: for the signal s'(t) containing the fault component, obtain Calculate the absolute value of the analytical signal z'(t) and obtain the envelope signal of the signal s'(t): Step S46: Through the envelope signal obtained above, compare the frequency component with the highest energy proportion in the envelope signal with the fault passing frequency of the monitored bearing. If the fault frequencies are the same, it can be determined that the bearing has a fault, and the accurate diagnosis of bearing faults is realized.

Images