A rolling bearing fault intelligent identification method based on envelope shape analysis

Abstract

A kind of rolling bearing fault intelligent identification method based on envelope form analysis, first, the original vibration signal of rolling bearing under normal state, outer ring fault, inner ring fault, rolling element fault state is obtained, standardization is handled to obtain original vibration signal standard envelope spectrum, sample library is constructed using original vibration signal standard envelope spectrum, sample library is divided into training set, verification set and test set;Then construct convolutional neural network model and train optimization model, obtain rolling bearing fault identification model;Again, the vibration signal in the working process of the rolling bearing to be measured is collected and converted into the standard envelope spectrum of the vibration signal to be measured;Finally, the standard envelope spectrum of the vibration signal to be measured is input into the rolling bearing fault identification model, and the state of the rolling bearing to be measured is obtained;The present application combines bearing envelope spectrum analysis with deep learning method, automatically identifies rolling bearing fault and type through image, and has high intelligent degree and strong universality.

Description

Technical Field

[0001] This invention relates to the field of rolling bearing fault diagnosis technology, and specifically to an intelligent identification method for rolling bearing faults based on envelope morphology analysis. Background Technology

[0002] As an important component of various rotating machinery, the health status of rolling bearings directly affects the normal operation of the entire equipment. Therefore, monitoring and diagnosing their operating status is of great significance.

[0003] Envelope spectrum analysis (Author: Song Xiaomei, Title: Research on Fault Diagnosis of Rolling Bearings Based on Envelope Demodulation Analysis, Journal: Instrumentation and Analysis Monitoring) is an effective method for diagnosing rolling bearing faults and has been widely used in practical diagnosis. For outer ring faults in rolling bearings, the envelope spectrum contains the characteristic frequency of the outer ring fault and its harmonics; for inner ring faults, the envelope spectrum contains the characteristic frequency of the inner ring fault and its harmonics, as well as sidebands modulated by the frequency conversion of the bearing's inner ring fault characteristic frequency; for rolling element faults, the envelope spectrum contains the characteristic frequency of the inner ring fault and its harmonics, as well as sidebands modulated by the cage fault characteristic frequency of the bearing's inner ring fault characteristic frequency. However, envelope spectrum analysis relies on human observation of the presence or absence of spectral peaks at the fault characteristic frequencies in the spectrum to diagnose faults, which suffers from the problems of subjectivity, high labor costs, and low level of automation.

[0004] In recent years, intelligent fault diagnosis methods such as deep learning (Publication No. CN108426713A, title: A weak fault diagnosis method for rolling bearings based on wavelet transform and deep learning; Publication No. CN108444708A, title: A method for establishing an intelligent diagnostic model for rolling bearings based on convolutional neural networks) have been applied to bearing fault diagnosis. The entire diagnosis process can be completed without human intervention and has a high degree of intelligence. However, training a deep learning model requires a large number of labeled samples, which are often difficult to obtain in practical applications. In addition, intelligent fault diagnosis methods such as deep learning usually assume that the training set and the test set data follow the same probability distribution. However, in actual application scenarios, due to the large differences in bearing models and operating conditions, the training set and the test set data are difficult to follow the same distribution, and the fault diagnosis effect is not ideal. It is often necessary to rebuild the model and retrain it, which is time-consuming and lacks scalability and universality. Summary of the Invention

[0005] In order to overcome the shortcomings of the prior art, the present invention aims to provide an intelligent identification method for rolling bearing faults based on envelope morphology analysis. By combining bearing envelope spectrum analysis with deep learning methods, it can achieve intelligent identification of rolling bearing faults, with a high degree of intelligence and strong versatility.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for intelligent identification of rolling bearing faults based on envelope morphology analysis includes the following steps:

[0008] Step 1: Obtain the original vibration signals of the rolling bearing under normal conditions, outer ring fault, inner ring fault, and rolling element fault conditions;

[0009] Step 2: Standardize the original vibration signal to obtain the standard envelope spectrum of the original vibration signal;

[0010] Step 3: Construct a sample library using the standard envelope spectrum of the original vibration signal to reflect the rolling bearing under normal, outer ring fault, inner ring fault, and rolling element fault conditions.

