A method for selecting sparse wavelet packet nodes of compressed acoustic impact signals

Abstract

The application discloses a method for selecting sparse wavelet packet nodes of compressed acoustic impact signals, and solves the problem of difficulty in compressing large data of periodic pulse acoustic signals in the prior art track. The method utilizes the clustering characteristics of wavelet packet nodes to perform wavelet packet decomposition on collected healthy signals and fault signals; then, each subband energy is normalized; and by calculating the difference value of the corresponding subband and arranging in descending order, a normalized damage sensitive energy coefficient vector is obtained; then, by calculating the difference value between adjacent coefficients, the maximum difference value position and the corresponding node number of the position in front of the maximum difference value position are obtained, that is, the wavelet packet node sensitive to the pulse signal; then, the wavelet packet related to the fault is retained, and the wavelet coefficients of other wavelet packet nodes are set to zero; finally, inverse wavelet packet transformation is performed to obtain output data; the method can meet the requirements of high compression ratio and high fidelity under the condition of retaining effective acoustic information, and is suitable for compression processing of acoustic impact signals of railway steel rails.

Description

Technical Field

[0001] This invention belongs to the field of railway acoustic impact signal processing technology, specifically relating to a sparse wavelet packet node selection method for compressed acoustic impact signals that selects effective information containing impact signals for data transmission based on the cluster characteristics of wavelet packet decomposition, achieving a high data compression ratio while preserving effective acoustic information. This method is applicable to the compression processing of railway rail acoustic impact signals. Background Technology

[0002] Currently, there are two methods for monitoring damage to rails or wheel treads. One method involves deploying sensors on the axle. The drawback of this method is that vibrations from components such as the car body, bogie, and frame are also transmitted to the sensors, leading to overlap in many vibration frequency domains and making it difficult to pinpoint the vibration source, thus reducing the accuracy of damage identification. The other method involves installing static condition monitoring equipment along the trackside, such as wheel impact load detector systems and acoustic-based trackside monitoring systems. The disadvantages of this method are that vibrations in the rail decay very quickly, making large-scale real-time monitoring impossible; furthermore, acoustic-based trackside monitoring systems are susceptible to interference from electromagnetic and environmental noise.

[0003] Sound has the advantages of long propagation distance and high speed in rails. If there is damage to the rail or wheel tread, a pulse impact signal will be generated when the wheel and rail interact. Therefore, theoretically, by listening to the impact sound signal in the rail, it is possible to detect the safety status of the rail or wheel over long distances. However, acoustic signals have a wide frequency range, long sampling period, and huge data volume, making direct remote transmission and subsequent analysis in a data center difficult. Therefore, it is necessary to conduct technical research on the compression processing of rail acoustic impact signal data. Summary of the Invention

[0004] This invention addresses the aforementioned problems by providing a sparse wavelet packet node selection method for compressed acoustic impact signals. This method selects effective information containing the impact signal for data transmission based on the cluster characteristics of wavelet packet decomposition, achieving a high data compression ratio while preserving effective acoustic information. It is applicable to the compression processing of acoustic impact signals from railway rails.

[0005] The technical solution adopted in this invention is as follows: the sparse wavelet packet node selection method for the compressed acoustic impulse signal includes the following steps:

[0006] Step 1: Perform Nyquist sampling on the sensor output signal to obtain the acquired health signal. ;

[0007] Step 2: Acoustic health signals and fault signals Perform wavelet packet decomposition to obtain the normalized energy vector for each subband. and ;

[0008] Step 3: Obtain the corresponding sub-band and The difference and descending order are used to obtain the normalized damage sensitivity energy coefficient vector. ;

[0009] Step 4: Calculate the vector The difference between adjacent elements in the vector is used to obtain the vector. ;

[0010] Step 5: Obtain the vector The wavelet packet node corresponding to the sensitive factor before position n, where the maximum difference is located, is the effective wavelet packet related to the fault.

