A non-iterative deconvolution method for bearing fault feature extraction under variable speed conditions

Abstract

The present invention relates to the technical field of rolling bearing fault diagnosis, and in particular to a non-iterative deconvolution method for extracting bearing fault features under variable speed conditions, comprising the following steps: S1, high-frequency sampling of bearing vibration signals; S2, calculating the cumulative relative rotation angle #imgabs0# of the inner and outer rings of the bearing using the speed signal to calculate the theoretical rotation angle Δθ corresponding to the interval between adjacent fault impacts; S3, constructing an alternative target vector t=[t1, t2, ..., t N ] T ; S4, the candidate target vector t i ,i=1,2,...,N t Substitute the multi-point D norm into the deconvolution solution formula as the objective function for solution; select the filtered signal with the maximum objective function value as the final deconvolution result #imgabs1#S5, resample the fault-related feature #imgabs2# to obtain the time domain stationary signal #imgabs3#S6, calculate the fault-related feature order envelope spectrum, and output the diagnosis result. The present invention can efficiently and accurately extract fault feature information and effectively diagnose faults under variable speed conditions.

Description

Technical Field

[0001] The invention relates to the technical field of rolling bearing fault diagnosis, in particular to a non-iterative deconvolution method for extracting bearing fault features under variable speed conditions. Background Art

[0002] As a key component of rotating equipment, rolling bearings are widely used in rotating machinery in various fields such as aircraft engines, wind turbines and high-speed trains. Their health status has a very important impact on the normal operation of the entire equipment. Therefore, it is necessary to develop effective bearing fault feature extraction technology for bearing fault detection and identification. In the initial stage of bearing failure, the local defects on the bearing surface are relatively small, so the amplitude of the impact characteristics generated cyclically at the fault location is usually low, making the fault characteristics weak and easily masked by large noise and strong interference components in the vibration signal. Therefore, the extraction and enhancement of bearing fault features is an important factor in determining whether bearing faults can be discovered in a timely and accurate manner.

[0003] Bearings are generally composed of an inner ring, an outer ring, rollers and a cage. Usually, during use, the inner ring is fixed and the outer ring rotates, or the outer ring is fixed and the inner ring rotates. When the inner ring and the outer ring move relative to each other, the rollers roll under radial loads. After a long period of load movement, the inner or outer ring raceway is prone to peeling, cracks, pitting and other faults. At this time, when the roller contacts the fault point, due to the change in contact stiffness, a fault feature will be introduced into the vibration signal. The frequency of this feature is positively correlated with the rotation speed of the inner or outer ring, and the ratio is only determined by the bearing structure parameters.

[0004] At present, a large number of signal processing methods have been proposed and used to extract and enhance the fault characteristics of rolling bearings from vibration signals. The deconvolution method is an important technical means to achieve this goal, and has also been widely studied and improved in recent years. This method can extract fault characteristics by setting a certain objective function, obtaining the filter signal and filter parameters corresponding to the maximized function. There are two types of deconvolution methods: iterative solution and non-iterative solution. The non-iterative deconvolution method can achieve fast solution of the algorithm and avoid the situation of local optimal solution. However, the existing non-iterative deconvolution method can only realize the extraction of rolling bearing features under constant speed conditions, and cannot achieve satisfactory results under the more common variable speed conditions. Summary of the invention

[0005] The purpose of the present invention is to address the problems existing in the background technology and to propose a non-iterative deconvolution method for bearing fault feature extraction under variable speed conditions, which can efficiently and accurately extract fault feature information and effectively diagnose faults under variable speed conditions.

[0006] The technical solution of the present invention is a non-iterative deconvolution method for extracting bearing fault features under variable speed conditions, comprising the following steps:

[0007] S1. Attach the vibration acceleration sensor to the bearing cover at the end of the rolling bearing to be tested, perform high-frequency sampling on the bearing vibration signal to obtain vibration data. According to the sampling frequency and rotation speed, intercept the signal within a period of time directly as the original time domain signal, recorded as x = [x1, x2, ..., x N ] T , and synchronously collect the bearing speed signal, denoted as v = [v1, v2, ..., v N ] T ;

[0008] S2. Calculate the cumulative relative rotation angle of the inner and outer rings of the bearing using the speed signal

[0009]

[0010] According to the characteristic order of bearing fault, the theoretical rotation angle Δθ corresponding to the interval between adjacent fault impacts is calculated:

[0011] Δθ=360 / FCO

[0012] Among them, FCO represents the fault characteristic order of the bearing;

[0013] S3. Based on S2, construct an alternative target vector t=[t1, t2, ..., t N ] T Indicates the relative position of the fault impulse:

[0014]

