Roller fault sound source localization method based on TDOA time delay estimation algorithm

Abstract

The present invention discloses a method for locating the sound source of a roller fault based on a TDOA time delay estimation algorithm. The method comprises the following steps: installing multiple sound sensors at set intervals next to a belt conveyor to collect the sound signals of the rollers of the belt conveyor; extracting effective sound information of the roller fault using VMD and envelope spectrum kurtosis methods; performing generalized cross-correlation delay estimation on selected IMF components of the extracted effective sound information of the roller fault using a TDOA time delay estimation algorithm, and determining the time delay by the time difference corresponding to the maximum cross-correlation coefficient in the time delay domain; and deriving the location of the fault sound source by combining the propagation speed of sound in air, the spatial geometric relationship between the location of the fault sound source, and the acoustic sensor. The present invention utilizes the above-mentioned method for locating the sound source of a roller fault based on a TDOA time delay estimation algorithm, which can extract periodic fault information of the roller fault from the coupled interference components of the transmission path, thereby accurately locating the sound source of the roller fault.

Description

Technical Field

[0001] The present invention relates to the technical field of belt conveyor roller fault location technology, and in particular to a roller fault sound source location method based on a TDOA time delay estimation algorithm. Background Art

[0002] Belt conveyors, the primary means of transport in today's coal-fired power plants, are subject to harsh conditions year-round, including high coal dust concentrations and humid air. These conditions can easily lead to problems such as roller jams, belt tears, and frame deformation, which can severely impact the operation of the conveying system. Therefore, it's essential to monitor and diagnose the operating conditions of belt conveyor rollers.

[0003] At present, belt conveyor roller fault monitoring mainly focuses on two methods based on roller temperature and sound, such as:

[0004] In "Research on Belt Conveyor Roller Fault Identification Algorithm Based on Fiber Optic Temperature Measurement Technology", Guo Qinghua disclosed the research on roller shaft temperature detection using a distributed fiber optic temperature measurement system installed on a belt conveyor. He proposed a roller fault identification algorithm based on the historical temperature rise characteristic curve of the damaged roller and a roller fault identification algorithm based on the two-dimensional distribution of the roller's daily temperature rise and time, realizing roller fault detection and positioning based on temperature rise characteristics.

[0005] In "Research and Application of Belt Conveyor Roller Fault Noise Monitoring and Fault Point Location Technology", Hua Shoutong, Zhang Wenzhong, Chen Yi, etc. disclosed that the temperature signal under the condition of roller fault noise can be obtained by inspection robots to realize the monitoring and location of roller fault noise.

[0006] In "Roller Fault Diagnosis Method Based on Sound Signals", Hao Hongtao et al. proposed a roller fault diagnosis method based on sound signals, which integrated multiple sound signal processing methods to determine whether the roller is faulty. Then, the empirical mode value decomposition method was used to estimate the area where the belt conveyor roller failure occurred. The combined method of wavelet packet decomposition and reconstruction and Hilbert envelope analysis was used to further extract the roller bearing fault frequency, thereby achieving accurate positioning of the faulty bearing.

[0007] Based on the "Roller Fault Diagnosis Method Based on Sound Signals" published by Su Yaorui, the "Research on Non-Contact Fault Identification Method of Remote Belt Conveyor Rollers", the "Research on Fault Detection Method of Remote Belt Conveyor Rollers" published by Ni Fanfan, and the "Remote Belt Conveyor Roller Fault Inspection Method" published by Hao Hongtao, Ni Fanfan, Chen Liang, etc., a set of roller fault online monitoring systems were developed based on LabVIEW.

[0008] The "Research on Intelligent Monitoring System for Mining Belt Conveyors" published by Mao Qinghua, Mao Jingen, Ma Hongwei, and others, and the "Fault Diagnosis of Rollers Based on Spectral Clustering Analysis" published by Song Tianxiang, Yang Mingjin, Yang Linshun, and others, further utilize intelligent algorithms such as clustering and artificial neural networks for fault detection based on roller sound signal processing. Specifically, the "Research on Intelligent Monitoring System for Mining Belt Conveyors" collects sound signals from rollers under different damage conditions, uses wavelet denoising to extract feature vectors, and then uses a neural network (BP-RBF) to perform fault detection on the extracted feature vectors. The "Fault Diagnosis of Rollers Based on Spectral Clustering Analysis" uses spectral clustering analysis to achieve roller fault diagnosis.