[0011] Step 6: Retain the wavelet packets related to the fault and set the wavelet coefficients of other wavelet packet nodes to zero;

[0012] Step 7: Perform inverse wavelet packet transform to obtain the output data.

[0013] Step two, fault signal Energy vector and Obtained through the following methods:

[0014] Obtain the wavelet packet decomposition coefficients, and obtain the energy of each subband according to equation (1);

[0015] (1)

[0016] In the formula: j is the wavelet packet decomposition level; N is the number of wavelet packet coefficients in the i-th sub-band; These are wavelet packet decomposition coefficients;

[0017] The fault signal is obtained according to equation (2). ;

[0018] (2)

[0019] In the formula: s(t) is the periodic impact component; i is the number of impact repetitions; The periodic impact frequency is the same as the wheel rotation frequency. It refers to the magnitude of periodic shocks; The resonant frequency; The attenuation coefficient; Added Gaussian white noise; The sampling frequency;

[0020] Health signals are obtained according to formula (3). Normalized energy vector and fault signals Normalized energy vector .

[0021] (3)

[0022] Step three, damage-sensitive energy coefficient vector Obtained through the following methods:

[0023] The energy vector related to the damage is obtained according to equation (4). ;

[0024] (4)

[0025] The damage-sensitive energy coefficient vector is obtained according to equation (5);

[0026] (5)

[0027] In the formula: This refers to the corresponding node information;

[0028] Sensitive factors Sort in descending order to obtain a vector .

[0029] In step four, the difference between adjacent elements Obtained through the following methods:

[0030] The difference between adjacent elements is obtained according to equation (6).

[0031] (6)

[0032] In step five, the position of the maximum difference n is... When it is the maximum value, the corresponding The number of nodes before the descending sort.

[0033] The specific method for retaining the fault-related wavelet packets in step six is ​​as follows: set all wavelet sensitivity factors that are close to 0 to 0; retain the remaining, larger wavelet sensitivity factors, which are the fault-related wavelet packets.

[0034] The specific method for obtaining the output data by inverse wavelet packet transform in step seven is as follows: after performing inverse wavelet transform on the fault-related wavelet packets retained in step six, the obtained data is the required output data.

[0035] The beneficial effects of this invention are as follows: This sparse wavelet packet node selection method for compressed acoustic impact signals utilizes the clustering characteristics of wavelet packet nodes to perform wavelet packet decomposition on the acquired healthy and fault signals; then, the energy of each sub-band is normalized; and by calculating the difference between the corresponding sub-bands and arranging them in descending order, a normalized damage-sensitive energy coefficient vector is obtained; furthermore, by solving for the difference between adjacent coefficients, the position of the maximum difference and the corresponding node number of the position before it are obtained, which are the wavelet packet nodes sensitive to the pulse signal; then, wavelet packets related to the fault are retained, and the wavelet coefficients of other wavelet packet nodes are set to zero; finally, an inverse wavelet packet transform is performed to obtain the output data. Based on the clustering characteristics of wavelet packet decomposition, this invention compresses massive acoustic impact signals using a sparse wavelet packet node selection method, and selects effective information containing the impact signal for data transmission. It can meet the requirements of high compression ratio and high fidelity while preserving effective acoustic information, and is suitable for the compression processing of acoustic impact signals from railway rails. Attached Figure Description

[0036] Figure 1 This is a flowchart of the method of the present invention.

[0037] Figure 2 These are time-domain waveform diagrams of the analog signals involved in the embodiments of the present invention; wherein: (a) is the time-domain waveform diagram of the normal signal after adding noise, and (b) is the time-domain waveform diagram of the fault signal after adding noise.

[0038] Figure 3 These are the energy spectrum diagrams of the signal processing results involved in the embodiments of the present invention; wherein: (a) the energy spectrum diagram of the normal signal, and (b) the energy spectrum diagram of the fault signal.

[0039] Figure 4 The waveform diagram of the energy coefficient obtained in the embodiment of the present invention is shown; wherein: (a) the energy coefficient of the fault signal, and (b) the sensitive energy factor of the fault signal.

