CN114201831A

CN114201831A - Rolling bearing working condition quantitative analysis method based on vibration signal real-time acquisition

Info

Publication number: CN114201831A
Application number: CN202111515069.4A
Authority: CN
Inventors: 王孝良; 李校舟
Original assignee: Dalian University of Technology
Current assignee: Dalian University of Technology
Priority date: 2021-12-13
Filing date: 2021-12-13
Publication date: 2022-03-18

Abstract

A rolling bearing working condition quantitative analysis method based on vibration signal real-time acquisition comprises the following steps: calibrating a vibration acceleration sensor by using a standard force sensor, and establishing a mathematical mapping relation between the readings of the vibration sensor and the load force; after calibration, installing the device on an industrial site, installing a temperature sensor, and performing real-time data acquisition; extracting the rotating speed from the vibration signal by combining a time domain and frequency domain signal processing method; suppressing the frequency conversion component in the vibration signal by using singular value decomposition, extracting load information from the reconstructed signal, and judging the load impact; reversely deducing a load value by using the calibrated characteristics of the vibration sensor, and quantizing each variable load process by using load characteristic parameters and average load intensity for working condition description; and finally, taking the readings of the temperature sensor and the material parameters of the bearing as auxiliary references. The method is used in an industrial field, is used as an intermediate layer of a bearing life prediction network model to improve training effect and recognition accuracy, and is beneficial to dynamic analysis and correction of the life prediction model.

Description

Rolling bearing working condition quantitative analysis method based on vibration signal real-time acquisition

Technical Field

The invention belongs to the field of equipment fault diagnosis and service life prediction, and relates to a rolling bearing working condition quantitative identification method based on real-time vibration data.

Background

The bearing is called an industrial joint, and the rolling bearing plays an important role in industry and is one of the parts which are easily damaged. The state of health, the remaining service life of the rolling bearing is therefore closely linked to the safe and reliable operation of the industrial production line. Through monitoring the operating condition of antifriction bearing, the remaining life of accurate prediction bearing can help the enterprise to reduce cost of maintenance and shut down the risk of production, avoids the incident because of the bearing damage leads to, has very important industrial value.

At present, the life prediction of the rolling bearing has attracted wide attention of domestic and foreign scholars. The patent CN113326590A proposes a method and a device for predicting the life of a rolling bearing based on a deep learning model, which uses deep learning to model and analyze time series data and predict the life of the rolling bearing; the patent CN113449465A proposes a life prediction method for a rolling bearing, which performs dimension reduction on the vibration signal characteristics of the bearing to obtain an initial characteristic matrix, performs nonlinear fitting, and then trains with a time convolution network model to predict the life of the bearing; patent CN113011463A proposes a method for predicting the life of a bearing of a chinese herbal medicine device based on improved GA-BP, which uses an improved genetic algorithm to optimize a BP neural network for performing life prediction on degradation indexes obtained from vibration signals. The methods are all based on the vibration data of specific rotating speed and stable working conditions for experimental research, and the actual working conditions are complex and random, so that the bearing life research without the working conditions has no practical significance. Patent CN112834222A proposes a method and an electronic device for dynamically monitoring the service life of a train bearing, which calculate the radial force and the axial force of the train bearing, equivalent to load equivalent, according to the triaxial acceleration data of the bearing, and calculate the service life of the train bearing according to a theoretical model of the fatigue life of the bearing. The method considers the working condition before the service life is predicted, has certain guiding significance, but the load of the bearing cannot be accurately recorded by installing a force sensor on a large industrial and mining operation site, and the load in the method mainly comes from the weight of a train and is not subjected to important analysis on complex and variable random loads, so that the universality is low.

In the existing bearing life prediction methods and technologies, no matter based on mechanism modeling or data modeling, the operation environment and the working condition of the bearing are not quantitatively evaluated, but the operation is carried out under the assumption of specific rotating speed, specific load and a full-steady process, so that under complex working conditions, various bearing life prediction methods cannot accurately predict the residual life of the bearing, or the deviation between a predicted value and an actual value is large. Essentially, the evolution of the remaining service life of the same bearing component is regulated differently under different operating modes, different load types and different load capacities. If the characteristic parameters of the working conditions in the bearing operating environment, such as the load type, the load capacity, the rotating speed of the main shaft, the lubricating state and the temperature, can be accurately measured or quantitatively evaluated, the bearing life prediction method can accurately predict the residual life of the bearing.

Disclosure of Invention

In order to solve the problem that the service life prediction of the bearing is inaccurate in complex working conditions, the invention provides a rolling bearing operation condition quantitative evaluation method based on real-time vibration signals.

According to a calculation formula of an L-P fatigue life theory and a correction coefficient thereof, the load, the material property and the service condition of a rolling bearing are all important influence factors of the service life, wherein the material is determined by the type selection of the bearing, and the service condition mainly considers the influence of the running speed under the working condition of the bearing and the lubrication degree under the working temperature condition on the service life of the bearing; the temperature is an important detection quantity capable of reflecting the lubrication degree, and the high temperature easily causes the damage of an oil film and influences the service life of the bearing; the load directly affects the contact stress borne by the bearing, the service life of the bearing is influenced, the bearing has different bearing capacity and different performance degradation speeds due to different types and sizes of loads; the working speed influences the vibration and the impact of the bearing to a certain extent, so that the stress impact generated when the bearing vibrates is different in magnitude, and the degradation influence on the bearing is different.

