Gearbox gear detection equipment detection data acquisition method and system

Abstract

The present application belongs to the field of data acquisition, and relates to a gearbox gear detection equipment detection data acquisition method and system. The method comprises the following steps: collecting gearbox vibration and synchronous speed signals, dividing the vibration signals into frames, and performing noise reduction processing on each frame; estimating the composite noise spectrum based on the synchronous speed signal and harmonic analysis, combining the time-frequency kurtosis to adaptively select a multi-window function to calculate the noisy power spectrum, using the transient characteristics and prior signal-to-noise ratio to determine the nonlinear subtraction factor, performing spectral subtraction to obtain the target power spectrum, comparing the deviation of the actual meshing frequency and the theoretical value to dynamically determine whether to reconstruct the phase spectrum; and reconstructing all processed data frames by overlapping addition to obtain the detection signal. The present application can comprehensively suppress harmonic and random noise, and conditionally reconstruct key phase information, thereby improving the recognition degree and diagnostic accuracy of early weak fault features of the gearbox.

Description

Technical Field

[0001] This invention belongs to the field of data acquisition, and in particular relates to a method and system for acquiring detection data from a gearbox gear testing device. Background Technology

[0002] As the core component of power transmission, the gearbox's internal gears' operating condition directly determines the overall reliability and service life of the equipment. During long-term meshing, gears are prone to failures such as tooth surface spalling, cracks, and tooth breakage due to wear, fatigue, and impact. If these failures are not diagnosed in time, they may trigger a chain reaction of failures, leading to equipment downtime, production interruptions, and significant economic losses. Therefore, achieving early diagnosis of gear failures is a crucial step in avoiding sudden equipment downtime and reducing maintenance costs.

[0003] Currently, non-destructive testing (NDT) technology based on vibration signal analysis has become the mainstream approach for gearbox gear fault diagnosis due to its advantages such as high real-time performance, non-invasiveness, and ability to reflect the dynamic characteristics of gear meshing. Its core principle is to collect vibration signals generated by gear meshing and extract characteristic information such as frequency, amplitude, and phase from the signals using industrial information and data processing technologies, thereby determining whether a gear fault exists and its type. However, under actual operating conditions, the vibration signals of a gearbox typically exhibit strong non-stationary characteristics and extremely low signal-to-noise ratios. The signals not only contain transient impact components reflecting localized gear damage but also harmonic and sideband interference from other rotating components such as shafts, bearings, and other gear pairs, as well as random noise generated by the background environment. These complex noise components often obscure early, subtle fault characteristics, posing a significant challenge to feature extraction in industrial information and data processing.

[0004] To address noise interference, existing technologies in industrial information and data processing often employ time-frequency analysis, adaptive filtering, and spectral subtraction to process gearbox vibration signals. The core assumption of traditional spectral subtraction is that the noise is a stationary signal, achieving noise reduction in the frequency domain through a pre-defined fixed noise spectrum model. However, gearbox harmonic noise dynamically changes with engine speed, and a fixed noise spectrum model cannot accurately match the actual noise distribution, resulting in a significant amount of residual harmonic interference after noise reduction. Furthermore, the spectral subtraction process destroys the phase information of the original signal, which is crucial for characterizing the transient impact characteristics of gear faults; phase distortion directly leads to misjudgment or missed detection of fault features. In the power spectrum estimation stage, commonly used methods such as the short-time Fourier transform suffer from spectral leakage, affecting frequency resolution and the accuracy of amplitude estimation, further impacting the feature extraction accuracy of industrial information and data processing. Adaptive filtering relies on a reference signal that matches the interference signal to achieve noise reduction. However, under gearbox operating conditions, it is difficult to obtain a reference signal that is completely synchronized with the harmonic noise, resulting in poor suppression of speed-related harmonics and easy over-filtering of fault signals, causing the loss of effective features required for industrial information and data processing.

[0005] This shows that the shortcomings of traditional vibration signal processing technology in areas such as non-stationary harmonic noise suppression, phase information preservation, and power spectrum estimation accuracy have become key bottlenecks restricting the development of early fault diagnosis technology for gearbox gears. Summary of the Invention

[0006] In view of this, the purpose of the present invention is to provide a method and system for acquiring test data of a gearbox gear testing device, so as to solve the technical problems in gearbox gear testing, such as low signal-to-noise ratio caused by mixed non-stationary harmonic noise and random background noise, easy masking of early fault characteristics, and the fact that traditional processing methods have problems such as inaccurate noise suppression, destruction of phase information, insufficient power spectrum estimation accuracy, and difficulty in balancing strong noise suppression and weak fault feature preservation.

[0007] This invention proposes a data acquisition method for gearbox gear testing equipment, which can effectively solve the problem of handling non-stationary harmonic noise, including the following steps:

[0008] S1, acquire the vibration signal and synchronous speed signal of the gearbox, and divide the vibration signal into multiple data frames;

[0009] S2, for each data frame, perform the following operations:

[0010] Based on the synchronous speed signal, combined with harmonic analysis and minimum control recursive averaging method, the composite noise spectrum containing random noise and harmonic noise is estimated.

[0011] The number of multi-window functions is adaptively determined based on the time-frequency kurtosis characteristics of the data frame, and a weighted multi-window spectrum estimation is performed on the data frame to obtain the noisy power spectrum.

[0012] Combining the transient characteristics of the noisy power spectrum with the prior signal-to-noise ratio, a nonlinear subtraction factor is determined for each frequency point. The noisy power spectrum is then subjected to spectral subtraction using the composite noise spectrum to obtain the denoised power spectrum.