[0040] Figure 5 This is a descending order diagram of sensitive factors in an embodiment of the present invention; wherein: (a) sensitive energy factors are arranged in descending order, and (b) the difference diagram of sensitive energy factors of adjacent nodes after descending order.

[0041] Figure 6 This is a waveform diagram of the reconstructed analog signal in an embodiment of the present invention.

[0042] Figure 7 The method of this invention is used to reconstruct the envelope spectrum of an analog signal; wherein: (a) IMF1 signal envelope spectrum, (b) IMF2 signal envelope spectrum, (c) IMF3 signal envelope spectrum; it can be seen from the figure that the impulse signal frequency is 120Hz and its harmonics, which can... Figure 7(a) Easily identifiable; in Figure 7 It can also be identified in (b) and (c), but it will be affected by noise.

[0043] Figure 8 The envelope spectrum of the analog signal is reconstructed using the DCT algorithm; where: (a) IMF1 signal envelope spectrum; (b) IMF2 signal envelope spectrum; (c) IMF3 signal envelope spectrum; it can be seen from the figure that the impulse signal frequency is 120Hz and its harmonics, which can... Figure 8 (a) Easily identifiable; in Figure 8 In (b) and (c), it is almost unrecognizable.

[0044] Figure 9 The envelope spectrum of the analog signal is reconstructed using wavelet compression algorithm; where: (a) envelope spectrum of IMF1 signal; (b) envelope spectrum of IMF2 signal; (c) envelope spectrum of IMF3 signal; From the figure, it can be seen that: the impulse signal frequency is 120Hz and its harmonics, in Figure 9 It cannot be recognized in Chinese. Detailed Implementation

[0045] To address the challenge of compressing large amounts of periodic impulse acoustic signals in railway tracks, this invention proposes a sparse wavelet packet node selection method for compressing acoustic impulse signals. Based on the cluster characteristics of wavelet packet decomposition, the proposed method selects effective information containing the impulse signal for data transmission. While preserving the effective acoustic information, it achieves a high data compression ratio, making it highly suitable for compressing and processing railway acoustic impulse signals.

[0046] The specific steps of this invention are described in detail. The method for selecting sparse wavelet packet nodes of the compressed acoustic impulse signal includes:

[0047] Step 1: Nyquist sampling; set the sampling frequency to 16kHz and the number of sampling points to 4096, perform Nyquist sampling on the sensor output signal to obtain the acquired health signal. .

[0048] Step two, acoustic health signals and fault signals Perform wavelet packet decomposition to obtain the normalized energy vector for each subband. and .

[0049] When there are no tire tread marks on the train wheels, the health signal collected from the rails contains two sine waves of different frequencies, representing the rotational frequency conditions of different components in a healthy state. Therefore, the health signal can be obtained by combining the two sine wave signals and adding white Gaussian noise. Health signals like Figure 2As shown in (a).

[0050] When a wheel has tread scratches, as the wheel rotates, it generates periodic impacts at the same frequency as the wheel's rotation. (Fault signal) Energy vector and Obtained through the following methods:

[0051] Obtain the wavelet packet decomposition coefficients, and obtain the energy of each subband according to equation (1).

[0052] (1)

[0053] In the formula: j is the wavelet packet decomposition level; N is the number of wavelet packet coefficients in the i-th sub-band; These are the wavelet packet decomposition coefficients.

[0054] The fault signal is obtained according to equation (2). ;

[0055] (2)

[0056] In the formula: s(t) is the periodic impact component; i is the number of impact repetitions; The periodic impact frequency is the same as the wheel rotation frequency. It refers to the magnitude of periodic shocks; The resonant frequency; The attenuation coefficient; Added Gaussian white noise; The sampling frequency.

[0057] Health signals are obtained according to formula (3). Normalized energy vector and fault signals Normalized energy vector .

