Locomotive gear fault detection method, device and server

Abstract

The application provides a locomotive gear fault detection method and device and a server, relates to the technical field of locomotive gear fault diagnosis, and comprises the following steps: performing median filtering and mean filtering on an original time-frequency distribution to obtain noise estimation values of each time in a time-frequency distribution matrix of the original time-frequency distribution, and determining a true signal component discrimination threshold of each time; determining a first time-frequency ridge line based on the noise estimation values and the true signal component discrimination threshold, and establishing a dynamic optimization penalty function according to the first time-frequency ridge line to determine a second time-frequency ridge line; connecting and processing discontinuous time-frequency ridge lines with endpoint acceleration values through linear interpolation fitting to determine a repaired second continuous time-frequency ridge line, so as to determine a target time-frequency ridge line; and determining a locomotive gear fault detection result by using the target time-frequency ridge line and a gear transmission ratio relationship. The application can significantly enhance the extraction capability of the time-frequency ridge line, and further improve the accuracy of gear fault detection.

Description

Technical Field

[0001] The present invention relates to the technical field of locomotive gear fault diagnosis, and in particular to a locomotive gear fault detection method, device and server. Background Art

[0002] Gears are key components of a locomotive's transmission system, and their health status has a direct impact on the locomotive's operating quality and safety. Therefore, it is necessary to regularly and effectively detect the health status of locomotive gears. Currently, relevant technologies have proposed that the vibration signal of the gearbox housing can be subjected to time-frequency analysis to obtain a time-frequency distribution, and that the real-time rotational frequency can be extracted from the time-frequency distribution through time-frequency ridge extraction methods such as the maximum search algorithm, the fast path optimization algorithm, and the ridge path reconstruction algorithm, and then converted into real-time rotational speed. However, the measurement accuracy of the above-mentioned time-frequency ridge extraction method is low, and the analysis adaptability is poor when the ridge line is discontinuous, which in turn affects the measurement results of the gear rotational speed, ultimately resulting in a decrease in the detection accuracy of gear faults. Summary of the Invention

[0003] In view of this, an object of the present invention is to provide a locomotive gear fault detection method, device and server, which can significantly enhance the time-frequency ridge extraction capability and thereby improve the accuracy of gear fault detection.

[0004] In a first aspect, an embodiment of the present invention provides a locomotive gear fault detection method, the method comprising: obtaining an original time-frequency distribution and a gear transmission ratio relationship of a locomotive gear vibration signal; performing median filtering and mean filtering on the original time-frequency distribution to obtain a noise estimation value at each moment in the time-frequency distribution matrix of the original time-frequency distribution, and performing threshold detection on the noise estimation value through a threshold detector to determine a true signal component discrimination threshold at each moment; based on the noise estimation value and the true signal component discrimination threshold, determining a first time-frequency ridge line through a maximum search model, and establishing a dynamic optimization penalty function based on the first time-frequency ridge line; determining a second time-frequency ridge line using the dynamic optimization penalty function, wherein the second time-frequency ridge line includes: an intermittent time-frequency ridge line and a first continuous time-frequency ridge line; connecting the intermittent time-frequency ridge lines with matching endpoint acceleration values ​​through linear interpolation fitting to determine a repaired second continuous time-frequency ridge line, and determining a target time-frequency ridge line using the first continuous time-frequency ridge line and the second continuous time-frequency ridge line; determining a locomotive gear fault detection result using the target time-frequency ridge line and the gear transmission ratio relationship.

[0005] In one embodiment, the noise estimation value includes: a first noise estimation value and a second noise estimation value, and the steps of performing median filtering and mean filtering on the original time-frequency distribution to obtain the noise estimation value at each moment in the time-frequency distribution matrix of the original time-frequency distribution, and performing threshold detection on the noise estimation value through a threshold detector to determine the real signal component discrimination threshold at each moment include: performing median filtering on the original time-frequency distribution to obtain the first noise estimation value; determining the noise discrimination threshold at each moment based on the first noise estimation value through the threshold detector; using the noise discrimination threshold and the first noise estimation value to remove local maximum points generated by the real signal component in the time-frequency distribution matrix to determine the residual signal time-frequency distribution; performing mean filtering on the residual signal time-frequency distribution to obtain the second noise estimation value; and performing threshold detection on the second noise estimation value through the threshold detector to determine the real signal component discrimination threshold at each moment.

[0006] In one embodiment, based on the noise estimate and the true signal component discrimination threshold, a first time-frequency ridge is determined through a maximum search model, and a dynamic optimization penalty function is established based on the first time-frequency ridge, including: using a second noise estimate and a true signal component discrimination threshold to screen the local maximum points of the original time-frequency distribution, thereby determining the time-frequency distribution of the local maximum points generated by the true signal; extracting the first time-frequency ridge from the amplitude of the local maximum point time-frequency distribution through the maximum search model, and establishing a dynamic optimization penalty function based on the first time-frequency ridge.

[0007] In one embodiment, the step of determining the second time-frequency ridge using a dynamic optimization penalty function includes: iteratively calculating the penalty function value of each point from the start time to the end time of the original time-frequency distribution for the dynamic optimization penalty function, and determining a penalty function set for each candidate local maximum point; summing the penalty function sets of each candidate local maximum point and calculating the maximum value, determining the penalty function summation matrix, and determining the frequency hopping matrix corresponding to the penalty function summation matrix; extracting the second time-frequency ridge through the penalty function summation matrix and the frequency hopping matrix.