The invention describes the working condition parameters of the bearing as the following parameters: load characteristic parameters, average load strength, spindle rotation speed, temperature and material parameters. The load characteristic parameters are used for describing the variable load process in a quantized mode and distinguishing whether the external load is a steady type or an impact type, for the variable load, the variable load starting and stopping time, the load peak value and the load change rate are used for describing, specifically, the variable load starting and stopping time, the load peak value describes the maximum instantaneous strength of the load, the load change rate describes the instantaneous impact degree of the external load, and the variable load starting and stopping time is used for representing the acting time of the load. The total load impact amount of the bearing in a certain bearing time can be converted into load strength, so that the average load strength is used for reflecting the load magnitude borne in the load changing process.

In many cases, force sensors and eddy current sensors are inconvenient to install in actual industrial and mining production, and vibration signals are more convenient to detect. Therefore, the bearing load and the working rotating speed are quantitatively identified by detecting the vibration signal, and a plurality of working condition parameters are combined to form a combined parameter to be provided for a subsequent life prediction model for use by combining the temperature detection data and the bearing material parameters.

In order to achieve the above object, the technical scheme adopted by the invention is as follows:

a rolling bearing operation condition quantitative evaluation method based on real-time vibration signals comprises the following steps:

step S1: and calibrating the vibration sensor, and establishing a mapping relation between the vibration acceleration and the load. Since the quantification of the load strength is relative, the sensor is calibrated to a zero value before quantification. And calibrating acceleration data of the vibration sensor on a bearing vibration test bed by using a standard force sensor, and measuring the mapping relation between the load force and the acceleration amplitude by loading.

The calibration adopts a bearing vibration monitoring test bed device, and the device comprises a driving motor, a frequency converter, a main shaft, a flexible coupling, a bearing seat, a rolling bearing, vibration acceleration sensors, a radial loading device and a temperature sensor which are respectively arranged in the horizontal direction and the vertical direction, and a standard force sensor used during calibration, or a set of loading device matched with a digital force display.

Step S2: and mounting the calibrated vibration sensor on a bearing working site, collecting vibration acceleration signals during the running period of the rolling bearing, and simultaneously collecting temperature data to obtain a monitoring data sequence.

Step S3: and identifying the rotating speed according to the vibration signal time domain waveform and the envelope spectrum. In a complex industrial field, a rotating speed sensor cannot be installed to directly measure the rotating speed of a main shaft where a bearing is located under many conditions, so that real-time analysis needs to be carried out through existing vibration signals.

Step S4: and carrying out impact recognition on the external load of the bearing. When the bearing works actually, the bearing is acted by external load at any time, the external load can be stable or impact type, the impact load happens frequently on an industrial site, and the performance of the bearing is greatly damaged, so that the bearing needs to be accurately identified. The impact property of the load is described quantitatively by using the load characteristic parameter provided by the invention.

Step S5: and quantitatively identifying the external load of the bearing. The difference of the load information can cause the difference of the vibration signals, in turn, the vibration characteristics also reflect the load information born by the bearing, and a certain mathematical relationship exists between the load and the vibration signals. And according to a load characteristic curve obtained by the standard force sensor and the vibration acceleration sensor in the S1, calculating the load information contained in the vibration signal in a reverse-deducing way, and quantitatively describing by using the load characteristic parameter and the average load intensity provided by the invention.

In summary, the present invention identifies the most important operating parameters of the rolling bearing during operation from step S3 to step S5: speed of rotation, load characteristics, load strength. The temperature signal can reflect working and lubricating states to a certain extent, so that real-time working condition quantitative data of the rolling bearing are formed by combining the temperature signal measured in real time on an industrial field and material parameters of the bearing.

Further, the calibration of the sensor in step S1 is the key to the accuracy of the method, and mainly includes:

s1-1: building a vibration monitoring test bed for a rolling bearing and installing a counter bearingThe radial loading device of (2) must ensure that the radial force loading direction is perpendicular to the axial direction of the bearing. And a standard force sensor is arranged on the basis, and the magnitude of the real-time radial load force is recorded in an explicit mode. And a vibration acceleration sensor is arranged on the bearing seat to acquire an acceleration signal. In a three-axis coordinate system, the directions x and y are recorded as radial directions, the direction z is recorded as an axial direction, and the acceleration vectors of the axes x and y and the radial acceleration recorded as a bearing are obtained. Radial load force is marked F_rThe radial acceleration is marked as a_r。

S1-2: keeping the shafting centered, not applying external load, acquiring force signals and vibration signals, wherein the working rotating speed is 80% of the bearing limit rotating speed. The force sensor measurement is now approximately 0 and if not, the force sensor needs to be calibrated against this value. The vibration signals are mixed with the rotating speed information, so that the vibration signals are subjected to singular value decomposition, the frequency conversion component is weakened, and the load information is highlighted. In the reconstructed vibration signal without the frequency conversion energy, the time domain average value of the acceleration in the unit time interval delta T is approximately 0, and if not, the acceleration sensor needs to be calibrated according to the time domain average value.

