Low-speed heavy-load bearing fault identification method, system, medium, equipment and terminal

Abstract

The present invention belongs to the technical field of rolling bearing fault identification, and discloses a method, system, medium, equipment and terminal for identifying low-speed and heavy-load bearing faults. The method performs filtering and decomposition on the signal, solves the characteristic quantities of the first three component signals obtained by the decomposition and constructs an eigenvalue matrix; adopts a distance evaluation technology to reduce the dimension of the eigenvalue matrix and screen out significant features; inputs the significant features into a BP neural network for training and testing, and realizes the identification of low-speed and heavy-load bearing faults. In view of the problem that the original signal has broadband, non-stationary and strong noise, the present invention focuses on analyzing the steps of filtering and decomposing the signal, solves the characteristic quantities of the first three component signals obtained by the decomposition and constructs an eigenvalue matrix, adopts a distance evaluation technology method to reduce the dimension of the eigenvalue matrix, screens out significant features, and inputs the significant features into a BP neural network for training and testing, thereby realizing the accurate identification of the fault type of the low-speed and heavy-load bearing.

Description

Technical Field

[0001] The present invention belongs to the technical field of rolling bearing fault identification, and in particular relates to a low-speed heavy-load bearing fault identification method, system, medium, equipment and terminal. Background Art

[0002] At present, rolling bearings are the main components to maintain the normal operation of mechanical equipment. Ensuring that they are in good condition plays a very critical role in the normal operation of the equipment. In most cases, mechanical equipment failures are closely related to rolling bearings. Therefore, the fault diagnosis of rolling bearings is of great significance to engineering safety and is an important research topic in the field of mechanical fault diagnosis. Once a rolling bearing fails, it often causes a chain reaction, resulting in varying degrees of damage to the entire mechanical equipment. In severe cases, it may cause equipment shutdown or even cause an accident. Therefore, during the operation of mechanical equipment, it is of great significance to use reasonable analysis methods to detect and diagnose its operating conditions, extract effective fault feature information, and quickly diagnose bearing faults to ensure safe production. However, traditional rolling bearing diagnosis technology based on vibration testing often assumes that the working conditions are stable and that the fault only comes from equipment degradation or failure. In fact, in industries such as machinery, energy, and national defense, many large-scale mechanical equipment generally operate under complex conditions of low speed and alternating heavy loads. Therefore, the fault diagnosis of rolling bearings under low speed and heavy load conditions is also a research focus in the engineering field.

[0003] The vibration signal of low-speed and heavy-load bearings presents broadband, non-stationary and strong noise characteristics. When a bearing fails, the bearing defect usually produces a periodic impact on the system, and the impact signal is modulated by the rotation frequency, so the characteristic frequency band of the bearing fault usually appears in a narrowband form. Since the impact duration is relatively short, the fault vibration signal presents a short-term impact feature, and the short-term impact signal spectrum is wide-band, resulting in non-stationary characteristics of the instantaneous amplitude and instantaneous frequency of the bearing vibration signal; at the same time, during the operation of the equipment, other components will also become a strong alternating vibration source as the working conditions change. This strong alternating vibration will be transmitted to the low-speed and heavy-load bearings, which will form a large noise pollution, resulting in strong noise characteristics in the vibration signal. As a key core equipment in the steel manufacturing industry, the safe and stable operation of the crane greatly affects the normal operation and processing costs of the steel plant. The low-speed and heavy-load bearing is the key load-bearing component of the crane, and it works in harsh environments and variable working conditions for a long time. Therefore, it is crucial to diagnose and identify the low-speed and heavy-load bearings of the crane. The fault identification of low-speed and heavy-load bearings is essentially a process of feature extraction and pattern recognition. In the engineering field, traditional time-frequency analysis methods are used to extract fault features from noise signals, but they are prone to frequency ambiguity. In addition, existing fault feature extraction methods are mostly used to extract narrowband components. Affected by extreme point interpolation functions or the Gibson effect, it is difficult to extract broadband fault feature information from non-stationary and strong noise signals.