[0008] In one embodiment, the step of extracting the second time-frequency ridge line through the penalty function summation matrix and the frequency hop matrix includes: determining the maximum value point at the end moment in the penalty function summation matrix as the starting calculation point, and iteratively calculating the frequency hop matrix along the time decreasing direction from the corresponding moment of the starting calculation point to determine the optimal frequency hop at each moment; through a preset time-frequency ridge line extraction model, the ridge line connection state is determined based on the optimal frequency hop at each moment, and the time-frequency ridge line is cyclically traversed and extracted to determine all the second time-frequency ridge lines, wherein, in any cyclic traversal process, when the ridge line connection state is discontinuous, the time-frequency ridge line extraction is stopped, a section of discontinuous time-frequency ridge line is generated, and after deleting the local maximum point corresponding to the discontinuous time-frequency ridge line, the next round of time-frequency ridge line extraction is performed.

[0009] In one embodiment, the step of determining the ridge line connection state includes: when the optimal frequency jump at any moment is greater than a preset frequency jump threshold, determining that the ridge line connection state is discontinuous.

[0010] In one embodiment, the step of determining a locomotive gear fault detection result by using the relationship between the target time-frequency ridge line and the gear transmission ratio includes: determining the rotational frequency information of a rotating shaft in a locomotive gearbox by using the relationship between the target time-frequency ridge line and the gear transmission ratio; performing order tracking processing on the original time-frequency distribution through the rotational frequency information to determine an angular domain signal and an angular domain order spectrum corresponding to the angular domain signal; if, in the angular domain order spectrum, sideband information corresponding to the rotational order of the locomotive gear exists at the meshing order of the locomotive gear, it is determined that a locomotive gear fault exists.

[0011] In a second aspect, an embodiment of the present invention further provides a locomotive gear fault detection device, which includes: a data acquisition module for acquiring the original time-frequency distribution of the locomotive gear vibration signal and the gear transmission ratio relationship; a filtering processing module for performing median filtering and mean filtering on the original time-frequency distribution to obtain the noise estimation value at each moment in the time-frequency distribution matrix of the original time-frequency distribution, and performing threshold detection on the noise estimation value through a threshold detector to determine the real signal component discrimination threshold at each moment; a function establishment module for determining the first time-frequency component through a maximum search model based on the noise estimation value and the real signal component discrimination threshold. A dynamic optimization penalty function is established based on the first time-frequency ridge line; a data analysis module uses the dynamic optimization penalty function to determine the second time-frequency ridge line, wherein the second time-frequency ridge line includes: a discontinuous time-frequency ridge line and a first continuous time-frequency ridge line; a target time-frequency curve confirmation module connects the discontinuous time-frequency ridge lines that match the endpoint acceleration values ​​through linear interpolation fitting to determine the repaired second continuous time-frequency ridge line, and uses the first continuous time-frequency ridge line and the second continuous time-frequency ridge line to determine the target time-frequency ridge line; a fault detection module uses the relationship between the target time-frequency ridge line and the gear transmission ratio to determine the locomotive gear fault detection result.

[0012] In a third aspect, an embodiment of the present invention further provides a server, comprising a processor and a memory, wherein the memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement any one of the methods provided in the first aspect.

[0013] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement any one of the methods provided in the first aspect.

[0014] The embodiments of the present invention bring the following beneficial effects:

[0015] The embodiments of the present invention provide a locomotive gear fault detection method, device and server, which obtain the original time-frequency distribution and gear ratio relationship of the locomotive gear vibration signal, perform median filtering and mean filtering on the original time-frequency distribution, obtain the noise estimation value at each moment in the original time-frequency distribution matrix, and perform threshold detection on the noise estimation value through a threshold detector to determine the real signal component discrimination threshold at each moment. Based on the noise estimation value and the real signal component discrimination threshold, the local maximum value points of the original time-frequency distribution are screened and processed, and then the maximum value search model is used to determine the first maximum value point. A time-frequency ridgeline is extracted, and a dynamic optimization penalty function is established based on the first time-frequency ridgeline. After determining the first time-frequency ridgeline, a second time-frequency ridgeline is determined using the dynamic optimization penalty function. The second time-frequency ridgeline includes a discontinuous time-frequency ridgeline and a first continuous time-frequency ridgeline. The discontinuous time-frequency ridgelines with matching endpoint acceleration values ​​are connected through linear interpolation fitting to determine a repaired second continuous time-frequency ridgeline. The first and second continuous time-frequency ridgelines are then used to determine a target time-frequency ridgeline. The relationship between the target time-frequency ridgeline and the gear transmission ratio is then used to determine the locomotive gear fault detection result. Embodiments of the present invention can significantly enhance the time-frequency ridgeline extraction capability, thereby improving the accuracy of gear fault detection.

[0016] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or understood by practicing the present invention. The purposes and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0017] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 A schematic flow chart of a locomotive gear fault detection method provided by an embodiment of the present invention;

[0020] Figure 2 A schematic diagram of a locomotive gear fault detection method provided by an embodiment of the present invention;

[0021] Figure 3 A schematic flow chart of another locomotive gear fault detection method provided by an embodiment of the present invention;

[0022] Figure 4 A schematic structural diagram of a locomotive gear fault detection device provided by an embodiment of the present invention;

[0023] Figure 5 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0025] Gears are key components of a locomotive's transmission system, and their health has a direct impact on the locomotive's operating quality and safety. Therefore, effective monitoring of the health of locomotive gears is essential. During actual locomotive operation, the speed changes in real time, and the gears operate accordingly under variable speed conditions. Due to the real-time speed variation, the vibration signals collected under variable speed conditions are frequency modulated, and their spectrum naturally exhibits spectral blurring. Traditional signal processing methods fail when processing vibration signals under variable speed conditions. The key to effectively analyzing gear vibration signals under variable speed conditions lies in obtaining the time-varying speed. The most direct approach is to install a tachometer on the axle to measure the real-time speed. However, for safety reasons, railway authorities do not allow the installation of additional sensors on the axle. Therefore, vibration sensors installed on the gearbox housing are typically used to indirectly estimate the real-time speed of the gears. Time-frequency analysis of the gearbox vibration signal yields a time-frequency distribution. The real-time rotational frequency is extracted from the time-frequency distribution using a time-frequency ridge extraction algorithm, and the rotational frequency is multiplied by 60 to convert it into the real-time speed.