[0009] In summary, the aforementioned methods are all based on patrol inspections, using acoustic sensors mounted on patrol equipment on the track. However, due to the long transport distances and large number of rollers on belt conveyors, a long patrol track must be constructed, resulting in high patrol inspection costs. Summary of the Invention

[0010] To solve the above problems, the present invention provides a method for locating the sound source of roller faults based on the TDOA time delay estimation algorithm. By installing sound sensors at a certain distance on the belt conveyor, and then using a method based on VMD combined with envelope spectrum kurtosis to extract roller fault characteristics from the sound signal, the time delay of the component containing the roller fault characteristic information is estimated, and the faulty roller can be accurately located according to the result of the time delay estimation and the spatial geometric relationship between the sound source and the acoustic sensor.

[0011] To achieve the above object, the present invention provides a method for locating the sound source of a roller fault based on a TDOA time delay estimation algorithm, comprising the following steps:

[0012] S1. Install multiple sound sensors at set intervals next to the belt conveyor to collect sound signals from the rollers of the belt conveyor;

[0013] S2. VMD and envelope spectrum kurtosis method are used to extract effective sound information of roller fault;

[0014] S3. Using the TDOA delay estimation algorithm, the IMF components selected from the extracted effective sound information of the roller fault are subjected to generalized cross-correlation delay estimation, and the delay is determined by the time difference corresponding to the maximum value of the cross-correlation coefficient in the delay domain;

[0015] S4. The location of the fault sound source is obtained by combining the propagation speed of sound in the air medium, the spatial geometric relationship between the location of the fault sound source and the acoustic sensor.

[0016] Preferably, step S2 specifically includes the following steps:

[0017] S21. VMD is used to solve the misclassification and aliasing problems caused by frequency band segmentation, and to obtain effective sound information of roller faults.

[0018] S22. The envelope spectrum kurtosis method is used to eliminate the interference introduced by the accidental vibration impact component in the effective sound information of the roller fault.

[0019] Preferably, step S21 specifically includes the following steps:

[0020] S211. Analyze the modal component u using Hilbert transform k (t), and obtain each IMF component {u k}、Analytical signal and corresponding unilateral spectrum, the calculation formula is as follows:

[0021]

[0022] Where δ(t) is the impulse function; j is the complex imaginary part; t is the time series;

[0023] S212, set the center frequency of each modal component to w k , the signal spectrum is shifted to baseband using the following formula:

[0024]

[0025] Where, It is an exponential signal;

[0026] S213. Construct a constrained variational model and calculate the bandwidth of each component signal:

[0027]

[0028] In the formula, {u k} are the IMF components, {w k} is the center frequency of each IMF component, k is the number of modes obtained by decomposition, and f(t) is the sound signal collected by the original acoustic sensor; is a partial differential.

[0029] Preferably, step S22 specifically includes the following steps:

[0030] S221, let the time domain signal be x(t), use VMD to decompose x(t) into several components x i (t), the analytical signal of the modal component is:

[0031]

[0032] Where, Represents x i(t), where i = 1, 2, 3, ..., so the envelope signal of the IMF component is:

[0033]

[0034] The corresponding envelope signal spectrum is:

[0035]

[0036] Where Env(n) is the envelope signal; m is the frequency domain signal sequence, N is the number of discrete points, n is the time domain signal sequence, and m = 0, 1, ..., N-1.

[0037] S222, let the roller bearing failure frequency be f r When the roller bearing fails, the envelope spectrum analysis of the components containing the fault components is performed to observe whether the fault frequency or its multiple frequency exists. Therefore, when calculating its spectrum kurtosis, only [f r ,4f r ] within the spectrum line NUM u , signal envelope spectrum kurtosis k Es for:

[0038]

[0039] Where NUM1 is the spectrum sequence; V 2 (m) represents the square of the envelope spectrum sequence; V 4 (m) represents the fourth power of the envelope spectrum sequence.

[0040] Preferably, step S3 specifically includes the following steps:

[0041] S31. Analyze and compare the time domain similarity between two sound signals using a basic cross-correlation algorithm;

[0042] Assuming that two acoustic sensors are placed at the head and tail ends of the belt conveyor, the model of the sound signal collected by them is:

[0043] x1(t)=S(t-τ1)+n1(t) (8)

[0044] x2(t)=S(t-τ2)+n2(t) (9)

[0045] Where: S(t-τ1) represents the fault source information received by the acoustic sensor at the head end of the belt conveyor; S(t-τ2) represents the fault source information received by the acoustic sensor at the tail end of the belt conveyor; n1(t) represents the Gaussian white noise of the acoustic sensor at the head end of the belt conveyor; n2(t) represents the Gaussian white noise of the acoustic sensor at the tail end of the belt conveyor; and S(t-τ1), S(t-τ2), n1(t) and n2(t) are independent of each other, τ1 and τ2 are the propagation times of the fault sound wave transmitted to the two acoustic sensors;

[0046] The cross-correlation function of x1(t) and x2(t) is expressed as:

[0047] R 12 (τ)=E[x1(t)x2(t-τ)] (10)

[0048] Where R 12 (τ) represents the cross-correlation function;

[0049] Substituting equations (8) and (9) into equation (10), we obtain:

[0050]

[0051] Where s(t-τ1-τ) and s(t-τ2-τ) represent the fault sound source information received by the acoustic sensors at the head and tail ends, respectively, and n2(t-τ) represents the noise;

[0052] Therefore, from formula (11):

[0053] R 12 (τ)=E[s(t-τ1)s(t-τ1-τ)]=R s (τ-(τ1-τ2)) (12)

[0054] Where R s (τ-(τ1-τ2)) represents the correlation function;

[0055] It shows that when τ-(τ1-τ2)=0, R 12 (τ) The τ corresponding to the peak value is the time delay between the fault sound wave propagating to the two acoustic sensors;

[0056] S32. Obtain a generalized cross-correlation function between the sound signals collected by the two acoustic sensors using a generalized cross-correlation algorithm:

[0057] S321, filtering the sound signals collected by the two acoustic sensors and then framing them using a finite-length window;

[0058] S322, performing Fourier transform on the sound signal processed in step S31 to obtain the cross-power spectrum of the two signals in this frame;

[0059] S323. After weighting the frame in the frequency domain using a weighting function, perform inverse Fourier transform to obtain the cross-correlation function of the frame, extract the time τ corresponding to the peak value in the cross-correlation function, and obtain the time delay τ between the sound wave propagating to the two acoustic sensors. 12 ;

[0060] S324. According to the relationship between the cross-correlation function and the cross-power spectrum, we obtain:

[0061]

[0062] Where R 12 (τ) represents the cross-correlation function; G 12 (w) is the cross-power spectrum of the signals x1(t) and x2(t) received by the two acoustic sensors;

[0063] S325. Perform weighted processing on equation (13) and obtain the cross-correlation function through inverse Fourier transform:

[0064]

[0065] Where, ψ 12 (w) represents the generalized cross-correlation weighting function.

[0066] The present invention has the following beneficial effects:

[0067] (1) VMD (Variational Mode Decomposition, VMD) and envelope spectrum kurtosis analysis are used to extract fault information components, solving the problem that the sound signals collected by sensors often contain a large amount of interference components coupled by transmission paths due to the complex operating environment of belt conveyors;

[0068] (2) By first performing generalized cross-correlation delay estimation on the selected IMF components of the sound signal, and then making a delay estimation based on the time difference corresponding to the maximum value of the cross-correlation coefficient in the delay domain, the position of the fault sound source is obtained by combining the propagation speed of sound in the air medium and the spatial geometric relationship between the sound source position and the acoustic sensor. The periodic fault information of the roller fault can be extracted from the coupled interference components of the transmission path, and the sound source of the roller fault can be accurately located.

[0069] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1Flowchart of the method for locating the sound source of a roller fault based on the TDOA time delay estimation algorithm of the present invention;

[0071] Figure 2 A schematic diagram of the interference of the transmission path on the sound source signal of the roller fault sound source localization method based on the TDOA time delay estimation algorithm of the present invention;

[0072] Figure 3 The signal collected by the acoustic sensor X0 arranged at the head end of the belt conveyor as described in the verification example of the present invention;

[0073] Figure 4 The signal collected by the acoustic sensor X1 arranged in the middle of the belt conveyor as described in the verification example of the present invention;

[0074] Figure 5 The signal collected by the acoustic sensor X2 arranged at the tail end of the belt conveyor as described in the verification example of the present invention;

[0075] Figure 6 This is a time domain diagram of the IMF2 component of the acoustic sensor X0 arranged at the head end of the belt conveyor described in the verification example of the present invention;

[0076] Figure 7 This is the envelope spectrum of the acoustic sensor X0 arranged at the head end of the belt conveyor as described in the verification example of the present invention;

[0077] Figure 8 This is a time domain diagram of the IMF2 component of the acoustic sensor X1 arranged in the middle of the belt conveyor as described in the verification example of the present invention;

[0078] Figure 9 This is the envelope spectrum of the acoustic sensor X1 arranged in the middle of the belt conveyor as described in the verification example of the present invention;

[0079] Figure 10 This is a time domain diagram of the IMF2 component of the acoustic sensor X2 arranged at the tail end of the belt conveyor described in the verification example of the present invention;

[0080] Figure 11 This is the envelope spectrum of the acoustic sensor X2 arranged at the tail end of the belt conveyor as described in the verification example of the present invention;

[0081] Figure 12 This is a diagram of the delay estimation result described in the verification example of the present invention;

[0082] Figure 13 This is a diagram of the delay estimation result according to an embodiment of the present invention. DETAILED DESCRIPTION

[0083] In order to make the purpose, technical solutions and advantages disclosed in the embodiments of the present invention clearer, the embodiments of the present invention are further described in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention and are not intended to limit the embodiments of the present invention. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, where the same or similar numbers throughout represent the same or similar elements or elements with the same or similar functions.