S1-3: on the basis of S12, the radial load is applied to the bearing in an increasing mode within the range of the radial bearing capacity of the bearing, and the loading range is started from 5% of the maximum load value, and every 2% of detection points are added until the maximum value is added. And respectively recording the calibration value of the force and the radial acceleration calibration value of the reconstructed signal after singular value decomposition of the vibration signal.

S1-4: repeating three groups of the tests, and taking the average value of three groups of measured values to obtain the radial load F_rWith radial vibration acceleration a_rThe mapping relationship of (2). Performing interpolation fitting on the two-dimensional data to obtain the load characteristic a of the vibration sensor_r-F_rCurve line.

Further, step S2 mainly includes:

and a calibrated vibration acceleration sensor and a calibrated temperature sensor are installed on the bearing seat in the industrial field, and the bearing operation data is monitored in real time. In many production sites of steel rolling, industrial and mining, etc., force sensors and eddy current sensors cannot be installed for the reasons of performance, safety, convenience, etc., which is also the reason why the vibration sensor is calibrated in step S1.

Further, step S3 mainly includes:

s3-1: and searching a target frequency band in the vibration signal envelope spectrogram. For the safe operation condition of the bearing, the frequency conversion signal in the measured signal is one of the main components, the energy of the frequency conversion signal is high, and the amplitude is reflected to be high in frequency domain analysis. Searching for the positioning frequency conversion in the effective frequency band of the envelope spectrum, and because other characteristic frequencies usually use the frequency conversion as fundamental frequency and the energy of the frequency conversion is large, searching for local amplitude from low frequency, the amplitude is obviously higher than surrounding sideband noise, and the frequency corresponding to the marked highest amplitude is f₀，f₀The target frequency is assumed, and according to a corresponding formula, the rotating speed of a main shaft where a bearing is located is R ═ f₀X 60, error in rotational speed:

in the formula (f)_cFor the sampling frequency, L is the number of FFT operation points. It can be seen that the way to reduce the error is the number of FFT points, but this makes the frequency components more dense and the effect is not ideal. Thus, with f₀And selecting a target frequency band for the center, and performing band-pass filtering on the signal.

S3-2: the original vibration signal is band-pass filtered. At f determined in S31₀As the center frequency, performing band-pass filtering with the frequency conversion error as the bandwidth dimension, i.e. the band-pass interval is [ f ]₀-ΔR/60,f₀+ΔR/60]And obtaining a main shaft frequency conversion time domain signal, wherein the waveform of the time domain signal is close to a sine wave in an ideal state.

S3-3: and calculating the rotation speed of the main shaft. Regarding the signal waveform of the band-pass filtered signal obtained in S32 as a sine wave, determining a sine period on the waveform, which is labeled as T, and then obtaining the spindle rotation speed:

further, step S4 mainly includes:

s4-1: and on the basis of S3, mining the load information in the vibration signal by using singular value decomposition. Under normal operation state, the conversion frequency in the vibration signal is the main component, the energy of the main component is obviously larger than that of the load signal, the original vibration signal is directly used for reflecting the load information, and therefore, the conversion frequency component f is required to be firstly converted₀And carrying out weakening and denoising, and displaying the characteristics of the submerged load signal. The vibration signal is decomposed and reconstructed by adopting Singular Value Decomposition (SVD), and the SVD is a matrix orthogonalization decomposition method and is widely applied to the field of signal processing. Because singular values are inherent characteristics of the matrix, the singular value eigenvectors have a point-to-point mapping relation with the loading state contained in the signal.

Singular value decomposition firstly converts the vibration signal 1h to be processed into a matrix m n form, satisfies h m n, constructs the matrix by using a mode of equal-length continuous truncation, and performs SVD (singular value decomposition) to obtain a singular value matrix S ═ diag (sigma-delta)₁,σ₂,...,σ_r) Where σ is₁≥σ₂≥...≥σ_rThe magnitude of each singular value reflects the energy of the vibration signal, the first i large singular values reflect the main components of the signal, and the remaining r-i smaller singular values reflect the noise components of the signal. Since the main component in the S3 spectrum is the frequency conversion, the first, i.e. largest, singular value is zeroed out, i.e. the frequency conversion component f is set₀And (3) setting the value to be 0, reconstructing the vibration signal and achieving the effect of weakening frequency conversion.

S4-2: the reconstructed acceleration signal is analyzed. Calculating the radial acceleration a in unit time delta T according to the reconstructed vibration signal_r，

Wherein, a_xAnd a_yRespectively representing x-direction and y-direction acceleration calibration values of the reconstructed acceleration signal. According to the load characteristic a of the vibration sensor calibrated in the step S1_r-F_rAnd the curve can reversely analyze the load quantity in the current vibration signal.

S4-3: repeating the procedures S41 and S42 within continuous acquisition time to obtain a series of channelsThe value of the directional acceleration, denoted as a_r[k]Correspondingly obtaining a series of F_r[k]Wherein k is 0,1, 2. Then, a group of load identification values with time sequence attributes are obtained in the real-time monitoring process, the time interval between adjacent observation values is a sampling interval, and F is obtained through drawing_rK two-dimensional images, which can visually express load change information in the monitoring time.