[0004] Through the above analysis, the problems and defects of the prior art are as follows:

[0005] (1) Traditional rolling bearing diagnostic technology based on vibration testing often assumes that the operating conditions are constant speed and constant load, and assumes that the fault only comes from equipment degradation or failure. It is not suitable for complex working conditions such as low speed and alternating heavy load, and has the problem of limited applicability.

[0006] (2) In the engineering field, the commonly used time-frequency analysis method is used to extract fault features from noise signals, but it is easy to produce frequency ambiguity and modal aliasing, resulting in errors in the signal decomposition results, affecting the accuracy of fault feature extraction.

[0007] (3) Existing fault feature extraction methods are mostly used to extract narrowband signal components. Affected by the extreme point interpolation function or the Gibson effect, they are prone to edge sharpening, resulting in a series of fluctuations in the extraction results. Therefore, it is difficult to accurately extract broadband fault feature information of signals from non-stationary and strong noise. Summary of the invention

[0008] In response to the problems existing in the prior art, the present invention provides a method, system, medium, device and terminal for identifying a low-speed and heavy-loaded bearing fault, and in particular, relates to a method, system, medium, device and terminal for identifying a low-speed and heavy-loaded bearing fault based on generalized wide-narrow band sparse decomposition.

[0009] The present invention is implemented as follows: a method for identifying a low-speed and heavy-load bearing fault, the method comprising:

[0010] The signal is filtered and decomposed, and the characteristic quantities of the first three component signals obtained by decomposition are solved and the eigenvalue matrix is ​​constructed; the distance evaluation technology is used to reduce the dimension of the eigenvalue matrix and screen out the significant features; the significant features are input into the BP neural network for training and testing to realize the low-speed and heavy-load bearing fault identification.

[0011] Furthermore, the low-speed and heavy-load bearing fault identification method further includes:

[0012] The generalized wide-narrowband sparse decomposition method is used to filter and decompose the collected vibration signals of low-speed and heavy-loaded bearings. The most representative signal components containing bearing fault information are determined before extracting eigenvalues ​​from the signal components. According to the inherent filtering characteristics of the generalized wide-narrowband sparse decomposition method, the first three signal components are selected to extract statistical parameter features. The distance evaluation technology method is used to calculate the eigenvalues ​​of the first three component signals after decomposition, and then the significant features matching the fault information are screened out. The significant features are input into the BP neural network for training and testing to realize the fault identification of low-speed and heavy-loaded bearings.

[0013] Further, the low-speed heavy-load bearing fault identification method comprises the following steps:

[0014] Step 1, propose a generalized wide-narrowband sparse decomposition method;

[0015] Step 2: Use distance evaluation technology to extract features;

[0016] Step three, use BP neural network to realize fault identification of low-speed and heavy-load bearings.

[0017] Further, the generalized wide-narrowband sparse decomposition method in step 1 includes generalized Fourier transform;

[0018] Generalized demodulation converts specific curve components in a multi-component signal whose time-frequency distribution is a number of curves into a linear straight line parallel to the time axis through a preset phase function. It is used for non-stationary signal analysis under low speed and alternating load. The essence of the generalized demodulation algorithm is the generalized Fourier transform.

[0019] For a real signal x(t), the generalized Fourier transform is defined as:

[0020]

[0021] In the formula, s0(t) is a real-valued function of t. The generalized Fourier transform is essentially Perform a standard Fourier transform on XG (f) Perform inverse Fourier transform to obtain x(t):

[0022]

[0023] Let X G (f) = δ(f-f0), then The generalized Fourier transform is to transform a signal with an instantaneous frequency of f(t) = f0 + s′0(t) into a signal with energy concentrated at f(t) = f0; for a signal x(t), if the time-frequency distribution is f0 + s′0(t), then find a phase function approximating s0(t) to perform generalized demodulation on the original signal, and the demodulation function obtained is The time-frequency distribution of is a straight line f(t)=f0 parallel to the time axis.