[0084] It should be noted that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or are inherent to these processes, methods, products or devices.

[0085] Like reference numerals and letters denote like items in the following drawings, and thus, once an item is defined in one drawing, it does not require further definition or explanation in subsequent drawings.

[0086] In the description of the present invention, it should be noted that the terms "upper", "lower", "inside", "outside", etc. indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, or are the orientations or positional relationships in which the inventive product is usually placed when in use. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, they should not be understood as limiting the present invention.

[0087] In the description of the present invention, it should also be noted that, unless otherwise expressly specified or limited, the terms "disposed," "installed," and "connected" should be understood broadly. For example, they may refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on the specific circumstances.

[0088] First of all, it is known from common knowledge in this field that when a roller bearing fails, the impact vibration caused by the periodic fault is transmitted in the form of sound waves. Therefore, two acoustic sensors are arranged at both ends of the belt conveyor (or at a certain distance), and assuming that there is no energy attenuation and noise interference components during the sound transmission process, the signals collected by the two sensors should be periodic impact signals of the same frequency with only phase differences. By performing generalized cross-correlation analysis on the signal, the time difference between the fault sound source being transmitted to the two acoustic sensors can be estimated in the time delay domain. The distance between the fault sound source position and the two sensors is calculated based on this time difference combined with the speed of sound propagation in the air medium.

[0089] However, the fault sound source signal will be affected by the transmission path coupling interference component during the transmission process. The transmission path interference mainly includes the non-periodic transient impact component caused by the surrounding noise and the abnormal sound of the machine. As a result, the signal finally collected by the acoustic sensor is actually as follows Figure 2 The image shows a superposition of periodic fault impulse components, non-periodic transient impulse components, and noise components. Directly estimating the time delay of the acquired signal would not effectively determine the phase difference between the two acoustic sensor signal waveforms, making it impossible to locate the fault source. Therefore, appropriate signal processing methods are required to extract the periodic fault impulse information of the fault source from the directly acquired sound signal.

[0090] like Figure 1 As shown, based on the above analysis, the present invention is designed, that is, a method for locating the sound source of a roller fault based on a TDOA time delay estimation algorithm, which includes the following steps:

[0091] S1. Install multiple sound sensors at set intervals next to the belt conveyor to collect sound signals from the rollers of the belt conveyor;

[0092] S2. VMD and envelope spectrum kurtosis method are used to extract effective sound information of roller fault;

[0093] Variational Mode Decomposition (VMD) solves the optimal modal components by constraining the variational model construction, which can overcome the misclassification and aliasing problems caused by frequency band segmentation. This method can effectively extract bearing fault information.

[0094] Preferably, step S2 specifically includes the following steps:

[0095] S21. VMD is used to solve the misclassification and aliasing problems caused by frequency band segmentation, and to obtain effective sound information of roller faults.

[0096] Preferably, step S21 specifically includes the following steps:

[0097] S211. Analyze the modal component u using Hilbert transform k (t), and obtain each IMF component {u k}、Analytical signal and corresponding unilateral spectrum, the calculation formula is as follows:

[0098]

[0099] Where δ(t) is the impulse function; j is the complex imaginary part; t is the time series;

[0100] S212, set the center frequency of each modal component to w k , the signal spectrum is shifted to baseband using the following formula:

[0101]

[0102] Where, It is an exponential signal;

[0103] S213. Construct a constrained variational model and calculate the bandwidth of each component signal:

[0104]

[0105] In the formula, {u k} are the IMF components, {w k} is the center frequency of each IMF component, k is the number of modes obtained by decomposition, and f(t) is the sound signal collected by the original acoustic sensor; is a partial differential.

[0106] S22. The envelope spectrum kurtosis method is used to eliminate the interference introduced by the accidental vibration impact component in the effective sound information of the roller fault.

[0107] The acoustic signal of a roller bearing fault exhibits distinct periodic and impact characteristics. Time-domain kurtosis is only sensitive to impact characteristics and cannot characterize the signal's periodicity. Therefore, it is susceptible to non-periodic transient impact components in the acoustic signal. Envelope spectrum kurtosis extends the kurtosis calculation process to the frequency domain, leveraging the periodic nature of fault information to eliminate interference introduced by occasional vibration impact components in the signal.