[0026] Currently, time-frequency ridge extraction algorithms include the following: maximum search algorithm, fast path optimization algorithm, and ridge path reorganization algorithm. The maximum search algorithm is the simplest time-frequency ridge extraction algorithm. This method searches for the point with the maximum amplitude at each moment in the time-frequency matrix and then connects the maximum values ​​at different moments. However, this method is only applicable to extracting single-component signals. If this method is used to extract multi-component signals, the continuity of the extracted ridges cannot be guaranteed. To extract multi-component signals while ensuring the continuity of the extracted ridges, some scholars have improved the maximum search algorithm by adding a search band limit to the original maximum search algorithm. However, the band setting generally relies on prior knowledge. A smaller search band may cause the true ridge to be outside the search band, while a larger band may introduce more noise and other signal components. The fast path optimization algorithm uses a cost function to find time-frequency ridges by globally considering all local maxima in the time-frequency distribution. However, this method is relatively crude in obtaining local maxima, and many of the local maxima obtained are irrelevant to the target ridge, resulting in poor subsequent optimization results. In addition, the cost function designed by the fast path optimization algorithm is not suitable for analyzing vibration signals with large speed fluctuations. The ridge path reconstruction algorithm first extracts all ridges from the time-frequency distribution using a band-limited maximum search algorithm. It then groups the ridges based on the frequency change rate of the ridges at their intersections, reconnecting ridges with similar change rates. However, because this method uses a maximum search algorithm to extract the initial ridges, its accuracy is relatively limited. Locomotive gears operate under time-varying conditions, and effective analysis of their health requires accurate real-time speed / frequency. The aforementioned time-frequency ridge extraction algorithms all have certain drawbacks, resulting in limited analysis results. Furthermore, existing methods cannot effectively analyze ridges that are both intersecting and discontinuous. Therefore, the locomotive gear fault detection method, device, and server provided by the present invention can significantly enhance the time-frequency ridge extraction capability, thereby improving the accuracy of gear fault detection.

[0027] See also Figure 1 The flowchart of a locomotive gear fault detection method shown in FIG. 1 mainly includes the following steps S102 to S112:

[0028] Step S102, obtaining the original time-frequency distribution of the locomotive gear vibration signal and the gear ratio relationship. In one embodiment, the time-frequency distribution of the locomotive gear vibration signal is obtained by short-time Fourier transform to obtain a time-frequency distribution matrix TFD(wi,tn), where i is the index of the frequency axis in the time-frequency distribution and n is the index of the time axis in the time-frequency distribution.

[0029] Step S104, perform median filtering and mean filtering on the original time-frequency distribution to obtain the noise estimation value at each moment in the time-frequency distribution matrix of the original time-frequency distribution, and perform threshold detection on the noise estimation value through a threshold detector to determine the real signal component discrimination threshold at each moment, wherein, median filtering is to take the middle value of the filtered data and then filter and output it, mean filtering is to take the average value of the filtered data and then filter and output it, Neyman-Pearson hypothesis test theory, that is, Neyman-Pearson detector, threshold detector, the threshold detector can be used for peak detection, the noise estimation value includes: a first noise estimation value and a second noise estimation value, in one embodiment, each column of the original time-frequency distribution TFD(wi,tn) is median filtered to obtain the first noise estimation value at each moment tn. Then, the noise discrimination threshold λ1 at each moment is obtained by the Neyman-Pearson hypothesis test theory, and the first noise estimation value is obtained. The time-frequency distribution of the remaining signal is obtained by removing the local maximum points generated by the real signal components in the original time-frequency distribution and the threshold λ1. right Use mean filtering to obtain the second noise estimate at time tn The threshold value detector is used again to obtain the true signal component discrimination threshold λ2 at each moment.

[0030] Step S106, based on the noise estimation value and the true signal component discrimination threshold, the first time-frequency ridge is determined by the maximum search model, and a dynamic optimization penalty function is established according to the first time-frequency ridge. In one embodiment, the second noise estimation value The local maximum points of TFD(wi,tn) are screened by the true signal component discrimination threshold λ2, and the false local maximum points generated by noise are removed. The time-frequency distribution MTFD(wm,tn) of the local maximum points generated only by the true signal components and the corresponding amplitude Am(tn) and frequency vm(tn) are obtained. A time-frequency ridge line is extracted from Am(tn) using the maximum search model, and the time-frequency ridge line is determined to be the first time-frequency ridge line wp. The dynamic optimization penalty function AF(m(tn),tn) is established based on the first time-frequency ridge line wp.