S4-4: on the basis of S43, each point is piecewise interpolated and fitted into a curve. The time period parallel to the horizontal axis is stable load, the curve is in a monotone ascending trend in the loading process, and the curve is in a monotone descending trend in the unloading process. Acting force is generated on the bearing in the loading and unloading processes, so the loading and unloading processes are regarded as variable-load processes. According to the definition of the impact load, when the load action time is short and the duration is less than half of the inherent vibration period of the stressed bearing, the load is regarded as the impact load.

Further, step S5 mainly includes:

s5-1: based on S4, the slope of the curve, i.e. the load change rate, in the time period of the load change process reflects the degree of the load instantaneous impact, and is recorded as mu_k(ii) a The maximum and minimum values of the curve in this time period reflect the maximum strength of the load, denoted F respectively_maxAnd F_min. The parameters are characterized by load characteristic parameters uniformly, if the combination parameter is M (k), then

M(k)＝f(t₁,t₂,μ_k,F_max,F_min)

In the formula, t₁,t₂To record the start and end times of the process described by m (k), with other parameters functioning as described above. If the kth sampling point is taken as the center and a sampling points are continuously taken as the time range of the description process, t is₁＝k-a/2，t₂＝k+a/2。

S5-2: according to S51, load characteristic parameter m (k) f (t)₁,t₂,μ_k,F_max,F_min) And calculating the average load strength. Statistically, the average load intensity is obtained by dividing the load integral value in the period of time by the time length and is recorded as S_kWhich reflects the bearing in this continuous periodThe overall strength under load is calculated by the formula:

where k represents the kth sample point, i.e., the time at this time satisfies t_kWhere a is a positive integer and represents the number of consecutive sampling intervals, the total duration of the described procedure can be denoted as a · Δ T.

The sampling is discrete, so the essence of integrating the above equation is to perform a rectangular or trapezoidal approximation, and the time interval needs to be close to infinity, which degrades the accuracy of calculating the average strength of the load. For the convenience of calculation, the average strength of the load is calculated according to an empirical formula according to the load change rule identified in step S4.

For the case that the loading rule is linear trend, it can be calculated as:

for the case that the loading rule is in a sine trend, the statistical time can be calculated as:

S(k)＝F_min+0.75(F_max-F_min)

for the condition that the load changes in a staged way in the statistical time, a bearing load empirical formula of NSK company is used as reference, and the statistical time a.DELTA.T is divided into T according to each stage₁，T₂，…，T_pP segments in total, assuming load F₁Has a working time of T₁At a rotational speed of R₁…, bearing a load F_pHas a working time of T_pAt a rotational speed of R_pThen the load average strength is calculated as:

in the formula, the ball bearing is defined as "e" 3, and the roller bearing is defined as "e" 10/3.

S5-3: the load characteristic parameters m (k), the average load strength s (k), the real-time rotation speed r (k), the temperature parameters w (k) collected by the temperature sensor, the bearing material parameters are recorded as α, one-dimensional rolling bearing working condition quantization parameters are formed, and the one-dimensional rolling bearing working condition quantization parameters are recorded as E, and then E (k) ═ m (k), s (k), r (k), w (k), and α are calculated. The result is the working condition quantification result obtained by the invention, and the result can be used as a bearing life prediction model modeling based on the working condition parameters and can also be used as a network model intermediate layer for bearing life prediction based on the original monitoring data. The quantitative prediction and representation of the residual life of the bearing are specifically analyzed by combining working condition parameters, and the actual significance and the prediction accuracy of the residual life prediction research work of the bearing are improved by using the method.

The invention has the following beneficial effects:

1) in the existing research, the rolling residual life is predicted under the stable working condition of specific rotating speed and specific load, and the complex and random actual working condition of the bearing cannot be represented. In order to carry out bearing service life prediction work with more practical significance, the invention provides a method for quantizing and identifying working conditions based on real-time vibration data, and the accurate quantization of the working conditions of the bearing is a necessary premise of prediction work.

2) The invention provides a method for identifying the rotating speed of a rotating shaft where a bearing is located according to real-time vibration signals, wherein the vibration responses caused by different rotating speeds of a main shaft are different, and the impacts on the bearing are different. However, many times, due to the convenience and safety of equipment installation in an industrial field, the data transmission performance requirement of a detection environment and the like, it is difficult to install a rotation speed sensor, and a method other than the rotation speed sensor is needed to obtain a real-time rotation speed. The invention utilizes a frequency domain analysis and signal band-pass filtering method to identify the rotating speed, and carries out the next analysis on the bearing working condition based on the rotating speed.

3) The method comprises the load characteristics and the average load strength. In the process of extracting load information contained in the vibration signal, the interference of main shaft frequency conversion in the original signal component and the limitations that frequency domain analysis is applicable to a linear system and not applicable to impact load are fully considered, singular value decomposition is adopted to weaken frequency conversion components, and the reconstructed signal is used for load identification. Meanwhile, before the vibration sensor is used, the acceleration sensor is calibrated by using a standard force sensor, and a mapping relation between an acceleration value and a load is established. After the calibrated vibration sensor is put into field use, the working condition characteristic information in the vibration signal is extracted in real time, the mapping relation between the real-time vibration data and the load characteristic parameter and the average load strength is established, and the load capacity is accurately calculated. The invention completely depends on the capabilities of a vibration sensor, a temperature sensor and a computer which are convenient for industrial field installation, is not limited by industrial environment, has strong operability and strong universality, has excellent effect on quantitative evaluation of the working condition of the bearing, and has very important guiding significance on early warning of bearing faults and prediction of service life.