[0108] Preferably, step S22 specifically includes the following steps:

[0109] S221, let the time domain signal be x(t), use VMD to decompose x(t) into several components x i (t), the analytical signal of the modal component is:

[0110]

[0111] Where, Represents x i (t), where i = 1, 2, 3, ..., so the envelope signal of the IMF component is:

[0112]

[0113] The corresponding envelope signal spectrum is:

[0114]

[0115] Where Env(n) is the envelope signal; m is the frequency domain signal sequence, N is the number of discrete points, n is the time domain signal sequence, and m = 0, 1, ..., N-1.

[0116] S222, let the roller bearing failure frequency be f r When the roller bearing fails, the envelope spectrum analysis of the components containing the fault components is performed to observe whether the fault frequency or its multiple frequency exists. Therefore, when calculating its spectrum kurtosis, only [f r ,4f r ] within the spectrum line NUM u , signal envelope spectrum kurtosis k Es for:

[0117]

[0118] Where NUM1 is the spectrum sequence; V 2 (m) represents the square of the envelope spectrum sequence; V 4 (m) represents the fourth power of the envelope spectrum sequence.

[0119] S3. Using the TDOA delay estimation algorithm, the IMF components selected from the extracted effective sound information of the roller fault are subjected to generalized cross-correlation delay estimation, and the delay is determined by the time difference corresponding to the maximum value of the cross-correlation coefficient in the delay domain;

[0120] Step S3 specifically includes the following steps:

[0121] S31. Time delay estimation is the key to locating the fault sound source, and its accuracy will directly affect the accuracy of positioning. First, acoustic sensors need to be placed at different locations to collect the sound source information of the roller fault. A time delay estimation algorithm is used to calculate the relative time difference between the fault signal and the corresponding sensor. Therefore, a basic cross-correlation algorithm is used to analyze and compare the time domain similarity between the two sound signals.

[0122] Assuming that two acoustic sensors are placed at the head and tail ends of the belt conveyor, the model of the sound signal collected by them is:

[0123] x1(t)=S(t-τ1)+n1(t) (8)

[0124] x2(t)=S(t-τ2)+n2(t) (9)

[0125] Where S(t-τ1) represents the fault source information received by the acoustic sensor at the head end of the belt conveyor; S(t-τ2) represents the fault source information received by the acoustic sensor at the tail end of the belt conveyor; n1(t) represents the Gaussian white noise of the acoustic sensor at the head end of the belt conveyor; n2(t) represents the Gaussian white noise of the acoustic sensor at the tail end of the belt conveyor; and S(t-τ1), S(t-τ2), n1(t) and n2(t) are independent of each other, τ1 and τ2 are the propagation times of the fault sound wave transmitted to the two acoustic sensors;

[0126] The cross-correlation function of x1(t) and x2(t) is expressed as:

[0127] R 12 (τ)=E[x1(t)x2(t-τ)] (10)

[0128] Where R 12 (τ) represents the cross-correlation function;

[0129] Substituting equations (8) and (9) into equation (10), we obtain:

[0130]

[0131] Where s(t-τ1-τ) and s(t-τ2-τ) represent the fault sound source information received by the acoustic sensors at the head and tail ends, respectively, and n2(t-τ) represents the noise;

[0132] Therefore, from formula (11):

[0133] R 12 (τ)=E[s(t-τ1)s(t-τ1-τ)]=R s (τ-(τ1-τ2)) (12)

[0134] Where R s (τ-(τ1-τ2)) represents the correlation function;

[0135] It shows that when τ-(τ1-τ2)=0, R 12 (τ) The τ corresponding to the peak value is the time delay between the fault sound wave propagating to the two acoustic sensors;

[0136] It can be seen that calculating R 12The time corresponding to the peak value of (τ) is the time difference between the fault sound source signal being transmitted to the two acoustic sensors. However, in practical applications, the peak value of the correlation function is easily interfered with by external environmental factors, causing the peak value to weaken or shift, thereby affecting the accuracy of the delay estimation.

[0137] S32. The main advantages of cross-correlation time estimation are simplicity and small computational complexity, but this method has obvious limitations. First, in the theoretical derivation process, it is assumed that there is no correlation between the signal and the noise, but in actual applications, the signals received by the acoustic sensor all come from the same fault sound source. Secondly, the signal in the formula is an infinitely long sequence, but in the actual production process, only a finite-length signal can be used instead. Moreover, the cross-correlation algorithm can only be used to process data that is an integer multiple of the sampling period. All of these will bring some errors to the delay estimation, making the peak of the correlation function not obvious, which affects the accuracy of the delay estimation. Therefore, the generalized cross-correlation algorithm is used to obtain the generalized cross-correlation function between the sound signals collected by the two acoustic sensors:

[0138] S321, filtering the sound signals collected by the two acoustic sensors and then framing them using a finite-length window;