[0024] Furthermore, the generalized wide-narrowband sparse decomposition method in step 1 further includes:

[0025] (1) Broadband and narrowband joint dictionary library

[0026] A joint dictionary consisting of two typical wideband signal dictionaries, square wave signal and sawtooth wave signal, and a narrowband signal dictionary is established to decompose the wideband square wave signal and narrowband sine signal from the noise signal. The final decomposition result is obtained by searching in the three dictionary libraries. The joint dictionary library is constructed as follows:

[0027] Dic1={A1square(ω1n+θ1,D1)};

[0028] Dic2={A2sawtooth(ω2n+θ2,D2)};

[0029] Dic3={A3(n)cos(ω3n+θ3(n))};

[0030] Among them, θ, ω and A represent the original phase, angular frequency and amplitude in the dictionary library respectively; in Dic3, the maximum angular frequency of A3(n) is limited to be much smaller than ω3, and θ3(n) is constrained to be a slowly changing function to ensure that the elements in Dic3 are narrowband signals.

[0031] (2) Iterative process of generalized wide-narrowband sparse decomposition method

[0032] Solving IMF using generalized wide-narrowband sparse decomposition method i Signal components include:

[0033] 1) Set r0(n) = x(n);

[0034] 2) Extract the DC component;

[0035]

[0036] r1(n)=x(n)-IMF0(n);

[0037] in, is the Fourier transform of r0(n), yes The inverse Fourier transform of

[0038] 3) Extracting generalized demodulated signal;

[0039] According to the characteristics of the original signal, a suitable phase function s0(n) is selected, and the generalized Fourier transform of the original signal is performed to obtain a generalized demodulated signal; X G (f) is the signal after generalized demodulation;

[0040] 4) Set i=1 and solve the following optimization problem Q1, which is constructed as follows:

[0041] Q1:Minimize T1(A1,ω1,θ1,D1),T2(A2,ω2,θ2,D2),T3[A3(n),ω3(n),θ3(n)]

[0042]

[0043]

[0044] Among them, T j is to adjust the smoothness function to obtain the sparsest decomposition of the original signal; D (2) is the quadratic differential form operator; λ>0 is the weight of residual smoothness, which is set to 1;

[0045] 5) Use the artificial chemical reaction optimization algorithm (ACROA) to solve Q1 and obtain the optimal solution T j(j＝1,2,3) , and find the minimum value T j , select T j The corresponding optimal solution IMF i (n);

[0046] 6) Update r1(n): r i+1 (n) = r i (n)-IMF i (n);

[0047] 7) If the following termination condition is met, the iteration process is terminated; otherwise, return to step 4);

[0048]

[0049] Furthermore, the distance evaluation technology in step 2 for feature extraction includes:

[0050] (1) Calculate eigenvalues

[0051] After the vibration signal of the low-speed and heavy-load bearing is decomposed by generalized wide-narrow band sparse decomposition, the decomposed component signal is subjected to feature extraction. Feature extraction requires the selection of appropriate feature parameters for characterization. When performing fault identification of the low-speed and heavy-load bearing, the statistical eigenvalues ​​in the time domain and frequency domain are selected to characterize the fault characteristics.

[0052] (2) Feature extraction

[0053] After calculating the eigenvalues ​​of the component signals, 22 statistical eigenvalues ​​that characterize the bearing vibration signals are obtained, and the eigenvalue matrix is ​​constructed using the distance evaluation technology method to screen out significant features. i,j,k is the jth statistical parameter of the kth sample in the i-th category, C and N i are the number of categories and samples respectively. j is the effective factor of the jth statistical parameter. The upper and lower bounds of the normalized effective factor are 0 and 1, and the normalized effective factor value is 0.6. The statistical parameters with effective factors greater than 0.6 are regarded as significant features.

[0054] The significant features are normalized as shown below:

[0055]

[0056] Among them, f i,j is the jth significant parameter of the i-th data sample, J′ is the number of significant features, and l is the number of samples; is the maximum absolute value of the j-th significant parameter of all categories.

[0057] Another object of the present invention is to provide a low-speed and heavy-load bearing fault identification system using the low-speed and heavy-load bearing fault identification method, the low-speed and heavy-load bearing fault identification system comprising:

[0058] Sparse decomposition module, used to propose a generalized wide- and narrow-band sparse decomposition method;

[0059] A feature extraction module, used for extracting features using distance evaluation technology;

[0060] The fault identification module is used to realize the fault identification of low-speed and heavy-load bearings by using BP neural network.

