Rolling bearing state monitoring method
By using information entropy and kurtitude criterion in SVMD to optimize the penalty coefficient α, combined with the LSTM network, the problem of penalty coefficient and state classification in the vibration signal decomposition of rolling bearings is solved, and the precise monitoring of the state of rolling bearings is achieved.
Patent Information
- Application Number
- CN202411354385.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-27
- Publication Date
- 2024-12-27
AI Technical Summary
When using successive variational modal decomposition (SVMD), it is difficult to determine the optimal problem and precise state classification of the penalty coefficient α, which affects the decomposition effect of the vibration signal of the rolling bearing.
The information entropy is used as the fitness function, and the penalty coefficient α of the SVMD algorithm is optimized by using the Walrus Optimizer (WO) algorithm, and the optimal modal component that best expresses the fault information is selected through the kurtitude criterion, and the status monitoring is performed in combination with the LSTM network.
Effectively identifying the bearing state improves the accuracy and stability of signal decomposition, avoids the problem of difficult to determine the modal number, and realizes accurate monitoring of the vibration signal of the rolling bearing.
Smart Images
Figure CN119202841A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of rolling bearing state monitoring, and in particular to a rolling bearing state monitoring method. Background Art
[0002] Rolling bearings are essential and particularly important parts in rotating mechanical equipment. When the equipment runs for a long time or works under high frequency and heavy load, the bearings are prone to failure, and the running condition of the rolling bearings will have a reverse effect on the entire equipment. If an abnormal situation occurs during the operation of the rolling bearing, it will not only cause economic losses, but also endanger the lives and health of workers, causing immeasurable consequences. Therefore, it is very important to monitor the condition of the bearings, stop the loss in time when there is a problem with the bearing, replace and repair it as soon as possible, minimize the economic loss and maximize production safety, which is of great significance for production that focuses on both economy and safety.
[0003] In signal processing and analysis, the decomposition of bearing vibration signals is usually handled using empirical mode decomposition (EMD). When decomposing signals by recursive mode decomposition, errors will accumulate during signal reconstruction, which directly affects the results and causes modal aliasing. Variational mode decomposition (VMD) uses a non-recursive decomposition method to decompose signals by constructing and solving constrained variational problems, which prevents modal aliasing. It has the advantages of strong algorithm robustness and high discrimination of each signal component. It is widely used in the process of mechanical fault feature extraction. However, this method has the problem that the number of modes and penalty factors of the VMD algorithm need to be manually selected, which cannot achieve the optimal result and affects the effect of signal decomposition.
[0004] In view of the shortcomings of these current signal processing methods, Nazari et al. optimized VMD and proposed successive variational mode decomposition (SVMD). By adding some criteria to the optimization problem of VMD, the modal decomposition can be adaptively achieved. Based on the concept of VMD, SVMD obtains an eigenmode with a center frequency around a predetermined mode by obtaining an approximate value of the modal center frequency. SVMD introduces a sequence optimization strategy to improve the accuracy and stability of the decomposition. The variational modes of the signal are gradually extracted in the form of iteration and optimization to achieve multi-component analysis of the signal. In the decomposition process, only the penalty coefficient needs to be determined, and the number of signal modes does not need to be known in advance. This allows the SVMD method to converge to the same number of modes as the VMD method when the K value is known when the decomposition number K is unknown, avoiding the difficulty of determining the number of modes. In addition, SVMD has higher computational efficiency, and its initial value has better robustness for the modal center frequency. However, when the successive variational mode decomposition is used to decompose the vibration signal of the rolling bearing, it is necessary to determine the optimal problem of the penalty coefficient α decomposed by the method and the accurate state classification. The existing technology cannot solve this problem well. Summary of the invention
[0005] The present invention provides a rolling bearing state monitoring method to solve the problem of determining the optimal penalty coefficient α decomposed by successive variational modal decomposition when decomposing a rolling bearing vibration signal and the problem of accurate state classification.