[0013] The theoretical meshing frequency is calculated based on the synchronous speed signal. The deviation between the instantaneous meshing frequency estimated based on the power spectrum after noise reduction and the theoretical meshing frequency is compared. When the deviation exceeds the dynamic threshold, the phase spectrum is reconstructed using an iterative algorithm to obtain the target phase spectrum. Otherwise, the original phase spectrum of the data frame is retained.

[0014] The denoised power spectrum and the target phase spectrum are combined to perform an inverse Fourier transform to obtain the processed data frame;

[0015] S3. All processed data frames are reconstructed using the overlapping and addition method to obtain the detection signal.

[0016] Further, in step S2, the method for obtaining the composite noise spectrum is as follows:

[0017] The fundamental frequency and the first 5 order rotational harmonic frequencies of the current data frame are determined using the synchronous rotational speed signal.

[0018] Based on the noisy power spectrum obtained by weighted multi-window spectrum estimation of the data frame, at each frequency point, the noisy power spectrum is smoothed by recursive averaging and the candidate values ​​of the noise spectrum are updated.

[0019] The minimum tracking noise estimate for each frequency point is updated by comparing the candidate values ​​with a preset minimum bias parameter.

[0020] By combining the fundamental frequency with the first 5 order rewind harmonic frequencies, the minimum tracking noise estimate is corrected, and the power of the harmonic components is superimposed on the corresponding frequency points to obtain the composite noise spectrum.

[0021] Further, in step S2, the method for obtaining the noisy power spectrum is as follows:

[0022] Calculate the spectral kurtosis values ​​of the data frame at different time points to obtain the kurtosis time series;

[0023] When the peak value in the kurtosis time series exceeds the kurtosis threshold, the number of multi-window functions is determined to be 2; otherwise, the number of multi-window functions is determined to be 5.

[0024] The discrete, elongated spherical sequence is used as the multi-window function to perform weighted summation and Fourier transform on the data frames to obtain the noisy power spectrum.

[0025] Further, in step S2, the method for obtaining the power spectrum after noise reduction is as follows:

[0026] Calculate each frequency point Prior signal-to-noise ratio Prior signal-to-noise ratio It is the ratio of the power spectrum of the noise-reduced data obtained after processing the previous data frame to the power spectrum of the composite noise of the current data frame.

[0027] Obtain spectral subtraction control parameters With noise spectrum lower limit factor ;

[0028] Based on the prior signal-to-noise ratio Determine the nonlinear subtraction factor that is inversely proportional to it. ;

[0029] The power spectrum after noise reduction is calculated using the following formula. :

[0030] ;

[0031] in, The power spectrum includes noise. This is the composite noise spectrum.

[0032] Further, in step S2, the method for obtaining the target phase spectrum is as follows:

[0033] Based on the number of gear teeth and synchronous speed signal Using the formula Calculate the theoretical meshing frequency ;

[0034] In the power spectrum after noise reduction, with the theoretical meshing frequency as the center, Search for the peak frequency within the frequency range and use the peak frequency as the instantaneous engagement frequency;

[0035] The dynamic threshold is set to 3% of the theoretical meshing frequency. When the absolute value of the deviation between the instantaneous meshing frequency and the theoretical meshing frequency exceeds the dynamic threshold, phase spectrum reconstruction is performed.

[0036] The phase spectrum is randomly initialized, and the inverse Fourier transform is performed on the power spectrum after noise reduction to obtain the time domain signal. The time domain signal is truncated and then subjected to Fourier transform again to extract a new phase spectrum. The process is iterated 100 times to obtain the target phase spectrum.

[0037] Furthermore, the technical solution of the transmission gear testing equipment data acquisition system provided by this invention is as follows:

[0038] A data acquisition system for a gearbox gear testing device includes the following units:

[0039] The acquisition unit acquires the vibration signal and synchronous speed signal of the gearbox, and divides the vibration signal into multiple data frames;

[0040] The processing unit performs the following operations for each data frame:

[0041] Based on the synchronous speed signal, combined with harmonic analysis and minimum control recursive averaging method, the composite noise spectrum containing random noise and harmonic noise is estimated.

[0042] The number of multi-window functions is adaptively determined based on the time-frequency kurtosis characteristics of the data frame, and a weighted multi-window spectrum estimation is performed on the data frame to obtain the noisy power spectrum.

[0043] Combining the transient characteristics of the noisy power spectrum with the prior signal-to-noise ratio, a nonlinear subtraction factor is determined for each frequency point. The noisy power spectrum is then subjected to spectral subtraction using the composite noise spectrum to obtain the denoised power spectrum.

[0044] The theoretical meshing frequency is calculated based on the synchronous speed signal. The deviation between the instantaneous meshing frequency estimated based on the power spectrum after noise reduction and the theoretical meshing frequency is compared. When the deviation exceeds the dynamic threshold, the phase spectrum is reconstructed using an iterative algorithm to obtain the target phase spectrum. Otherwise, the original phase spectrum of the data frame is retained.

[0045] The denoised power spectrum and the target phase spectrum are combined to perform an inverse Fourier transform to obtain the processed data frame;

[0046] The detection signal is obtained by reconstructing all processed data frames using an overlapping and adding method.

[0047] Furthermore, the processing unit obtains the composite noise spectrum using the following method:

[0048] The fundamental frequency and the first 5 order rotational harmonic frequencies of the current data frame are determined using the synchronous rotational speed signal.

[0049] Based on the noisy power spectrum obtained by weighted multi-window spectrum estimation of the data frame, at each frequency point, the noisy power spectrum is smoothed by recursive averaging and the candidate values ​​of the noise spectrum are updated.

