CN111947927B

CN111947927B - A fault detection method for rolling bearings based on chromaticity theory

Info

Publication number: CN111947927B
Application number: CN202010688970.0A
Authority: CN
Inventors: 刘剑慰; 姜斌; 杨蒲
Original assignee: Nanjing University of Aeronautics and Astronautics
Current assignee: Nanjing University of Aeronautics and Astronautics
Priority date: 2020-07-16
Filing date: 2020-07-16
Publication date: 2021-09-07
Anticipated expiration: 2040-07-16
Also published as: CN111947927A

Abstract

The invention relates to a rolling bearing fault detection method based on chromaticity theory. According to the collected bearing vibration information, the chromaticity algorithm is used to calculate the RGB-like characteristics of the bearing, and the support vector data description algorithm SVDD is used for classification to realize fault detection. The invention belongs to the technical field of bearing detection. The fault detection method described in the present invention is to analyze the vibration signal that is easily obtained by the bearing, and the fast Fourier transform converts the vibration sequence data into frequency domain data, and uses three overlapping digital filters to convert the frequency domain data into class data. RGB data, and classify it with SVDD as a classifier to obtain a representation of normal and fault states, respectively. The method can provide basis for bearing fault detection.

Description

Rolling bearing fault detection method based on chromaticity theory

Technical Field

The invention relates to a rolling bearing fault state analysis device and method based on a chromaticity theory, and belongs to the technical field of fault diagnosis of electromechanical systems.

Background

Ball bearings have been widely used in various ac/dc motor transmission and mechanical rotation applications as a common support and rotation isolation device in electromechanical systems. With the improvement of production requirements, an electromechanical system is developed towards the direction of high power and high rotating speed, and is more and more applied to severe working conditions such as humidity, high temperature, strong impact and the like, so that the aging and abrasion of a bearing are aggravated, and the production stop loss and even the accident risk caused by the bearing fault are increased. Therefore, the real-time fault detection of the rolling bearing has important economic and social values.

The traditional bearing fault detection mainly adopts a manual detection method, and changes of temperature, sound, oil color and the like of the bearing are observed by using human senses so as to judge whether the bearing normally operates. Most of the methods rely on the working experience of operators, so that the diagnosis efficiency is low, and the potential faults of the bearings are difficult to detect in time.

In order to monitor the working state of the bearing, companies at home and abroad develop products for on-line monitoring of the bearing state, for example, the Entek-IRD company in America adopts a peak energy technology, excites the self structure of a machine to generate self vibration according to mechanical peak energy, and modulates the self vibration frequency as a carrier frequency and a fault frequency to obtain a peak energy spectrum so as to judge the fault; the Peak Vue technology researched by Emerson company collects and monitors the stress wave of a bearing, obtains the Peak value and the frequency of the stress wave, and converts the Peak value into a frequency spectrum for analysis; the Shock Pulse technique (Shock Pulse Method) developed by SKF corporation, sweden monitors the vibration acceleration of the rolling bearing during operation, and then analyzes the vibration acceleration after envelope detection demodulation to determine the running state of the bearing. A large number of students in China also use vibration signals of bearings to detect faults, and the variable mode decomposition method and the resonance demodulation technology are used for detecting early faults of the rolling bearings compared with how to detect the early faults of the rolling bearings; yuanhongeast and the like use vibration energy to diagnose the fault characteristics and distinguish the state of the bearing; identifying early fault types of inner and outer rings of a bearing by using morphological difference filtering and differential entropy; the Cipei cloud and the like adopt an EEMD method to decompose the vibration characteristics of the rolling bearing and use a support vector machine to carry out mode identification to realize fault diagnosis; li Shi Zhi and the like adopt a fine composite multi-scale diffusion entropy and support vector machine method to carry out fault diagnosis on the rolling bearing. The methods all achieve certain effects, but as the rolling bearing is an object with a complex working state and serious interference in a working site, how to quickly and effectively obtain the fault characteristics of the rolling bearing according to the bearing operation data is still a difficult problem attracting a plurality of students to research.

Given the problems discussed above, it is desirable to devise a method and apparatus that can collect and quickly detect rolling bearing faults online, which can detect bearing anomalies based on the vibration conditions of the bearing.

