A gearbox fault diagnosis method based on embedded acceleration sensor

Abstract

A kind of gearbox fault diagnosis method based on embedded acceleration sensor, first utilize gearbox embedded acceleration sensor data signal, separate gravity acceleration signal by low-pass filter, eliminate gravity influence;Then by extracting local maximum and cubic spline interpolation reduce the signal amplitude variation influence brought by embedded acceleration sensor with gear rotation;Again by maximum average kurtosis deconvolution obtains filtering result, realizes gearbox fault diagnosis based on embedded acceleration sensor;The method reduces the limitation of traditional method relying on external sensor information, carries out fault diagnosis by gear internal embedded acceleration sensor information, eliminates multipath coupling and signal attenuation effect;Realize the information enhancement processing of gearbox embedded acceleration sensor, make fault feature more obvious.

Description

Technical Field

[0001] This invention belongs to the field of mechanical fault diagnosis technology, specifically relating to a gearbox fault diagnosis method based on an embedded acceleration sensor. Background Technology

[0002] Gear transmission is an important transmission form in many important fields such as wind power generation, aerospace, and engineering machinery. Currently, fault diagnosis and life prediction methods based on monitoring external vibration, oil, and temperature have been implemented both domestically and internationally. However, these monitoring methods reflect the comprehensive response characteristics of the transmission system and can only diagnose mid-to-late-stage faults. This is because the external measurement signals of planetary gear trains face complex multi-path coupling and attenuation effects, making it difficult to diagnose early, weak faults and identify multi-source coupled information. Embedded signal measurement can effectively overcome the effects of multi-path coupling and attenuation. However, existing embedded measurement methods both domestically and internationally are still in the early stages of development, and due to the harsh internal environment and complex motion of gearboxes, there are relatively few embedded signal measurement and processing methods specifically for gearboxes.

[0003] On the one hand, traditional gearbox fault diagnosis technology mainly relies on data acquired by external gearbox sensors to achieve fault diagnosis (Ning Shaohui, Han Zhennan, Wu Xuefeng, et al. Gear crack fault diagnosis using embedded sensors [J]. Vibration and Shock, 2018, 37(11):6.DOI:10.13465 / j.cnki.jvs.2018.11.007.). This process faces complex multipath coupling and signal attenuation effects during signal transmission, making it difficult to acquire effective state information. On the other hand, there are currently few embedded signal measurement methods for gearboxes, and a lack of embedded signal processing methods for gearboxes. Therefore, the measurement and processing of embedded acceleration signals from gearboxes remains a challenging problem. Summary of the Invention

[0004] In order to overcome the shortcomings of the prior art, the present invention aims to provide a gearbox fault diagnosis method based on an embedded acceleration sensor, so as to realize the measurement and processing of embedded acceleration signals of the gearbox.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A gearbox fault diagnosis method based on an embedded accelerometer sensor is proposed. First, the data signal from the embedded accelerometer sensor in the gearbox is used to separate the gravitational acceleration signal through a low-pass filter to eliminate the influence of gravity. Then, the influence of signal amplitude changes caused by the embedded accelerometer sensor rotating with the gear is reduced by extracting local maxima and using cubic spline interpolation. Finally, the filtering result is obtained by deconvolution with maximum average kurtosis, thus realizing gearbox fault diagnosis based on the embedded accelerometer sensor.

[0007] A gearbox fault diagnosis method based on an embedded acceleration sensor includes the following steps:

[0008] Step 1: Embed the vibration acceleration sensor onto the gear under test, collect the vibration acceleration signal of the gear under test in real time, and truncate and remove the mean of the vibration acceleration signal to obtain the processed vibration signal x. V (n);

[0009] Step 2: From vibration signal x V Separate the gravitational acceleration component g(n) from (n), and separate the vibration signal x V (n) Filtering is performed using a low-pass filter, which is set as an FIR filter f(l) of length L. Where sinc represents the Singer function, l represents the l-th coefficient of the low-pass filter, and win(l) is the window function. This is the cutoff frequency of the low-pass filter. The estimated actual operating frequency of the gear;

[0010] The gravitational acceleration component g(n) is obtained through convolution filtering: g(n) = x V (n)*f(l);

[0011] Step 3: Perform phase compensation on the separated gravitational acceleration component g(n). Where g s (n) represents the gravitational acceleration component after the phase compensation stage, L represents the filter length, and END represents the signal length;

