A method for tracing the source of a reducer howl based on blind source separation

Abstract

The application provides a reducer howling tracing method based on blind source separation, comprising the following steps: collecting noise signals in different vibration transmission path directions of the reducer; performing short-time Fourier transform to obtain signal frequency spectrum, and then performing sound pressure level calculation; obtaining intermediate bandwidth, upper critical bandwidth and lower critical bandwidth with each frequency point of the frequency spectrum as center frequency, and combining the sound pressure level calculation to highlight the ratio; setting a highlight ratio threshold to perform filtering processing, and reserving the amplitude and phase of the intermediate bandwidth frequency range corresponding to the highlight ratio greater than the threshold; performing inverse Fourier transform on the reserved frequency band, and performing amplitude normalization processing; using blind source separation technology to perform independent component separation; selecting a separated signal with greater energy to perform frequency analysis, screening a plurality of frequencies with a larger energy proportion, calculating corresponding orders, and obtaining orders causing howling. The application considers the connection between noises in different vibration transmission path directions of the reducer, and realizes comprehensive and accurate howling tracing.

Description

Technical Field

[0001] This invention belongs to the field of reducer whistling source tracing technology, specifically involving a reducer whistling source tracing method based on blind source separation. Background Technology

[0002] New energy is the development trend of engineering machinery and heavy vehicles, and high-speed precision electric drive reducers are core components of these two types of major equipment. Electric drive reducers are characterized by high speed. Under high-speed and heavy-load conditions, the lack of engine noise masking effect makes the gear order noise of the electric drive reducer prominent. At the same time, the transmission system's natural frequencies are dense, and the number of local modes and potential resonance points increases, resulting in noise that seriously affects driving comfort. Whistling is the main manifestation of noise in gear transmission systems. Therefore, it is necessary to trace the source of whistling in electric drive reducers.

[0003] Currently, the source tracing of gearbox whistling mainly relies on spectral analysis of the gearbox acceleration noise signal based on acoustic signals to identify prominent orders. Simultaneously, to better evaluate whistling noise, the critical frequency band is analyzed, and comparisons are made between the critical frequency band and adjacent frequency bands to assess the whistling noise. Summary of the Invention

[0004] The applicant of this application discovered that reducer whistling is a rigid-flexible coupling response generated by the elastic system of the gearbox under dynamic excitation load. The vibration transmission path within the gearbox is complex, and different transmission paths have different attenuation or gain effects on the vibration, thus affecting the noise level in different directions of the reducer. Existing methods for tracing the source of reducer whistling all perform source analysis on noise in a single direction, neglecting the connection between noise in different vibration transmission paths of the reducer. This makes the whistling source analysis one-sided and unable to achieve accurate and comprehensive source tracing of reducer whistling.

[0005] The purpose of this invention is to address the problem of inaccurate source tracing results in existing methods for tracing the source of whistling in reducers based on acoustic signals, which analyze noise from individual directions separately. This invention provides a method for tracing the source of whistling in reducers based on blind source separation. This method uses noise signals from different vibration transmission paths of the reducer, and combines a prominence ratio threshold to filter and reconstruct the whistling frequency band. A blind source separation algorithm is used to independently separate the filtered signals, extracting and analyzing the main influencing frequencies of the separated signals according to their energy proportions. This achieves accurate overall whistling source tracing analysis of the reducer, providing strong data support for whistling suppression.

[0006] To achieve the above objectives, the technical solution provided by this invention is:

[0007] 1. A method for tracing the source of whistling in a reducer based on blind source separation, characterized by the following steps:

[0008] Step 1: Arrange multiple microphones around the reducer to collect noise time-domain signals in different vibration transmission path directions of the reducer.

[0009] Step 2: Perform short-time Fourier transform on the noise time-domain signals of the collected reducer in each direction to convert the noise time-domain signals into noise frequency-domain signals, obtain the spectrum of the frequency-domain signals in each direction, retain the amplitude and phase of the spectrum, and calculate the sound pressure level based on the amplitude.