[0011] Step 4: Divide the sample library into training set, validation set and test set according to a certain ratio;

[0012] Step 5: Construct a two-dimensional convolutional neural network model, train the network model using the training set, optimize and update the network model based on the training results, and obtain the rolling bearing fault identification model.

[0013] Step 6: Collect vibration signals during the operation of the rolling bearing under test. Obtain the standard envelope spectrum of the vibration signal under test according to Step 2.

[0014] Step 7: Input the standard envelope spectrum of the vibration signal to be tested into the trained rolling bearing fault identification model to obtain the state of the rolling bearing to be tested.

[0015] In step one, a vibration acceleration sensor is used to collect the original vibration signals of the rolling bearing under normal, outer ring fault, inner ring fault, and rolling element fault conditions, respectively, and the signal length L of each collection is determined:

[0016] L = f s ×T

[0017] Among them, f s Where is the sampling frequency, and T is the sampling time. Let T = 4s. The signal length must be greater than or equal to this result.

[0018] Step two specifically includes:

[0019] 2.1) The optimal center frequency f of the original vibration signal is selected using the fast spectral kurtosis method. z and bandwidth B w ;

[0020] 2.2) The original vibration signal was bandpass filtered using a Butterworth bandpass filter, with the filter bandwidth selected as [f]. z -B w / 2,f z +Bw [2], to obtain the filtered signal, the squared amplitude function of the Butterworth filter is:

[0021]

[0022] Where j is the order of the filter, w c The cutoff frequency;

[0023] 2.3) Perform Hilbert demodulation on the filtered signal to obtain the demodulated signal;

[0024] 2.4) Perform a Fourier transform on the demodulated signal to obtain the envelope spectrum;

[0025] 2.5) Input the fault characteristic frequency of the bearing and its related frequencies as the search center frequency, set the frequency search error δ, and extract the point with the largest amplitude (F) in the search center frequency error range of the envelope spectrum. i ,a i The search center frequency includes: the outer ring fault characteristic frequency and its harmonics nf. o Inner ring fault characteristic frequency and its harmonics nf i And the modulation frequency nf centered on these frequencies, with sideband size equal to the turn rate. i ±mf r ; characteristic frequencies of rolling element failures and their harmonics nf b And the modulation frequency nf centered at these frequencies, with sideband sizes equal to the cage fault characteristic frequencies. b ±mf c ;

[0026] Where 1≤n≤3, 1≤m≤2, 1≤i≤33 and n, m, and i are all integers, f o f i f b f c These are the characteristic frequencies of failure in the outer ring, inner ring, rolling elements, and cage of a rolling bearing, respectively. r Let δ be the frequency, and δ be the search error, taken as δ = 2Hz;

[0027] 2.6) Calculate the average spectral amplitude u within the envelope spectrum frequency range [0, s]:

[0028]

[0029] In the formula: s=max(3.5fi,3.5fo,3.5fb), S(k) is the envelope spectrum sequence, k=1,2,...,K,K is the number of data points in the envelope spectrum frequency range [0,s];

[0030] Set a threshold p, where p = βu, β is the threshold coefficient, representing the ratio between the threshold and the average amplitude of the spectrum, and take β = 2; when the amplitude of the extracted point is greater than or equal to the threshold, it is considered a valid point; when the amplitude of the extracted point is less than the threshold, it is considered an invalid point, and its amplitude is set to zero.

[0031] 2.7) For point (F) i ,a i Amplitude normalization is performed.

[0032]

[0033] Where 1≤i≤33, i is an integer, a imax For a i The maximum value in;

[0034] 2.8) Establish a new coordinate system with a vertical coordinate range of 0-1 and an horizontal coordinate range of 0-34, with an interval of 1. From left to right, the system is divided into three fault characteristic regions: outer ring fault, inner ring fault, and rolling element fault, corresponding to horizontal coordinates 1-3, 4-18, and 19-33, respectively. The normalized points (F...) i A i Place it within the corresponding fault characteristic area, according to F i Arrange the lines in ascending order to their corresponding horizontal coordinate positions to generate a new spectral diagram, which serves as the standard envelope spectrum of the original vibration signal.