[0006] The present invention provides a rolling bearing condition monitoring method, comprising the following steps:
[0007] Step 1, obtaining a vibration signal of a rolling bearing with a 0.014-inch fault on the outer ring;
[0008] Step 2, using information entropy as the fitness function, and using the WO algorithm to find the penalty coefficient α of the SVMD algorithm;
[0009] Step 3, decomposing the vibration signal of the rolling bearing using a successive variational mode decomposition method;
[0010] Step 4, select the best modal component that can best express the fault information by using the kurtosis criterion;
[0011] Step 5, calculating nine time domain characteristics of the optimal modal component under the working condition; wherein the nine time domain characteristics include mean, variance, waveform factor, peak value, peak factor, kurtosis, effective value, pulse factor and margin factor;
[0012] Step 6, perform the same operations as steps 1 to 5 for the other nine working conditions;
[0013] Step 7, divide the nine time domain features of the ten best modal components into a test set and a training set, with the ratio of the test set to the training set being 2:8; input the samples of the training set into the LSTM network for training;
[0014] Step 8: Input the samples of the test set into the trained LSTM monitoring model for testing to achieve the condition monitoring of the rolling bearing.
[0015] Furthermore, the step 2 comprises:
[0016] Step 2.1, initialize the relevant parameters of the WO algorithm, where the population size is set to 15 and the maximum number of iterations is 20; WO optimizes the penalty coefficient α of SVMD so that the selection range of the penalty parameter α is [500,3000];
[0017] Step 2.2, calculate the fitness of each walrus, find the optimal fitness value, use the Levy function to modify and update the population, calculate the fitness value and update the optimal position;
[0018] Step 2.3, save the result of step 2.2, if the specified number of iterations is not reached, re-cycle the calculation and update the position, if it is reached, proceed to the next step.
[0019] Furthermore, the step 3 comprises:
[0020] Step 3.1, add the criteria to the optimization problem of VMD to achieve adaptive modal decomposition; SVMD is based on the VMD concept, and obtains an eigenmode with a center frequency around the predetermined mode by obtaining the approximate value of the modal center frequency;
[0021] In step 3.2, the vibration signal of the rolling bearing with a 0.014-inch fault on the outer ring is decomposed into 8 modal components according to the WO optimized SVMD, and the corresponding optimal penalty coefficient α is generated.
[0022] Furthermore, the step 4 comprises:
[0023] Step 4.1, calculate the kurtosis of the eight modal components. Under normal circumstances, the kurtosis value is close to 3, and the kurtosis value of the standard normal distribution is 3; if the kurtosis value deviates significantly from 3, it indicates that there is a fault problem in the mechanical equipment, which is defined as follows:
[0024]
[0025] K is the kurtosis value; n is the number of signal data points; X is the value of the ith data point; μ is the mean of the signal.
[0026] Step 4.2, check the kurtosis of each IMF and select the one that deviates most from 3 as the best IMF.
[0027] Furthermore, the step 5 comprises:
[0028] Step 5.1, calculate the nine time domain characteristics of the optimal IMF obtained in step 4.2.
[0029] Further, the step 6 comprises:
[0030] Repeat steps 1 to 5 for the vibration signals of the remaining 9 working conditions, and finally sort out the ten best modal components.
[0031] Furthermore, the step 7 comprises:
[0032] Step 7.1, 80% of the time domain feature vectors of each optimal IMF are used as training sets and input into the LSTM network for state classification to train the monitoring system;
[0033] Step 7.2: When the classification accuracy of the LSTM network reaches the target accuracy, the training of the LSTM network is completed.
[0034] Further, the step 8 comprises:
[0035] Step 8.1, 20% of the time domain feature vectors of each optimal IMF are used as a test set and input into the LSTM network for testing to determine the state of each vibration signal.
[0036] The present invention has the following beneficial effects: a rolling bearing state monitoring method of the present invention obtains bearing vibration signals and classifies and organizes them; uses information entropy as fitness function and adopts WO algorithm to find penalty coefficient α of SVMD algorithm; uses successive variational modal decomposition method to decompose the vibration signal of rolling bearing; in view of the situation that the fault information may not be clearly expressed during successive variational modal decomposition, the present invention adopts kurtosis criterion to select the best modal component that can best express the fault information; calculates nine time domain features of the best modal component under each state; divides the nine time domain features into test set and training set, and the ratio of the two is 2:8; inputs samples of the training set into the model to obtain a trained LSTM state monitoring model; inputs samples of the test set into the trained LSTM monitoring model for testing, and achieves state monitoring of rolling bearing. The method can effectively identify the bearing state through the vibration signal of the rolling bearing, and has achieved good results. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] In order to more clearly illustrate the technical solution of the present invention, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.