[0012] Step 4: In the vibration signal x V Remove the gravitational acceleration component from (n), x s (n)=x V (n)-g s (n), x s (n) represents the signal after removing the gravitational acceleration component;

[0013] Step 5: In signal x s (n) Extract the local maxima of the signal at M points and fit a cubic spline interpolation to obtain the result x. M (n);

[0014] Step Six: In signal x s By reducing the influence of amplitude variation in (n), we obtain signal x(n), x(n) = x s (n) / x M (n);

[0015] Step 7: Perform maximum mean kurtosis deconvolution on the signal x(n) to obtain the filtered signal x. re Envelope spectrum analysis was performed.

[0016] The steps of the maximum average kurtosis deconvolution method are as follows:

[0017] 7.1) Instantaneous rotation angle calculation: First, calculate the cumulative rotation angle θ(t) of the shaft where the gear being measured is located based on the theoretical period T:

[0018]

[0019] Where: t0—initial time of the signal to be processed; T—theoretical period, in seconds;

[0020] 7.2) Signal Segmentation: Based on the cumulative rotation angle θ(t) obtained from equation (1), the input signal x(n) is continuously segmented using a set of non-overlapping rectangular window functions:

[0021]

[0022] in

[0023]

[0024] In the formula: S n —The nth signal slice; W n —Used for interception S The time-domain rectangular window function of n; N m ——S m With S m+1 The boundary point, for any m, and The corresponding angular domain interval bandwidth is a constant; FCO—fault characteristic order, i.e., the cyclic order of the impact characteristic; the subscript of t indicates the sampling point number;

[0025] 7.3) Given the maximum number of iterations, filter length, and iteration threshold, initialize the filter coefficients to f0;

[0026] 7.4) Calculate the filtered signal y using the input signal x(n) and the filter coefficients f0;

[0027] 7.5) Update the filter coefficients f according to the iterative equation for deconvolution with maximum mean kurtosis. The iterative equation is:

[0028]

[0029] Where: P — number of signal slices, T m —The m-th signal slice S mL is the set of indices of all sampling points in the input signal x(n), where L is the filter length.

[0030] 7.6) Calculate the relative deviation of the filter coefficients before and after the last iteration. If the deviation is less than the iteration threshold or the maximum number of iterations is reached, terminate the iteration; otherwise, return to step 7.4) and continue iterating.

[0031] 7.7) The iteration terminates, and the optimal filter coefficients and the corresponding filtered signal x are output. re .

[0032] Compared with the prior art, the present invention has the following advantages:

[0033] a) The method of the present invention reduces the limitation of traditional gearbox fault diagnosis methods that rely on external sensor information, and performs fault diagnosis by using information from an embedded acceleration sensor inside the gear, thus eliminating multipath coupling and signal attenuation effects;

[0034] b) The method of the present invention reduces the influence of signal amplitude changes caused by the embedded accelerometer rotating with the gear by eliminating the influence of gravitational acceleration; finally, the filtered signal is obtained by deconvolution with maximum average kurtosis, thereby realizing the information enhancement processing of the embedded accelerometer in the gearbox and making the fault characteristics more obvious. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the acceleration sensor connection structure in an embodiment of the present invention.

[0036] Figure 2 This is a flowchart of the method of the present invention.

[0037] Figure 3 The vibration signal x(n) is measured by the embedded accelerometer in an embodiment of the present invention.

[0038] Figure 4 This is the envelope spectrum of the vibration signal x(n) in an embodiment of the present invention.

[0039] Figure 5 The vibration signal s(n) is measured by an external acceleration sensor in an embodiment of the present invention.

[0040] Figure 6 This is the envelope spectrum of the vibration signal s(n) in an embodiment of the present invention.

[0041] Figure 7 The signal is the vibration signal x(n) after signal processing in an embodiment of the present invention.

[0042] Figure 8 The envelope spectrum of the vibration signal x(n) after signal processing is shown in the embodiment of the present invention.

[0043] Figure 9 This is the signal after the vibration signal s(n) in the embodiment of the present invention has been deconvolved with maximum average kurtosis.

