A method and system for extracting the fault components of rotating machinery based on the misalignment superposition method

Abstract

The present invention provides a method and system for extracting the fault components of a rotating machine based on the misalignment superposition method, including the following steps: extracting the fault feature points to determine the positions of the impact fault components; correcting the optimal offset of the superimposed signal; separating the abnormal sound fault components in the rotating machine through the automatic separation method of the fault components. The method described in the present invention is simple and easy to implement. By writing the algorithm into the hardware platform through software programming, the abnormal sound signal can be processed in real time. Moreover, the result of the method described in the present invention is highly accurate. This method can retain the frequency components of the fault components, and after multiple superpositions, the extraction effect is remarkable.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and in particular, to a method and system for extracting the impact fault components of a rotating machine based on a dislocation superposition algorithm. Background Art

[0002] With the development of industrial production, rotating machines are widely used in modern industrial equipment, accounting for about 80% of mechanical equipment. They are mainly applied in engineering fields such as wind power generation, transportation machinery, mining machinery, aerospace, and marine vessels. Common rotating mechanical equipment includes electric motors, wind turbines, steam turbines, gearboxes, and engines, etc. However, rotating mechanical equipment often operates under relatively harsh environmental and working conditions, such as high-temperature, high-humidity environments, corrosive environments, high-load and variable-load working conditions, etc., which can easily cause problems such as wear, fracture, and aging of its rotating components, resulting in economic losses and some major accidents. When a rotating machine fails, due to mechanical vibration and the system's own errors, its acoustic signal will present a non-stationary periodic fault signal. However, extracting the fault components from the fault signal is often the most important and difficult problem because the fault components in the fault signal are easily interfered by internal and external factors of the system. Therefore, it is necessary to perform noise reduction extraction on the fault components of the rotating equipment, improve the signal-to-noise ratio of the signal, highlight the state information in the signal, further improve the accuracy of extracting the fault information of the mechanical equipment, and thus improve the reliability and accuracy of mechanical fault characteristics and fault diagnosis.

[0003] In view of the diversity of the working environment and the complexity of the interference signals during the operation of rotating machines, the commonly used methods for extracting the fault components in abnormal sounds are wavelet transform (WT) and empirical mode decomposition (EMD). As a widely used signal processing tool, wavelet transform has powerful multi-resolution analysis capabilities in both the time domain and the frequency domain. In the research on wavelet noise reduction theory and applications, common noise reduction methods include wavelet coefficient correlation noise reduction, wavelet modulus maximum noise reduction, and wavelet threshold noise reduction, etc. However, although wavelet noise reduction has good processing capabilities for non-stationary signals, it is very difficult to select the wavelet basis function. Selecting different wavelet basis functions will result in different signal noise reduction effects. EMD is a method proposed by Huang et al. for processing non-stationary signals. It effectively decomposes non-linear and non-stationary signals without the need to preset any basis functions, solving the problem of selecting the basis function in wavelet transform and being widely used in the field of mechanical fault diagnosis. However, it also has deficiencies such as endpoint effects, under-envelope phenomena, and mode mixing. Summary of the Invention

[0004] In view of the technical problems mentioned in the above background art, a method and system for extracting the fault components of rotating machinery based on the misalignment superposition method are provided. The present invention includes a method for noise reduction and extraction of the impact fault components of rotating machinery based on acoustic signals, comprising the following steps:

[0005] S1: Extract the fault feature points to determine the positions of the impact fault components; select similar basis functions according to the waveforms of the impact fault components to perform continuous wavelet transform on the original fault acoustic signals, and extract the periodic impact fault feature points of the fault signals by setting the wavelet hard threshold function and the rectangular window function;

[0006] S2: Correct the optimal offset of the superimposed signal; expand the signal segment containing the fault components according to the length of the impact fault components and the positions of the fault feature points, and obtain the optimal offset by means of traversal, and correct the superimposed signal;

[0007] S3: Separate the abnormal sound fault components in the rotating machinery by the automatic separation method of fault components; divide the corrected signal into two groups on average according to the parity of the serial numbers, the segments with odd serial numbers are the first group, and the segments with even serial numbers are the second group; taking the starting superposition point as the reference, perform superposition operations on the segments in each group in ascending order of the serial numbers, and finally calculate the correlation coefficients of the signals after superposition of the odd-numbered group and the even-numbered group, and determine the effect of superposition noise reduction by setting the threshold M.