[0015] Among them, K represents The rounded value of θ0 represents the initial position of the first fault impact. Since the actual value of θ0 is uncertain, a set of θ0 is set at equal intervals in the range of 0 to Δθ to generate the candidate target vector sequence t i ,i=1,2,...,N t , the specific number of θ0 is:

[0016]

[0017] S4, the candidate target vector t i ,i=1,2,...,N t Substitute it into the deconvolution solution formula with multi-point D norm as the objective function to solve it. The multi-point D norm is expressed as:

[0018]

[0019] Where y represents the input vibration signal x=[x1,x2,...,x N ] T After the filter f=[f1,f2,...,f L ] T filtering a filtered signal obtained by filtering;

[0020] Select the filtered signal with the maximum objective function value as the final deconvolution result

[0021] S5: Fault-related features extracted in S4 Resampling is performed, and the number of sampling points per unit time is inversely proportional to the instantaneous speed, and a time-domain stable signal is obtained.

[0022] S6. Yes After the Hilbert transform, a fast Fourier transform is performed to calculate the fault-related characteristic order envelope spectrum and output the diagnosis results.

[0023] Preferably, the solution formula using multi-point D norm maximization as the optimization condition is as follows:

[0024]

[0025]

[0026]

[0027] Compared with the prior art, the present invention has the following beneficial technical effects:

[0028] The present invention makes up for the shortcomings of the existing deconvolution method, accurately extracts fault feature information under variable speed conditions while ensuring high efficiency, effectively diagnoses the fault, and has good robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a schematic diagram of the structure of a test bench according to an embodiment of the present invention;

[0030] Figure 2 This is a rolling bearing inner ring fault diagram according to an embodiment of the present invention;

[0031] Figure 3 is a flow chart of a method according to an embodiment of the present invention;

[0032] Figure 4 is the original time domain speed signal of the embodiment of the present invention;

[0033] Figure 5 is the original time domain vibration signal of the embodiment of the present invention;

[0034] Figure 6 is the envelope order spectrum of the original time domain signal of an embodiment of the present invention;

[0035] Figure 7 Fault-related features extracted by the embodiment of the present invention;

[0036] Figure 8 This is the fault-related feature envelope order spectrum extracted by the embodiment of the present invention.

[0037] Figure numerals: 1. driving motor; 2. planetary gearbox; 3. fixed-axis gearbox; 4. torque sensor; 5. bearing seat; 6. rolling bearing; 7. brake. DETAILED DESCRIPTION

[0038] Taking a rolling bearing fault simulation test bench of a wind turbine transmission system as an example, the rolling bearing fault simulation test bench includes a drive motor 1, a planetary gearbox 2, a fixed axis gearbox 3, a torque sensor 4, a bearing seat 5, a rolling bearing 6 and a brake 7. Figure 1 As shown, the driving motor 1 drives the planetary gearbox 2 and the fixed axis gearbox 3 to transmit, and the output end of the fixed axis gearbox 3 is connected to the brake 7 through the rolling bearing 6 installed in the bearing seat 5. The fault type of the rolling bearing 6 is the inner ring crack fault, such as Figure 2 The vibration sensor M is installed on the upper part of the bearing seat 5, and the key phase sensor N is installed on the output end of the fixed axis gearbox 3. The test system performs high-frequency sampling and data storage on the rolling bearing vibration signal. The sampling process frequency is 20000Hz and the sampling time is 15s.

[0039] The outer ring of the rolling bearing on the test bench of this embodiment is fixed, and the inner ring is driven to rotate by the rotating shaft. The specific fault characteristic parameters are shown in Table 1.

[0040] Table 1: Fault characteristic order of each component of rolling bearing

[0041]

[0042] The present invention is used to analyze the original data for rolling bearing fault diagnosis to determine the bearing fault type. Figure 3 As shown, the present invention proposes a non-iterative deconvolution method for extracting bearing fault features under variable speed conditions, comprising the following steps:

[0043] S1. Use the key phase sensor and vibration sensor to synchronously collect and obtain the inner ring speed and bearing vibration signals of the rolling bearing. According to the sampling frequency, intercept 1s of data as the original signal, record the speed signal as v, record the vibration signal as x, and the original speed signal as Figure 4 As shown, the original time domain vibration signal is Figure 5 As shown, the envelope order spectrum of the original time domain signal is Figure 6 As shown in the figure, no obvious fault features can be found from the time domain signal and envelope order spectrum.