[0010] Step 3: Based on the spectrum results obtained in Step 2, obtain the intermediate bandwidth, upper critical bandwidth, and lower critical bandwidth with each frequency point of the spectrum as the center frequency, and calculate the prominence ratio at each frequency point by combining the sound pressure level in Step 2.

[0011] Step 4: Set a prominence ratio threshold to filter the noise frequency domain signal obtained in Step 2, and retain the amplitude and phase of the intermediate bandwidth frequency range corresponding to the prominence ratio greater than the prominence ratio threshold.

[0012] Step 5: Reconstruct the signal frequency band retained in Step 4 using inverse Fourier transform to obtain the filtered noise time-domain signal.

[0013] Step 6: Perform amplitude normalization processing on the signal obtained in Step 5;

[0014] Step 7: Use blind source separation technology to separate the normalized noise signal into independent components, and obtain the separated noise signals sorted by energy level;

[0015] Step 8: Select several separated noise signals with relatively high energy for frequency analysis, and filter out multiple frequencies with a large energy proportion.

[0016] Step 9: Calculate the order corresponding to the frequency selected in Step 8 based on the rotational speed of the reducer input shaft, thereby obtaining the order that causes the reducer to whine.

[0017] Furthermore, in step 1, a microphone is arranged on the upper, left, and right sides of the reducer, respectively.

[0018] Furthermore, in step 3, the center frequency range for calculating the salience ratio is 90Hz to 7284Hz, and the corresponding intermediate bandwidth, upper critical bandwidth, and lower critical bandwidth are calculated at 1Hz intervals.

[0019] Furthermore, in step 4, the salience ratio threshold is set to 15 dB.

[0020] Furthermore, in step 7, the FastICA algorithm is used to separate the independent components of the noise signal.

[0021] Further, step 7 includes:

[0022] Step 7.1: Center the normalized noise signal;

[0023] Step 7.2: Whiten the centered noise signal;

[0024] Step 7.3: Solve for the separation matrix of the whitened noise signal;

[0025] Step 7.4: Use a separation matrix to separate the noise signal.

[0026] Furthermore, in step 8, the five separated noise signals with the top five energy percentages are selected for frequency analysis, and the five frequencies with the top five energy percentages are screened out.

[0027] The advantages of this invention are:

[0028] This invention presents a method for tracing the source of reducer whistling based on blind source separation. It collects noise signals from different vibration transmission paths of the reducer, uses a prominence ratio evaluation standard combined with short-time Fourier transform and inverse Fourier transform to adaptively filter the noise signals in each direction, and employs blind source separation technology to independently separate the components of the filtered signals in each direction, extracting the characteristic frequencies that contribute most to the reducer whistling. By analyzing the dominant frequency order, the source of the reducer whistling is traced. Therefore, this invention considers the relationship between noise in different vibration transmission paths of the reducer, making the whistling source analysis comprehensive and enabling accurate, integrated whistling source tracing. Attached Figure Description

[0029] The features and advantages of the invention will become more readily understood from the following description with reference to the accompanying drawings, in which:

[0030] Figure 1 This is a flowchart of the reducer whistling source tracing method based on blind source separation according to the present invention;

[0031] Figure 2 This is the noise signal prominence ratio in the three directions of the reducer (upper side, left side, and right side) in the example of the present invention;

[0032] Figure 3 This is a diagram showing the source tracing results of howling in an example of the present invention. Detailed Implementation

[0033] The present invention will now be described in detail with reference to the accompanying drawings and exemplary embodiments thereof. It should be noted that the following detailed description of the present invention is for illustrative purposes only and is not intended to limit the scope of the invention.

[0034] This invention provides a method for tracing the source of a reducer's howling based on blind source separation, which takes into account the relationship between noises in different vibration transmission paths of the reducer and traces the source of the howling of the reducer.