[0008] Compared with the prior art, the present invention has the following advantages:

[0009] The method and system for extracting the fault components of rotating machinery based on CDSM disclosed in the present invention have the following beneficial effects: 1. Simple and easy to implement. By writing the algorithm into the hardware platform through software programming, the abnormal sound signals can be processed in real time. 2. High result accuracy. This method can retain the frequency components of the fault components, and after multiple superpositions, the extraction effect is remarkable. Description of the Drawings

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0011] Figure 1 It is the schematic diagram of the acquisition and processing platform for the abnormal sound signals of the rotating machinery of the present invention;

[0012] Figure 2 It is the schematic diagram of the "abrupt change area" and the "stable area" of the impact-type fault components of the present invention;

[0013] Figure 3 is the offset τ of the mutation region in different cycles of the fault signal of the present invention ;

[0014] Figure 4 is the signal W of the present invention s range (n s ) and schematic diagram;

[0015] Figure 5 is the block diagram of the fault component extraction system of the present invention;

[0016] Figure 6 is the original signal in the embodiment of the present invention;

[0017] Figure 7 is the extraction result in the embodiment of the present invention. Specific implementation manners

[0018] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all 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 shall fall within the protection scope of the present invention.

[0019] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that such used data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order different from those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0020] As Figure 1-7 shown, the present invention provides a noise reduction extraction method for impact fault components of a rotating machine based on acoustic signals, including the following steps:

[0021] Step S1: Extract fault feature points to determine the positions of impact fault components; select similar basis functions according to the waveforms of the impact fault components to perform continuous wavelet transform on the original fault acoustic signal, and extract the periodic impact fault feature points of the fault signal by setting a wavelet hard threshold function and a rectangular window function;

[0022] Step S2: Correct the optimal offset of the superimposed signal; obtain the signal segment containing the fault component according to the length of the impact fault component and the position expansion of the fault characteristic points, obtain the optimal offset through traversal, and correct the superimposed signal;

[0023] Step S3: Separate the abnormal sound fault components in the rotating machinery through the automatic separation method of fault components; divide the corrected signal into two groups on average according to the parity of the serial numbers, the segments with odd serial numbers are the first group, and the segments with even serial numbers are the second group; taking the starting superimposed point as the reference, perform the superimposition operation on the segments in each group in ascending order of the serial number, and finally calculate the correlation coefficient of the signals after the odd and even groups are superimposed, and determine the effect of superimposed noise reduction by setting the threshold M.

[0024] According to the characteristics of the impact fault components, they are artificially divided into an "abrupt change area" and a "stable area". Since in the "abrupt change area" stage, the impact fault source generates intense vibrations and releases a large amount of energy, the signals in this stage have characteristics such as shortness, periodicity, and large energy, contain more fault components, and are not easily buried by noise. Therefore, in order to improve the calculation efficiency, the "abrupt change area" containing more energy and information is separated as the impact fault component information. The impact fault components essentially belong to the sudden release of energy, and the time of the entire impact process is relatively short, making the impact signal respond violently in the initial stage and then gradually decay with the increase of time. According to the characteristics of the impact fault components, they are artificially divided into an "abrupt change area" and a "stable area", as Figure 2 shown, where the "abrupt change area" is in the starting stage of the impact fault component, and the "stable area" is in the middle and late stages of the impact fault component. In the "abrupt change area" stage, the impact fault source generates intense vibrations and releases a large amount of energy, the signals in this stage have characteristics such as shortness, periodicity, and large energy, contain more fault components, and are not easily buried by noise; in the "stable area" stage, the vibrations generated by the impact fault source gradually decrease, the energy gradually decreases, and the amplitude of the generated sound signal also continuously decreases, and the information components it contains are less.