[0050] The minimum tracking noise estimate for each frequency point is updated by comparing the candidate values ​​with a preset minimum bias parameter.

[0051] By combining the fundamental frequency with the first 5 order rewind harmonic frequencies, the minimum tracking noise estimate is corrected, and the power of the harmonic components is superimposed on the corresponding frequency points to obtain the composite noise spectrum.

[0052] Furthermore, the processing unit obtains the noisy power spectrum using the following method:

[0053] Calculate the spectral kurtosis values ​​of the data frame at different time points to obtain the kurtosis time series;

[0054] When the peak value in the kurtosis time series exceeds the kurtosis threshold, the number of multi-window functions is determined to be 2; otherwise, the number of multi-window functions is determined to be 5.

[0055] The discrete, elongated spherical sequence is used as the multi-window function to perform weighted summation and Fourier transform on the data frames to obtain the noisy power spectrum.

[0056] Furthermore, the processing unit obtains the power spectrum after noise reduction as follows:

[0057] Calculate each frequency point Prior signal-to-noise ratio Prior signal-to-noise ratio It is the ratio of the power spectrum of the noise-reduced data obtained after processing the previous data frame to the power spectrum of the composite noise of the current data frame.

[0058] Obtain spectral subtraction control parameters With noise spectrum lower limit factor ;

[0059] Based on the prior signal-to-noise ratio Determine the nonlinear subtraction factor that is inversely proportional to it. ;

[0060] The power spectrum after noise reduction is calculated using the following formula. :

[0061] ;

[0062] in, The power spectrum includes noise. This is the composite noise spectrum.

[0063] Furthermore, the processing unit acquires the target phase spectrum using the following method:

[0064] Based on the number of gear teeth and synchronous speed signal Using the formula Calculate the theoretical meshing frequency ;

[0065] In the power spectrum after noise reduction, with the theoretical meshing frequency as the center, Search for the peak frequency within the frequency range and use the peak frequency as the instantaneous engagement frequency;

[0066] The dynamic threshold is set to 3% of the theoretical meshing frequency. When the absolute value of the deviation between the instantaneous meshing frequency and the theoretical meshing frequency exceeds the dynamic threshold, phase spectrum reconstruction is performed.

[0067] The phase spectrum is randomly initialized, and the inverse Fourier transform is performed on the power spectrum after noise reduction to obtain the time domain signal. The time domain signal is truncated and then subjected to Fourier transform again to extract a new phase spectrum. The process is iterated 100 times to obtain the target phase spectrum.

[0068] The beneficial effects of this invention are:

[0069] This invention estimates the composite noise spectrum by combining harmonic analysis and the minimum control recursive averaging method, simultaneously suppressing strong harmonic interference from other rotating components and background random noise, thus solving the problem of handling non-stationary harmonic noise in existing technologies. Simultaneously, this invention adaptively determines the number of multi-window functions for weighted multi-window spectrum estimation based on the time-frequency kurtosis characteristics of the data frame, reducing spectral leakage and improving the accuracy of power spectrum estimation. The use of a nonlinear spectral subtraction strategy reduces noise generation and minimizes excessive attenuation of the fault impact signal amplitude. Furthermore, this invention conditionally reconstructs the phase spectrum, mitigating the severe damage to signal phase information caused by spectral subtraction, effectively preserving and enhancing key phase information representing transient impact characteristics. The reconstructed detection signal has a high signal-to-noise ratio, and both amplitude-frequency and phase characteristics are guaranteed, improving the identification and diagnostic accuracy of early weak fault characteristics in the transmission. Attached Figure Description

[0070] Figure 1 This is a flowchart illustrating the steps of a data acquisition method for a gearbox gear testing device according to the present invention.

[0071] Figure 2 This is a schematic diagram of signal acquisition and framing.

[0072] Figure 3 This is a schematic diagram of composite noise spectrum estimation;

[0073] Figure 4 This is a schematic diagram of adaptive multi-window spectral estimation.

[0074] Figure 5 This is a schematic diagram of the nonlinear spectral subtraction process;

[0075] Figure 6 This is a schematic diagram of phase reconstruction decision based on meshing frequency deviation. Detailed Implementation

[0076] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0077] A specific embodiment of the data acquisition method for a gearbox gear testing device provided by the present invention:

[0078] like Figure 1 As shown, a method for acquiring test data from a gearbox gear testing device includes the following steps:

[0079] S1, acquire the vibration signal and synchronous speed signal of the gearbox, and divide the vibration signal into multiple data frames.

[0080] In this step, vibration signals are acquired using an accelerometer mounted on the gearbox housing, and synchronous speed signals are acquired using an optical encoder or Hall sensor mounted on the input or output shaft of the gearbox. The acquired continuous vibration signals are then segmented; for example, a data frame length of 4096 sampling points is obtained, with a 50% overlap between adjacent data frames, resulting in 2048 sampling points. Each data frame is multiplied by a window function, such as a Hamming window, to reduce spectral leakage. Figure 2 As shown.

[0081] S2, for each data frame, perform the following operations:

[0082] S21, Based on the synchronous speed signal, combined with harmonic analysis and minimum control recursive averaging method, the composite noise spectrum containing random noise and harmonic noise is estimated;

[0083] S22, based on the time-frequency kurtosis characteristics of the data frame, adaptively determine the number of multi-window functions, perform weighted multi-window spectrum estimation on the data frame, and obtain the noisy power spectrum;

[0084] S23, combining the transient characteristics of the noisy power spectrum with the prior signal-to-noise ratio, determine the nonlinear subtraction factor for each frequency point, and use the composite noise spectrum to perform spectral subtraction on the noisy power spectrum to obtain the noise-reduced power spectrum;

[0085] S24, calculate the theoretical meshing frequency based on the synchronous speed signal, compare the deviation between the instantaneous meshing frequency estimated based on the power spectrum after noise reduction and the theoretical meshing frequency, and when the deviation exceeds the dynamic threshold, reconstruct the phase spectrum using an iterative algorithm to obtain the target phase spectrum; otherwise, retain the original phase spectrum of the data frame.