[0139] S322, performing Fourier transform on the sound signal processed in step S31 to obtain the cross-power spectrum of the two signals in this frame;

[0140] S323. After weighting the frame in the frequency domain using a weighting function, perform inverse Fourier transform to obtain the cross-correlation function of the frame, extract the time τ corresponding to the peak value in the cross-correlation function, and obtain the time delay τ between the sound wave propagating to the two acoustic sensors. 12 ;

[0141] S324. According to the relationship between the cross-correlation function and the cross-power spectrum, we obtain:

[0142]

[0143] Where R 12 (τ) represents the cross-correlation function; G 12 (w) is the cross-power spectrum of the signals x1(t) and x2(t) received by the two acoustic sensors;

[0144] S325. Perform weighted processing on equation (13) and obtain the cross-correlation function through inverse Fourier transform:

[0145]

[0146] Where, ψ 12 (w) represents the generalized cross-correlation weighting function.

[0147] The selection of the above function is based on two aspects: first, the surrounding noise and reflection conditions. Different weighting functions can be selected on site according to the actual situation. Second, the weighting function is introduced to make R 12 (τ) has a relatively sharp peak, which improves the accuracy of delay estimation.

[0148] S4. The location of the fault sound source is obtained by combining the propagation speed of sound in the air medium, the spatial geometric relationship between the location of the fault sound source and the acoustic sensor.

[0149] Verification example:

[0150] Acoustic sensors were placed every 6 meters along a 12-meter-long conveyor belt. That is, one sensor was placed at each end and in the middle of the conveyor belt, with all three sensors aligned. The sampling frequency for all three sensors was set to 25 kHz, and the sampling time was set to 1 second. The fault sound source was located 2 meters from the conveyor belt end.

[0151] To verify the ability of the proposed method to identify the periodic pulse characteristics of the faulty roller and the preprocessing effect in delay estimation, the sound signals collected in the experiment were analyzed. The steps are as follows:

[0152] (1) The three acoustic sensors collect the roller fault sound signals x1(t), x2(t), and x3(t), and observe their time domain waveforms;

[0153] (2) Use VMD to decompose the three groups of signals into several modal components c ji (j=1, 2, 3; i=1, 2, ..., n), calculate the envelope spectrum kurtosis value of the above modal components, and select the component with the highest kurtosis value in each group of data as the optimal component.

[0154] (3) Assume that the optimal components of the three groups of original sound signals in step (2) are c 11 、c 21 、c 31 , then the corresponding envelope spectrum kurtosis value is K Es (c 11 ), K Es (c 21 ), K Es (c 31 ), compare the envelope spectrum kurtosis values corresponding to the three components, and select the components with the maximum and second largest envelope spectrum kurtosis values.

[0155] (4) Perform a generalized cross-correlation analysis on the components with the maximum and second-largest kurtosis values of the envelope spectrum selected in (3) to obtain a time delay estimate. Combined with the speed of sound propagation in air, the spatial relationship between the sound source of the roller bearing fault and the acoustic sensor, the faulty roller is accurately located.

[0156] like Figure 3-Figure 5 As shown in the figure, since the acoustic sensor at the front end of the belt conveyor is closest to the fault sound source, a significant impulse component with a larger amplitude is observed in its time domain signal. Due to energy attenuation during sound transmission, the impulse amplitude in the time domain signal collected by the acoustic sensor in the middle of the belt conveyor is reduced. The acoustic sensor at the rear end of the belt conveyor is farthest from the fault sound source, and its time domain waveform has a less significant impulse component.

[0157] The three groups of signals are decomposed by VMD with 4 decomposition layers. The envelope spectrum kurtosis value of each IMF component after decomposition is calculated.

[0158] Table 1 Envelope spectrum kurtosis values of IMF components of three groups of signals after VMD decomposition

[0159] i 1 2 3 4 <![CDATA[K Es (c 0i )]]> 23.809 53.262 45.584 34.428 <![CDATA[K Es (c 1i )]]> 5.359 64.891 45.342 37.145 <![CDATA[K Es (c 2i )]]> 11.221 31.010 13.691 24.341

[0160] As can be seen from Table 1, the higher the kurtosis value, the higher the proportion of fault components in the component. By selecting the IMF component with the maximum envelope spectrum kurtosis in each group of signals from Table 1, that is, selecting c 02 、c 12 、c 22 Three modal components.