[0025] In the process of extracting the fault characteristic points, select the db5 wavelet in the dbN wavelet system with a relatively high waveform similarity to the "abrupt change area" of the rotating machinery impact fault signal as the wavelet basis function for the continuous wavelet transform of the impact fault signal S(n); then perform the continuous wavelet transform on the impact fault sound signal S(n) of the rotating machinery to obtain the wavelet coefficients (W ψ S)(a, b), and perform hard thresholding on the obtained wavelet coefficients, and filter the wavelet coefficients containing noise coefficients in the wavelet time-frequency diagram through formula (1) to obtain the threshold wavelet coefficients η H (W ψS(a, b), λ), and then find the starting point of the impact fault component; the mathematical model of the hard thresholding is:

[0026]

[0027] Where η H (W ψ S(a, b), λ) represents the wavelet coefficient after being processed by the hard threshold function, denoted as η H ; W ψ S(a, b) represents the wavelet coefficient; λ represents the set hard threshold;

[0028] Then, through the window function processing of the wavelet coefficients η H of the hard thresholded wavelet time-frequency diagram and using the fault feature point extraction algorithm, extract the feature points of the fault component for each cycle.

[0029] The steps to extract the periodic fault feature points from the hard thresholded wavelet time-frequency diagram are:

[0030] Step SA: Select two diagonally opposite time-frequency points (t1, f1), (t2, f2) to form a rectangular window R, and frame out a time-frequency block from the wavelet time-frequency diagram after the hard threshold processing, and denote it as η H (a0, b0);

[0031] Where, t1 < t2, f1 < f2, a0 = t1, t1 + 1 / f s ,..., t2, b0 = f1, f1 + fs / 2a,..., f2, f s represents the sampling frequency, and a represents the scale factor of the continuous wavelet transform;

[0032] Take the position of |η H (a0, b0) and denote it as H | max Keep the frequency [f1, f1 + f / 2a,..., f2] unchanged, and let the rectangular window R be translated along the time axis by t2 each time, continuously frame out one time-frequency area after another η s (a H (a i , b i ), and denote the position of extracting |η H (a i , b i ) as H | max If the length of the signal taken is not an integer multiple of the length of the rectangular window, then the maximum value of i is:

[0033] ​

[0034] where \(i\) is the number of translation times, \(i = 0, 1, 2,\cdots\); \([\cdot]\) is the integer function; \(n\) tot is the total number of sampling points of the signal \(S(n)\); \(f\) s is the sampling frequency;

[0035] The rectangular window \(R\) includes the time-frequency points \((t_1, f_1)\) and \((t_2, f_2)\) that are diagonally opposite. Then the length \(l\) of the rectangular window \(R\) is \(l=t_2 - t_1\), and the width \(w\) is \(w = f_2 - f_1\); considering the frequency distribution of the impact fault signal and the length \(\sigma\) of the impact fault component; the time of the mutation region is:

[0036] \(t\) σ \(=\frac{\sigma}{f}\) s (3);

[0037] The rectangular window starts at \(t = 0\), so \(t_1 = 0\); according to the general distribution of the characteristic frequencies of general impact fault sound signals in the medium and low frequencies, \(f_1\) can be set between \(0\) and \(5000Hz\), and \(f_2\leq5000Hz\); if \(t_2\) is selected too large, it will cause the rectangular window to enclose multiple fault signal periods, which is not conducive to extracting the period of the fault signal. If it is selected too small, it will increase the calculation amount. Therefore, generally, the value range of \(l\) is \(2t\sigma\leq l\leq4t\sigma\), that is, corresponding to \(2\sigma\) to \(4\sigma\) sampling points;

[0038] Step SB: Let the point corresponding wavelet coefficient be For Set two conditions to extract the characteristic points;

[0039] Condition 1, if and then is called the characteristic point of the impact fault component; Condition 2, when , take The corresponding point \(C\) is the characteristic point of the impact fault component; finally, all the characteristic points of the impact fault component are assigned to \(Q\) j \((t\) jj , \(f\) jj )), so the point set of the characteristic points of each period of the impact fault component is \([Q_1, Q_2,\cdots, Q\) j , where \(j = 1, 2, 3,\cdots\);

[0040] Step SC: By using the positions of the characteristic points of the impact fault signal in each period, estimate the period of the fault signal. Then the period of the fault signal is approximately \(T_1=t\) 22 \(-t\) 11 , \(T_2=t\) 33 \(-t\) 22 , \(\cdots\), \(T\) j-1 \(=t\) jj \(-t\) (j-1)(j-1) .