[0086] S25, perform an inverse Fourier transform on the power spectrum after noise reduction and the target phase spectrum to obtain the processed data frame.

[0087] Specifically, in step S21, the fundamental frequency of the input or output shaft is calculated based on the synchronous rotational speed signal corresponding to the current data frame, thereby determining the theoretical frequency positions of all harmonic components. The energy of these harmonic components is identified and extracted from the noisy power spectrum to form a harmonic noise spectrum. Simultaneously, a minimum-value controlled recursive averaging algorithm is used to process the power spectra of multiple consecutive data frames, estimating the power spectrum of random background noise by tracking and smoothing the minimum value of each frequency point over a period of time. The harmonic noise spectrum and the random noise spectrum are then added point-by-point in the frequency domain to obtain the composite noise spectrum of the data frame, such as... Figure 3 As shown.

[0088] In an optional embodiment, the composite noise spectrum is obtained as follows:

[0089] The fundamental frequency and the first 5 order rotational harmonic frequencies of the current data frame are determined using the synchronous speed signal obtained in step S1.

[0090] Based on the noisy power spectrum obtained by weighted multi-window spectrum estimation of the above data frames, at each frequency point, the noisy power spectrum is smoothed by recursive averaging and the candidate values ​​of the noise spectrum are updated.

[0091] The minimum tracking noise estimate for each frequency point is updated by comparing the above candidate values ​​with a preset minimum bias parameter.

[0092] By combining the fundamental frequency and the first 5 order resonant frequencies, the minimum tracking noise estimate is corrected, and the power of the harmonic components is superimposed on the corresponding frequency points to obtain the composite noise spectrum.

[0093] The synchronous speed signal is a time series reflecting the real-time speed of a certain shaft in the gearbox. The average speed value of each data frame to be processed within the time span is calculated and denoted as d (in revolutions per minute). The average speed d is converted to a rotational frequency in Hertz by dividing by 60, which is the fundamental frequency of the current data frame. Its calculation formula is The first 5 order reciprocating harmonic frequencies Then by using the base frequency Multiplying by the corresponding harmonic order q (q=1, 2, 3, 4, 5) respectively yields the following: .

[0094] In a preferred embodiment of the present invention, each frequency point The minimum value tracking noise estimation update is an iterative process. In each data frame, the smoothed noisy power spectrum, i.e., the candidate value... With a preset minimum bias parameter Multiply by the current minimum tracking noise estimate The product is compared. If the candidate value Less than the product, that is This indicates that the energy at the current frequency point may have reached a new noise trough. At this point, the minimum value is used to track the noise estimation. Update directly to the smaller candidate value, i.e. Conversely, the current frequency point may contain signal components or noise is slowly increasing, requiring minimum tracking noise estimation. Increase slowly at a preset, small growth rate to ensure that it can track slow changes in non-stationary noise while avoiding being mistakenly boosted by sudden signal energy.

[0095] Based on the determined fundamental frequency and the first 5 harmonic frequencies In the frequency domain, the frequency points or frequency bands corresponding to these harmonic frequencies are located. The power value of the noisy power spectrum obtained from the original weighted multi-window spectrum estimation at the harmonic frequency points is taken as the power contribution of the harmonic noise at that location. The harmonic power values ​​are superimposed or replaced to the minimum value to track the noise estimation. At the corresponding positions. A preferred implementation is at each harmonic frequency point. The corrected composite noise spectrum will be displayed at this location. The value of ) is directly set to the value of the noisy power spectrum. ,Right now At all other non-harmonic frequency points At that point, its original minimum tracking estimate is maintained, i.e. The obtained composite noise spectrum It contains both a smooth random noise floor and sharp harmonic noise peaks.

[0096] For example, if the synchronous speed signal measures a speed of 1800 revolutions per minute, the fundamental frequency is 30 Hz, locking in five harmonic frequencies: 30 Hz, 60 Hz, 90 Hz, 120 Hz, and 150 Hz. When processing the noisy power spectrum, for non-harmonic frequency points, such as 45 Hz, assuming the recursively smoothed noise spectrum candidate value is 0.02 units of power, and the minimum tracking noise estimate is 0.015 units of power, with a preset bias parameter of 1.2, since 0.02 is greater than 0.018, the noise estimate at the frequency point is slowly increased to accurately track changing random background noise. After completing the initial estimation of random noise, corrections are made for the determined harmonic frequencies. In the above example, the minimum tracking algorithm might suppress the noise power at 90 Hz to 0.03 units, but in reality, harmonics are inherent noise from the steady-state operation of the equipment, with an energy of 0.5 units in the original noisy power spectrum. At this point, this 0.5 unit power is re-superimposed onto the noise estimate at the 90 Hz frequency, resulting in a composite noise spectrum value of 0.53 units at that point. This effectively tracks changes in random noise while preserving the inherent harmonic noise information of the rotating machinery, avoiding misinterpretation of harmonics generated during normal equipment operation as fault signals.