Drawings

FIG. 1 is an overall flow chart of the present invention.

Fig. 2 is a structural diagram of the bearing vibration test bench of the invention.

Fig. 3 is a schematic view of the three-axis coordinate directions of the present invention.

FIG. 4 is a diagram illustrating a calibration load characteristic curve according to the embodiment.

FIG. 5 is a waveform diagram of vibration signals of the rolling bearing used in the embodiment.

FIG. 6 is a graph showing the rotational speed of the vibration signal used in the embodiment.

FIG. 7 is a partially enlarged view of a rotational speed identification diagram of a vibration signal used in the embodiment.

FIG. 8 is a diagram of a reconstructed signal after singular value decomposition of a vibration signal used in an embodiment.

Fig. 9 is a schematic view of a load recognition process of a vibration signal used in the embodiment.

Fig. 10 is a schematic diagram of the variable load according to different variation laws of the present invention.

FIG. 11 is a schematic diagram of a quantitative evaluation structure of the working conditions used in the present invention.

In fig. 2: 1, driving a motor in a variable frequency mode; 2, a main shaft; 3, coupling; 4 speed/torque sensors; 5 rolling bearings; 6 a temperature sensor; 7 radial force loading means; 8, bearing seats; 9 horizontal/vertical acceleration sensor.

Detailed Description

The following describes an embodiment of the present invention with reference to the drawings.

Example 1:

FIG. 1 shows a rolling bearing working condition quantitative analysis method based on real-time vibration signals, which specifically comprises the following steps:

step S1: the vibration sensor is initially calibrated, and then an acceleration-load curve is measured by using the vibration sensor.

S1-1: FIG. 2 shows the structure of a vibration monitoring test bed for a rolling bearing, wherein a radial loading device for the bearing is arranged according to the direction shown in the figure, and a hydraulic loading mode is selected. The radial force loading direction is ensured to be vertical to the axial direction of the bearing. And a standard force sensor is arranged on the basis, and the size of the radial load is recorded in real time. And a vibration acceleration sensor is arranged on the bearing seat to acquire an acceleration signal. Fig. 3 shows the direction identification of a three-axis coordinate system, which records the x and y directions as radial directions and the z direction as axial directions, and records the x and y acceleration synthesis as the radial acceleration of the bearing. Radial load force is marked F_rThe radial acceleration is marked as a_r。

S1-2: keeping the shafting centered, setting the external load to 0N, adjusting the frequency converter to make the main shaft rotate at 80% of the bearing limit speed, and recording the frequency as f_p. And collecting force signals and vibration signals in real time. The force sensor measurement at this time is approximately 0, if not, denoted as F₀The force sensor needs to be calibrated according to this value. The vibration signal is mixed with the rotating speed information, so that the vibration signal is subjected to singular value decomposition to weaken the frequency conversion component f_pAnd highlighting the load information. In the reconstructed vibration signal without frequency conversion energy, the main components are load information and noise. Because the acceleration is not loaded, the time domain average value of the acceleration in the unit time interval is approximate to 0, if not, the acceleration values in the x direction and the y direction are respectively recorded as a_x0And a_y0The acceleration sensor needs to be calibrated according to the value.

S1-3: within the range of the radial bearing capacity of the bearing, the radial load is applied to the bearing in an incremental manner, and the loading range isStarting at 5% of the maximum load value, every 2% of the test points until the maximum value is reached. And respectively recording the calibration value of the force and the calibration value of the radial acceleration of the vibration signal reconstructed by singular value decomposition. Radial acceleration is noted

Load is noted as F_r[k]＝F'_r[k]-F₀Wherein k is 0,1, …, N-1, N is the length of the collected signal, F'_r[k]、a_x[k]、a_y[k]Respectively, a standard force sensor reading and a vertical acceleration sensor reading and a horizontal acceleration sensor reading.

S1-4: repeating three groups of the tests, and taking the average value of three groups of measured values to obtain the radial load F_rWith radial vibration acceleration a_rThe mapping relationship of (2). Performing interpolation fitting on the two-dimensional data to obtain the load characteristic a of the vibration sensor_r-F_rThe curves are shown in figure 4.

Step S2: and (3) mounting the calibrated vibration acceleration sensor on an industrial field bearing seat, radially mounting and fixing the vibration acceleration sensor according to the figure 2, and simultaneously mounting a temperature sensor as shown in the figure 2 to acquire a vibration acceleration signal and a temperature signal of the rolling bearing during operation. Acquiring a bearing radial vibration signal sequence in operation as a_r[k]And the temperature signal is denoted as W [ k ]]K is 0,1, …, N-1, N is the length of the vibration signal.

Step S3: according to the radial vibration signal a of the bearing_r[k]And calculating the rotating speed of the main shaft where the bearing is located. The rotational speed is recorded as R [ k ]]The frequency of rotation is denoted as f₀[k]. The calculation scheme refers to example 2.

Step S4: and judging the load change trend and identifying the load impact. The method specifically comprises the following steps:

s4-1: according to the conclusion of S3, the singular value decomposition is used to convert the frequency component f₀And (4) weakening and denoising, and fully mining load information submerged in the vibration signal. SVD decomposing the vibration signal to convert the frequency component f₀And after the corresponding singular value is set to be zero, reconstructing the signal. The procedure of this step is as described in example 2.