[0031] Step S108, using a dynamic optimization penalty function to determine the second time-frequency ridge, wherein the second time-frequency ridge includes: an intermittent time-frequency ridge and a first continuous time-frequency ridge. In one embodiment, a dynamic optimization penalty function is used to calculate the penalty function value from the starting time t1 to the end time tN, and obtain the penalty function sum matrix U(m(tn), tn) and the corresponding frequency jump matrix J(m(tn), tn), wherein U(m(tn), tn) represents the cumulative value of the penalty function AF(m(tn), tn) calculated from the starting time t1 to the time tn, and J(m(tn), tn) represents the frequency jump situation of the local maximum point at the time tn pointing to the local maximum point at the time tn-1, and then the maximum value U(m(tN), tN) of the matrix U(m(tn), tn) at the time tN is selected as the starting point for obtaining the time-frequency ridge, and the time-frequency ridge is iteratively calculated from the time tN in the direction of decreasing time. n, where if the frequency jump value of J(mc(tn),tn) at time tk exceeds the given threshold wλ, the ridge line is considered to be discontinuous at this time, and the ridge line is stopped, thereby obtaining a discontinuous time-frequency ridge line. The local maximum point corresponding to the discontinuous time-frequency ridge line is deleted from MTFD(w(tn),tn), and the dynamic optimization penalty function summation matrix of the remaining local maximum points and its corresponding frequency jump matrix are recalculated. Starting from the end time tN of the remaining time-frequency matrix, the next time-frequency ridge line is extracted again; repeat the above steps until all local maximum points in MTFD(w(tn),tn) are deleted, and finally z1 time-frequency ridge lines are extracted. This set of time-frequency ridge lines is the second time-frequency ridge line.

[0032] In step S110, the discontinuous time-frequency ridges with matching endpoint acceleration values ​​are connected through linear interpolation fitting to determine the repaired second continuous time-frequency ridge, and the target time-frequency ridge is determined using the first continuous time-frequency ridge and the second continuous time-frequency ridge. In one embodiment, the acceleration values ​​of each time-frequency ridge in the second time-frequency ridge at adjacent moments are calculated. If the acceleration values ​​of the endpoints of two discontinuous ridges match, the two ridges are considered to be discontinuous ridges, and linear interpolation fitting is performed on them through the acceleration values ​​of the two end points to make them a complete ridge. The final number of extracted ridges is z2, including the first continuous time-frequency ridge and the second continuous time-frequency ridge, and this set of time-frequency ridges is determined as the target time-frequency ridge.

[0033] Step S112: Determine the locomotive gear fault detection result using the target time-frequency ridge and the gear transmission ratio relationship. In one embodiment, the rotational frequency of the gearbox shaft is obtained according to the gear transmission ratio relationship. The order of the original time-domain vibration signal is tracked using the obtained rotational frequency. The gear meshing order and its harmonics are observed in the order spectrum to see whether there are sideband features corresponding to the gear fault. If the corresponding sideband features are present, it can be determined that the locomotive gear is damaged.

[0034] The locomotive gear fault detection method provided by the embodiment of the present invention can significantly enhance the time-frequency ridge extraction capability, thereby improving the accuracy of gear fault detection.

[0035] The present invention also provides a method for detecting a locomotive gear failure. Figure 2 A schematic diagram of a locomotive gear fault detection method shown and Figure 3 The flowchart of another locomotive gear fault detection method shown in FIG. 1 specifically includes the following (1) to (5):

[0036] (1) Use short-time Fourier transform to obtain the original time-frequency distribution of the locomotive gear vibration signal and the gear ratio relationship, and adjust the basic settings, where the basic settings include: false alarm probability PFA1, false alarm probability PFA2 and frequency jump threshold wλ.

[0037] (2) The original time-frequency distribution is subjected to median filtering to obtain a first noise estimation value. Based on the first noise estimation value, the noise discrimination threshold value at each moment is determined by a threshold detector. The local maximum points generated by the real signal component in the time-frequency distribution matrix are removed by using the noise discrimination threshold value and the first noise estimation value to determine the remaining signal time-frequency distribution. The remaining signal time-frequency distribution is subjected to mean filtering to obtain a second noise estimation value. The second noise estimation value is subjected to threshold detection by a threshold detector to determine the real signal component discrimination threshold value at each moment. In a real-time manner, the original signal time-frequency distribution TFD (w i ,t n ):

[0038] TFD(w i ,t n )=True(w i ,t n )+False(w i ,t n ) (1)

[0039] Where i is the frequency index, i∈[1,M]. True(w i ,t n ) represents the time-frequency distribution generated by the real signal component, False(w i,t n ) represents the time-frequency distribution caused by noise interference. The time-frequency distribution at time tn is defined as the double hypothesis test:

[0040] H0: TFD(w i ,t n )=False(w i ,t n ) (2)

[0041] H1: TFD(w i ,t n )=True(w i ,t n )+False(w i ,t n ) (3)

[0042] Where H0 means that the time-frequency distribution is caused by noise interference, and H1 means that the time-frequency distribution is jointly generated by noise and real signal components.

[0043] Perform median filtering on each column of the time-frequency distribution TFD(wi,tn) of the original signal to obtain the first noise estimate at time tn

[0044]

[0045] Where Median(·) represents the median filter operation, That is False(w i ,t n ).

[0046] Based on Neyman-Pearson hypothesis testing theory, the random variable Perform hypothesis testing:

[0047]

[0048] if is greater than the noise discrimination threshold λ1, indicating that hypothesis H1 is satisfied; if It is less than the threshold λ1, which means that the hypothesis H0 is satisfied.