[0035] Reference Figure 1 The method for tracing the source of a reducer whistling sound based on blind source separation, as an exemplary embodiment of the present invention, includes the following steps:

[0036] Step S1: Arrange multiple microphones around the reducer to collect noise time-domain signals in different vibration transmission path directions of the reducer.

[0037] Step S2: Perform short-time Fourier transform on the noise time-domain signals of the collected reducer in each direction to convert the noise time-domain signals into noise frequency-domain signals, obtain the spectrum of the frequency-domain signals in each direction, retain the amplitude and phase of the spectrum, and calculate the sound pressure level based on the amplitude.

[0038] Step S3: Based on the spectrum results obtained in step S2, obtain the intermediate bandwidth, upper critical bandwidth, and lower critical bandwidth with each frequency point of the spectrum as the center frequency, and calculate the prominence ratio at each frequency point by combining the sound pressure level in step S2.

[0039] Step S4: Set a prominence ratio threshold to filter the noise frequency domain signal obtained in step S2, and retain the amplitude and phase of the intermediate bandwidth frequency range corresponding to the prominence ratio greater than the prominence ratio threshold.

[0040] Step S5: The signal frequency band retained in step S4 is reconstructed using inverse Fourier transform to obtain the filtered noise time-domain signal.

[0041] Step S6: Perform amplitude normalization processing on the signal obtained in step S5;

[0042] Step S7: Use blind source separation technology to separate the normalized noise signal into independent components, and obtain the separated noise signals sorted by energy level.

[0043] Step S8: Select several separated noise signals with relatively high energy for frequency analysis, and filter out multiple frequencies with a relatively high energy proportion.

[0044] Step S9: Calculate the order corresponding to the frequency selected in step S8 based on the rotational speed of the reducer input shaft, thereby obtaining the order that causes the reducer to whine.

[0045] The method for tracing the source of whistling noise in a reducer according to the present invention is as follows: Multiple microphones are arranged at a certain distance from the reducer to collect noise signals along different vibration transmission paths of the reducer; the collected noise signals are converted from the time domain to the frequency domain using a short-time Fourier transform, with a rectangular window applied to the noise signals in each direction during the short-time Fourier transform; the amplitude and phase of the spectral data in each direction are retained, and the sound pressure level is calculated using the amplitude to obtain the sound pressure level at each frequency point. The sound pressure level at each frequency point is used to calculate the overlap ratio; therefore, a threshold for the overlap ratio needs to be defined. Overlap ratios below the threshold are considered satisfactory, and the amplitude and phase of frequency bands above the overlap ratio threshold are retained. An inverse Fourier transform is used to reconstruct the retained frequency bands, achieving adaptive filtering of the original signal. The filtered signal undergoes amplitude normalization to eliminate the influence of amplitude on subsequent processing. Each channel's filtered signal contains the dominant frequency of the reducer's whistling. Therefore, blind source separation technology is used to independently separate the noise signals in each direction, extract the comprehensive noise components, sort them according to their energy proportion, select the signals with the highest energy for frequency analysis, and combine the reducer's input shaft speed for main order analysis, thereby achieving the source tracing of the reducer's whistling.

[0046] In step S1, microphones are arranged around the measuring surface of the electric drive reducer to synchronously collect steady-state operating speed and synchronous audio data of the reducer under load. The microphones can be 1000mm away from the measuring surface. In some embodiments, one microphone is arranged on the upper side, left side and right side of the reducer, for a total of three microphones.

[0047] In step S2, short-time Fourier transforms are performed on the noise data of the reducer in each direction. A rectangular window can be selected during data segmentation to retain the amplitude and phase of the spectrum obtained after the short-time Fourier transform, and the sound pressure level corresponding to each frequency amplitude is calculated. Because short-time Fourier transform is used, the problem of reducer noise signals being easily interfered with by external noise is effectively solved, thus achieving effective suppression of external interference noise.