[0041] As a preferred embodiment, in the present application, during the correction process of the superimposed signal offset, according to the characteristic point Q of the impact fault component j (t jj , f jj ), the number of sampling points corresponding to the characteristic point is obtained as follows:

[0042] n jj = t jj · f s (4)

[0043] Wherein, n jj represents the sampling point corresponding to the fault characteristic point; t jj represents the time of the characteristic point of the impact fault component, and f s represents the sampling frequency;

[0044] According to the position of the characteristic point of the impact fault component, (σ - 1) sampling points are extended forward and backward respectively from the n jj th sampling point, and the extended signal is denoted as Then the signal contains the impact fault component; where n j ∈[n jj - σ + 1, n jj + σ - 1], and σ is the length of the mutation region, that is, the length of the impact fault component;

[0045] The empirical formula for the length σ of the mutation region is:

[0046]

[0047] Wherein, f s represents the sampling frequency of the acoustic signal;

[0048] Due to the interference of background noise, the position of the fault component characteristic point in the mutation region of each period has a deviation, resulting in different degrees of position offset τ in the mutation region W in different signal periods. Therefore, it cannot be used j directly for superimposed noise reduction

[0049] To solve the superimposition problem caused by the offset of the mutation region, a method of traversing and calculating the optimal offset τopt relative to the reference signal is proposed, and further find the relatively optimal starting superimposition point for superimposed noise reduction. The specific method is as follows:

[0050] (1) Take any section containing the fault component as the reference signal, denoted as W s range (n(n s)。

[0051] (2) According to the range of the offset τ being 0 ≤ τ ≤ σ - 1, make Expand (σ - 1) sampling points forward and backward respectively to obtain an extended signal segment

[0052] (3) From Continuously intercept (2σ - 1) sampling points starting from the first sampling point as the signal segment Set τj As the offset on , The offset signal is recorded as τ j The subset of is represented as L, and set L to {0:2σ - 2}. Search for the optimal starting stacking point, that is, search for The optimal offset of the offset τj in where the stacking length is (2σ - 1) sampling points.

[0053] (4) Calculate W s range (n s ) and The correlation coefficient of And traverse all values of τj in L. Take The offset corresponding to the maximum value is the optimal offset relative to the reference signal The equation of this method is as follows:

[0054]

[0055] Among them, Is The optimal offset relative to the reference signal; argmax[] is the set of maximum independent variable points; W s range (n s ) is the reference signal segment; Is The signal segment after offsetting τj; Is W s range (n s ) and The Pearson correlation of.

[0056] Preferably, the superimposed signal includes the following steps:

[0057] Step S11: Select 2K consecutive Signals, and according to the parity of the serial number j Divide into odd and even groups, and perform different numbers of superposition operations on each group from low to high according to the serial number. The superposition formula is as follows:

[0058]

[0059] in, is the signal after the odd-numbered groups are superimposed; is the signal after the even number of groups are superimposed; K is the number of superpositions; is an odd-numbered signal containing fault components; is the even-array signal containing fault components; ρopt(K) is the Pearson correlation between the signal after the odd-array superposition and the signal after the even-array superposition.

[0060] Step S12: Determine the number of superpositions by setting a threshold value M. When ρopt(K)≥M, stop the iteration and output the superposition result. The output superposition signal is the rotating machinery impact fault component. In order to achieve a good noise reduction effect on the fault component and save time, M should be set between 0.75 and 0.85, and here M=0.80.

[0061] As a preferred embodiment, the present invention also includes a rotating machinery fault component extraction system based on CDSM, comprising:

[0062] A signal acquisition unit for receiving abnormal sound signals of rotating machinery;

[0063] A computing unit that uses CDSM to process abnormal noise signals of rotating machinery.

[0064] Embodiment 1:

[0065] In the process of building a signal acquisition system, including a rotating machine, an acoustic sensor, an encoder, an acquisition card, etc., the principle of the abnormal sound signal acquisition and processing platform is as follows: Figure 1 As shown:

[0066] The acoustic sensor is placed directly above the engine to collect abnormal sound signals. The acquisition card connects the acoustic signal to the computer for processing.

[0067] In the process of processing the collected signal by using CDSM in the present invention, the process of processing the collected signal by using CDSM can be described as follows:

[0068] In the "sudden change zone" stage, the impact fault source generates violent vibrations and releases a large amount of energy. The signal in this stage is short-lived, periodic, and has high energy. It contains more fault components and is not easily buried by noise. In order to improve the calculation efficiency, the "sudden change zone" containing more energy and information is separated as the impact fault component information. Figure 2The waveform in the catastrophic area selects the db5 wavelet as the wavelet basis function for the continuous wavelet transform of the impact fault signal. Then, the continuous wavelet transform is performed on the impact fault sound signal S(n) of the rotating machinery to obtain the wavelet coefficients (W ψ S)(a, b), and the obtained wavelet coefficients are processed by hard thresholding. The mathematical model of the hard threshold function is:

[0069]

[0070] where η H (W ψ S(a, b), λ) is the wavelet coefficient after being processed by the hard threshold function, denoted as η H ; W ψ S(a, b) is the wavelet coefficient; λ is the set hard threshold.