[0049]

[0050] Given the false alarm probability PFA1, the threshold λ1 can be determined by the following formula:

[0051]

[0052] Where, It indicates that the probability density function of the real signal component is judged to be generated by noise, and it statistically satisfies the chi-square distribution:

[0053]

[0054] Therefore, the hypothesis test of formula (6) is equivalent to:

[0055]

[0056] According to formula (9), the local maximum value generated by the real signal component can be removed. The time-frequency distribution TFD (wi (tn), tn) at time tn is recorded as

[0057] right Use mean filtering to get the second noise estimate

[0058]

[0059] Where Average(·) represents the mean filtering operation.

[0060] Based on the given false alarm probability PFA2, the noise estimate is obtained The true signal component discrimination threshold λ2 under:

[0061]

[0062] (3) Using the second noise estimation value and the true signal component discrimination threshold, the local maximum points of the original time-frequency distribution are screened to determine the time-frequency distribution of the local maximum points generated by the true signal, and the first time-frequency ridge is extracted from the amplitude of the time-frequency distribution of the local maximum points through the maximum search model, and a dynamic optimization penalty function is established based on the first time-frequency ridge. In one embodiment, the noise value at each moment and the true signal component discrimination threshold λ2 for the original signal time-frequency distribution TFD(w i ,t n ) is used to filter out the false local maximum points caused by noise, and the time-frequency distribution MTFD(wm,tn) of the filtered local maximum points and the corresponding amplitude Am(tn) and frequency vm(tn) are obtained:

[0063]

[0064] In the formula, NaN stands for Not a Number.

[0065] Use the maximum search algorithm to extract a first time-frequency ridge from Am(tn):

[0066] δ p (t n )=argmax(A m (t n )) (13)

[0067]

[0068] Where argmax means finding the maximum value, δp(tn) represents the index of the ridge point on the first time-frequency ridge line on the frequency axis, and wp(tn) represents the instantaneous frequency of the initial time-frequency ridge line.

[0069] (4) Establish a dynamic optimization penalty function, and for the dynamic optimization penalty function, iteratively calculate the penalty function value of each point from the start time to the end time of the original time-frequency distribution, determine the penalty function set of each candidate local maximum point, sum the penalty function sets of each candidate local maximum point and calculate the maximum value of each time point, thereby determining the penalty function sum matrix, and determining the frequency hopping matrix corresponding to the penalty function sum matrix, and extracting the second time-frequency ridge line through the penalty function sum matrix and the frequency hopping matrix. In one embodiment, a dynamic optimization penalty function is established based on the first time-frequency ridge line:

[0070]

[0071]

[0072] Δξ(m(t n ),t n )=w m (t n )-w m (t n-1 ) (17)

[0073] Among them, AF[m(t n ),t n ] represents the established dynamic optimization penalty function. When n = 1, the penalty function is established by taking the logarithm of the amplitude of the local maximum point. When n>1, the relationship between the amplitude of the local maximum point and the frequency of the adjacent time points is comprehensively considered, and the corresponding penalty function is established by reasonably weighting the amplitude and frequency, so that the algorithm can extract more accurate time-frequency ridges. ε(m(t n ),t n ) represents the constraint of the dynamic optimization penalty function on frequency hopping. n ),t n ) represents the difference matrix between the frequency wm(tn) of all local maxima at time tn and the frequency wm(tn-1) of all local maxima at time tn-1.

[0074] In one embodiment, the maximum point at the terminal moment in the penalty function sum matrix is ​​determined as the starting calculation point, and the frequency jump matrix is ​​iteratively calculated from the starting calculation point in the time-decreasing direction to determine the optimal frequency jump at each moment; through a preset time-frequency ridge extraction model, the connection state of the ridge is determined based on the optimal frequency jump at each moment, and the time-frequency ridge is cyclically traversed and extracted to determine all second time-frequency ridges, wherein, in any cyclic traversal process, when the ridge connection state is discontinuous, the time-frequency ridge extraction is stopped, a section of discontinuous time-frequency ridge is generated, and after deleting the local maximum point corresponding to the discontinuous time-frequency ridge, the next round of time-frequency ridge extraction is performed, wherein, when the optimal frequency jump at any moment is greater than the preset frequency jump threshold, the ridge connection state is determined to be discontinuous. In actual applications, the penalty function AF of all candidate local maximum points is iteratively calculated along the time axis from t1 to tn, and the maximum value is calculated after summing them, which is represented by the matrix U:

[0075]

[0076] When n = 1, the U matrix is ​​the logarithm of the amplitude of the local maximum point. When n > 1, the U matrix is ​​the cumulative value of the penalty function AF(m(tn), tn) calculated from the starting time t1 to the time tn for all paths.

[0077] Calculate the ridge frequency jump J(m(tn),tn) determined by the dynamic optimization penalty function from time tn to time tn-1:

[0078]

[0079] When n = 1, the J matrix is ​​an all-zero matrix. When n > 1, J(m(tn), tn) represents the frequency hopping from all local maximum points at time tn to all local maximum points at time tn-1.

[0080] The maximum dynamic optimization penalty function value U(m(tN),tN) at time tN is determined based on the penalty function sum matrix, and the frequency of the previous moment is reversed to the left based on the ridge frequency jump matrix:

[0081]

[0082] m c (t n-1 )=J (m c (t n ),t n ) (twenty one)

[0083] Wherein, mc(tn) represents the frequency index corresponding to the ridge point extracted at time point tn.

[0084] The optimal frequency jump of each time adjacent local maximum point is obtained by iteration from the time axis direction from tNtime to the left according to formula (21), if the frequency jump at tktime is greater than the given threshold wλ, it is considered that the time-frequency ridge line is interrupted at tktime, the forward calculation is stopped, and the final ridge line estimation result is obtained:

[0085] {R (t k ),…,R (t N )}={v (m c (t k )),…,v (m c (t N ))} (22)

[0086] R(tn) is the final extracted time-frequency ridge line.