[0044] S2. Calculate the cumulative relative rotation angle of the inner and outer rings of the bearing using the speed signal

[0045]

[0046] According to the characteristic order of bearing fault, the theoretical rotation angle Δθ corresponding to the interval between adjacent fault impacts is calculated:

[0047] Δθ=360 / FCO

[0048] Among them, FCO represents the fault characteristic order of the bearing;

[0049] S3. Based on S2, construct an alternative target vector t=[t1, t2, ..., t N ] T Indicates the relative position of the fault impulse:

[0050]

[0051] Among them, K represents The rounded value of θ0 represents the initial position of the first fault impact. Since the actual value of θ0 is uncertain, a set of θ0 is set at equal intervals in the range of 0 to Δθ to generate the candidate target vector sequence t i ,i=1,2,...,N t , the specific number of θ0 is:

[0052]

[0053] S4, the candidate target vector t i ,i=1,2,...,N t Substitute it into the deconvolution solution formula with multi-point D norm as the objective function to solve it. The multi-point D norm is expressed as:

[0054]

[0055] Where y represents the input vibration signal x=[x1,x2,...,x N ] T After the filter f=[f1,f2,...,f L ] T The solution formula for the filtered signal obtained by filtering with multi-point D norm maximization as the optimization condition is as follows:

[0056]

[0057]

[0058]

[0059] Select the filtered signal with the maximum objective function value as the final deconvolution result

[0060] S5: Fault-related features extracted in S4 Resampling is performed, and the number of sampling points per unit time is inversely proportional to the instantaneous speed, and a time-domain stable signal is obtained.

[0061] S6. Yes After the Hilbert transform, a fast Fourier transform is performed to calculate the fault-related characteristic order envelope spectrum and output the diagnosis results.

[0062] The fault-related features and fault-related feature envelope order spectrum obtained by the present invention are as follows: Figure 7 and Figure 8 As shown. Very obvious impact features can be observed from the time domain, and the higher the speed, the smaller the feature interval, indicating that the impact features are related to the bearing fault. It can also be seen from the extracted feature envelope order spectrum that there are obvious peaks at the inner ring fault order and its harmonics, which are consistent with the fault characteristics, and accurate diagnosis of rolling bearing faults is achieved. At the same time, due to the non-iterative properties of the method used, the average usage time for 50 repeated calculations using MATLAB R2020b software on a computer configured with Intel i7-6700K CPU and 16 GB RAM is only 0.3075 seconds, indicating that the method of the present invention has very high efficiency.

[0063] The present invention makes up for the shortcomings of the existing deconvolution method, accurately extracts fault feature information under variable speed conditions while ensuring high efficiency, effectively diagnoses the fault, and has good robustness.

[0064] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A non-iterative deconvolution method for bearing fault feature extraction under variable speed conditions, characterized in that: The steps include: S1. Attach the vibration acceleration sensor to the bearing cover at the end of the rolling bearing to be tested, perform high-frequency sampling on the bearing vibration signal to obtain vibration data. According to the sampling frequency and rotation speed, intercept the signal within a period of time directly as the original time domain signal, recorded as x = [x1, x2, ..., x N ] T , and synchronously collect the bearing speed signal, denoted as v = [v1, v2, ..., v N ] T ; S2. Calculate the cumulative relative rotation angle of the inner and outer rings of the bearing using the speed signal According to the characteristic order of bearing fault, the theoretical rotation angle Δθ corresponding to the interval between adjacent fault impacts is calculated: Δθ=360 / FCO Among them, FCO represents the fault characteristic order of the bearing; S3. Based on S2, construct an alternative target vector t=[t1, t2, ..., t N ] T Indicates the relative position of the fault impulse: Among them, K represents The rounded value of θ0 represents the initial position of the first fault impact. Since the actual value of θ0 is uncertain, a set of θ0 is set at equal intervals in the range of 0 to Δθ to generate the candidate target vector sequence t i ,i=1,2,...,N t , the specific number of θ0 is: S4, the candidate target vector t i ,i=1,2,...,N t Substitute it into the deconvolution solution formula with multi-point D norm as the objective function to solve it. The multi-point D norm is expressed as: Where y represents the input vibration signal x=[x1,x2,...,x N ] T After the filter f=[f1,f2,...,f L ] T filtering a filtered signal obtained by filtering; Select the filtered signal with the maximum objective function value as the final deconvolution result S5: Fault-related features extracted in S4 Resampling is performed, and the number of sampling points per unit time is inversely proportional to the instantaneous speed, and a time-domain stable signal is obtained. S6. Yes After the Hilbert transform, a fast Fourier transform is performed to calculate the fault-related characteristic order envelope spectrum and output the diagnosis results.

2. The non-iterative deconvolution method for bearing fault feature extraction under variable speed conditions according to claim 1 is characterized in that: The solution formula with multi-point D norm maximization as the optimization condition is as follows:

Images