[0161] like Figures 6-11 As shown, it can be seen that: (1) The roller fault information is mainly concentrated in the IMF2 component, that is, the frequency band range of the roller fault component will not change during the propagation of the sound signal. (2) The impact component appearing at the 0.2 second and 0.8 second positions will gradually attenuate as the distance increases. Therefore, in practical applications, it is necessary to pay attention to the effective distance between the acoustic sensor and the fault sound source to ensure that the fault information can be collected. (3) By analyzing the envelope spectra of the three groups of signals, it can be seen that the characteristic frequency of the roller fault is 3.9Hz. In addition, the amplitude of the signal frequency of about 8Hz is also quite obvious, and 8Hz is about twice the characteristic frequency of the roller fault. Therefore, the method of VMD combined with envelope spectrum kurtosis can effectively determine the component with high fault components.

[0162] At the same time Figure 12 As shown, due to K Es (c 02 )>K Es (c 12 )>K Es (c22 ), that is, c 02 、c 12 Contains more roller fault information than c 22 High, so we can preliminarily determine the range of the sound source fault, that is, between the first end acoustic sensor and the middle position acoustic sensor, and then 02 and c 12 A generalized cross-correlation analysis of the two component signals shows that the maximum peak is detected at τ = 0.0074s, resulting in the maximum cross-correlation coefficient. Given the sound propagation speed V = 340m / s and the installation distance L = 6m between the acoustic sensor at the head end and the acoustic sensor at the middle position, the location of the sound source is estimated according to Equation (15):

[0163]

[0164] It can be calculated that L1 = 1.922 m. This result is very close to the actual location of the fault sound source of the roller, which is 2 meters, indicating that the method of the present invention can accurately determine the location of the fault sound source.

[0165] Example:

[0166] The acoustic sensor used was the BSWA MPA416, with a sensitivity of -50 Mv / Pa and a frequency response range of 20 Hz to 20 kHz. The data acquisition card used was the NET8814-8, with a resolution of 24 bits, a sampling frequency of 8 Hz to 204.8 kHz, and a sampling range of ±5.5 V. To avoid complex on-site wiring that could affect the operation of the conveyor belt and the normal work of inspectors, all three acoustic sensors were installed 4 meters above the ground, with adjacent sensors spaced 30 meters apart. The first acoustic sensor was used as the reference point (0 meters), and the faulty roller was 53 meters from the reference point. The sampling frequency was set to 25 kHz, and the sampling time was 2 seconds.

[0167] In order to effectively extract the fault information of the roller bearing, VMD decomposition and envelope spectrum kurtosis calculation are performed on the three groups of sound signals collected by the three acoustic sensors. The results are shown in Table 2.

[0168] Table 2 Envelope spectrum kurtosis values of each IMF component after VMD decomposition of the three groups of signals

[0169] i 1 2 3 4 <![CDATA[K Es (c 0i )]]> 2.85 2.21 3.18 2.56 <![CDATA[K Es (c 1i )]]> 3.51 3.28 5.12 2.37 <![CDATA[K Es (c 2i )]]> 3.18 3.15 7.19 1.92

[0170] Select the IMF components with the largest and second largest envelope spectrum kurtosis values in the three groups of sound signals (c 23 、c 13 ), since these two groups of components are collected by the second acoustic sensor and the third acoustic sensor, it can be determined that the fault sound source is between these two acoustic sensors.

[0171] like Figure 13 As shown in the figure, a generalized cross-correlation analysis is performed on the selected components, and the time delay estimation reaches a peak at t = -0.1442. Combined with the spatial geometric relationship between the roller fault sound source and the acoustic sensor, it is found that the roller fault is located at a position 45 meters away from the reference point.

[0172] Therefore, the present invention adopts the above-mentioned roller fault sound source localization method based on the TDOA time delay estimation algorithm, which can extract the periodic fault information of the roller fault from the transmission path coupling interference component and realize the accurate localization of the roller fault sound source.