S4-2: to pairAnd reconstructing the acceleration signal for analysis. Calculating the radial acceleration a in unit time according to the reconstructed vibration signal_r，

Wherein, a_xAnd a_yRespectively representing x-direction and y-direction acceleration calibration values of the reconstructed acceleration signal. According to the load characteristic a of the vibration sensor calibrated in the step S1_r-F_rAnd the curve can reversely analyze the load value reflected by the current vibration signal according to the acceleration value.

S4-3: repeatedly carrying out the processes S41 and S42 within continuous acquisition time to obtain a series of radial acceleration values marked as a_r[k]Correspondingly obtaining a series of F_r[k]Wherein k is 0,1, 2. Then, a group of load identification values with time sequence attributes are obtained in the real-time monitoring process, the time interval between adjacent observation values is a sampling interval, and F is obtained through drawing_rAnd the k two-dimensional scatter diagram can visually express the load change information in the monitoring time.

S4-4: on the basis of S43, each point is piecewise interpolated and fitted into a curve. The time period parallel to the horizontal axis is a steady load, the curve is in a monotone ascending trend in the loading process, the curve is in a monotone descending trend in the unloading process, and under the random variable load condition, one possible situation is shown in figure 9. Acting force is generated on the bearing in the loading and unloading processes, and the loading and unloading processes are regarded as loading acting processes. According to the definition of the impact load, when the load action time is short and the duration is less than half of the inherent vibration period of the stressed bearing, the impact load is considered, and the natural frequency of the radial vibration of the bearing ring with the most obvious vibration performance is as follows:

in the formula, n is the vibration order, takes the positive integer that is greater than 2, and E is elastic modulus, and I is the moment of inertia of lasso cross section, and D is the neutral axis diameter of lasso cross section, and M is the quality of lasso unit length, to steel, substitutes the constant and gets:

where h and b represent the height and thickness of the ferrule, respectively.

Step S5: and carrying out quantitative calculation on the characteristic parameters of the bearing load and the average load strength to obtain quantitative parameters of the operation condition of the bearing.

S5-1: the data and curves obtained from S4 are shown in fig. 9. The slope of the curve, i.e. the load rate of change, during the period of the load change process reflects the degree of transient impact of the load, as shown by μ in fig. 9_a、μ_b、μ_c(ii) a The maximum and minimum values of the curve in a described process reflect the maximum strength of the load, as indicated by F in FIG. 9_maxAnd F_min1. Describing the above process with the load characteristic parameter m (k), there is m (k) ═ f (t)₁,t₂,μ_k,F_max1,F_min1). As shown in FIG. 9, t₁,t₂Starting and stopping time, mu, for describing the process described by M at this moment_kSet of slopes [ mu ] for each observation over a time period_a,...,μ_b,...μ_c,...]I.e. single-valued for linear processes, as shown in fig. 9 where M (k') ═ f (t)₄,t₅,μ_d,F_max2,F_min2) The described process is shown. If the kth sampling point is taken as the center and a sampling points are continuously taken as the time range of the description process, t is₁＝k-a/2，t₂＝k+a/2。

S5-2: according to the load characteristic parameter M (k) ═ f (t)₁,t₂,μ_k,F_max,F_min) The average load strength is calculated. Statistically, the average load intensity is obtained by dividing the load integral value in the period of time by the time length and is recorded as S_kThe integral strength of the bearing load in the continuous time is reflected, and the calculation formula is as follows:

where k represents the sampling point at the kth sampling interval Δ T, i.e., at this timeTime satisfies t_kWhere a is a positive integer and represents the number of consecutive sampling intervals, the total time interval of the whole process can be denoted as a · Δ T. M (k') ═ f (t) in fig. 9₄,t₅,μ_d,F_max2,F_min2) And the centroid of the shadow part is obtained.

Because the samples are discrete, the essence of integrating the above equation is to perform a rectangular or trapezoidal approximation of the extreme summation operation. In order to facilitate calculation, the average load strength is calculated according to an empirical formula by combining the identified load change rule. As shown in fig. 10, the calculation formulas for the common linear law, sinusoidal law, and variable load varying in stages are summarized as follows.

For the case that the loading rule is linear trend, it can be calculated as:

S(k)＝F_min+0.75(F_max-F_min)

for the case that the load changes in stages within the statistical time, the whole process time is divided into T₁，T₂，…，T_pN segments in total, assuming load F₁Has a working time of T₁At a rotational speed of R₁…, bearing a load F_pHas a working time of T_pAt a rotational speed of R_pThen the load average strength is calculated as:

For the variable load and variable speed working condition, the result is as follows:

s5-3: and (3) combining the load characteristic parameters M (k), the average load strength S (k), the real-time rotating speed R (k) and the parameters W (k) collected by the temperature sensor and the bearing material parameter alpha to form a one-dimensional rolling bearing working condition quantitative parameter, wherein the parameters E (k) are [ M (k), S (k), R (k), W (k) and alpha ]. The result is the quantization result obtained by the present invention, and the analysis structure of the present invention is shown in fig. 11.

Example 2:

the present embodiment further describes the rotation speed extraction in step S3 and the singular value decomposition in step S41 with reference to fig. 5 to 8.