[0097] In step S22, a kurtosis threshold is set, and the temporal kurtosis value of the current data frame is calculated. For example, if the kurtosis threshold is set to 4, and the temporal kurtosis value of the current data frame is greater than 4, it indicates that the signal may contain strong impulse components. In this case, a smaller number of window functions, such as 3, is selected to retain more transient details. If the temporal kurtosis value of the current data frame is less than or equal to 4, a larger number of window functions, such as 5, is selected to obtain a smoother spectral estimate. After determining the number of window functions, a corresponding number of orthogonal discrete spherical sequences are generated as multiple window functions. These are multiplied by the data frame, and a Fourier transform is performed on each result to calculate the power spectrum. The weighted average of the multiple power spectra is then used to obtain the noisy power spectrum, such as... Figure 4 As shown.

[0098] In an optional embodiment, the noisy power spectrum is obtained as follows:

[0099] Calculate the spectral kurtosis values ​​of the data frame at different time points to obtain the kurtosis time series;

[0100] When the peak value in the kurtosis time series exceeds the kurtosis threshold, the number of the multi-window functions is determined to be 2;

[0101] Otherwise, the number of the multi-window functions is determined to be 5;

[0102] The discrete, elongated spherical sequence is used as the multi-window function to perform weighted summation and Fourier transform on the data frames to obtain the noisy power spectrum.

[0103] When monitoring the vibration signal of a gearbox, if the gear runs smoothly and there is no obvious impact component in the vibration signal, the peak value of the calculated kurtosis time series may be 3.5. Since 3.5 does not exceed the set kurtosis threshold, such as 4.0, the vibration signal is judged to be in a stationary state. In order to obtain more refined frequency information to identify weak fault characteristics, five discrete flattened spherical sequences are selected as multi-window functions for spectral estimation, which can suppress spectral leakage and improve frequency resolution. Among them, the discrete flattened spherical sequences are also called Slepian sequences. Assuming the sequence length N=256 and the time-half-bandwidth product NW=4, a set of discrete flattened spherical sequences can be generated. The first-order sequence (q=1) has a shape very close to a Gaussian window or Hanning window in the time domain, with energy highly concentrated in the center and smoothly decaying to zero at both ends. Its main lobe in the frequency domain is the narrowest, and its sidelobe suppression capability is extremely strong. The second-order sequence (q=2) will have a zero-crossing point near the center of the time domain, and its shape is antisymmetric, with a slightly wider main lobe in the spectrum. Higher-order sequences (such as q=5) will have more zero-crossings, more oscillating time-domain waveforms, and wider main lobes in the spectrum, but still maintain strict orthogonality. In adaptively determining the number of window functions, this invention selects the first two or five of the discrete elongated spherical sequences as the window function set based on the characteristics of the vibration signal.

[0104] When a gear experiences localized spalling or tooth breakage, a strong transient impact will appear in the vibration signal. At this time, the calculated kurtosis time series peak value may surge to 7.2, exceeding the threshold of 4.0. Based on the peak value data, it is determined that the signal contains an important transient event. To obtain the timing of the impact, the number of window functions is reduced to 2. Using fewer window functions improves the temporal resolution, although it sacrifices some frequency resolution. This is crucial for pinpointing the timing of impact failures, ensuring that the optimal spectral estimation strategy can be employed regardless of whether it is smooth wear or a sudden impact.

[0105] In step S23, the prior signal-to-noise ratio (SNR) at each frequency point is estimated using a decision-oriented method. Simultaneously, the local energy concentration of the noisy power spectrum is analyzed to determine its transient characteristics. A nonlinear function related to the prior SNR is designed to calculate the basic subtraction factor. The nonlinear function has a smaller subtraction force when the SNR is high and a larger subtraction force when the SNR is low. The basic subtraction factor is corrected based on the transient characteristics. For frequency points where transient impacts are detected, the basic subtraction factor is further reduced to protect the impact information. After correction, the basic subtraction factor becomes a nonlinear subtraction factor. This nonlinear subtraction factor is multiplied by the composite noise spectrum, and the result is subtracted from the noisy power spectrum. This process removes the noise component from the original noisy power spectrum, leaving a clean, denoised power spectrum. In this process, to prevent meaningless negative values ​​from appearing in the calculated power spectrum after noise reduction due to overestimation of noise, thus introducing distortion, a positive lower limit for the spectrum is set. If the result of spectral subtraction is less than this lower limit, then this preset lower limit is directly used as the final result. Figure 5 As shown.

[0106] In one optional embodiment, the power spectrum after noise reduction is obtained as follows:

[0107] Calculate each frequency point Prior signal-to-noise ratio That is, the ratio of the power spectrum of the noise-reduced data obtained after processing the previous data frame to the power of the composite noise spectrum of the current data frame.

[0108] Obtain spectral subtraction control parameters With noise spectrum lower limit factor ;

[0109] Based on the prior signal-to-noise ratio Determine the nonlinear subtraction factor that is inversely proportional to it. ;

[0110] The power spectrum after noise reduction is calculated using the following formula. :

[0111] ;

[0112] in, The power spectrum includes noise. This is the composite noise spectrum.

[0113] For example, suppose at a frequency point At this location, the signal energy is very weak, and the prior signal-to-noise ratio is calculated to be 0.5, indicating that... The location is likely pure noise. A large nonlinear subtraction factor is calculated based on the calculated value, for example... It equals 2.5. A preferred implementation is to use the Sigmoid function or a similar sigmoid function, for example... ,in This is the prior signal-to-noise ratio converted to decibels, and a and b are adjustment constants for the steepness and center offset of the control function curve. If the spectral subtraction control parameters... If we set it to 1.5, then the amount of noise subtracted from the noisy power spectrum at the frequency point is 1.5 multiplied by 2.5 and then multiplied by the noise spectrum power at the frequency point, thus achieving noise suppression.