[0173] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A roller fault sound source localization method based on the TDOA time delay estimation algorithm is characterized by: The following steps are involved: S1. Install multiple sound sensors at set intervals next to the belt conveyor to collect sound signals from the rollers of the belt conveyor; S2. VMD and envelope spectrum kurtosis method are used to extract effective sound information of roller fault; S3. Using the TDOA delay estimation algorithm, the IMF components selected from the extracted effective sound information of the roller fault are subjected to generalized cross-correlation delay estimation, and the delay is determined by the time difference corresponding to the maximum value of the cross-correlation coefficient in the delay domain; Step S3 specifically includes the following steps: S31. Analyze and compare the time domain similarity between two sound signals using a basic cross-correlation algorithm; Assuming that two acoustic sensors are placed at the head and tail ends of the belt conveyor, the model of the sound signal collected by them is: x1(t)=S(t-τ1)+n1(t) (8) x2(t)=S(t-τ2)+n2(t) (9) Where: S(t-τ1) represents the fault source information received by the acoustic sensor at the head end of the belt conveyor; S(t-τ2) represents the fault source information received by the acoustic sensor at the tail end of the belt conveyor; n1(t) represents the Gaussian white noise of the acoustic sensor at the head end of the belt conveyor; n2(t) represents the Gaussian white noise of the acoustic sensor at the tail end of the belt conveyor; and S(t-τ1), S(t-τ2), n1(t) and n2(t) are independent of each other, τ1 and τ2 are the propagation times of the fault sound wave transmitted to the two acoustic sensors; The cross-correlation function of x1(t) and x2(t) is expressed as: R 12 (τ)=E[x1(t)x2(t-τ)] (10) Where R 12 (τ) represents the cross-correlation function; Substituting equations (8) and (9) into equation (10), we obtain: Where s(t-τ1-τ) and s(t-τ2-τ) represent the fault sound source information received by the acoustic sensors at the head and tail ends, respectively, and n2(t-τ) represents the noise; Therefore, from formula (11): R 12 (τ)=E[s(t-τ1)s(t-τ1-τ)]=R s (τ-(τ1-τ2)) (12) Where R s (τ-(τ1-τ2)) represents the correlation function; It shows that when τ-(τ1-τ2)=0, R 12 (τ) The τ corresponding to the peak value is the time delay between the fault sound wave propagating to the two acoustic sensors; S32. Obtain a generalized cross-correlation function between the sound signals collected by the two acoustic sensors using a generalized cross-correlation algorithm: S321, filtering the sound signals collected by the two acoustic sensors and then framing them using a finite-length window; S322, performing Fourier transform on the sound signal processed in step S31 to obtain the cross-power spectrum of the two signals in this frame; S323. After weighting the frame in the frequency domain using a weighting function, perform inverse Fourier transform to obtain the cross-correlation function of the frame, extract the time τ corresponding to the peak value in the cross-correlation function, and obtain the time delay τ between the sound wave propagating to the two acoustic sensors. 12 ; S324. According to the relationship between the cross-correlation function and the cross-power spectrum, we obtain: Where R 12 (τ) represents the cross-correlation function; G 12 (w) is the cross-power spectrum of the signals x1(t) and x2(t) received by the two acoustic sensors; S325. Perform weighted processing on equation (13) and obtain the cross-correlation function through inverse Fourier transform: Where, ψ 12 (w) represents the generalized cross-correlation weighting function; S4. The location of the fault sound source is obtained by combining the propagation speed of sound in the air medium, the spatial geometric relationship between the location of the fault sound source and the acoustic sensor.

2. The method for locating the sound source of a roller fault based on the TDOA time delay estimation algorithm according to claim 1 is characterized in that: Step S2 specifically includes the following steps: S21. VMD is used to solve the misclassification and aliasing problems caused by frequency band segmentation, and to obtain effective sound information of roller faults. S22. The envelope spectrum kurtosis method is used to eliminate the interference introduced by the accidental vibration impact component in the effective sound information of the roller fault.

3. The method for locating the sound source of a roller fault based on the TDOA time delay estimation algorithm according to claim 2 is characterized in that: Step S21 specifically includes the following steps: S211. Analyze the modal component u using Hilbert transform k (t), and obtain each IMF component {u k }、Analytical signal and corresponding unilateral spectrum, the calculation formula is as follows: Where δ(t) is the impulse function; j is the complex imaginary part; t is the time series; S212, set the center frequency of each modal component to w k , the signal spectrum is shifted to baseband using the following formula: Where, It is an exponential signal; S213. Construct a constrained variational model and calculate the bandwidth of each component signal: In the formula, {u k } are the IMF components, {w k } is the center frequency of each IMF component, k is the number of modes obtained by decomposition, and f(t) is the sound signal collected by the original acoustic sensor; is a partial differential.

4. The method for locating the sound source of a roller fault based on the TDOA time delay estimation algorithm according to claim 2, characterized in that: Step S22 specifically includes the following steps: S221, let the time domain signal be x(t), use VMD to decompose x(t) into several components x i (t), the analytical signal of the modal component is: Where, Represents x i (t), where i = 1, 2, 3, ..., so the envelope signal of the IMF component is: The corresponding envelope signal spectrum is: Where Env(n) is the envelope signal; m is the frequency domain signal sequence, N is the number of discrete points, n is the time domain signal sequence, and m = 0, 1, ..., N-1; S222, let the roller bearing failure frequency be f r When the roller bearing fails, the envelope spectrum analysis of the components containing the fault components is performed to observe whether the fault frequency or its multiple frequency exists. Therefore, when calculating its spectrum kurtosis, only [f r ,4f r ] within the spectrum line NUM u , signal envelope spectrum kurtosis k Es for: Where NUM1 is the spectrum sequence; V 2 (m) represents the square of the envelope spectrum sequence; V 4 (m) represents the fourth power of the envelope spectrum sequence.

Images