[0058] (3)

[0059] Health signals with noise Figure 2 (a) and fault signals with noise Figure 2 (b) The compressed analog signal is used for algorithmic calculation.

[0060] Decomposing the two signals into fifth-level wavelet packets yields the normalized energy vectors of the normal analog signal and the fault analog signal, which are respectively labeled as follows: and Their normalized wavelet packet energy spectrum is as follows Figure 3 (a) and Figure 3As shown in (b). Normalization: turning the data into decimals between (0,1) is mainly proposed for the convenience of data processing. It maps the data to the range of 0 to 1 for processing, so that indicators of different units or magnitudes can be compared and weighted.

[0061] Step 3: Obtain the corresponding sub-band and The difference and descending order are used to obtain the normalized damage sensitivity energy coefficient vector. .

[0062] Damage-sensitive energy coefficient vector Obtained through the following methods:

[0063] The energy vector related to the damage is obtained according to equation (4). .

[0064] (4)

[0065] like Figure 4 As shown in (a), the nodes can be determined as 1, 9 and 25 based on the horizontal coordinates.

[0066] The damage-sensitive energy coefficient vector is obtained according to equation (5).

[0067] (5)

[0068] In the formula: This refers to the corresponding node information.

[0069] like Figure 4 As shown in (b), the fault-sensitive energy factor vectors corresponding to nodes 1, 9, and 25 can be determined. .

[0070] Will Figure 4 The sensitive energy factors in (b) are arranged in descending order to obtain a vector. ,like Figure 5 (a).

[0071] Step four, calculate the vector The difference between adjacent elements in the vector is used to obtain the vector. .

[0072] Difference between adjacent elements Obtained through the following methods:

[0073] The difference between adjacent elements is obtained according to equation (6).

[0074] (6)

[0075] like Figure 5The multiple nodes in (b) are the difference vectors obtained by subtracting the nodes after sorting in descending order. .

[0076] Step 5, obtain the vector The wavelet packet node corresponding to the sensitive factor preceding the maximum difference position n is the valid wavelet packet related to the fault. When it is the maximum value, the corresponding The number of nodes before the descending sort.

[0077] Figure 5 (b) represents a vector The maximum value in the vector λ is located in the third position, which indicates that the first three nodes in the vector λ are the effective wavelet packet nodes related to the fault, and the sensitive wavelet packet nodes are nodes 1, 9 and 25.

[0078] Step 6: Retain the wavelet packets related to the fault and set the wavelet coefficients of other wavelet packet nodes to zero. The specific method for retaining the wavelet packets related to the fault is as follows: set all wavelet sensitivity factors that are close to 0 to 0; retain the remaining wavelet sensitivity factors that are larger, which are the wavelet packets related to the fault.

[0079] Step 7: Perform inverse wavelet packet transform to obtain the output data. The specific method for obtaining the output data using inverse wavelet packet transform is as follows: After performing an inverse wavelet transform on the fault-related wavelet packets retained in Step 6, the resulting data is the required output data, such as... Figure 6 As shown.

[0080] Compared with traditional DCT and wavelet compression algorithms, the method of this invention is as follows: Figure 7 , 8 9 and Table 1.

[0081] Table 1 Comparison of the three compression methods

[0082] Compression ratio 10.67 1.21 1.14 Peak signal-to-noise ratio 51.94 160.37 31.44

[0083] As shown in Table 1, the simulation data processed by the method of this invention has the highest compression rate, while the compression rates of the DCT algorithm and the wavelet compression algorithm are almost the same. The PSNR of the analog signal processed by the method of this invention is 51.94, which is significantly lower than the PSNR of 160.37 after processing by the DCT algorithm, but higher than the PSNR of 31.44 after processing by the wavelet compression algorithm. This is because the algorithm of this invention only retains sensitive nodes and completely ignores insensitive nodes. Figure 7 , Figure 8 , Figure 9 The envelope spectrum characteristics of the signals processed by the three compression algorithms are presented respectively. Figure 7(a) The 120Hz impulse frequency and its harmonics can be clearly identified. Figure 7 In (b) and (c), the impulse frequency can be identified, but it is severely affected by noise. The 120Hz impulse frequency and its harmonics can also be found in... Figure 8 It is clearly identified in (a). Figure 8 (b) Figure 8 (c) Figure 9 (a) Figure 9 (b) and Figure 9 In (c), the 120Hz impulse frequency and its harmonics cannot be determined. Experimental results show that, compared with the DCT algorithm and wavelet compression algorithm, the method of the present invention can better preserve the frequency characteristics of the pulse signal.

