Bearing fault diagnosis method based on vibration envelope spectrum characteristics

Abstract

The invention relates to the technical field of bearing fault diagnosis, in particular to a bearing fault diagnosis method based on vibration envelope spectrum characteristics, and the method comprises the steps: collecting the surface damage and normal vibration signals of a rolling bearing; feature extraction is carried out on the vibration signals after noise reduction, and a plurality of feature peak intensities ranked before and feature peak frequencies corresponding to the feature peak intensities are extracted in an ascending order according to the feature peak frequencies within the optimal feature frequency range of the rolling bearing surface damage; inputting the characteristic peak intensity and the characteristic peak frequency into a naive Bayes classifier, and performing fault prediction; and evaluating the prediction result of the naive Bayes classifier by using the evaluation index. The problem that precision and algorithm complexity of an existing fault diagnosis method cannot be considered at the same time is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of bearing fault diagnosis, and in particular to a bearing fault diagnosis method based on vibration envelope spectrum characteristics. Background Art

[0002] Rolling bearings are key components of rotating machinery and play an indispensable role in the operation of equipment. Various small defects may occur in rolling bearings during operation. These defects will gradually expand with the increase of service time and eventually lead to bearing failure. Bearing failure will cause it to enter abnormal working conditions, causing a series of vibration shocks and shock attenuation responses. The collected bearing vibration signals show a certain degree of non-stationary and periodic shock characteristics.

[0003] The time domain analysis method has been widely used in many simple mechanical fault diagnosis due to its simple calculation; such as wavelet analysis, empirical mode decomposition, ReliefF algorithm and other methods; frequency domain analysis, converting the signal into a series of discrete frequency components, commonly used frequency domain analysis methods include: power spectrum analysis, peak spectrum analysis, envelope spectrum analysis.

[0004] Fault identification is also called fault pattern recognition. Commonly used algorithms include K-nearest neighbor algorithm, random forest, support vector machine, convolutional neural network, recurrent neural network (RNN), etc. However, these algorithms require larger storage space, more computing resources and computing time, which leads to increased hardware setup and operation costs.

[0005] M.Alonso-González et al. proposed a new scheme for bearing fault diagnosis using envelope analysis and machine learning methods. This method detects bearing faults with minimal observation data and minimal features, thereby ensuring the accuracy of fault diagnosis while reducing the complexity of calculations. However, since the highest peak frequency of rolling element faults does not coincide with the actual fault characteristic frequency, this method cannot achieve rolling element fault diagnosis.

[0006] The current fault diagnosis methods have the problem that accuracy and algorithm complexity cannot be taken into account at the same time. Summary of the invention

[0007] In view of the shortcomings of the existing methods, the present invention takes the frequency and intensity of characteristic peaks as characteristics and adopts a naive Bayes classifier to realize fault diagnosis, which reduces the complexity of the algorithm from two aspects of feature extraction and fault identification with high accuracy, thereby improving the diagnostic efficiency and enhancing the timeliness of real-time industrial bearing fault diagnosis.

[0008] The technical solution adopted by the present invention is: a bearing fault diagnosis method based on vibration envelope spectrum characteristics comprises the following steps:

[0009] Step 1: Collect surface damage and normal vibration signals of rolling bearings;

[0010] As a preferred embodiment of the present invention, the vibration signal of the surface damage includes: outer ring fault, inner ring fault, and rolling element fault.

[0011] Step 2: noise reduction processing of the vibration signal;

[0012] As a preferred implementation of the present invention, the noise reduction process uses Hilbert transform to demodulate the envelope of the signal to obtain an envelope signal.

[0013] As a preferred embodiment of the present invention, the envelope signal retains the oscillation part of the original signal and the harmonics at the fault frequency.

[0014] Step 3: extract features from the vibration signal after noise reduction, and extract the top several characteristic peak intensities and the characteristic peak frequencies corresponding to the characteristic peak intensities in ascending order according to the characteristic peak frequencies within the optimal characteristic frequency range of rolling bearing surface damage;

[0015] As a preferred embodiment of the present invention, the first several are 4.