[0044] Figure 10 The envelope spectrum of the vibration signal s(n) in this embodiment of the invention is obtained after deconvolution with maximum average kurtosis. Detailed Implementation

[0045] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0046] The embodiment uses a gearbox failure test bench, such as Figure 1 As shown, the gearbox fault test bench includes a gearbox 4 and a gear 1 under test inside it. The gear 1 under test is mounted on a rotating shaft 5, which extends out of the gearbox 4 and is connected to a signal acquisition system. An embedded acceleration sensor 2 is embedded in the gear 1 under test. An external acceleration sensor 3 is fixed on the gearbox 4. The embedded acceleration sensor 2 is connected to the external acceleration sensor 3 and the signal acquisition system to measure the vibration signal of the gear 1 under test.

[0047] like Figure 2 As shown, a gearbox fault diagnosis method based on an embedded acceleration sensor includes the following steps:

[0048] Step 1: Embed the embedded accelerometer 2 onto the gear 1 under test, collect the vibration acceleration signal of the gear 1 under test in real time, and truncate and remove the mean of the vibration acceleration signal to obtain the processed vibration signal x. V (n);

[0049] Step 2: From vibration signal x V Separate the gravitational acceleration component g(n) from (n), and separate the vibration signal x V (n) Filtering is performed using a low-pass filter, which is set as an FIR filter f(l) of length L. Where sinc represents the Singer function, l represents the l-th coefficient of the filter, and win(l) is the window function. This is the cutoff frequency of the low-pass filter. In this embodiment, L is 512, representing the estimated actual operating rotational frequency of the gear 1 under test. The theoretical frequency is 4.44, and k is 1.5;

[0050] The gravitational acceleration component g(n) is obtained through convolution filtering: g(n) = x V (n)*f(l);

[0051] Step 3: Perform phase compensation on the gravitational acceleration component g(n). Where gs (n) represents the gravitational acceleration component after the phase compensation stage, L represents the filter length, and END represents the signal length;

[0052] Step 4: In the vibration signal x V Remove the gravitational acceleration component from (n), x s (n)=x V (n)-g s (n), x s (n) represents the signal after removing the gravitational acceleration component;

[0053] Step 5: To reduce the impact of signal amplitude changes caused by the embedded accelerometer rotating with the gear 1 being measured, in signal x s (n) Extract the local maxima of the signal at M points and fit a cubic spline interpolation to obtain the result x. M (n), in this embodiment, M is 60;

[0054] Step Six: In x s To reduce the influence of amplitude changes in the signal (n), we obtain signal x(n), where x(n) = x s (n) / x M (n);

[0055] Step 7: Perform maximum mean kurtosis deconvolution on the signal x(n) to obtain the filtered signal x. re Envelope spectrum analysis was performed.

[0056] The steps of the maximum average kurtosis deconvolution method are as follows:

[0057] 7.1) Instantaneous rotation angle calculation: First, calculate the cumulative rotation angle of the shaft 5 where the tested gear 1 is located based on the theoretical period T:

[0058]

[0059] Where: t0—initial time of the signal to be processed; T—theoretical period, in seconds;

[0060] 7.2) Signal Segmentation: Based on the cumulative rotation angle θ(t) obtained from equation (1), the input signal x(n) is continuously segmented using a set of non-overlapping rectangular window functions:

[0061]

[0062] in

[0063]

[0064] In the formula: S n —The nth signal slice; W n —Used to extract Sn The time-domain rectangular window function; N m ——S m With S m+1 The boundary point, for any m, and The corresponding angular domain interval bandwidth is a constant; FCO—fault characteristic order, i.e., the cyclic order of the impact characteristic; the subscript of t indicates the sampling point number;

[0065] 7.3) Given the maximum number of iterations, filter length, and iteration threshold, initialize the filter coefficients to f0;

[0066] 7.4) Calculate the filtered signal y using the input signal x(n) and the filter coefficients f0;

[0067] 7.5) Update the filter coefficients f according to the iterative equation for deconvolution with maximum mean kurtosis. The iterative equation is:

[0068]

[0069] Where: P — number of signal slices, T m —The m-th signal slice S m L is the set of indices of all sampling points in the input signal x(n), where L is the filter length.

[0070] 7.6) Calculate the relative deviation of the filter coefficients before and after the last iteration. If the deviation is less than the iteration threshold or the maximum number of iterations is reached, terminate the iteration; otherwise, return to step 7.4) and continue iterating.

[0071] 7.7) The iteration terminates, and the optimal filter coefficients and the corresponding filtered signal x are output. re .