The invention adopts two working signals of the rotating speed omega and the vibration x to carry out fault analysis. And fusing the rotating speed omega and the vibration signal to form data z (k), calculating an RGB-like signal related to the bearing state by using z (k), and performing fault analysis according to the change of RGB. The fault detection method designed by the invention is convenient to use and can be carried out on line.

Disclosure of Invention

The invention aims to provide a method for rapidly detecting faults of a rolling bearing according to rotating speed and vibration information. A method flow diagram is shown in fig. 1.

In order to achieve the purpose, the invention adopts the following technical scheme:

1. the real-time fault detection device consists of a computer, a corresponding data acquisition card, a vibration sensor and a rotating speed sensor. The fault detection module is installed in the computer in the form of software. The computer sends the instruction signal to the motor system, and transmits the current rotating speed and vibration signal back to the data acquisition card, and then converts the current rotating speed and vibration signal into a digital signal to the computer. And the computer calculates the RGB value of the class corresponding to the data sequence according to the operation parameters, and then analyzes the change of the health state of the bearing according to the change of the value.

2. The class RGB calculation module calculates a class RGB value according to the rolling bearing vibration time sequence data z (k), wherein k is sampling time, and the class RGB value is classified according to the class RGB value so as to judge the change of the health state of the bearing, and the method comprises the following specific steps:

step 1), calculating the length L of a reference sequence according to the rotation speed signal omega rpm and the vibration data sampling frequency f Hz:

wherein, floor (.) is to round the data by rounding off the tail. A is typically between 3 and 10.

And 2) segmenting the rolling bearing vibration time sequence data z (k) by taking the length L of the reference sequence as the length and beta as the overlapping proportion. Form an array Z

Z＝[z₁ z₂...z_i...] (2)

Wherein each z is_iComprising L data, z_iThe initial data of (a) is the i flow (L (1- β)) data of the original sequence.

As the measurements continue, the array Z continues to expand.

Step 3), calculating each z_iComprises the following steps:

step 3.1), sequence z in the array is compared_iPerforming Fast Fourier Transform (FFT) to convert into a sequence p in a frequency domain_i. Each p_iThe number of the data is L/2+ 1.

Step 3.2), 3 digital filter arrays R with overlapping bandwidth ranges are set_f、G_f、B_fWherein, in the step (A),

the last 0 in the array has L/4+ 1;

the first 0 in the array has L/4+1

For each p_iAnd calculating the similar RGB values by vector multiplication respectively:

where T denotes transposing the array.

Each group Ri, Gi, Bi constitutes a 3-dimensional point.

The new bearing that has just started to be put into operation is considered to be in a normal state. For the measured initial vibration data, RGB-like values are obtained through steps 1 to 3 as normal state data.

And 4) calculating a hypersphere containing all points for the space point set of the RGB data in the normal state, namely the hypersphere in the normal state, and obtaining the sphere center c and the radius r. The algorithm for computing hypersphere can use Support Vector Data Description (SVDD) method, or other machine learning method.

And 5) calculating the similar RGB value of the data to be detected and the distance d between the data to be detected and the sphere center c in the RGB space. And judging whether the bearing is in a normal state according to whether the radius of the spherical surface exceeds lambda times of the radius of the hypersphere. If the radius of the hypersphere is more than lambda times, the bearing is considered to start to have a fault.

Wherein, λ is selected according to the system environment condition, and can be 2 times in general.

3. Under the condition that the bearing normal state data set and multiple fault state data sets are provided, the normal state and fault state RGB values can be calculated according to the steps 1-3, and a normal state classifier and a fault state classifier are trained in the step 4 respectively to obtain a plurality of hypersphere surfaces. And judging whether the fault occurs and the fault type according to the hypersphere in which the data to be detected falls.

4. To reduce the interference effect, multiple sensors can be mounted on the same bearing, and multiple vibration data x can be obtained simultaneously_i(k) 1., m. And performing wavelet transformation on the m groups of data, reconstructing a low-frequency part of wavelet decomposition, and averaging to obtain a new vibration data sequence z (k).