[0071] By performing a window function processing on the wavelet coefficients η H of the hard thresholded wavelet time-frequency diagram and using a fault feature point extraction algorithm to extract the feature points of the fault components in each cycle. The steps to extract the periodic fault feature points from this hard thresholded wavelet time-frequency diagram are as follows:

[0072] (1) Select two diagonally opposite time-frequency points (t1, f1), (t2, f2) to form a rectangular window R, and frame out a time-frequency block from the hard thresholded wavelet time-frequency diagram, and denote it as η H (a0, b0). Where t1 < t2, f1 < f2, a0 = t1, t1 + 1 / fs,..., t2, b0 = f1, f1 + fs / 2a,..., f2, fs is the sampling frequency, and a is the scale factor of the continuous wavelet transform. Take the position of |η H | H in the time-frequency block η max , denoted as Keep the frequency [f1, f1 + fs / 2a,..., f2] unchanged, and let the rectangular window R be translated along the time axis by t2 each time, continuously framing out one time-frequency area after another η H (a i , b i ), and extract the position of |η H (a i , b i ) and denote it as H | max as If the length of the signal taken is not an integer multiple of the length of the rectangular window, the maximum value of i is:

[0073]

[0074] where \(i\) is the number of translation times, \(i = 0, 1, 2,\cdots\); \([\cdot]\) is the integer function; \(n\) tot is the total number of sampling points of the signal \(S(n)\); \(f\) s is the sampling frequency.

[0075] The rectangular window \(R\) is composed of two obliquely opposite time-frequency points \((t_1, f_1)\) and \((t_2, f_2)\). Then the length \(l\) of the rectangular window \(R\) is \(l = t_2 - t_1\), and the width \(w\) is \(w = f_2 - f_1\). Considering the frequency distribution of the impact fault signal and the length \(\sigma\) of the impact fault component. The time of the mutation area is:

[0076] \(t\) σ \(=\frac{\sigma}{f}\) s

[0077] The rectangular window starts from the moment \(t = 0\), so \(t_1 = 0\). According to the general distribution of the characteristic frequencies of general impact fault sound signals in the medium and low frequencies, \(f_1\) can be set between \(0\) and \(5000Hz\), and \(f_2\leq5000Hz\). If \(t_2\) is selected too large, it will cause the rectangular window to enclose multiple fault signal periods, which is not conducive to extracting the period of the fault signal. If it is selected too small, it will increase the calculation amount. Therefore, generally, the value range of \(l\) is \(2t\sigma\leq l\leq4t\sigma\), that is, corresponding to \(2\sigma\) to \(4\sigma\) sampling points.

[0078] (2) Let the point corresponding wavelet coefficient be For Set two conditions for extracting the characteristic points. Condition 1, if and then is called the characteristic point of the impact fault component. Condition 2, if When, take The corresponding point \(C\) is the characteristic point of the impact fault component. Finally, all the characteristic points of the impact fault component are assigned to \(Q_j(t_{jj}, f_{jj})\). Therefore, the point set of the characteristic points of each period of the impact fault component is \([Q_1, Q_2,\cdots, Q\) j , where \(j = 1, 2, 3,\cdots\).

[0079] (3) By using the characteristic point positions of the impact fault signals in each period, estimate the period of the fault signal. Then the period of the fault signal is approximately \(T_1 = t\) 22 -t 11 , \(T_2 = t\) 33 -t 22 , \(\cdots\), \(T\) j-1 \(= t\) jj -t (j-1)(j-1) .

[0080] According to the characteristic points \(Q_j(t_{jj}, f_{jj})\) of the impact fault component, the sampling points corresponding to the characteristic points are:

[0081] n jj = t jj ·f s

[0082] Among them, njj is the sampling point corresponding to the fault feature point; tjj is the time of the feature point of the impact fault component, and fs is the sampling frequency.