[0016] Step 4: Input the characteristic peak intensity and the corresponding characteristic peak frequency into the naive Bayes classifier to perform fault prediction;

[0017] As a preferred embodiment of the present invention, the construction of the naive Bayes classifier includes:

[0018] Set the bearing fault sample x = {y1, y2, ..., y n},{y1,y2,...,y n} is the fault characteristic value;

[0019] Calculate the fault sample belongs to fault class c t The probability is:

[0020]

[0021] In the formula, p(c t |y1,y2,…,y n ) is used to classify and predict the fault class c t The posterior probability of For a given fault class c t The fault characteristic value is {y1,y2,...,y n}, p(c t ) is the fault type c t The prior probability, p(y1,y2,…,y n ) is the prior probability of the fault characteristic value.

[0022] Step 5: Use the evaluation index to evaluate the prediction results of the naive Bayes classifier;

[0023] As a preferred embodiment of the present invention, the evaluation indicators include: accuracy, sensitivity and specificity.

[0024] As a preferred embodiment of the present invention, a bearing fault diagnosis system based on vibration envelope spectrum characteristics includes: a memory for storing instructions executable by a processor; a processor for executing instructions to implement a bearing fault diagnosis method based on vibration envelope spectrum characteristics.

[0025] As a preferred embodiment of the present invention, a computer readable medium stores a computer program code, and when the computer program code is executed by a processor, a bearing fault diagnosis method based on vibration envelope spectrum characteristics is implemented.

[0026] Beneficial effects of the present invention:

[0027] 1. The periodic impact vibration caused by bearing fault usually forms a regular fault characteristic peak on the envelope spectrum. The frequency of the characteristic peak reflects the frequency corresponding to the impact moment, and the intensity of the characteristic peak reflects the intensity of the impact. Using frequency or intensity alone cannot fully reflect the fault situation. Both are needed to reflect the fault characteristics from different dimensions. The frequency and intensity of the characteristic peak of the envelope spectrum are extracted as features, and Naive Bayes is used as the fault classifier to reduce the complexity of the algorithm from two aspects: feature extraction and fault identification.

[0028] 2. The bearing fault data of CWRU and MFPT are used to verify that the accuracy of the present invention is 99.90% and 100.00% respectively. Compared with the current bearing fault classification methods based on deep learning, data-driven and other technologies, the algorithm complexity and calculation amount are greatly reduced while ensuring the accuracy, and the classification efficiency is improved. It is suitable for real-time diagnosis of industrial bearing faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a logical block diagram of bearing fault diagnosis based on vibration envelope spectrum characteristics of the present invention;

[0030] Figure 2 are vibration signals of the rolling bearing of the present invention in different states;

[0031] Figure 3 (a)-(d) are envelope spectra of vibration signals of the healthy state, rolling element fault, inner ring fault and outer ring fault of the rolling bearing of the present invention respectively;

[0032] Figure 4 is the classification accuracy of different feature indicators;

[0033] Figure 5 (a)-(d) are respectively the envelope spectrum peak extractions of the health state, rolling element fault, inner race fault and outer race fault of the present invention;

[0034] Figure 6 (a)-(c) take 3, 4, and 5 groups of maximum envelope spectrum characteristic peak frequencies and intensities as the confusion matrix of fault features respectively;

[0035] Figure 7 (a)-(d) are the confusion matrices of the CWRU dataset when the load is 0-3;

[0036] Figure 8 is the confusion matrix on the MFPT dataset;

[0037] Fig. 9 (a)-(b) are the confusion matrix comparison diagrams of the characteristic peak intensity and the characteristic peak intensity of 4 groups of maximum envelope spectra used separately on the MFPT dataset. DETAILED DESCRIPTION

[0038] The present invention is further described below in conjunction with the accompanying drawings and embodiments. This figure is a simplified schematic diagram, which only illustrates the basic structure of the present invention in a schematic manner, and therefore it only shows the components related to the present invention.

[0039] like Figure 1 , a bearing fault diagnosis method based on vibration envelope spectrum characteristics includes the following steps:

[0040] Step 1: Collect vibration signals of surface damage of rolling bearings;

[0041] Rolling bearings are generally composed of an inner ring, an outer ring, rolling elements and a cage. During the operation of mechanical equipment, assembly errors, failure of components such as the inner ring, outer ring and rolling elements, and local degradation can all cause bearing vibrations.