[0072] like Figure 3 , Figure 4 As shown, Figure 3 For embedded accelerometer signals, Figure 4 For the corresponding signal envelope spectrum; such as Figure 5 , 6 As shown, Figure 5 For external accelerometer signals, Figure 6 For the corresponding signal envelope spectrum; from Figure 3 , 4 As can be seen from points 5 and 6, the fault characteristics of the embedded accelerometer 2 are more obvious in the time domain of the vibration signal x(n), and it is not affected by multipath coupling and signal attenuation effects. Figure 7 This is the filtered signal obtained after processing the embedded accelerometer sensor signal using the method of this invention. Figure 8 The corresponding signal envelope spectrum; Figure 9This is the filtered signal after the external accelerometer signal has undergone maximum mean kurtosis deconvolution processing. Figure 10 For the corresponding signal envelope spectrum; from Figure 7 , 8 As can be seen from points 9 and 10, the method of this invention can effectively enhance fault characteristics. A significant increase in the rotational frequency and its higher harmonics in the envelope spectrum indicates the presence of a minor local fault. The external acceleration signal exhibits some fault characteristics in the time domain, with a significant increase in the rotational frequency and its higher harmonics in the envelope spectrum. Based on the above experimental results, the gearbox fault diagnosis method based on an embedded accelerometer of this invention not only has the advantage of being unaffected by multipath coupling and signal attenuation effects compared to traditional measurement methods in the acquired vibration acceleration signal, but also has the advantage of enhancing fault characteristics.

Claims

1. A gearbox fault diagnosis method based on an embedded accelerometer, characterized in that, Includes the following steps: Step 1: Embed the vibration acceleration sensor onto the gear under test, collect the vibration acceleration signal of the gear under test in real time, and truncate and remove the mean of the vibration acceleration signal to obtain the processed vibration signal. ; Step 2: From vibration signals Separate the gravitational acceleration component from the middle , to vibration signal Filtering is performed using a low-pass filter, which is configured as an FIR filter of length L. , , in Represents the Singer function. The first term of the low-pass filter One coefficient, For window functions, This is the cutoff frequency of the low-pass filter. , The estimated actual operating frequency of the gear; gravitational acceleration components Obtained through convolution filtering: ; Step 3: Separate the gravitational acceleration components Perform phase compensation. ,in This refers to the gravitational acceleration component after the phase compensation phase. END is the filter length, and END is the signal length; Step 4: Vibration signal Remove the gravitational acceleration component from the middle. , The signal after removing the gravitational acceleration component; Step 5: In the signal The local maxima of the signal at M points are extracted and fitted with cubic spline interpolation to obtain the results. ; Step Six: In the signal To reduce the influence of amplitude variation, the signal is obtained. , ; Step 7: Signal The filtered signal is obtained by performing maximum mean kurtosis deconvolution. Envelope spectrum analysis was performed. The steps of the maximum average kurtosis deconvolution method described in step seven are as follows: 7.1) Instantaneous rotation angle calculation: First, calculate the cumulative rotation angle of the shaft where the gear being measured is located based on the theoretical period T. : (1) In the formula: —The initial time of the signal to be processed; —Theoretical period, in seconds; 7.2) Signal segmentation: Based on the cumulative rotation angle obtained by equation (1) Using a set of non-overlapping rectangular window functions to process the input signal Perform continuous segmentation: (2) in (3) In the formula: —The nth signal slice; —Used for interception The time-domain rectangular window function; —— and The boundary point, for any m, and The corresponding angular domain interval bandwidth is a constant; FCO—fault characteristic order, i.e., the cyclic order of the impact characteristic; the subscript of t indicates the sampling point number; 7.3) Given the maximum number of iterations, filter length, and iteration threshold, initialize the filter coefficients as follows: ; 7.4) Utilizing input signals and filter coefficients Calculate the filtered signal y; 7.5) Update the filter coefficients according to the iterative equation of maximum mean kurtosis deconvolution. The iterative equation is: (4) In the formula: P —Number of signal slices, ——No. m a signal slice All sampling points in the input signal The set of indices in L —Filter length; 7.6) Calculate the relative deviation of the filter coefficients before and after the last iteration. If the deviation is less than the iteration threshold or the maximum number of iterations is reached, terminate the iteration; otherwise, return to step 7.4) and continue iterating. 7.7) The iteration terminates, and the optimal filter coefficients and the corresponding filtered signal are output. .

Images