[0087] (5) The rotation frequency of the locomotive gear box is determined by using the target time-frequency ridge line and the gear transmission ratio relationship, the order tracking of the original time domain vibration signal is performed through the rotation frequency information, the angular domain signal and the angular domain order spectrum corresponding to the angular domain signal are determined, if there is a side frequency band corresponding to the rotation order of the locomotive gear at the meshing order of the locomotive gear in the angular domain order spectrum, the existence of the fault of the locomotive gear can be judged. In practical application, after the target time-frequency ridge line is determined, the rotation frequency of the gear box is obtained according to the transmission ratio relationship of the gear, the angular domain signal x(θ) is obtained by angular domain resampling of the original time domain signal x(t) through the rotation frequency information, and the angular domain order spectrum is obtained by Fourier transform:

[0088] X=FFT(x(θ)) (23)

[0089] Wherein, FFT(x(θ)) represents the Fourier transform of x(θ), X is the obtained angular domain order spectrum, whether there is a side frequency band corresponding to the rotation order of the gear at the meshing order of the gear in the order spectrum X is observed, if there is a related feature, the gear fault can be judged.

[0090] In summary, this application proposes a local maximum point screening criterion, based on the theory of mathematical statistics, and according to the Neyman-Pearson hypothesis (threshold detector) test to filter out the false local maximum points caused by noise interference. Compared with the maximum value judgment method in mathematics where the first-order derivative is equal to zero and the second-order derivative is less than zero, the judgment accuracy is significantly improved; this application also proposes a mathematical model for estimating cross-interrupted ridges based on a dynamic optimization penalty function. Starting from the vibration signal itself, by reasonably selecting key parameters, the model adaptability is improved, repeated parameter adjustment is avoided, and the time-frequency ridge line is extracted. In the process, the amplitude of the local maximum point in the time-frequency matrix and the frequency jump limit are comprehensively considered to ensure that the ridge line can still be correctly extracted when the speed fluctuation is large. When the extracted ridge lines are crossed, the associated discontinuous ridge lines can be accurately judged by a given threshold and the discontinuous ridge lines can be connected to ensure the continuity of the extracted ridge lines. When the extracted ridge lines are discontinuous, the extraction can be adaptively stopped. Compared with the existing algorithms, the algorithm is advanced in terms of accuracy, analysis of large speed fluctuation conditions, adaptability, and analysis of crossing and discontinuous ridge lines.

[0091] Regarding the locomotive gear fault detection method provided in the above embodiment, the present invention provides a locomotive gear fault detection device, see Figure 4 The schematic diagram of the structure of a locomotive gear fault detection device shown in FIG. 1 includes the following parts:

[0092] The data acquisition module 402 acquires the original time-frequency distribution of the locomotive gear vibration signal and the gear ratio relationship;

[0093] The filtering processing module 404 performs median filtering and mean filtering on the original time-frequency distribution to obtain the noise estimation value at each moment in the time-frequency distribution matrix of the original time-frequency distribution, and performs threshold detection on the noise estimation value through a threshold detector to determine the real signal component discrimination threshold at each moment;

[0094] Function establishment module 406, based on the noise estimation value and the real signal component discrimination threshold, determines the first time-frequency ridge line through the maximum search model, and establishes a dynamic optimization penalty function according to the first time-frequency ridge line;

[0095] The data analysis module 408 determines a second time-frequency ridge line using a dynamic optimization penalty function, wherein the second time-frequency ridge line includes: a discontinuous time-frequency ridge line and a first continuous time-frequency ridge line;

[0096] The target time-frequency curve confirmation module 410 connects the discontinuous time-frequency ridges that match the endpoint acceleration values ​​through linear interpolation fitting to determine the repaired second continuous time-frequency ridge, and uses the first continuous time-frequency ridge and the second continuous time-frequency ridge to determine the target time-frequency ridge.

[0097] The fault detection module 412 determines a locomotive gear fault detection result using the target time-frequency ridge and the gear ratio relationship.

[0098] The above-mentioned data processing device provided in the embodiment of the present application can significantly enhance the ability to extract time-frequency ridges, thereby improving the accuracy of gear fault detection.

[0099] In one embodiment, the noise estimation value includes: a first noise estimation value and a second noise estimation value. When performing median filtering and mean filtering on the original time-frequency distribution to obtain the noise estimation value at each moment in the time-frequency distribution matrix of the original time-frequency distribution, and performing threshold detection on the noise estimation value through a threshold detector to determine the real signal component discrimination threshold at each moment, the above-mentioned filtering processing module 404 is further used to: perform median filtering on the original time-frequency distribution to obtain the first noise estimation value; determine the noise discrimination threshold at each moment based on the first noise estimation value through the threshold detector; use the noise discrimination threshold and the first noise estimation value to remove local maximum points generated by the real signal component in the time-frequency distribution matrix to determine the residual signal time-frequency distribution; perform mean filtering on the residual signal time-frequency distribution to obtain the second noise estimation value; perform threshold detection on the second noise estimation value through the threshold detector to determine the real signal component discrimination threshold at each moment.

[0100] In one embodiment, when performing the steps of determining the first time-frequency ridge line based on the noise estimation value and the true signal component discrimination threshold through the maximum search model, and establishing a dynamic optimization penalty function based on the first time-frequency ridge line, the above-mentioned function establishment module 406 is also used to: use the second noise estimation value and the true signal component discrimination threshold to screen the local maximum points of the original time-frequency distribution, so as to determine the time-frequency distribution of the local maximum points generated by the true signal; extract the first time-frequency ridge line from the amplitude of the time-frequency distribution of the local maximum points through the maximum search model, and establish a dynamic optimization penalty function based on the first time-frequency ridge line.