[0042] Rolling bearing failure forms are divided into two categories: surface damage and local wear; the latter is a progressive failure, which is far less harmful than surface damage. Therefore, the focus of bearing fault diagnosis is surface damage failure; rolling bearing surface damage failure types mainly include outer ring failure (output category 4), inner ring failure (output category 3), rolling element failure (output category 2), and the healthy state is (output category 1).

[0043] Excluding the influence of structural and environmental factors, there are two main reasons for rolling bearing vibration: one is the high-frequency natural vibration frequency caused by the elastic factor of the bearing itself; the second is the fault or damage of the bearing element itself, which manifests as impact vibration. Its frequency is called the bearing natural frequency, which is the bearing fault characteristic frequency that needs to be extracted; it should be pointed out here that the first form of vibration will inevitably occur when the bearing is working, while the second form of vibration will only appear after a fault occurs, and the vibration characteristics of each fault are different.

[0044] Figure 2 The vibration signal diagram of different operating conditions of the CWRU bearing database with a speed of 1797 rpm, a fault diameter of 0.007 inches and 0 load; Figure 2 It can be seen that the bearing vibration signal shows the characteristics of periodic changes. Under normal circumstances, the bearing signal amplitude is small and shows periodic attenuation. After a fault occurs, the vibration signal shows significant impact modulation characteristics and is related to the fault type. In addition, due to the different impact excitation energy and the way the energy reaches the sensor, the vibration signal peaks of different fault types are different.

[0045] By observation Figure 2 Vibration signals in different states can be seen Figure 2 (b) The impulse of the rolling element fault signal is lower than that of other faults and is Figure 2 (a) The signals of health status are similar; although Figure 2 (c) Inner race fault and Figure 2 (d) The amplitude of the outer race fault signal is greater than that of the healthy state, but the two signals are similar, which makes it difficult to perform fault diagnosis directly through vibration signal analysis.

[0046] The present invention obtains vibration signals from healthy bearings and bearings with different faults, and uses vibration analysis technology to implement rolling bearing fault classification.

[0047] Step 2: noise reduction processing of the vibration signal;

[0048] When a bearing fails, there is often a lot of low-frequency noise interference in the vibration signal spectrum. Through demodulation, the fault characteristic components modulated at high frequencies are demodulated to low frequencies, reducing low-frequency interference and improving the signal-to-noise ratio.

[0049] Hilbert transform is used to realize envelope demodulation of the signal. After filtering, the envelope signal is obtained. The envelope signal retains the oscillation part of the original signal and contains the harmonics at the fault frequency.

[0050] Specifically include:

[0051] 1. Perform Hilbert transform on the vibration signal y(t) and obtain:

[0052]

[0053] In the formula, is the envelope demodulation signal obtained by convolution; 1 / πt is the impulse response of convolution; τ is the time axis integration variable, and y(τ) is the corresponding vibration signal.

[0054] 2. Construct the analytical signal s(t) of y(t):

[0055]

[0056] In the formula, is the imaginary part of the analytical signal, and y(t) is the vibration signal.

[0057] 3. The Hilbert transform envelope signal Y(t) of the signal y(t) is:

[0058]

[0059] After filtering Y(t), a fast Fourier transform (FFT) is performed to obtain the envelope spectrum of the original vibration signal y(t) and extract characteristic information from it.

[0060] Figure 3 This is the envelope spectrum of vibration signals in different operating states with a speed of 1797 rpm, a fault diameter of 0.007 inches and 0 load in the CWRU database. It can be seen from the figure that there are obvious differences in the peak values ​​and corresponding frequencies of the envelope spectra of different operating states.

[0061] Step 3: Extract features from the noise-reduced vibration signal;

[0062] Feature selection is an important task in any classification problem, because the performance of the classification model depends on the selection of appropriate features; the impact vibration caused by the fault usually forms characteristic peaks on the envelope spectrum, and the intensity of these peaks and their corresponding frequencies reflect the type and severity of the fault.

[0063] The characteristic peak intensity is the envelope spectrum peak value (amplitude), and the characteristic peak frequency is the frequency corresponding to the envelope spectrum peak value.

[0064] The fault diagnosis accuracy is compared with the characteristic indicators that have been widely used in mechanical fault diagnosis, including the average value (mea), root mean square value (var), kurtosis (kur), skewness (ske), root mean square (rms), and fai is the characteristic peak intensity and the corresponding characteristic peak frequency of the present invention.