[0048] In step S3, calculating the prominence ratio requires determining the intermediate bandwidth, upper critical bandwidth, and lower critical bandwidth corresponding to the center frequency. This bandwidth can be calculated according to the T / CSAE 176-2021 standard. The prominence ratio at each frequency is calculated based on the known sound pressure level of each channel's signal spectrum. In a specific embodiment, the center frequency range for calculating the prominence ratio is 90Hz to 7284Hz, and the corresponding intermediate bandwidth, upper critical bandwidth, and lower critical bandwidth are calculated at 1Hz intervals. The prominence ratio is the decibel difference between the average of the intermediate bandwidth and the upper and lower critical bandwidths. After obtaining the sound pressure level of each channel, the prominence ratio of each channel from 90Hz to 7284Hz can be calculated.

[0049] A threshold for the enhancement ratio needs to be set to evaluate howling noise; the lower the enhancement ratio, the less noticeable the howling. The amplitude and phase of frequency bands with enhancement ratios greater than the threshold are retained. In an optional embodiment of the invention, the enhancement ratio threshold is set to 15dB. When a channel has enhancement ratios greater than the threshold between 90Hz and 7284Hz, the corresponding frequency band with the intermediate bandwidth of that frequency is retained, i.e., the amplitude and phase of the corresponding frequency band are retained. Because a howling evaluation standard and threshold screening method based on enhancement ratio are adopted, the problem of irrelevant frequency components increasing algorithm complexity is effectively solved, thereby improving the accuracy of howling source tracing.

[0050] In step S5, the inverse Fourier transform is performed on the spectrum retained by each channel, and the reconstructed signal is the time-domain signal after adaptive filtering of the original signal.

[0051] Then, step S6 performs amplitude normalization on the filtered signal to avoid the influence of amplitude on subsequent processing steps. In some embodiments, the amplitude and phase retained by each channel need to be padded with negative frequency components. The negative frequency components are conjugate symmetrical with the retained frequency bands, and the remaining unretained frequency bands are padded with 0. The new spectrum is then subjected to an inverse Fourier transform to obtain the real part of the signal, thus obtaining the reconstructed signal. The length of the reconstructed signal is 65536. At this time, the amplitudes of the signals in each channel are not the same. Amplitude normalization ensures that the amplitude range is between [-1, 1].

[0052] For step S7, the blind source separation algorithm effectively solves the problem of the connection between noises in different directions of the reducer, thereby realizing the independent component separation of noises in different vibration transmission path directions of the reducer. The FastICA algorithm can be used to perform blind source separation on the reconstructed signals of each channel and extract characteristic frequencies. In a preferred embodiment, the reconstructed signals of each channel are constructed as vector signals X, and the signals are separated using FastICA. Before performing ICA, the signals need to be centered and whitened. Centering involves subtracting the mean value from the reconstructed signal vector X of each channel, turning it into a zero-mean vector. Assume the centered reconstructed signals of each channel are... The reconstructed signals of each channel after whitening are Z. The blind source separation problem requires obtaining the separation matrix W, that is, finding a separation matrix that maximizes the independence of the separated signals. The separation matrix W is obtained by maximizing the non-Gaussianity, and then Y = WZ is solved. Y is the separated noise signal, and the signals are sorted according to their energy.

[0053] The above algorithm is merely an example and is not intended to limit the present invention. The present invention can also employ other blind source separation algorithms for signal separation, such as Independent Vector Analysis (IVA) and Non-negative Matrix Factorization (NMF).

[0054] The energy percentage of a single isolated signal in the overall isolated signal indicates the contribution of that isolated noise signal to the overall howling noise. Therefore, in step S8, isolated signals with high energy percentages are selected for frequency analysis. In particular, the top five isolated signals with the highest energy percentages can be selected for frequency analysis. The frequencies are sorted by amplitude. If there are the same frequencies, they are counted according to the highest amplitude.