[0035] The sample library construction method in step three is as follows: save the obtained original vibration signal standard envelope spectrum as a JPG format grayscale image with a resolution of 1568×764px; and match all the original vibration signal standard envelope spectrum images of the bearing under normal conditions and under outer ring fault, inner ring fault, and rolling element fault conditions with the bearing conditions to form a sample library.

[0036] The two-dimensional convolutional network neural model in step five includes convolutional layers, pooling layers, batch normalization layers, flattening layers, and fully connected layers. The convolutional layers are used for feature extraction, the pooling layers are used to reduce the spatial dimension of the feature map, the batch normalization layers are used to prevent overfitting and accelerate convergence, the flattening layers are used to flatten the multidimensional feature map into a one-dimensional vector, and the fully connected layers are used for classification and output prediction.

[0037] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0038] 1. This invention is based on the envelope spectrum morphology of typical rolling bearing faults and uses a convolutional neural network model to automatically identify rolling bearing faults and types through images, without the need for manual observation of spectrum characteristics, and has a high degree of intelligence.

[0039] 2. This invention proposes a standardized processing method for rolling bearing vibration signals and constructs a general sample library reflecting rolling bearing fault types. It can identify fault types of different bearing models, has strong versatility, and overcomes the shortcomings of deep learning methods, such as the need for a large number of labeled samples, large differences in the distribution of test sets and training sets across bearing models, and poor diagnostic performance. Attached Figure Description

[0040] Figure 1 This is a flowchart of the present invention.

[0041] Figure 2 This is a flowchart of the process of generating the standard envelope spectrum of the original vibration signal in step two of this invention.

[0042] Figure 3 (a) is the standard envelope spectrum of the original vibration signal of a normal rolling bearing in the embodiment; (b) is the standard envelope spectrum of the original vibration signal of a rolling bearing with an outer ring fault in the embodiment; (c) is the standard envelope spectrum of the original vibration signal of a rolling bearing with an inner ring fault in the embodiment; and (d) is the standard envelope spectrum of the original vibration signal of a rolling element fault in the embodiment.

[0043] Figure 4 This is a schematic diagram of the specific structural parameters of the convolutional neural network in the embodiment.

[0044] Figure 5 This is a confusion matrix diagram of the classification results of the vibration signal of the rolling bearing under test in the embodiment. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0046] In this embodiment, the original vibration signal of a rolling bearing of model 6205-2RS was collected to construct a sample library. The sampling frequency was 12kHz and the sampling time was 4s. Taking a set of data with a bearing rotation frequency of 29.95Hz as an example, the characteristic frequency of the outer ring fault was 107.36Hz, the characteristic frequency of the inner ring fault was 162.19Hz, the characteristic frequency of the rolling element fault was 141.17Hz, and the characteristic frequency of the cage fault was 4.71Hz.

[0047] like Figure 1 As shown, a method for intelligent identification of rolling bearing faults based on envelope morphology analysis includes the following steps:

[0048] Step 1: Use a vibration acceleration sensor to collect the original vibration signals of the rolling bearing under normal conditions, outer ring fault conditions, inner ring fault conditions, and rolling element fault conditions. In this embodiment, the collected signal length is 48000.

[0049] Step two involves standardizing the original vibration signal to obtain its standard envelope spectrum, such as... Figure 2 As shown, it specifically includes:

[0050] 2.1) The optimal center frequency f of the original vibration signal is selected using the fast spectral kurtosis method. z 1875Hz, bandwidth B w 1250Hz;

[0051] 2.2) The original vibration signal was bandpass filtered using a Butterworth bandpass filter, with the filter bandwidth selected as [f]. z -B w / 2,f z +B w / 2], in this embodiment, the filter frequency band is selected as [1250, 2500], and the filtered signal is obtained. The squared amplitude function of the Butterworth filter is:

[0052]

[0053] Where j is the order of the filter, w c The cutoff frequency;

[0054] 2.3) Perform Hilbert demodulation on the filtered signal to obtain the demodulated signal;