[0065] Figure 4The accuracy of fault diagnosis on two data sets using different indicators as fault features is demonstrated; in the CWRU data set, the accuracy of variance (var), kurtosis (kur), root mean square (rms) and the characteristic indicator of the present invention (fai) can all reach 90%; among them, the characteristic indicator of the present invention (fai) has the highest accuracy, reaching 99.6%; in the MFPT data set, only kurtosis (kur) and the indicator of the present invention (fai) reach an accuracy of 90%, and both reach a high accuracy of 100.0%; considering the diagnostic accuracy of the two data sets comprehensively, the characteristic indicator of the present invention (fai) becomes the most effective fault characteristic indicator.

[0066] Figure 5 The characteristic indicators of different bearing conditions are displayed. If the global envelope spectrum is traversed to find multiple characteristic peaks, the amount of calculation will be greatly increased; therefore, it is necessary to find the optimal characteristic frequency range for feature extraction;

[0067] The optimal characteristic frequency range refers to the range that contains f oc 、f ic 、f bc ;The natural frequency of a normal vibration signal is usually less than about 50HZ.

[0068] First, calculate the outer ring fault characteristic frequency of the rolling bearing Inner race fault characteristic frequency Rolling element fault characteristic frequency

[0069] Where Z is the number of balls; d is the rolling element diameter; D is the bearing pitch diameter; α is the rolling element contact angle; f r =v / 60, v is the shaft rotation speed.

[0070] Depend on Figure 3 It can be seen that the fault characteristics of the low frequency band are particularly obvious, especially in the 0-200Hz frequency band; by calculation, it can also be found that the rolling element, outer ring and inner ring fault characteristic frequencies corresponding to the selected CWRU 0 load data set are 69.60, 159.93 and 105.87Hz respectively; therefore, the present invention selects the envelope spectrum of the first 200Hz as the traversal object for finding fault characteristics, that is, 200HZ within the optimal characteristic frequency range; extracts the four peaks of the envelope spectrum and their frequencies as fault characteristic values, such as Figure 5 As shown, the fault characteristic values ​​are arranged from small to large in frequency, and the first 4, i.e., the first 4 with the largest characteristic peak intensities and the corresponding characteristic peak intensities are selected; the complexity of the feature extraction method of the present invention is much lower than that of commonly used feature extraction methods, such as wavelet packet transform, principal component analysis, singular value decomposition, etc.

[0071] Since the inherent characteristic peak amplitude of the bearing may exceed the fault characteristic peak amplitude, it is necessary to extract multiple sets of maximum values; after experimental verification, such as Figure 6 As shown, the accuracy of group 3 is 93.3%, and the accuracy of groups 4 and 5 is 99.7%; however, more feature values ​​will affect the complexity of the classifier, so 4 groups are the most appropriate.

[0072] Step 4: Input the feature extraction into the Naive Bayes classifier for fault prediction;

[0073] The core idea of ​​the Naive Bayes classifier is to use the Bayesian theorem for classification, that is, to calculate the posterior probability of each category under given feature conditions and select the category with the highest posterior probability as the classification result. Compared with other classification algorithms, the Naive Bayes classifier has the advantages of simple algorithm, high efficiency and strong robustness.

[0074] The naive Bayes classifier assumes that the class conditions between attribute variables are independent of each other, and each attribute node Y i Only related to class node C; let U = {Y1, Y2, ..., Y n ,C} is a finite set of discrete random variables, where Y1,Y2,…,Y n is the fault characteristic variable, y i is the fault feature Y i The value of the fault type C ranges from {c1,c2,...,c m}.

[0075] A bearing fault sample X = {y1, y2, ..., y n},{y1,y2,...,y n} is the fault characteristic value, that is, the characteristic peak and the characteristic peak frequency; according to Bayesian theorem, this fault sample belongs to fault class c t The probability is:

[0076]

[0077] In the formula, p(c t |y1,y2,…,y n ) is used to classify and predict the fault class c t The posterior probability of For a given fault class c t The fault characteristic value is {y1,y2,...,y n}, p(c t ) is the fault type c t The prior probability, p(y1,y2,…,y n ) is the prior probability of the fault feature value; the posterior probability reflects the sample data for the fault class c timpact.