[0055] Then, in step S9, the order corresponding to each frequency is calculated based on the input shaft speed of the electric drive reducer, thereby obtaining the maximum order causing the reducer's squealing. Specifically, the meshing order of the meshing gear pairs on the input shaft, intermediate shaft, and output shaft can be calculated first based on the reducer structure and the number of teeth in the meshing gear pairs. Then, the harmonic orders corresponding to each order are calculated sequentially. The maximum order obtained from tracing the squealing source is matched with the calculated order to complete the source tracing.

[0056] As described above, the reducer howling source tracing method based on blind source separation of the present invention collects noise signals from different vibration transmission paths of the reducer, uses a prominence ratio evaluation standard combined with short-time Fourier transform and inverse Fourier transform to adaptively filter the noise signals in each direction, and uses blind source separation technology to independently separate the components of the filtered signals in each direction, extracting the characteristic frequencies that contribute the most to the reducer howling. By analyzing the dominant frequency order, the source of the reducer howling is traced. Therefore, the present invention takes into account the relationship between noise in different vibration transmission paths of the reducer, making the howling source tracing analysis comprehensive and enabling a comprehensive and accurate howling source tracing.

[0057] To more clearly illustrate the reducer whistling source tracing method based on blind source separation of the present invention, the following examples are provided:

[0058] The test object was a two-stage fixed-axis reducer with 29 teeth on the input shaft gear, 80 teeth on the intermediate shaft large gear, 21 teeth on the intermediate shaft small gear, and 76 teeth on the output shaft gear. The input shaft speed was constant at 3000 rpm, and the load was 120 Nm. Microphones were placed at a distance of 1 meter from the reducer, on its top, left, and right sides.

[0059] After acquiring the noise time-domain signals in different vibration transmission path directions, short-time Fourier transforms were performed on the acoustic signals of the three channels respectively. The data segmentation adopted a rectangular window with a window length of 65536, and the corresponding frequency resolution at a sampling rate of 20480Hz was 20480 / 65536Hz.

[0060] Then, according to the T / CSAE 176-2021 standard, the bandwidth corresponding to each center frequency is calculated. The center frequency range for calculating the standout ratio is 90Hz to 7284Hz. The corresponding intermediate bandwidth, upper critical bandwidth, and lower critical bandwidth are calculated at 1Hz intervals. After obtaining the sound pressure level at each frequency point of each channel, the standout ratio of each channel from 90Hz to 7284Hz can be calculated. Figure 2 The noise signal prominence ratios are shown in three directions: top, left, and right of the reducer. It can be seen that the prominence ratio amplitudes differ in different directions. If the reducer squealing is traced only in a single direction, the tracing results will be inaccurate.

[0061] Next, filtering is performed, with a surge ratio threshold set at 15dB. The amplitude and phase of frequency bands with a surge ratio greater than this threshold are retained. Then, the retained spectra of each channel are subjected to inverse Fourier transform, and after amplitude normalization, FastICA is used to perform blind source separation on the reconstructed signals of each channel, resulting in separated noise signals sorted by energy magnitude.

[0062] Finally, the top 5 separation signals with the largest energy proportions in blind source separation were selected, and the dominant frequencies of each separation signal were analyzed using Fourier transform. The frequencies in each separation signal were then sorted according to their amplitude. If there were the same frequency, the one with the highest amplitude was used. After obtaining the top 5 main influencing frequencies, the order corresponding to each frequency can be calculated based on the input shaft rotation speed.