[0114] At another frequency point At this frequency point, there is a distinct vibration peak caused by early bearing failure, with a calculated prior signal-to-noise ratio as high as 15. This indicates that at this frequency point... It has a strong signal component. Therefore, a very small nonlinear subtraction factor is calculated, for example... The value equals 0.1. At this point, the noise amount subtracted from the noisy power spectrum is only 1.5 multiplied by 0.1 and then by the noise spectrum power, resulting in very little attenuation of the original signal and completely preserving the amplitude information of the fault characteristics. Simultaneously, a lower limit factor for the noise spectrum is set. A value of 0.02 ensures that even in extremely weak frequency bands, a small amount of energy is retained after spectral subtraction, avoiding unnatural musical noise or data gaps in analysis, thus guaranteeing the continuity and quality of the output signal.

[0115] Preferably, the spectral reduction control parameter With noise spectrum lower limit factor Through offline experiments and optimized pre-settings, a sample dataset containing gearbox noise and fault signals under typical operating conditions was selected, and a series of settings were established. and Candidate value combinations, for example, The range can be set to [1, 5]. The range can be set to [0.01, 0.1]. Different parameter combinations are used to process the sample data, and then the processing effect under different parameter settings is comprehensively evaluated through a combination of objective evaluation indicators and subjective evaluation. The set that best represents the fault characteristics while introducing the least audible distortion is selected. and The value is used as a fixed configuration parameter in this invention.

[0116] In step S24, the theoretical meshing frequency is calculated based on the synchronous speed signal and the gear tooth count parameters. On the denoised power spectrum, an energy peak point is searched around the theoretical meshing frequency, and this peak point is used as the estimated instantaneous meshing frequency. A dynamic threshold is obtained, for example, 2% of the theoretical meshing frequency. If the deviation between the estimated instantaneous meshing frequency and the theoretical meshing frequency is greater than the dynamic threshold, the original phase is considered severely contaminated. At this point, the Griffin-Lin iterative algorithm is initiated, using the denoised power spectrum as an amplitude constraint. Starting from a random phase, through repeated Fourier forward and inverse transforms and time-domain constraints, iterates for, for example, 100 times to reconstruct a target phase spectrum that matches the target power spectrum. If the deviation does not exceed the dynamic threshold, the phase spectrum obtained by performing a Fourier transform on the original data frame is directly used as the target phase spectrum. Figure 6 As shown.

[0117] In an optional embodiment, the target phase spectrum is obtained as follows:

[0118] Based on the number of gear teeth and synchronous speed signal Using the formula Calculate the theoretical meshing frequency ;

[0119] In the power spectrum after noise reduction, with the theoretical meshing frequency as the center, Search for the peak frequency within the frequency range and use the peak frequency as the instantaneous engagement frequency;

[0120] The dynamic threshold is set to 3% of the theoretical meshing frequency. When the absolute value of the deviation between the instantaneous meshing frequency and the theoretical meshing frequency exceeds the dynamic threshold, the Griffin-Lim algorithm is used to perform phase spectrum reconstruction.

[0121] The phase spectrum is randomly initialized, and the inverse Fourier transform is performed on the power spectrum after noise reduction to obtain the time domain signal. The time domain signal is truncated and then subjected to Fourier transform again to extract a new phase spectrum. The process is iterated 100 times to obtain the target phase spectrum.

[0122] For example, a gear with 40 teeth operates at 1200 rpm, and the theoretical meshing frequency is calculated to be 800 Hz. The dynamic threshold is set to 3% of 800 Hz, or 24 Hz. Analysis of the post-denoised power spectrum reveals that within the search range of 785 Hz to 815 Hz, the energy peak occurs at 770 Hz. The absolute value of the deviation between the instantaneous meshing frequency and the theoretical meshing frequency is 30 Hz, exceeding the dynamic threshold of 24 Hz. This significant frequency shift is determined to be caused by severe tooth wear or broken teeth, rendering the phase information of the original signal unreliable after strong denoising. Therefore, phase spectrum reconstruction is triggered, and the Griffin-Lim algorithm is initiated. The Griffin-Lim algorithm is based on an amplitude spectrum enhanced around 800 Hz, and through 100 iterations, it continuously adjusts a randomly initialized phase spectrum until the amplitude spectrum of the generated time-domain signal when transformed back to the frequency domain is highly consistent with the target amplitude spectrum, thus obtaining a target phase spectrum that matches the enhanced amplitude spectrum.

[0123] In step S25, the square root of each frequency point of the power spectrum after noise reduction is performed to obtain the target amplitude spectrum. The target amplitude spectrum is combined with the target phase spectrum obtained in step S24 to form a complete complex spectrum. An inverse fast Fourier transform is performed on the complex spectrum to convert the signal from the frequency domain back to the time domain, resulting in a processed data frame.

[0124] S3. All processed data frames are reconstructed using the overlapping and addition method to obtain the detection signal.

[0125] In this step, the first processed data frame is used as the starting segment of the reconstructed signal. The beginning of the second processed data frame is aligned with the overlapping portion of the first data frame, and the two segments are added point by point. The non-overlapping portions are then joined to the next segment. This process is repeated until all processed data frames are joined together in their original time order, by adding the overlapping regions point by point, to form a continuous time-series signal, which is the detection signal.

[0126] The present invention provides a method for acquiring detection data in a gearbox gear testing device, which can obtain a detection signal with high signal-to-noise ratio and guaranteed amplitude-frequency and phase characteristics. It effectively solves the problem of difficult processing of non-stationary harmonic noise and overcomes the defect of traditional noise reduction methods that damage key phase information, thereby improving the accuracy of gearbox fault diagnosis.