Claims

1. A method for selecting sparse wavelet packet nodes of compressed acoustic impulse signals, characterized in that, Includes the following steps: Step 1: Perform Nyquist sampling on the sensor output signal to obtain the acquired health signal. ; Step 2: Acoustic health signals and fault signals Perform wavelet packet decomposition to obtain the normalized energy vector for each subband. and ; Step 3: Obtain the corresponding sub-band and The difference and descending order are used to obtain the normalized damage sensitivity energy coefficient vector. ; Step 4: Calculate the vector The difference between adjacent elements in the vector is used to obtain the vector. ; Step 5: Obtain the vector The wavelet packet node corresponding to the sensitive factor before position n, where the maximum difference is located, is the effective wavelet packet related to the fault. Step 6: Retain the wavelet packets related to the fault and set the wavelet coefficients of other wavelet packet nodes to zero; Step 7: Perform inverse wavelet packet transform to obtain the output data.

2. The method for selecting sparse wavelet packet nodes of compressed acoustic impulse signals according to claim 1, characterized in that: Step two, fault signal Energy vector and Obtained through the following methods: Obtain the wavelet packet decomposition coefficients, and obtain the energy of each subband according to equation (1); （1） In the formula: j is the wavelet packet decomposition level; N is the number of wavelet packet coefficients in the i-th sub-band; These are wavelet packet decomposition coefficients; The fault signal is obtained according to equation (2). ; （2） In the formula: s(t) is the periodic impact component; i is the number of impact repetitions; The periodic impact frequency is the same as the wheel rotation frequency. It refers to the magnitude of periodic shocks; The resonant frequency; The attenuation coefficient; Added Gaussian white noise; The sampling frequency; Health signals are obtained according to formula (3). Normalized energy vector and fault signals Normalized energy vector ; （3）。 3. The method for selecting sparse wavelet packet nodes of compressed acoustic impulse signals according to claim 1, characterized in that: Step three, damage-sensitive energy coefficient vector Obtained through the following methods: The energy vector related to the damage is obtained according to equation (4). ; （4） The damage-sensitive energy coefficient vector is obtained according to equation (5); （5） In the formula: This refers to the corresponding node information; Sensitive factors Sort in descending order to obtain a vector. .

4. The method for selecting sparse wavelet packet nodes of compressed acoustic impulse signals according to claim 1, characterized in that: In step four, the difference between adjacent elements Obtained through the following methods: The difference between adjacent elements is obtained according to equation (6); （6）。 5. The method for selecting sparse wavelet packet nodes of compressed acoustic impulse signals according to claim 1, characterized in that: In step five, the position of the maximum difference n is... When it is the maximum value, the corresponding The number of nodes before the descending sort.

6. The method for selecting sparse wavelet packet nodes of compressed acoustic impulse signals according to claim 1, characterized in that: The specific method for retaining the fault-related wavelet packets in step six is ​​as follows: set all wavelet sensitivity factors that are close to 0 to 0; retain the remaining, larger wavelet sensitivity factors, which are the fault-related wavelet packets.

7. The method for selecting sparse wavelet packet nodes of compressed acoustic impulse signals according to claim 1, characterized in that: The specific method for obtaining the output data by inverse wavelet packet transform in step seven is as follows: after performing inverse wavelet packet transform on the fault-related wavelet packets retained in step six, the obtained data is the required output data.

Images