[0055] 2.4) Perform a Fourier transform on the demodulated signal to obtain the envelope spectrum;

[0056] 2.5) Input the fault characteristic frequency of the bearing and its related frequencies as the search center frequency, set the frequency search error δ = 2Hz, and extract the point with the largest amplitude (F) in the search center frequency error range of the envelope spectrum. i ,a iIn this embodiment, the search center frequencies include: the outer ring fault characteristic frequency of 107.36Hz and its harmonics of 214.72Hz and 322.08Hz; the inner ring fault characteristic frequency and its harmonics of 162.19Hz, 324.38Hz, and 486.57Hz; and modulation frequencies centered on these frequencies with sideband sizes equal to the turnaround frequency of 102.29Hz, 132.24Hz, 192.14Hz, 222.09Hz, 264.48Hz, 294.43Hz, 354.33Hz, 384.28Hz, 426.67Hz, and 456Hz. 0.62Hz, 516.52Hz, 546.67Hz; rolling element fault characteristic frequencies and their harmonics 141.17Hz, 282.34Hz, 423.51Hz, and modulation frequencies centered on these frequencies with sideband sizes equal to the cage fault characteristic frequencies 131.75Hz, 136.46Hz, 145.88Hz, 150.59Hz, 277.63Hz, 272.92Hz, 287.05Hz, 296.47Hz, 418.8Hz, 414.09Hz, 428.22Hz, 432.93Hz;

[0057] 2.6) Calculate the average spectral amplitude u within the envelope spectrum frequency range [0, s]:

[0058]

[0059] In the formula: s=max(3.5fi,3.5fo,3.5fb), S(k) is the envelope spectrum sequence, k=1,2,...,K,K is the number of data points in the envelope spectrum frequency range [0,s];

[0060] In this embodiment, s = 567.67 Hz, u = 0.0018;

[0061] A threshold p is set, where p = βu, α is the threshold coefficient, representing the ratio between the threshold and the average amplitude of the spectrum. β = 2 is taken, and in this embodiment, p = 0.0036. When the amplitude of the extracted point is greater than or equal to the threshold, it is considered a valid point; when the amplitude of the extracted point is less than the threshold, it is considered an invalid point, and its amplitude is set to zero.

[0062] 2.7) For point (F) i ,a i Amplitude normalization is performed.

[0063]

[0064] Where 1≤i≤33, i is an integer, a imax For a i The maximum value in;

[0065] In this embodiment, A i The values ​​are: 0.025, 0, 0.075, 0.411, 0.154, 1, 0.067, 0.282, 0.333, 0.047, 0.169, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.035, 0, 0, 0.075, 0.047, 0.034, 0, 0, 0, 0, 0;

[0066] 2.8) Establish a new coordinate system with a vertical coordinate range of 0-1 and an horizontal coordinate range of 0-34, with an interval of 1. From left to right, the system is divided into three fault characteristic regions: outer ring fault, inner ring fault, and rolling element fault, corresponding to horizontal coordinates 1-3, 4-18, and 19-33, respectively. The normalized points (F...) i A i Place it within the corresponding fault characteristic area, according to F i Arrange the lines in ascending order to their corresponding horizontal coordinate positions to generate a new spectral diagram, which serves as the standard envelope spectrum of the original vibration signal.

[0067] Step 3: Construct a sample library using the standard envelope spectrum of the original vibration signal to reflect the rolling bearing under normal, outer ring fault, inner ring fault, and rolling element fault conditions. The sample library is constructed as follows: save the obtained standard envelope spectrum of the original vibration signal as a grayscale image in JPG format with a resolution of 1568×764px; and match all the standard envelope spectrum images of the original vibration signal under normal, outer ring fault, inner ring fault, and rolling element fault conditions with the bearing conditions to form the sample library.

[0068] The standard envelope spectra of the original vibration signals of the rolling bearing in this embodiment under normal condition, outer ring fault condition, inner ring fault condition, and rolling element fault condition are as follows: Figure 3 As shown in (a) to (d);

[0069] Step 4: Divide the sample library into training set, validation set and test set in a ratio of 7:2:1;

[0070] In this embodiment, the sample library contains 634 samples, including 64 normal samples, 276 outer ring fault samples, 144 inner ring fault samples, and 150 rolling element fault samples; it is divided into a training set of 444 samples, a validation set of 127 samples, and a test set of 63 samples.