[0083] Since the mother wavelet is selected according to the waveform in the mutation region during continuous wavelet transform, and the feature points of the impact fault component are determined based on the mutation region of the impact fault signal, therefore, the feature points of the impact fault component are at a certain position in the mutation region of the fault component. According to the position of the feature points of the impact fault component, expand (σ - 1) sampling points forward and backward from the njj-th sampling point respectively, and the expanded signal is denoted as Then the signal contains the impact fault component. Where n j ∈[n jj-σ + 1, n jj+σ-1 , σ is the length of the mutation region, that is, the length of the impact fault component.

[0084] The length σ of the mutation region is generally the length of the high-amplitude region of the artificially selected fault component. According to the literature review, the general empirical formula for σ is:

[0085]

[0086] Among them, fs is the sampling frequency of the acoustic signal.

[0087] Due to the interference of background noise, the positions of the feature points of the fault component in each period are different in the mutation region, resulting in different degrees of position offset τ in the mutation region Wj of different signal periods. As shown in Figure 3 Figure 3 , the offset between the mutation region Wj-1 in of the (j - 1)-th period and the mutation region Wj in of the j-th period is τj, and the offset between the mutation region Wj+1 in of the (j + 1)-th period and the mutation region Wj in of the j-th period is τj+1. Further calculated based on the position of the fault feature point and the length of the fault component (the length of the mutation region), the range of the offset τ is 0 ≤ τ ≤ σ - 1. If the position of the mutation region is not adjusted and the superposition calculation is directly performed, the offset τ will affect the misaligned superposition effect, resulting in inaccurate superposition results and even damaging the impact fault component.

[0088] To solve the superposition problem caused by the offset of the cusp region, this paper proposes a method to calculate the optimal offset τopt relative to the reference signal through traversal, and further find the relatively optimal starting superposition point for superposition and noise reduction. The specific method is as follows:

[0089] (1) Take any segment containing fault components as the reference signal, denoted as W s range (n s ).

[0090] (2) According to the range of the offset τ being 0 ≤ τ ≤ σ - 1, make extend (σ - 1) sampling points forward and backward respectively to obtain the extended signal segment

[0091] (3) Continuously intercept (2σ - 1) sampling points from the first sampling point in and denote it as the signal segment Set τj as the offset on , The offset signal is recorded as The subset of τj is represented as L, and set L as {0:2σ - 2}. Finding the optimal starting superposition point is to find the optimal offset of the offset τj in where the superposition length is (2σ - 1) sampling points.

[0092] (4) Calculate the correlation coefficient between W s range (n s ) and , and traverse all values of τj in L. As shown in Figure 4 , where W s range (n s ) is the selected reference signal, the signal in the red window is i.e., when τj = 0 The signal in the blue window is when τj = d Take the offset corresponding to the maximum value as the optimal offset relative to the reference signal The equation of this method is as follows:

[0093]

[0094] Among them, is the optimal offset relative to the reference signal; argmax[] is the set of maximum independent variable points; Ws range (n s ) is the reference signal segment; for The signal segment after the offset τj; It is W s range (n s )and Pearson correlation.

[0095] —Signal automatic superposition method:

[0096] Select 2K consecutive signal, according to the parity of sequence number j Divide into odd and even groups, and perform different numbers of superposition operations on each group from low to high according to the serial number. The superposition formula is as follows:

[0097]

[0098] in, is the signal after the odd-numbered groups are superimposed; is the signal after the even number of groups are superimposed; K is the number of superpositions; is an odd-numbered signal containing fault components; is the even-array signal containing fault components; ρopt(K) is the Pearson correlation between the signal after the odd-array superposition and the signal after the even-array superposition.

[0099] Finally, the number of superpositions is determined by setting a threshold value M. When ρopt(K)≥M, the iteration is stopped and the superposition result is output. The output superposition signal is the rotating machinery impact fault component. However, according to the literature, when the number of superpositions reaches a certain level, the correlation coefficient ρ generally increases to around 0.9 and tends to be stable, and there is no obvious increase trend. In order to achieve a good noise reduction effect for the fault component and save time, M should be set between 0.75 and 0.85, and here M=0.80 is taken.

[0100] By using CDSM and the encoder to process the abnormal sound signal, the fault component can be finally separated.