[0038] Figure 1 This is a flow chart of the rolling bearing condition monitoring method of the present invention.
[0039] Figure 2 This is the vibration signal diagram of the rolling bearing with a 0.014-inch fault on the outer ring. Figure 2 (a) is the time domain diagram, Figure 2 (b) is the frequency domain diagram.
[0040] Figure 3 It is an iterative process curve of WO optimization SVMD in the embodiment.
[0041] Figure 4 The time domain and frequency domain diagrams of each IMF after the vibration signal of the rolling bearing with a 0.014-inch fault on the outer ring is decomposed by SVMD.
[0042] Figure 5 This is the accuracy curve of the training process for monitoring ten types of faults using the training set.
[0043] Figure 6 This is the loss function curve of the training process for monitoring ten types of faults using the training set.
[0044] Figure 7 The following are the test results of monitoring ten types of faults using the test set. DETAILED DESCRIPTION
[0045] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. The technical solutions provided by the embodiments of the present invention are described in detail below in conjunction with the drawings.
[0046] The embodiment of the present invention uses the vibration signal of a rolling bearing with a 0.014-inch fault on the outer ring as experimental data for analysis and processing. A drive-end bearing, model 6205-2RS JEM SKF, deep groove ball bearing is used. The test bearing supports the motor shaft, the motor operating condition is 3hp, and the test bearing is subjected to electrospark machining to introduce a single-point fault, and the fault diameter is 7mils. The vibration data is collected using an acceleration sensor, and the accelerometer is connected to a housing with a magnetic base. The accelerometer is placed at the 12 o'clock position on the drive end of the motor housing. A 16-channel DAT recorder is used to collect vibration signals, and post-processing is performed in a Matlab environment. The drive-end bearing experiment is collected at 12,000 samples / second. All fan-end bearing data are collected at a rate of 12,000 samples / second, and the speed and horsepower data are collected by a torque sensor / encoder and recorded manually.
[0047] See also Figure 1 The specific implementation steps of the rolling bearing condition monitoring method of the present invention are as follows:
[0048] Step 1: Obtain the vibration signal of the rolling bearing with a 0.014-inch fault on the outer ring.
[0049] Specifically, the vibration data of the rolling bearing with a 0.014-inch fault on the outer ring is obtained by an acceleration sensor. The time domain diagram and spectrum diagram when the fault occurs are as follows: Figure 2 shown.
[0050] Step 2: Using information entropy as the fitness function, the WO algorithm is used to find the penalty coefficient α of the SVMD algorithm.
[0051] Specifically, the step 2 includes:
[0052] Step 2.1, initialize the parameters of the Walrus Optimizer (WO) algorithm, where the population size is set to 15 and the maximum number of iterations is 20; WO optimizes the penalty coefficient α of SVMD so that the selection range of the penalty parameter α is [500, 3000];
[0053] Step 2.2, calculate the fitness of each walrus, find the optimal fitness value, use the Levy function to modify and update the population, calculate the fitness value and update the optimal position;
[0054] Step 2.3, save the result of step 2.2, if the specified number of iterations is not reached, re-cycle the calculation and update the position, if it is reached, proceed to the next step.
[0055] After 20 iterations, the optimal penalty coefficient was obtained and the position was updated. The optimal penalty coefficient was 1840.89. The information entropy value increased from 0.6848 to 0.6833 after three iterations. It remained at this value in the subsequent iterations, which reflects the strong optimization ability and fast convergence speed of the optimization algorithm. Figure 3 shown.
[0056] Step 3: Decompose the vibration signal of the rolling bearing using the successive variational mode decomposition method.
[0057] Specifically, step 3 includes:
[0058] In step 3.1, the criteria are added to the optimization problem of VMD to achieve adaptive modal decomposition. SVMD is based on the VMD concept and obtains an eigenmode with a center frequency around a predetermined mode by obtaining an approximate value of the modal center frequency.
[0059] SVMD does not need to set the K value. Some criteria can be added to the VMD optimization problem to achieve modal decomposition adaptively. Substitute the penalty coefficient obtained by WO optimization SVMD into it and decompose the vibration signal. The decomposition result is as follows: Figure 4 shown.