[0071] Step 5: Construct a two-dimensional convolutional neural network model, train the network model using the training set, optimize and update the network model based on the training results, and obtain the rolling bearing fault identification model.

[0072] In this embodiment, the two-dimensional convolutional neural network model uses Adam as the optimizer, sparse_categorical_crossentropy as the loss function, and accuracy and loss value as performance evaluation metrics for network model training. The two-dimensional convolutional neural network model is trained iteratively. The batch_size is set to 4, and the epochs are set to 40. The test set is used to verify the detection accuracy of the trained two-dimensional convolutional neural network model in actual tests. Accuracy and loss value are used as performance evaluation metrics, and the two-dimensional convolutional neural network model is iteratively optimized.

[0073] Where batch_size represents the number of samples in each mini-batch during a training session, and epochs represents the number of training iterations for the entire training set.

[0074] The specific network structure parameters of the final network model are as follows: Figure 4 As shown, the algorithm consists of two convolutional layers, a batch normalization layer, a pooling layer, a flattening layer, and two fully connected layers. The convolutional layers are used for feature extraction, the pooling layer reduces the spatial dimensionality of the feature map, the batch normalization layer prevents overfitting and accelerates convergence, the flattening layer flattens the multi-dimensional feature map into a one-dimensional vector, and the fully connected layers are used for classification and output prediction. The convolutional layer filters have a size of (3, 3) and a stride of (1, 1), use the same padding method, and employ ReLU as the activation function. The first convolutional layer has 32 filters. The second convolutional layer has 64 filters. Each of the two convolutional layers is followed by a batch normalization layer and a pooling layer (max pooling). The filter size is (2, 2), the stride is (2, 2), and the same padding is used. The output is then flattened by a flattening layer and mapped to the output class by two fully connected layers. The first fully connected layer has 32 neurons and uses ReLU as the activation function, while the second fully connected layer has 4 neurons and uses softmax as the activation function.

[0075] Step 6: Collect vibration signals during the operation of the rolling bearing under test. Obtain the standard envelope spectrum of the vibration signal under test according to Step 2.

[0076] In this embodiment, the rolling bearing model under test is 6203-2RS, which is different from the bearing model 6205-2RS used to construct the sample library, to illustrate the effectiveness of this invention in identifying faults in different bearing models. A total of 364 vibration signals were collected, including 64 normal signals, 104 outer ring fault signals, 96 inner ring fault signals, and 100 rolling element fault signals.

[0077] Step 7: Input the standard envelope spectrum of the vibration signal to be tested into the trained rolling bearing fault identification model to obtain the state of the rolling bearing to be tested;