[0078] Since p(y1,y2,…,y n ) is a constant for all fault types, so Maximum p(c t |y1,y2,…,y n ) value is the largest, the prior probability p(c t )for:

[0079]

[0080] Where: t = 1, 2, ..., m; N is the total number of bearing fault training samples; is the fault class c in the training sample t The number of samples that appear.

[0081] Based on each fault characteristic value y1,y2,...,y n The prerequisite that they are independent of each other, the conditional probability is:

[0082]

[0083] Among them, p(y i ∣c t ) indicates fault type c t The fault characteristic value is y i The conditional probability can be calculated from the training samples, that is:

[0084]

[0085] Where: is the fault class c in the training sample t and fault characteristic value y i The number of samples that appear at the same time, is the fault class c in the training sample t The number of samples that appear, if is 0, then the probability is:

[0086]

[0087] Through equations (5)-(9), we can get the fault sample belongs to a certain fault class c in fault type C. i The probability p(c t |Y), if the probability value of belonging to a certain fault class is the largest, then the bearing fault sample belongs to this bearing fault class.

[0088] Step 5. The performance of fault diagnosis results is evaluated based on three statistical parameters, namely accuracy (Acc), sensitivity (Sen) and specificity (Spe), which are calculated from the respective confusion matrix of each classification result.

[0089]

[0090] Among them, true positives (TP) and true negatives (TN) represent the number of correctly classified cases, while false positives (FP) and false negatives (FN) represent the number of misclassified cases.

[0091] Experimental results:

[0092] The performance of the proposed method is evaluated on the Case Western Reserve University (CWRU) bearing dataset and the Mechanical Failure Prevention Technology (MFPT) bearing dataset.

[0093] CWRU bearing dataset:

[0094] A 12k drive-end bearing and 0-3 load vibration data with a fault diameter of 0.007 inches were selected to generate four data sets; the bearing status selections were healthy state, rolling element fault, inner ring fault and outer ring fault at 6 o'clock; the CWRU vibration data were processed with a sliding window of 2048 steps and 400 steps, and a total of about 4*299 sample data were obtained; 70% of them were selected as training sets to train the model, and 30% were selected as test sets to verify the effect of the model.

[0095] MFPT bearing dataset:

[0096] The healthy state (H) with a load of 270 pounds and a sampling rate of 97656 Hz, and the 3-second vibration data of the outer race fault (ORF) and inner race fault (IRF) with a load of 250 pounds and a sampling rate of 48828 Hz were selected to generate a data set; the MFPT vibration data was processed with a sliding window of 2048 steps and a step size of 400 to obtain a total of about 3*362 sample data; 80% of them were selected as training sets to train the model, and 20% were selected as test sets to verify the effect of the model.

[0097] The selected fault features are input into the Naive Bayes classifier for bearing fault classification, thus completing the entire process of bearing fault diagnosis.

[0098] Figure 7 The confusion matrix of the diagnosis results of the proposed method for bearing faults with different loads in the CWRU bearing dataset is shown.

[0099] The performance of bearing fault classification is shown in Table 1. Except for the classification accuracy value of load 1, which is 99.58%, the classification accuracy values ​​of the other three groups are as high as 100%; the average accuracy value, sensitivity value and specificity value are 99.90%, 99.90% and 99.97% respectively.

[0100] Table 1. Evaluation of fault detection results

[0101]

[0102] In order to verify the simplicity and efficiency of the naive Bayes algorithm, the vibration signal data of 0 to 3 loads are used to compare the accuracy and classification time with the other two classic machine learning algorithms, and the average value of the diagnosis results of the four groups of data is taken as the comparison basis, as shown in Table 2. It can be seen from Table 2 that in terms of accuracy, the naive Bayes algorithm is similar to the random forest and support vector machine algorithms, and the training and testing time is much lower than the random forest algorithm and slightly lower than the support vector machine algorithm.