[0101] The written programs are imported into the single chip microcomputer, and then connected to the signal acquisition system. The whole engine fault component extraction system block diagram is as follows Figure 5 shown.

[0102] In this embodiment, a typical rotating machinery gearbox is taken as an example, a broken tooth fault of the intermediate shaft of the gearbox is set, and CDSM is used to reduce the noise of the abnormal sound to extract the components of the impact fault.

[0103] Take the collected abnormal sound signal as the test signal, as followsFigure 6 as shown

[0104] Through CDSM analysis, the optimal offset is calculated The results are shown in Table 1

[0105] Table 1 Intermediate shaft Optimal offset

[0106]

[0107] Select 2K consecutive signals and set the threshold M = 0.8. According to the parity of the serial number j, they are divided into odd and even groups, and different numbers of superposition operations are performed on each group in ascending order of the serial number. After 5 times of superposition by the CDSM algorithm, the impact fault components of the gear tooth breakage are obtained. At this time, the correlation between the odd and even groups is 0.9531, and it can be considered that the separation result is effective. The results are as Figure 7 shown

[0108] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. In the above embodiments of the present invention, the descriptions of each embodiment have their own emphases. For the parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments. In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways

[0109] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention

Claims

1. A noise reduction extraction method for the impact fault components of rotating machinery based on acoustic signals, characterized in that, It includes the following steps: S1: Extract the fault feature points to determine the positions of the impact fault components; select similar basis functions according to the waveforms of the impact fault components to perform continuous wavelet transform on the original fault sound signal, and extract the periodic impact fault feature points of the fault signal by setting the wavelet hard threshold function and the rectangular window function; S2: Correct the optimal offset of the superimposed signal; expand the signal segment containing the fault components according to the length of the impact fault components and the positions of the fault feature points, obtain the optimal offset by traversing, and correct the superimposed signal; S3: Separate the abnormal sound fault components in the rotating machinery by the automatic separation method of fault components; evenly divide the corrected signal into two groups according to the parity of the serial numbers, the segments with odd serial numbers are the first group, and the segments with even serial numbers are the second group; with the starting superimposed point as the reference, perform superimposition operations on the segments in each group in ascending order of the serial numbers, and finally calculate the correlation coefficients of the signals after the odd and even groups are superimposed, and determine the effect of superimposition noise reduction by setting the threshold M.

2. A noise reduction extraction method for impact fault components of a rotating machinery based on sound signals according to claim 1, wherein In the process of extracting the fault feature points, the db5 wavelet in the dbN wavelet system with a relatively high similarity to the waveform in the mutation area of the impact fault signal of the rotating machinery is selected as the wavelet basis function for the continuous wavelet transform of the impact fault signal S(n); then, the continuous wavelet transform is performed on the impact fault sound signal S(n) of the rotating machinery to obtain the wavelet coefficients (W ψ S)(a, b), and the obtained wavelet coefficients are subjected to hard thresholding processing. The wavelet coefficients containing noise coefficients in the wavelet time-frequency diagram are filtered through formula (1) to obtain the threshold wavelet coefficients η H (W ψ S(a, b), λ), and then the starting point of the impact fault component is found; the mathematical model of the hard thresholding is as follows: Among them, η H (W ψ S(a, b), λ) represents the wavelet coefficient after being processed by the hard threshold function, denoted as η H ; W ψ S(a, b) represents the wavelet coefficient; λ represents the set hard threshold; Then, by performing window function processing on the wavelet coefficients η of the hard-thresholded wavelet time-frequency diagram and using a fault feature point extraction algorithm, the feature points of the fault components in each cycle are extracted. H Window function processing and the use of a fault feature point extraction algorithm are carried out to extract the feature points of the fault components in each cycle.