[0127] A specific embodiment of the data acquisition system for a gearbox gear testing device provided by the present invention:

[0128] A data acquisition system for a gearbox gear testing device includes the following units:

[0129] The acquisition unit acquires the vibration signal and synchronous speed signal of the gearbox, and divides the vibration signal into multiple data frames;

[0130] The processing unit performs the following operations for each data frame:

[0131] Based on the synchronous speed signal, combined with harmonic analysis and minimum control recursive averaging method, the composite noise spectrum containing random noise and harmonic noise is estimated.

[0132] The number of multi-window functions is adaptively determined based on the time-frequency kurtosis characteristics of the data frame, and a weighted multi-window spectrum estimation is performed on the data frame to obtain the noisy power spectrum.

[0133] Combining the transient characteristics of the noisy power spectrum with the prior signal-to-noise ratio, a nonlinear subtraction factor is determined for each frequency point. The noisy power spectrum is then subjected to spectral subtraction using the composite noise spectrum to obtain the denoised power spectrum.

[0134] The theoretical meshing frequency is calculated based on the synchronous speed signal. The deviation between the instantaneous meshing frequency estimated based on the power spectrum after noise reduction and the theoretical meshing frequency is compared. When the deviation exceeds the dynamic threshold, the phase spectrum is reconstructed using an iterative algorithm to obtain the target phase spectrum. Otherwise, the original phase spectrum of the data frame is retained.

[0135] The denoised power spectrum and the target phase spectrum are combined to perform an inverse Fourier transform to obtain the processed data frame;

[0136] The detection signal is obtained by reconstructing all processed data frames using an overlapping and adding method.

[0137] It should be noted that the data acquisition process of the acquisition unit is the same as the processing process of step S1 in the embodiment of the above-mentioned method for acquiring data of a gearbox gear testing device, and the processing process of the processing unit is the same as the processing process of steps S2-S3 in the embodiment of the above-mentioned method for acquiring data of a gearbox gear testing device. These will not be described in detail here.

[0138] In addition, the transmission gear testing equipment detection data acquisition system of the present invention also includes a processor and a memory. The memory stores a computer program, and the processor executes the computer program to implement the transmission gear testing equipment detection data acquisition method in the above embodiment.

[0139] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as resistive random access memory (RRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (EDRAM), high-bandwidth memory (HBM), hybrid memory cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented by computer-readable / executable instructions stored or otherwise maintained on such a computer-readable medium.

[0140] While various embodiments of the invention have been shown and described in this specification, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention.

Claims

1. A method for acquiring test data from a gearbox gear testing device, characterized in that, Includes the following steps: S1, acquire the vibration signal and synchronous speed signal of the gearbox, and divide the vibration signal into multiple data frames; S2, for each data frame, perform the following operations: S21, Based on the synchronous speed signal, combined with harmonic analysis and minimum control recursive averaging method, the composite noise spectrum containing random noise and harmonic noise is estimated; S22, based on the time-frequency kurtosis characteristics of the data frame, adaptively determine the number of multi-window functions, perform weighted multi-window spectrum estimation on the data frame, and obtain the noisy power spectrum; S23, combining the transient characteristics of the noisy power spectrum with the prior signal-to-noise ratio, determine the nonlinear subtraction factor for each frequency point, and use the composite noise spectrum to perform spectral subtraction on the noisy power spectrum to obtain the noise-reduced power spectrum; S24, calculate the theoretical meshing frequency based on the synchronous speed signal, compare the deviation between the instantaneous meshing frequency estimated based on the power spectrum after noise reduction and the theoretical meshing frequency, and when the deviation exceeds the dynamic threshold, reconstruct the phase spectrum using an iterative algorithm to obtain the target phase spectrum; otherwise, retain the original phase spectrum of the data frame. S25, perform an inverse Fourier transform on the power spectrum after noise reduction and the target phase spectrum to obtain the processed data frame; S3. All processed data frames are reconstructed using the overlapping and addition method to obtain the detection signal.

2. The data acquisition method for a gearbox gear testing device according to claim 1, characterized in that, In step S21, the method for obtaining the composite noise spectrum is as follows: The fundamental frequency and the first 5 order rotational harmonic frequencies of the current data frame are determined using the synchronous rotational speed signal. Based on the noisy power spectrum obtained by weighted multi-window spectrum estimation of the data frame, at each frequency point, the noisy power spectrum is smoothed by recursive averaging and the candidate values ​​of the noise spectrum are updated. The minimum tracking noise estimate for each frequency point is updated by comparing the candidate values ​​with a preset minimum bias parameter. By combining the fundamental frequency with the first 5 order rewind harmonic frequencies, the minimum tracking noise estimate is corrected, and the power of the harmonic components is superimposed on the corresponding frequency points to obtain the composite noise spectrum.

3. The data acquisition method for a gearbox gear testing device according to claim 2, characterized in that, In step S22, the method for obtaining the noisy power spectrum is as follows: Calculate the spectral kurtosis values ​​of the data frame at different time points to obtain the kurtosis time series; When the peak value in the kurtosis time series exceeds the kurtosis threshold, the number of multi-window functions is determined to be 2; otherwise, the number of multi-window functions is determined to be 5. The discrete, elongated spherical sequence is used as the multi-window function to perform weighted summation and Fourier transform on the data frames to obtain the noisy power spectrum.

4. The data acquisition method for a gearbox gear testing device according to claim 3, characterized in that, In step S23, the method for obtaining the power spectrum after noise reduction is as follows: Calculate each frequency point Prior signal-to-noise ratio Prior signal-to-noise ratio It is the ratio of the power spectrum of the noise-reduced data obtained after processing the previous data frame to the power spectrum of the composite noise of the current data frame. Obtain spectral subtraction control parameters With noise spectrum lower limit factor ; Based on the prior signal-to-noise ratio Determine the nonlinear subtraction factor that is inversely proportional to it. ; The power spectrum after noise reduction is calculated using the following formula. : ； in, The power spectrum includes noise. This is the composite noise spectrum.