[0078] The sample library was divided into a training set and a validation set in a 7:3 ratio. The standard envelope spectrum of the vibration signal to be tested was directly used as the test set and input into the trained rolling bearing fault identification model to obtain the test set classification results. Figure 5 The confusion matrix diagram for the test set classification results shows that the predicted category represents the identification result of the vibration signal. Normal indicates that the bearing has no fault, while outer ring fault, inner ring fault, and rolling element fault indicate the corresponding fault types. The accuracy of the test set is 97.53%.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments or equivalent substitutions can be made to some of the technical features. Such modifications or equivalent substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent identification of rolling bearing faults based on envelope morphology analysis, characterized in that, Includes the following steps: Step 1: Obtain the original vibration signals of the rolling bearing under normal conditions, outer ring fault, inner ring fault, and rolling element fault conditions; Step two involves standardizing the original vibration signal to obtain its standard envelope spectrum; specifically, this includes: 2.1) The optimal center frequency of the original vibration signal is selected using the fast spectral kurtosis method. and bandwidth ; 2.2) Use a Butterworth bandpass filter to perform bandpass filtering on the original vibration signal, and select the appropriate filter frequency band. The filtered signal is obtained, and the squared amplitude function of the Butterworth filter is: in, Let the order be the filter order. The cutoff frequency; 2.3) Perform Hilbert demodulation on the filtered signal to obtain the demodulated signal; 2.4) Perform a Fourier transform on the demodulated signal to obtain the envelope spectrum; 2.5) Input the fault characteristic frequency of the bearing and its related frequencies as the search center frequency, and set the frequency search error. Extract the point with the largest amplitude within the search center frequency error range of the envelope spectrum. The search center frequency includes: the outer ring fault characteristic frequency and its harmonics. Inner ring fault characteristic frequencies and their harmonics And modulation frequencies centered on these frequencies, with sideband sizes equal to the revolving frequency. characteristic frequencies of rolling element failures and their harmonics And modulation frequencies centered at these frequencies, with sideband sizes equal to the cage fault characteristic frequencies. ; in, , , and , , All are integers. , , , These are the characteristic frequencies of failures in the outer ring, inner ring, rolling elements, and cage of the rolling bearing, respectively. For frequency conversion, To account for search error, take ; 2.6) Calculate the frequency range of the envelope spectrum. Average spectral amplitude within : In the formula: , It is an envelope spectrum sequence. , Envelope spectrum frequency range The number of data points within; Set threshold ,in, , The threshold coefficient represents the ratio between the threshold and the average amplitude of the spectrum. When the amplitude of an extracted point is greater than or equal to the threshold, it is considered a valid point; when the amplitude of an extracted point is less than the threshold, it is considered an invalid point, and its amplitude is set to zero. 2.7) Point Perform amplitude normalization: in, , Integer, for The maximum value in; 2.8) Establish a new coordinate system with a vertical coordinate range of 0-1 and an horizontal coordinate range of 0-34, with an interval of 1. From left to right, the system is divided into three fault characteristic regions: outer ring fault, inner ring fault, and rolling element fault, corresponding to horizontal coordinates 1-3, 4-18, and 19-33, respectively. The normalized points... Placed within the corresponding fault characteristic area, according to Arrange the lines in ascending order to their corresponding horizontal coordinate positions to generate a new spectral diagram, which serves as the standard envelope spectrum of the original vibration signal. Step 3: Construct a sample library using the standard envelope spectrum of the original vibration signal to reflect the normal state, outer ring fault state, inner ring fault state, and rolling element fault state of the rolling bearing. Step 4: Divide the sample library into training set, validation set and test set according to the proportions; Step 5: Construct a two-dimensional convolutional neural network model, train the network model using the training set, optimize and update the network model based on the training results, and obtain the rolling bearing fault identification model. Step 6: Collect vibration signals during the operation of the rolling bearing under test. Obtain the standard envelope spectrum of the vibration signal under test according to Step 2. Step 7: Input the standard envelope spectrum of the vibration signal to be tested into the trained rolling bearing fault identification model to obtain the state of the rolling bearing to be tested.

2. The method according to claim 1, characterized in that: In step one, a vibration acceleration sensor is used to collect the original vibration signals of the rolling bearing under normal, outer ring fault, inner ring fault, and rolling element fault conditions, respectively, and the length of the signal collected each time is determined. : in, Sampling frequency, For sampling time, take .

3. The method according to claim 1, characterized in that, The method for constructing the sample library in step three is as follows: save the obtained standard envelope spectrum of the original vibration signal as a grayscale image in JPG format with a resolution of 1568. 764 px; A sample library is formed by mapping the standard envelope spectrum images of all original vibration signals of the bearing under normal conditions and under outer ring fault, inner ring fault, and rolling element fault conditions to the bearing conditions.

4. The method according to claim 1, characterized in that: The two-dimensional convolutional network neural model in step five includes convolutional layers, pooling layers, batch normalization layers, flattening layers, and fully connected layers. The convolutional layers are used for feature extraction, the pooling layers are used to reduce the spatial dimension of the feature map, the batch normalization layers are used to prevent overfitting and accelerate convergence, the flattening layers are used to flatten the multidimensional feature map into a one-dimensional vector, and the fully connected layers are used for classification and output prediction.

Images