3. A noise reduction extraction method for impact fault components of a rotating machinery based on sound signals according to claim 2, wherein: The steps of extracting periodic fault feature points from the hard-thresholded wavelet time-frequency diagram are: SA: Select two diagonally opposite time-frequency points \((t1, f1)\) and \((t2, f2)\) to form a rectangular window \(R\), and select a time-frequency block from the wavelet time-frequency diagram after the hard threshold processing, and denote it as \(\eta\) H (a0, b0); where t1 < t2, f1 < f2, a0 = t1, t1 + 1 / f s ,..., t2, b0 = f1, f1 + fs / 2a,..., f2, f s denotes the sampling frequency, and a denotes the scale factor of the continuous wavelet transform; Take the time-frequency block η H (a0,b0) where |η H | max The position of is denoted as Keep the frequency [f1,f1 + f s / 2a,...,f2] unchanged, and let the rectangular window R be translated along the time axis from the starting position of the signal every t2. The translation distance is the same as the length of the rectangular window itself. Each time the rectangular window R is translated, a time-frequency region will be framed out. Therefore, one time-frequency region η H (a i ,b i ) will be framed out continuously until the selected signal is completely framed; extract |η H (a i ,b i ) from the time-frequency block η H | max The position of is denoted as If the length of the selected signal is not an integer multiple of the length of the rectangular window, then the maximum value of i is: where i is the number of translation times, i = 0, 1, 2...; [] is the rounding function; n tot is the total number of sampling points of the signal S(n); f s is the sampling frequency; The rectangular window R includes: obliquely opposite time-frequency points (t1, f1), (t2, f2), then the length l of the rectangular window R = t2 - t1, and the width w = f2 - f1; considering the frequency distribution of the impact fault signal and the length σ of the impact fault components; the time of the abrupt change region is: t σ = σ / f s (3); The rectangular window starts at time t = 0, so t1 = 0; according to the characteristic frequency of the general impact fault sound signal, which is generally distributed in the middle and low frequencies, f1 can be set between 0 and 5000 Hz, and f2 ≤ 5000 Hz; the length of the rectangular window, l, ranges from 2t σ ≤ l ≤ 4t σ , that is, corresponding to 2σ to 4σ sampling points; SB: Set point The corresponding wavelet coefficient is For Set two conditions for feature point extraction; Condition 1, if and then it is called as the characteristic point of the impact fault component; Condition 2, if at this time, take the corresponding point C as the characteristic point of the impact fault component; Finally, all the characteristic points of the impact fault component are assigned to Q j (t jj , f jj ), so the point set of the characteristic points of each cycle of the impact fault component is [Q1, Q2, ···, Q j , where j = 1, 2, 3, ···; SC: By using the characteristic point positions of the shock fault signals in each cycle to estimate the period of the fault signals, the period of the fault signals is approximately T1 = t 22 -t 11 , T2 = t 33 -t 22 ,..., T j-1 = t jj -t (j-1)(j-1) .

4. A noise reduction extraction method for impact fault components of a rotating machinery based on sound signals according to claim 3, wherein: During the correction process of the superimposed signal offset, according to the characteristic point Q of the impact fault component j (t jj ,f jj ), the number of sampling points corresponding to the characteristic point is obtained as follows: n jj = t jj · f s (4) Among them, n jj represents the sampling point corresponding to the fault feature point; t jj represents the time of the feature point of the impact fault component, and f s represents the sampling frequency; According to the positions of the characteristic points of the impact fault components, expand (σ - 1) sampling points forward and backward respectively from the n jj -th sampling point. The expanded signal is denoted as Then the signal contains the impact fault components; where n j ∈[n jj - σ + 1, n jj + σ - 1], and σ is the length of the mutation region, that is, the length of the impact fault components; The empirical formula for the length σ of the abrupt change region is: where f s represents the sampling frequency of the acoustic signal.

5. A method for noise reduction and extraction of impact fault components of rotating machinery based on acoustic signals according to claim 4, characterized in that: The steps of the superimposed signal include: S11: Select 2K consecutive signals. Divide them into odd-numbered groups and even-numbered groups according to the parity of the sequence number j, and perform superposition operations with different numbers of times on each group in ascending order of the sequence number. The superposition formula is as follows: ​ Among them, is the signal after odd - group superposition; is the signal after even - group superposition; K is the number of superposition times; is the odd - group signal containing fault components; is the even - group signal containing fault components; ρ opt (K) is the Pearson correlation between the signal after odd - group superposition and the signal after even - group superposition; S12: Determine the number of superimpositions by setting the threshold M. When ρopt(K) ≥ M, stop the iteration and output the superimposed result. Then the output superimposed signal is the impact fault component of the rotating machinery.

6. A rotating machinery fault component extraction system based on CDSM, which applies the method described in any one of claims 1-5, characterized in that It includes: A signal acquisition unit for receiving the abnormal sound signal of the rotating machinery; A calculation unit for processing the abnormal sound signal of the rotating machinery using CDSM.

Images