[0101] In one embodiment, when performing the step of determining the second time-frequency ridge using a dynamic optimization penalty function, the above-mentioned data analysis module 408 is also used to: for the dynamic optimization penalty function, iteratively calculate the penalty function value of each point from the start time to the end time of the original time-frequency distribution, and determine the penalty function set of each candidate local maximum point; sum the penalty function sets of each candidate local maximum point and then calculate the maximum value, determine the penalty function summation matrix, and determine the frequency hopping matrix corresponding to the penalty function summation matrix; extract the second time-frequency ridge through the penalty function summation matrix and the frequency hopping matrix.

[0102] In one embodiment, when performing the step of extracting the second time-frequency ridge line through the penalty function summation matrix and the frequency hop matrix, the above-mentioned data analysis module 408 is also used to: determine the maximum value point at the end moment in the penalty function summation matrix as the starting calculation point, and iteratively calculate the frequency hop matrix in the time decreasing direction from the corresponding moment of the starting calculation point to determine the optimal frequency hop at each moment; through the preset time-frequency ridge line extraction model, based on the optimal frequency hop at each moment, determine the ridge line connection state, and perform a cyclic traversal extraction on the time-frequency ridge line to determine all the second time-frequency ridge lines, wherein, in any cyclic traversal process, when the ridge line connection state is discontinuous, the time-frequency ridge line extraction is stopped, a section of discontinuous time-frequency ridge line is generated, and after deleting the local maximum point corresponding to the discontinuous time-frequency ridge line, the next round of time-frequency ridge line extraction is performed.

[0103] In one embodiment, when performing the step of determining the ridge connection state, the data analysis module 408 is further configured to: determine that the ridge connection state is discontinuous when the optimal frequency jump at any moment is greater than a preset frequency jump threshold.

[0104] In one embodiment, when performing the step of determining the locomotive gear fault detection result by using the target time-frequency ridge line and the gear transmission ratio relationship, the above-mentioned fault detection module 412 is further used to: determine the rotational frequency information of the rotating shaft in the locomotive gearbox by using the target time-frequency ridge line and the gear transmission ratio relationship; perform order tracking processing on the original time-frequency distribution through the rotational frequency information to determine the angular domain signal and the angular domain order spectrum corresponding to the angular domain signal; if in the angular domain order spectrum, there is sideband information corresponding to the rotation order of the locomotive gear at the meshing order of the locomotive gear, it is determined that there is a locomotive gear fault.

[0105] The device provided in the embodiment of the present invention has the same implementation principle and technical effects as those in the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the device embodiment, reference can be made to the corresponding content in the aforementioned method embodiment.

[0106] An embodiment of the present invention provides an electronic device. Specifically, the electronic device includes a processor and a storage device. The storage device stores a computer program, and when the computer program is executed by the processor, it executes the method described in any one of the above-mentioned embodiments.

[0107] Figure 5 This is a structural diagram of an electronic device provided in an embodiment of the present invention. The electronic device 100 includes: a processor 50, a memory 51, a bus 52 and a communication interface 53. The processor 50, the communication interface 53 and the memory 51 are connected via the bus 52; the processor 50 is used to execute an executable module stored in the memory 51, such as a computer program.

[0108] The memory 51 can include a high-speed random access memory (RAM), and can also include a non-volatile memory, such as at least one disk memory. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 53 (which can be wired or wireless), and the Internet, a wide area network, a local area network, a metropolitan area network, etc. can be used.

[0109] The bus 52 can be an ISA bus, a PCI bus, an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 Only one bidirectional arrow is used in the figure, but it does not mean that there is only one bus or only one type of bus.

[0110] The memory 51 is used to store programs, and the processor 50 executes the programs after receiving execution instructions. The method executed by the device defined by the flow process disclosed in any of the embodiments of the present application can be applied to the processor 50 or implemented by the processor 50.

[0111] The processor 50 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method may be completed by hardware integrated logic circuits or software instructions in the processor 50. The processor 50 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It may implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in conjunction with the embodiments of the present invention may be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium well-known in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or the like. The storage medium is located in the memory 51 , and the processor 50 reads the information in the memory 51 and completes the steps of the above method in combination with its hardware.

[0112] The computer program product of the readable storage medium provided in the embodiment of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the method described in the previous method embodiment. The specific implementation can be referred to the previous method embodiment and will not be repeated here.

[0113] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0114] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A locomotive gear fault detection method, characterized in that: The method comprises: Obtain the original time-frequency distribution of the locomotive gear vibration signal and the gear ratio relationship; Performing median filtering and mean filtering on the original time-frequency distribution to obtain noise estimation values ​​at each moment in the time-frequency distribution matrix of the original time-frequency distribution, and performing threshold detection on the noise estimation values ​​through a threshold detector to determine a true signal component discrimination threshold at each moment; Based on the noise estimation value and the true signal component discrimination threshold, a first time-frequency ridge line is determined through a maximum search model, and a dynamic optimization penalty function is established according to the first time-frequency ridge line; Determine a second time-frequency ridge line by using the dynamic optimization penalty function, wherein the second time-frequency ridge line includes: a discontinuous time-frequency ridge line and a first continuous time-frequency ridge line; The discontinuous time-frequency ridge lines that match the endpoint acceleration values ​​are connected by linear interpolation fitting to determine a repaired second continuous time-frequency ridge line, and the first continuous time-frequency ridge line and the second continuous time-frequency ridge line are used to determine a target time-frequency ridge line; The target time-frequency ridge and the gear ratio relationship are used to determine a locomotive gear fault detection result.