Compared with the prior bearing fault detection technology, the invention has the beneficial effects that:

(1) the invention converts the vibration signal into the similar RGB value, is relatively convenient to calculate, can be visually displayed by chromaticity and is convenient for manual observation.

(2) After the quasi-RGB value is calculated, the space lattice can be obtained, the machine learning artificial intelligence method is convenient to classify, and the method has good expansibility.

Drawings

FIG. 1 is a flow chart of an embodiment of real-time fault detection;

FIG. 2 is the calculation of bearing RGB values (bearing class RGB in normal and fault conditions);

FIG. 3 shows the comparison of the bearing RGB and the distance between the center of the hypersphere and the radius of the hypersphere (comparing the distance between the normal data and the center of the hypersphere with the detection threshold)

FIG. 4 comparison of fault data to circle center distance to fault detection threshold

Detailed Description

The following will further explain the implementation of the present invention by taking the fault detection of bearing fault test data of the university of Keiss Caesar as an example with reference to FIG. 2.

Embodiment 1:

the bearing testing experimental platform manufactured by the university of Keiss Sichu comprises a 2-horsepower motor, a torque sensor, a power meter and electronic control equipment. The bearing under test supports the motor shaft. A single point of failure is placed on the bearing using electrical discharge machining techniques. In the experiment, an acceleration sensor is used for collecting vibration signals, and the acceleration sensor is respectively arranged at the driving end of the motor shell and the 12 o' clock position of the fan end. The vibration signals were collected by a 16 channel DAT recorder. The vibration signal sampling frequency is f-12 kHz. Bearing model 6205-2RS JEM SKF.

In this case, a test data set under normal conditions without load is selected, the RGB-like values are calculated, and the hypersphere under normal conditions is calculated to obtain the center of a circle and the radius. And then selecting a test data set with a fault at the driving end, and calculating the RGB-like value and the distance between the RGB-like value and the center of the normal state hypersphere. And comparing the distance value with the radius, and judging whether a fault occurs. The process of computing class RGB is as follows:

step 1), according to the rotation speed signal ω 1797 rpm, taking α 10 to obtain the length L of the reference sequence:

wherein, floor (.) is to round the data by rounding off the tail.

And 2) segmenting the sequence z (k) by taking the length L of the reference sequence as the length and the overlapping proportion of beta, wherein the beta can be selected from 0 to 90 percent, and the beta can be 20 percent. Forming a super array Z

Z＝[z₁ z₂...z_i...]

As the measurements continue, the superset Z continues to expand.

Wherein the element z of the superset group_iThe initial data of (a) is the number of i flow (L (1- β)) of the original sequence.

Step 3), calculating the class RGB values corresponding to each array element, comprising the following steps:

step 3.1), for z in the array_iThe sequence is subjected to Fast Fourier Transform (FFT) to be converted into a sequence p of a frequency domain_i. Each p_iThe number of the data is L/2+ 1.

the last 0 in the array has L/4+ 1;

the first 0 in the array has L/4+1

where T denotes transposing the array.

Each group Ri, Gi, Bi constitutes a 3-dimensional point. The set of spatial points of the RGB data is calculated for the normal case, as shown by the triangular lattice of fig. 2.

And 4) calculating a hypersphere containing all points, namely a hypersphere in a normal state, by using a Support Vector Data Description (SVDD) method for the normal state data set to obtain a sphere center c ═ 3.3931, 0.812, 0.1518, and a radius r ═ 0.0001116. Other machine learning methods can also be used for the algorithm for computing the hypersphere. And judging whether the bearing is in a normal state according to whether the radius of the spherical surface exceeds lambda times of the radius of the hypersphere. Here, λ is 2, i.e., the failure detection threshold 0.0002223. As shown in fig. 3.

And 5) calculating the similar RGB values of the data to be detected according to the steps 1 to 3, as shown in the circular lattice of FIG. 2. The distance d between the RGB point and the center c of the sphere is then calculated. If the distance exceeds the threshold, the bearing is considered to begin to fail.

As shown in fig. 4, since the distance between the failure points exceeds the threshold, it is possible to correctly determine that the failure is caused.