[0103] Table 2 Comparison of detection results and time of different classifiers

[0104]

[0105] Figure 8 The confusion matrix of the MFPT bearing dataset is shown when the characteristic peak intensity and characteristic peak frequency are combined at 0 load. Fig. 9 This is the confusion matrix of the MFPT bearing data set when the characteristic peak intensity is used alone and the characteristic peak frequency is used alone at 0 load. It can be seen that the effect is not good when used alone. When the two are used together, there is a fusion effect on each other, and the accuracy is significantly improved.

[0106] In addition, the accuracy and algorithm complexity of the method of the present invention are compared with those of the literature published in the past five years using the CWRU bearing database for fault identification. The results are shown in Table 3. Since the algorithm complexity of deep learning is much higher than that of machine learning, the algorithm complexity of the fault diagnosis method based on deep learning is not listed in detail here.

[0107] Table 3 Comparison of accuracy and algorithm complexity of different methods

[0108]

[0109] It can be seen from Table 3 that the accuracy of the method of the present invention is similar to that of other algorithms, and the algorithm complexity is much lower than that of other methods; although the complexity of the improved envelope analysis and FKNN algorithms is similar to that of the method proposed in the present invention, they can only realize three-category fault diagnosis of healthy state, inner ring fault and outer ring fault.

[0110] The above results show that the method of the present invention has a high classification accuracy, while greatly reducing the complexity of the algorithm, improving the classification efficiency, and is more suitable for industrial bearing fault diagnosis and maintenance.

[0111] Based on the above ideal embodiments of the present invention, the relevant staff can make various changes and modifications without departing from the technical concept of the present invention through the above description. The technical scope of the present invention is not limited to the contents of the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A bearing fault diagnosis method based on vibration envelope spectrum characteristics, characterized in that: The following steps are involved: Step 1: Collect surface damage and normal vibration signals of rolling bearings; Step 2: noise reduction processing of the vibration signal; Step 3: extract features from the vibration signal after noise reduction, and extract the top several characteristic peak intensities and the characteristic peak frequencies corresponding to the characteristic peak intensities in ascending order according to the characteristic peak frequencies within the optimal characteristic frequency range of rolling bearing surface damage; Step 4: Input the characteristic peak intensity and the corresponding characteristic peak frequency into the naive Bayes classifier to perform fault prediction; Step 5: Use the evaluation indicators to evaluate the prediction results of the naive Bayes classifier.

2. The bearing fault diagnosis method based on vibration envelope spectrum characteristics according to claim 1 is characterized in that: The first several are 4.

3. The bearing fault diagnosis method based on vibration envelope spectrum characteristics according to claim 1 is characterized in that: The construction of the Naive Bayes classifier includes: Set the bearing fault sample X = {y1, y2, ..., y n },{y1,y2,...,y n } is the fault characteristic value; Calculate the fault sample belongs to fault class c t The probability is: In the formula, p(c t |y1,y2,…,y n ) is used to classify and predict the fault class c t The posterior probability of For a given fault class c t The fault characteristic value is {y1,y2,...,y n }, p(c t ) is the fault type c t The prior probability, p(y1,y2,…,y n ) is the prior probability of the fault characteristic value.

4. The bearing fault diagnosis method based on vibration envelope spectrum characteristics according to claim 1 is characterized in that: The vibration signals of surface damage include: outer ring failure, inner ring failure, and rolling element failure.

5. The bearing fault diagnosis method based on vibration envelope spectrum characteristics according to claim 1 is characterized in that: The noise reduction process uses Hilbert transform to demodulate the envelope of the signal to obtain an envelope signal.

6. The bearing fault diagnosis method based on vibration envelope spectrum characteristics according to claim 5 is characterized in that: The envelope signal retains the oscillatory portion of the original signal and the harmonics at the fault frequency.

7. The bearing fault diagnosis method based on vibration envelope spectrum characteristics according to claim 1 is characterized in that: Evaluation metrics include: Accuracy, sensitivity, and specificity.

8. Bearing fault diagnosis system based on vibration envelope spectrum characteristics, characterized in that: include: a memory for storing instructions executable by a processor; A processor, used to execute instructions to implement the bearing fault diagnosis method based on vibration envelope spectrum characteristics as described in any one of claims 1-7.

9. A computer readable medium storing a computer program code, characterized in that: When the computer program code is executed by a processor, the bearing fault diagnosis method based on vibration envelope spectrum characteristics as described in any one of claims 1 to 7 is implemented.

Images