This example 2 uses the bearing data set of the university of kas sierra bearing, the tested bearing is a SKF6205 deep groove ball bearing, the number of rollers is 9, and the bearing supports the rotating shaft of the motor. The sampling frequency is 12000Hz, the motor loads 3 horsepower, 1 horsepower (manufactured by HP English) is 0.7457 Kilowatts (KW), and the motor rotates at about 1730 rpm.

Step S3: and determining the real-time rotating speed according to the vibration signal. The method specifically comprises the following steps:

s3-1: computing an amplitude spectrum Y [ k ] based on S2]And then calculating the envelope spectrum, checking the frequency distribution condition, and directly calculating the envelope spectrum by using an envelope function of MATLAB software. Obtaining original signal x [ k ] by fast Fourier transform calculation]The spectral FFT transformation formula of (a):

wherein, Y [ k ]]Representing a vibration signal a_r[n]The corresponding complex frequency domain amplitude, k is 0,1, …, N-1, i is the imaginary unit, and N is the length of the vibration signal.

As shown in FIG. 5, the low band image is amplified by searching for the frequency components at the beginning of the low band, and the image has a high amplitude frequency component with an amplitude several times higher than the peripheral sideband, which is defined as the frequency f₀According to fig. 5, the initial value is 30.01 Hz.

S3-2: according to

Calculating the sideband width of 0.6866Hz, and pairing f on the original signal₀-ΔR/60,f₀+ΔR/60]The intervals being band-pass filtered, i.e. [29.3234Hz,30.6966Hz]The filtering method adopts a 2 nd order Butterworth filter and the sampling frequency f_cAt 12kHz, L is 524288 points of FFT operation. And obtaining a main shaft frequency conversion time domain signal, wherein the waveform of the time domain signal is close to a sine wave in an ideal state.

S3-3: on the filtered time domain signal waveform diagram obtained in S32, as shown in fig. 6, the signal is more regular than the original signal, and is partially amplified as shown in fig. 7, the signal waveform is regarded as a sine wave, the sine period of the waveform is determined, and when T is 0.6230-0.5889, which is 0.0341S, the spindle rotation speed is:

in FIG. 7, the conversion frequency is 30.01Hz, and the conversion speed is 1800 rpm. The data set reference speed was 1750 rpm. The rotating speed calculated by the method is closer to a real value, and the effect is more accurate than that of directly taking a frequency point value.

S4-1: and carrying out singular value decomposition and reconstruction on the original vibration signal. According to the conclusion of S3, the singular value decomposition is used to convert the frequency component f₀And carrying out weakening denoising. SVD decomposing the vibration signal to convert the frequency component f₀And after the corresponding singular value is set to be zero, reconstructing the signal. Fig. 8 shows a time domain waveform and a frequency spectrum of the reconstructed signal, and compared with the original signal of fig. 5, the frequency spectrum component has obvious change and is more uniformly distributed than the amplitude value in the original signal, because the frequency conversion component has a larger proportion and is obviously represented, and is suppressed by filtering after reconstruction.

The invention can quantitatively identify and describe the rotating speed, the load characteristics and the load strength of the bearing in operation based on the vibration signal, quantitatively describe the characteristics of the variable load process, effectively inhibit the interference of unnecessary signal components in each link by adopting a signal processing technology and singular value decomposition in the identification process, and improve the quantitative analysis precision. Through accurate calibration and a large amount of engineering use, a set of statistical rules is extracted, the load characteristic curve of the vibration sensor and the coefficient of an average load strength empirical formula are corrected, the calculation precision of working condition quantization parameters is higher, and more important help is provided for the accuracy and the practicability of the bearing service life prediction work.

Claims

1. A rolling bearing operation condition quantitative analysis method based on vibration signal real-time acquisition is characterized by comprising the following steps:

step S1: initially calibrating a vibration sensor, and establishing a mapping relation between vibration acceleration and load; the calibration adopts a bearing vibration monitoring test bed device, which comprises a driving motor, a frequency converter, a main shaft, a flexible coupling, a bearing seat, a rolling bearing, vibration acceleration sensors, a radial loading device and a temperature sensor which are respectively arranged in the horizontal direction and the vertical direction, and a standard force sensor used in the calibration or a set of loading devices matched with a digital force display; the specific process comprises the following steps:

s1-1: building a vibration monitoring test bed of the rolling bearing, and mounting a radial loading device on the bearing; installing a standard force sensor, and recording the size of a radial load in real time; mounting a vibration acceleration sensor on the bearing seat and collecting an acceleration signal;

s1-2: under the working condition that the external load is zero, carrying out zero calibration and calibration on the standard force sensor and the vibration sensor;

s1-3: in the radial bearing range of the bearing, a radial load is applied to the bearing in an incremental manner, and the calibration value of the force and the calibration value of the radial acceleration in the reconstruction signal for filtering the frequency conversion component are recorded respectively;

s1-4: repeating the above test, and averaging multiple groups of measured values to obtain radial load F_rWith radial vibration acceleration a_rThe mapping relationship of (2); performing interpolation fitting on the two-dimensional data to obtain the load characteristic a of the vibration sensor_r-F_rA curve;