2. The locomotive gear fault detection method according to claim 1, characterized in that: The noise estimation value includes: a first noise estimation value and a second noise estimation value; the steps of performing median filtering and mean filtering on the original time-frequency distribution to obtain noise estimation values ​​at each moment in the time-frequency distribution matrix of the original time-frequency distribution, and performing threshold detection on the noise estimation values ​​by a threshold detector to determine a true signal component discrimination threshold at each moment include: Performing median filtering on the original time-frequency distribution to obtain the first noise estimation value; determining, by a threshold detector, a noise discrimination threshold at each moment based on the first noise estimation value; Using the noise discrimination threshold and the first noise estimation value, removing local maximum points generated by real signal components in the time-frequency distribution matrix to determine the remaining signal time-frequency distribution; performing mean filtering on the time-frequency distribution of the residual signal to obtain the second noise estimation value; The second noise estimation value is subjected to threshold detection by a threshold detector to determine the real signal component discrimination threshold at each moment.

3. The locomotive gear fault detection method according to claim 1, characterized in that: The step of determining a first time-frequency ridge line based on the noise estimation value and the true signal component discrimination threshold by using a maximum search model, and establishing a dynamic optimization penalty function according to the first time-frequency ridge line includes: Using the second noise estimate and the true signal component discrimination threshold, the local maximum points of the original time-frequency distribution are screened to determine the time-frequency distribution of the local maximum points generated by the true signal; A first time-frequency ridge line is extracted from the amplitude of the time-frequency distribution of the local maximum point through a maximum search model, and the dynamic optimization penalty function is established according to the first time-frequency ridge line.

4. The locomotive gear fault detection method according to claim 1, characterized in that: The step of determining the second time-frequency ridge line by using the dynamic optimization penalty function includes: For the dynamic optimization penalty function, iteratively calculate the penalty function value of each point from the start time to the end time of the original time-frequency distribution to determine the penalty function set of each candidate local maximum point; Calculate the maximum value by summing the penalty function sets of each candidate local maximum point, determine a penalty function sum matrix, and determine a frequency hopping matrix corresponding to the penalty function sum matrix; The second time-frequency ridge is extracted using the penalty function sum matrix and the frequency hopping matrix.

5. The locomotive gear fault detection method according to claim 4, characterized in that: The step of extracting the second time-frequency ridge line by using the penalty function sum matrix and the frequency hopping matrix includes: Determining the maximum value point at the end moment in the penalty function sum matrix as the starting calculation point, and iteratively calculating the frequency hopping matrix along the time decreasing direction from the moment corresponding to the starting calculation point to determine the optimal frequency hopping at each moment; By presetting the time-frequency ridge extraction model, based on the optimal frequency jump at each moment, the ridge connection state is determined, and the time-frequency ridge is cyclically traversed and extracted to determine all the second time-frequency ridges. In any cyclic traversal process, when the ridge connection state is discontinuous, the time-frequency ridge extraction is stopped, and a section of discontinuous time-frequency ridge is generated. After deleting the local maximum point corresponding to the discontinuous time-frequency ridge, the next round of time-frequency ridge extraction is performed.

6. The locomotive gear fault detection method according to claim 5, characterized in that: The step of determining the ridge line connection state includes: When the optimal frequency jump at any moment is greater than a preset frequency jump threshold, the ridge line connection state is determined to be discontinuous.

7. The locomotive gear fault detection method according to claim 1, characterized in that: The step of determining a locomotive gear fault detection result by using the target time-frequency ridge line and the gear transmission ratio relationship includes: Determining the rotational frequency information of a rotating shaft in a locomotive gearbox by using the target time-frequency ridge and the gear ratio relationship; Performing order tracking processing on the original time-frequency distribution using the frequency conversion information to determine an angular domain signal and an angular domain order spectrum corresponding to the angular domain signal; If, in the angular domain order spectrum, sideband information corresponding to the rotation order of the locomotive gear exists at the meshing order of the locomotive gear, it is determined that the locomotive gear has a fault.

8. A locomotive gear fault detection device, characterized in that: The device comprises: Data acquisition module, which obtains the original time-frequency distribution of the locomotive gear vibration signal and the gear ratio relationship; a filtering processing module, performing median filtering and mean filtering on the original time-frequency distribution to obtain a noise estimation value at each moment in the time-frequency distribution matrix of the original time-frequency distribution, and performing threshold detection on the noise estimation value through a threshold detector to determine a true signal component discrimination threshold at each moment; a function establishment module, which determines a first time-frequency ridge line based on the noise estimation value and the true signal component discrimination threshold through a maximum search model, and establishes a dynamic optimization penalty function according to the first time-frequency ridge line; A data analysis module is configured to determine a second time-frequency ridge line using the dynamic optimization penalty function, wherein the second time-frequency ridge line includes: a discontinuous time-frequency ridge line and a first continuous time-frequency ridge line; a target time-frequency curve confirmation module, connecting the discontinuous time-frequency ridge lines that match the endpoint acceleration values ​​through linear interpolation fitting, determining a repaired second continuous time-frequency ridge line, and determining a target time-frequency ridge line using the first continuous time-frequency ridge line and the second continuous time-frequency ridge line; The fault detection module determines a locomotive gear fault detection result by using the target time-frequency ridge and the gear transmission ratio relationship.

9. A server, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement the method according to any one of claims 1 to 7.

Images