[0063] like Figure 3 The figure shows the calculated comprehensive source of the whistling noise, with the dominant order being 15.2. First, based on the reducer structure and the number of teeth in the meshing gear pairs, the meshing orders of the two pairs of meshing gears on the input shaft and intermediate shaft, and the intermediate shaft and output shaft, are calculated to be 29 and 7.6125, respectively. Analysis reveals that the dominant order of 15.2 is the second harmonic of the meshing order of the gear pairs on the intermediate shaft and output shaft. Therefore, the source of the whistling noise can be identified as this particular gear pair. To reduce the whistling noise, targeted optimization of this gear pair is necessary.

[0064] In summary, the experiments have verified the applicability and effectiveness of the reducer whistling source tracing method based on blind source separation provided by this invention.

[0065] Finally, it should be noted that the features mentioned and / or shown in the above description of exemplary embodiments of the present invention can be combined in the same or similar manner with one or more other embodiments, combined with features in other embodiments, or substituted for corresponding features in other embodiments. These combined or substituted technical solutions should also be considered to be included within the scope of protection of the present invention.

Claims

1. A method for tracing the source of screeching from a reducer based on blind source separation, characterized in that, Includes the following steps: Step 1: Arrange multiple microphones around the reducer to collect noise time-domain signals in different vibration transmission path directions of the reducer. Step 2: Perform short-time Fourier transform on the noise time-domain signals of the collected reducer in each direction to convert the noise time-domain signals into noise frequency-domain signals, obtain the spectrum of the frequency-domain signals in each direction, retain the amplitude and phase of the spectrum, and calculate the sound pressure level based on the amplitude. Step 3: Based on the spectrum results obtained in Step 2, obtain the intermediate bandwidth, upper critical bandwidth, and lower critical bandwidth with each frequency point of the spectrum as the center frequency, and calculate the prominence ratio at each frequency point by combining the sound pressure level in Step 2. Step 4: Set a prominence ratio threshold to filter the noise frequency domain signal obtained in Step 2, and retain the amplitude and phase of the intermediate bandwidth frequency range corresponding to the prominence ratio greater than the prominence ratio threshold. Step 5: Reconstruct the signal frequency band retained in Step 4 using inverse Fourier transform to obtain the filtered noise time-domain signal. Step 6: Perform amplitude normalization processing on the signal obtained in Step 5; Step 7: Use blind source separation technology to separate the normalized noise signal into independent components, and obtain the separated noise signals sorted by energy level; Step 8: Select several separated noise signals with relatively high energy for frequency analysis, and filter out multiple frequencies with a large energy proportion. Step 9: Calculate the order corresponding to the frequency selected in Step 8 based on the rotational speed of the reducer input shaft, thereby obtaining the order that causes the reducer to whine.

2. The method for tracing the source of reducer whistling based on blind source separation according to claim 1, characterized in that: In step 1, a microphone is placed on the upper, left, and right sides of the reducer, respectively.

3. The method for tracing the source of reducer whistling based on blind source separation according to claim 1 or 2, characterized in that: In step 3, the center frequency range for calculating the salience ratio is 90Hz to 7284Hz, and the corresponding intermediate bandwidth, upper critical bandwidth, and lower critical bandwidth are calculated at 1Hz intervals.

4. The method for tracing the source of reducer whistling based on blind source separation according to claim 1 or 2, characterized in that: In step 4, the salience ratio threshold is set to 15 dB.

5. The method for tracing the source of reducer whistling based on blind source separation according to claim 1 or 2, characterized in that: In step 7, the FastICA algorithm is used to separate the independent components of the noise signal.

6. The method for tracing the source of reducer whistling based on blind source separation according to claim 5, characterized in that, Step 7 includes: Step 7.1: Center the normalized noise signal; Step 7.2: Whiten the centered noise signal; Step 7.3: Solve for the separation matrix of the whitened noise signal; Step 7.4: Use a separation matrix to separate the noise signal.

7. The method for tracing the source of reducer whistling based on blind source separation according to claim 1 or 2, characterized in that: In step 8, the five separated noise signals with the top five energy percentages are selected for frequency analysis, and the five frequencies with the top five energy percentages are screened out.

Images