[0060] In step 3.2, the vibration signal of the rolling bearing with a 0.014-inch fault on the outer ring is decomposed into 8 modal components according to the WO optimized SVMD, and the corresponding optimal penalty coefficient α is generated.
[0061] After SVMD decomposition, it can be seen that the original signal is decomposed into 8 intrinsic mode functions (IMFs). The center frequency of each IMF component from top to bottom is decomposed from low to high and evenly distributed on each frequency, and no modal aliasing occurs. The frequency band occupied by each component is relatively narrow, which corresponds well to the original signal spectrum. Therefore, the SVMD optimized based on WO is well applied to the decomposition of the signal, ensuring the accuracy of the decomposition results.
[0062] Step 4: Select the best modal component that can best express the fault information based on the kurtosis criterion.
[0063] Specifically, step 4 includes:
[0064] Step 4.1, calculate the kurtosis of the eight modal components. Under normal circumstances, the kurtosis value is close to 3, and the kurtosis value of the standard normal distribution is 3; if the kurtosis value deviates significantly from 3, it indicates that there is a fault problem in the mechanical equipment, which is defined as follows:
[0065]
[0066] K is the kurtosis; n is the number of signal data points; X is the value of the ith data point; μ is the mean of the signal.
[0067] Step 4.2, check the kurtosis of each IMF and select the one that deviates most from 3 as the best IMF.
[0068] The kurtosis of each IMF after calculation is shown in Table 1. The kurtosis value of IMF6 deviates most from 3, so IMF6 is the best IMF.
[0069] Table 1 Kurtosis values of 8 IMFs after decomposing case fault signals
[0070]
[0071] Step 5, calculating nine time domain characteristics of the optimal modal component under the working condition; wherein the nine time domain characteristics include mean, variance, waveform factor, peak value, peak factor, kurtosis, effective value, pulse factor and margin factor.
[0072] Specifically, the step 5 includes:
[0073] Step 5.1, calculate the nine time domain characteristics of the optimal IMF obtained in step 4.2.
[0074] For the optimal IMF obtained in step 4.2, nine time domain features, including mean, variance, waveform factor, peak value, peak factor, kurtosis, effective value, pulse factor and margin factor, are calculated as new feature vectors to prepare for the next step.
[0075] Step 6: Perform the same operations as steps 1 to 5 for the other nine working conditions.
[0076] Specifically, the step 6 includes:
[0077] Repeat steps 1 to 5 for the vibration signals of the remaining 9 working conditions, and finally sort out the ten best modal components.
[0078] The other nine states follow the above steps to decompose the information of each type of fault, and find the best IMF through kurtosis information, and then obtain the nine time domain features of the ten IMFs to prepare for input into the LSTM network. The best penalty coefficients for decomposing each type of fault are shown in Table 2.
[0079] Table 2 Optimal penalty coefficients for various types of faults in decomposition cases
[0080]
[0081] In step 7, the nine time domain features of the ten best modal components are divided into a test set and a training set, with the ratio of the test set to the training set being 2:8; the samples of the training set are input into the LSTM network for training.
[0082] Specifically, the step 7 includes:
[0083] Step 7.1, 80% of the time domain feature vectors of each optimal IMF are used as training sets and input into the LSTM network for state classification to train the monitoring system;
[0084] Step 7.2: When the classification accuracy of the LSTM network reaches the target accuracy, the training of the LSTM network is completed.
[0085] The nine time domain features of the ten best modal components are divided into a test set and a training set, with a ratio of 2:8. The samples of the training set are input into the LSTM network for training. The training situation is observed through the training accuracy change curve and loss function curve of the training set. The training set accuracy and loss function curves are shown in Figure 5 , Figure 6 shown.
[0086] Step 8: Input the samples of the test set into the trained LSTM monitoring model for testing to achieve the condition monitoring of the rolling bearing.
[0087] Specifically, the step 8 includes:
[0088] Step 8.1, 20% of the time domain feature vectors of each optimal IMF are used as a test set and input into the LSTM network for testing to determine the state of each vibration signal.
[0089] The samples of the test set are input into the trained LSTM monitoring model for testing to achieve the condition monitoring of the rolling bearing. Figure 7 As shown, the accuracy rate reached 99%. The present invention was applied to the Case Western Reserve University bearing data set for verification and analysis, proving that the proposed method can effectively and accurately identify the bearing state through the vibration signal of the rolling bearing, and achieved good results.