step S2: installing the calibrated vibration sensor on a bearing working site, collecting vibration acceleration signals during the running period of the rolling bearing, and simultaneously collecting temperature data to obtain a monitoring data sequence;

step S3: determining a target frequency band of the frequency conversion according to the numerical characteristics of the vibration signal frequency domain diagram, then carrying out band-pass filtering, and identifying the rotating speed according to the oscillogram of the filtered signal;

step S4: carrying out singular value decomposition on the vibration acceleration signal, inhibiting a frequency conversion component, highlighting load information, and carrying out impact identification on the external load of the bearing in a reconstructed signal;

step S5: carrying out quantitative identification on the external load of the bearing; and (3) establishing a mathematical relation between the load and the vibration signal in a time sequence, quantitatively summarizing each variable load process by using the load characteristic parameters, and calculating the average load intensity to quantitatively describe the load bearing intensity.

2. The rolling bearing operation condition quantitative analysis method according to claim 1, wherein the load is a radial external load, the radial direction refers to the radial direction of the bearing, the acceleration is detected respectively in the horizontal direction and the vertical direction, the vector sum of the acceleration and the vector sum is obtained, and the method for filtering the frequency conversion component is singular value decomposition.

3. The rolling bearing operating condition quantitative analysis method according to claim 1 or 2, wherein the step S3 includes:

s3-1: searching a target frequency band in a vibration signal envelope spectrogram; searching local amplitude from low frequency, the amplitude is obviously higher than surrounding sideband noise, and marking the frequency f corresponding to the highest amplitude₀，f₀The target frequency is assumed, and according to a corresponding formula, the rotating speed of a main shaft where a bearing is located is R ═ f₀X 60, error in rotational speed:

in the formula (f)_cTaking the sampling frequency as well as L as the number of FFT operation points; with f₀Selecting a target frequency band for the center, and performing band-pass filtering on the signal;

s3-2: performing band-pass filtering on the original vibration signal; at f determined in S31₀Band-pass filtering, i.e. band-pass, with the frequency-transfer error as a measure of bandwidth for the centre frequencyInterval is [ f₀-ΔR/60,f₀+ΔR/60]Obtaining a main shaft frequency conversion time domain signal, wherein the waveform of the time domain signal is close to a sine wave in an ideal state;

s3-3: calculating the rotation speed of the main shaft; regarding the signal waveform of the band-pass filtered signal obtained in S32 as a sine wave, determining a sine period on the waveform, which is labeled as T, and then obtaining the spindle rotation speed:

4. the rolling bearing operation condition quantitative analysis method according to claim 3, wherein the step S4 comprises:

s4-1: load information in the vibration signals is mined by using singular value decomposition; firstly, weakening and denoising the frequency conversion component, and displaying the characteristics of the submerged load signal;

s4-2: analyzing the reconstructed acceleration signal; calculating the radial acceleration a of the bearing from the reconstructed vibration signal_rReference is made to the load characteristic a of the vibration sensor before being put into use in the industrial field_r-F_rThe curve is inverted to analyze the load characteristics reflected by the current vibration signal;

s4-3: repeating the processes S41 and S42 within the continuous acquisition time to obtain a series of radial acceleration values marked as a_r[k]The corresponding analysis result shows a series of F_r[k]Where k is 0,1,2, i.e. a set of load identification values with time-series attributes is obtained during real-time monitoring, plotted to obtain F_rK two-dimensional images, which can visually express load change information within the monitoring time;

s4-4: on the basis of the two-dimensional scatter diagram, each point is subjected to piecewise interpolation fitting to be drawn into a curve; and judging the load impact property according to the curve characteristics.

5. The rolling bearing operation condition quantitative analysis method based on vibration signal real-time acquisition according to claim 1,2 or 4The method is characterized in that the reconstructed acceleration signals are respectively reconstructed and calibrated in the radial horizontal direction and the radial vertical direction, then the modulus of the vector sum is worked out as an acceleration value,

wherein, a_xAnd a_yRespectively representing x-direction and y-direction acceleration calibration values of the reconstructed acceleration signal.

6. The rolling bearing operation condition quantitative analysis method according to claim 5, wherein the step S5 comprises:

s5-1: on a load time sequence diagram obtained by detection and quantification, mathematically describing each variable load process by using a load characteristic parameter;

s5-2: calculating the average load intensity of each variable load process according to the load characteristic parameters; statistically, the load integral value in the period of time divided by the time length is the average load intensity, which reflects the intensity of the load borne by the bearing in the continuous period of time;

s5-3: the calculated load characteristic parameters, average load strength and real-time rotating speed are combined with the temperature value acquired by the temperature sensor and the bearing material parameter alpha to form one-dimensional rolling bearing working condition quantitative parameters, the result is the quantitative result required by the method, the quantitative result can be used as the input of a bearing service life prediction model based on the working condition parameters to carry out model training, and can also be used as a network model intermediate layer for carrying out bearing service life prediction based on the original acquired data, the residual service life of the quantitative representation bearing is required to be specifically analyzed in combination with the working condition parameters, and the method is used for improving the actual significance and the prediction accuracy of the bearing residual service life prediction research work.

7. The rolling bearing operation condition quantitative analysis method according to claim 6, wherein the load characteristic parameters mainly comprise five parts, namely start time, end time, load change rate, load minimum value and load maximum value of a variable load process.