5. The data acquisition method for a gearbox gear testing device according to claim 4, characterized in that, In step S24, the method for obtaining the target phase spectrum is as follows: Based on the number of gear teeth and synchronous speed signal Using the formula Calculate the theoretical meshing frequency ; In the power spectrum after noise reduction, with the theoretical meshing frequency as the center, Search for the peak frequency within the frequency range and use the peak frequency as the instantaneous engagement frequency; The dynamic threshold is set to 3% of the theoretical meshing frequency. When the absolute value of the deviation between the instantaneous meshing frequency and the theoretical meshing frequency exceeds the dynamic threshold, phase spectrum reconstruction is performed. The phase spectrum is randomly initialized, and the inverse Fourier transform is performed on the power spectrum after noise reduction to obtain the time domain signal. The time domain signal is truncated and then subjected to Fourier transform again to extract a new phase spectrum. The process is iterated 100 times to obtain the target phase spectrum.

6. A data acquisition system for a gearbox gear testing device, characterized in that, Includes the following units: The acquisition unit acquires the vibration signal and synchronous speed signal of the gearbox, and divides the vibration signal into multiple data frames; The processing unit performs the following operations for each data frame: Based on the synchronous speed signal, combined with harmonic analysis and minimum control recursive averaging method, the composite noise spectrum containing random noise and harmonic noise is estimated. The number of multi-window functions is adaptively determined based on the time-frequency kurtosis characteristics of the data frame, and a weighted multi-window spectrum estimation is performed on the data frame to obtain the noisy power spectrum. Combining the transient characteristics of the noisy power spectrum with the prior signal-to-noise ratio, a nonlinear subtraction factor is determined for each frequency point. The noisy power spectrum is then subjected to spectral subtraction using the composite noise spectrum to obtain the denoised power spectrum. The theoretical meshing frequency is calculated based on the synchronous speed signal. The deviation between the instantaneous meshing frequency estimated based on the power spectrum after noise reduction and the theoretical meshing frequency is compared. When the deviation exceeds the dynamic threshold, the phase spectrum is reconstructed using an iterative algorithm to obtain the target phase spectrum. Otherwise, the original phase spectrum of the data frame is retained. The denoised power spectrum and the target phase spectrum are combined to perform an inverse Fourier transform to obtain the processed data frame; The detection signal is obtained by reconstructing all processed data frames using an overlapping and adding method.

7. The data acquisition system for a gearbox gear testing device according to claim 6, characterized in that, The processing unit obtains the composite noise spectrum using the following method: The fundamental frequency and the first 5 order rotational harmonic frequencies of the current data frame are determined using the synchronous rotational speed signal. Based on the noisy power spectrum obtained by weighted multi-window spectrum estimation of the data frame, at each frequency point, the noisy power spectrum is smoothed by recursive averaging and the candidate values ​​of the noise spectrum are updated. The minimum tracking noise estimate for each frequency point is updated by comparing the candidate values ​​with a preset minimum bias parameter. By combining the fundamental frequency with the first 5 order rewind harmonic frequencies, the minimum tracking noise estimate is corrected, and the power of the harmonic components is superimposed on the corresponding frequency points to obtain the composite noise spectrum.

8. The data acquisition system for a gearbox gear testing device according to claim 7, characterized in that, The processing unit obtains the noisy power spectrum as follows: Calculate the spectral kurtosis values ​​of the data frame at different time points to obtain the kurtosis time series; When the peak value in the kurtosis time series exceeds the kurtosis threshold, the number of multi-window functions is determined to be 2; otherwise, the number of multi-window functions is determined to be 5. The discrete, elongated spherical sequence is used as the multi-window function to perform weighted summation and Fourier transform on the data frames to obtain the noisy power spectrum.

9. The data acquisition system for a gearbox gear testing device according to claim 8, characterized in that, The method by which the processing unit obtains the power spectrum after noise reduction is as follows: Calculate each frequency point Prior signal-to-noise ratio Prior signal-to-noise ratio It is the ratio of the power spectrum of the noise-reduced data obtained after processing the previous data frame to the power spectrum of the composite noise of the current data frame. Obtain spectral subtraction control parameters With noise spectrum lower limit factor ; Based on the prior signal-to-noise ratio Determine the nonlinear subtraction factor that is inversely proportional to it. ; The power spectrum after noise reduction is calculated using the following formula. : ； in, The power spectrum includes noise. This is the composite noise spectrum.

10. The data acquisition system for a gearbox gear testing device according to claim 6, characterized in that, The method by which the processing unit obtains the target phase spectrum is as follows: Based on the number of gear teeth and synchronous speed signal Using the formula Calculate the theoretical meshing frequency ; In the power spectrum after noise reduction, with the theoretical meshing frequency as the center, Search for the peak frequency within the frequency range and use the peak frequency as the instantaneous engagement frequency; The dynamic threshold is set to 3% of the theoretical meshing frequency. When the absolute value of the deviation between the instantaneous meshing frequency and the theoretical meshing frequency exceeds the dynamic threshold, phase spectrum reconstruction is performed. The phase spectrum is randomly initialized, and the inverse Fourier transform is performed on the power spectrum after noise reduction to obtain the time domain signal. The time domain signal is truncated and then subjected to Fourier transform again to extract a new phase spectrum. The process is iterated 100 times to obtain the target phase spectrum.

Images