[0090] The above-mentioned embodiments may be selected or replaced. The embodiments are merely analyses of preferred examples of the present invention and do not limit the concept and scope of the present invention. Without departing from the concept of the present invention, various changes and improvements made to the technical solution of the present invention by other professional and technical personnel in the field shall fall within the protection scope of the present invention.
Claims
1. A rolling bearing condition monitoring method, characterized in that: The steps include: Step 1, obtaining a vibration signal of a rolling bearing with a 0.014-inch fault on the outer ring; Step 2, using information entropy as the fitness function, and using the WO algorithm to find the penalty coefficient α of the SVMD algorithm; Step 3, decomposing the vibration signal of the rolling bearing using a successive variational mode decomposition method; Step 4, select the best modal component that can best express the fault information by using the kurtosis criterion; Step 5, calculating nine time domain characteristics of the optimal modal component under the working condition; wherein the nine time domain characteristics include mean, variance, waveform factor, peak value, peak factor, kurtosis, effective value, pulse factor and margin factor; Step 6, perform the same operations as steps 1 to 5 for the other nine working conditions; Step 7, divide the nine time domain features of the ten best modal components into a test set and a training set, with the ratio of the test set to the training set being 2:8; input the samples of the training set into the LSTM network for training; Step 8: Input the samples of the test set into the trained LSTM monitoring model for testing to achieve the condition monitoring of the rolling bearing.
2. The rolling bearing condition monitoring method according to claim 1, characterized in that: The step 2 comprises: Step 2.1, initialize the relevant parameters of the WO algorithm, where the population size is set to 15 and the maximum number of iterations is 20; WO optimizes the penalty coefficient α of SVMD so that the selection range of the penalty parameter α is [500,3000]; Step 2.2, calculate the fitness of each walrus, find the optimal fitness value, use the Levy function to modify and update the population, calculate the fitness value and update the optimal position; Step 2.3, save the result of step 2.2, if the specified number of iterations is not reached, re-cycle the calculation and update the position, if it is reached, proceed to the next step.
3. The rolling bearing condition monitoring method according to claim 1, characterized in that: The step 3 comprises: Step 3.1, add the criteria to the optimization problem of VMD to achieve adaptive modal decomposition; SVMD is based on the VMD concept, and obtains an eigenmode with a center frequency around the predetermined mode by obtaining the approximate value of the modal center frequency; In step 3.2, the vibration signal of the rolling bearing with a 0.014-inch fault on the outer ring is decomposed into 8 modal components according to the WO optimized SVMD, and the corresponding optimal penalty coefficient α is generated.
4. The rolling bearing condition monitoring method according to claim 1, characterized in that: The step 4 comprises: Step 4.1, calculate the kurtosis of the eight modal components. Under normal circumstances, the kurtosis value is close to 3, and the kurtosis value of the standard normal distribution is 3; if the kurtosis value deviates significantly from 3, it indicates that there is a fault problem in the mechanical equipment, which is defined as follows: K is the kurtosis; n is the number of signal data points; X is the value of the ith data point; μ is the mean of the signal. Step 4.2, check the kurtosis of each IMF and select the one that deviates most from 3 as the best IMF.
5. The rolling bearing condition monitoring method according to claim 4, characterized in that: The step 5 comprises: Step 5.1, calculate the nine time domain characteristics of the optimal IMF obtained in step 4.
2.
6. The rolling bearing condition monitoring method according to claim 1, characterized in that: The step 6 comprises: Repeat steps 1 to 5 for the vibration signals of the remaining 9 working conditions, and finally sort out the ten best modal components.
7. The rolling bearing condition monitoring method according to claim 1, characterized in that: The step 7 comprises: Step 7.1, 80% of the time domain feature vectors of each optimal IMF are used as training sets and input into the LSTM network for state classification to train the monitoring system; Step 7.2: When the classification accuracy of the LSTM network reaches the target accuracy, the training of the LSTM network is completed.
8. The rolling bearing condition monitoring method according to claim 1, characterized in that: The step 8 comprises: Step 8.1, 20% of the time domain feature vectors of each optimal IMF are used as a test set and input into the LSTM network for testing to determine the state of each vibration signal.