Embodiment 2:

a data set of a bearing test experiment platform of the university of Keys Sichuang is adopted, and the data set comprises 4 kinds of state data, namely Normal state Normal, inner ring fault Faultln, rolling ball fault FaultBill, outer ring fault FaultOut and the like.

Steps 1 to 3 are the same as steps 1 to 3 in embodiment 1, and the RGB-like values of 4 states are calculated.

And 4) calculating the hypersphere by respectively adopting a support vector data description method for the 4 state data sets to obtain 4 sphere centers and radiuses. The normal state hypersphere center c1 is [3.3931, 0.812, 0.1518], and λ is 2, that is, the fault detection threshold 0.0002223. The supersphere center c1 of the inner ring fault state is [31.3754, 6.4175 and 0.4353], and λ is 1, namely the threshold value is 0.0261. The hypersphere center c1 is [7.0926, 1.2328, 0.1373] and λ is 1, i.e. the threshold value 0.0003906. The outer ring fault state hypersphere center c1 is [38.6083, 7.7409, 0.212], and λ is 1, i.e., the threshold 0.0943.

And 5) calculating the similar RGB value of the data to be detected according to the steps 1 to 3. And then calculating the distance d between the RGB point and each hypersphere sphere center c. If the distance is less than the corresponding threshold, the class belongs to. If the fault does not belong to any class, the fault is judged to be of an unknown type. Therefore, the method can not only detect the faults, but also classify the faults.

Claims

1. A rolling bearing fault detection method based on chromaticity theory is characterized as follows:

According to the vibration time series data z(k) of the rolling bearing, k is the sampling time, calculate the RGB-like value of the bearing, and analyze the change of the fault state of the bearing according to the RGB-like value, including the following specific steps:

Step 1), according to the rotational speed signal ω rev/min, the vibration data sampling frequency f Hz, calculate the reference sequence length L:

Among them, floor(.) is to round off the data; α is between 3 and 10;

Step 2), with the reference sequence length L as the length and β as the overlap ratio, segment the rolling bearing vibration time series data z(k) to form a sequence group Z

Z={z ₁ z ₂ ... z _i ... } (2)

Among them, each _zi includes L data, and its starting data is the i*floor(L*(1-β)) data of the original time series;

As the measurement continues, the sequence group Z continues to expand;

Step 3), calculate the class RGB value of each _zi , including the following steps:

Step 3.1), carry out fast Fourier transform FFT to each _zi sequence, convert into the sequence _pi of frequency domain, each _pi contains L/2+1 data;

Step 3.2), set 3 digital filter arrays R _f , G _f , B _f with overlapping bandwidth ranges, among which,

The last 0 in the array has L/4+1;

There are L/4+1 0s at the beginning of the array;

For each p _i , compute the RGB-like values separately using vector multiplication:

R _i =2p _i *R _f ^T

G _i = p _i *G _f ^T (4)

B _i =2p _i *B _f ^T

Among them, T represents the transpose of the array;

Each group of Ri, Gi, Bi constitutes a 3-dimensional point;

The new bearing that has just been put into operation is considered to be in a normal state, and for the measured initial vibration data, the RGB-like value is obtained through the above steps, which is the normal state data;

Step 4), for the spatial point set of the RGB data in the normal state, calculate a hypersphere containing all points, that is, the hypersphere in the normal state, and obtain the center c and the radius r; the algorithm for calculating the hypersphere can be described by support vector data ( SVDD) method, or other machine learning methods;

Step 5), the data to be detected, calculate the class RGB value, and the distance d from the center of the sphere c in the RGB space, and judge whether the bearing is in a normal state according to whether it exceeds λ times the radius of the hypersphere; If the radius is λ times, it is considered that the bearing begins to fail;

Among them, λ is selected according to the system environmental conditions, and can be more than 2 times.

2. A method for detecting faults of rolling bearings based on chromaticity theory according to claim 1, characterized in that: in the case of a bearing fault state data set and a normal state data set, according to steps 1 to 3, it is possible to calculate Normal state and fault state class RGB values, in step 4, the normal state classifier and the fault state classifier are trained respectively, multiple hyperspheres; according to which hypersphere the data to be detected falls in, it can be judged whether there is a fault, and the type of fault .