[0061] Another object of the present invention is to provide a computer device, the computer device comprising a memory and a processor, the memory storing a computer program, and when the computer program is executed by the processor, the processor performs the following steps:

[0062] The signal is filtered and decomposed, and the characteristic quantities of the first three component signals obtained by decomposition are solved and the eigenvalue matrix is ​​constructed; the distance evaluation technology is used to reduce the dimension of the eigenvalue matrix and screen out the significant features; the significant features are input into the BP neural network for training and testing to realize the low-speed and heavy-load bearing fault identification.

[0063] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the following steps:

[0064] The signal is filtered and decomposed, and the characteristic quantities of the first three component signals obtained by decomposition are solved and the eigenvalue matrix is ​​constructed; the distance evaluation technology is used to reduce the dimension of the eigenvalue matrix and screen out the significant features; the significant features are input into the BP neural network for training and testing to realize the low-speed and heavy-load bearing fault identification.

[0065] Another object of the present invention is to provide an information data processing terminal, which is used to implement the low-speed and heavy-load bearing fault identification system.

[0066] In combination with the above technical solutions and the technical problems solved, please analyze the advantages and positive effects of the technical solutions to be protected by the present invention from the following aspects:

[0067] First, in view of the problems existing in the above-mentioned prior art and the difficulty of solving the problems, the technical solutions to be protected by the present invention and the results and data during the research and development process are closely combined to analyze in detail and deeply how the technical solutions of the present invention solve the technical problems, and some creative technical effects brought about after solving the problems. The specific description is as follows:

[0068] The key to low-speed and heavy-load bearing fault diagnosis is to extract fault features from vibration signals. The low-speed and heavy-load bearing fault diagnosis method based on vibration signals can generally be divided into three steps. The first is the collection of vibration signals, the second is the processing of vibration signals and the extraction of fault features, and the last is the identification and diagnosis of fault modes. Among them, the key to low-speed and heavy-load bearing fault diagnosis based on vibration signals is to use appropriate signal analysis methods to extract fault information based on the study of fault mechanism.

[0069] In view of the problem that the existing analysis methods are difficult to accurately extract fault features when diagnosing faults of low-speed and heavy-loaded bearings and have limited applicability under complex working conditions, the present invention proposes a low-speed and heavy-loaded bearing fault identification method based on generalized wide-narrowband sparse decomposition. The main idea of ​​the generalized wide-narrowband sparse decomposition method is to search for the sparsest solution of the signal in a joint dictionary library containing IMF (intrinsic mode function), transform the signal decomposition into an optimization problem, use ACROA (artificial chemical reaction optimization algorithm) to optimize the parameters of the filter, and realize the adaptive filtering decomposition of the vibration signal in the optimization process. It avoids the generation of frequency confusion and makes the components have clearer physical meanings. On the basis of generalized wide-narrowband sparse decomposition of the original signal, the characteristic quantities of the first three component signals are calculated and the eigenvalue matrix is ​​constructed. The distance evaluation technology method is used to realize feature dimensionality reduction, and representative significant features are screened out. The basic idea of ​​the distance evaluation technology is to use effective factors to select characteristic quantities with small intra-class changes and large inter-class changes in a targeted manner, in order to facilitate comparison and normalization, so as to better realize the fault feature extraction of low-speed and heavy-loaded bearings. Finally, the extracted significant features are input into the BP neural network for training and testing. The BP neural network is a multi-layer feedforward network trained by the error back propagation algorithm, and is one of the most widely used neural network models. The error passes through the output layer, and the weights of each layer are corrected according to the error gradient descent method, and then back-propagated to the hidden layer and input layer layer by layer. The repetitive process of forward information propagation and error back propagation is the process of continuous adjustment of the weights of each layer, and also the process of neural network learning and training. This process continues until the error of the network output is reduced to an acceptable level, or the number of learning times is pre-set. It can learn and store a large number of input-output pattern mapping relationships without giving a mathematical equation to describe this mapping relationship. Therefore, it can be used to identify the fault types of low-speed and heavy-load bearings.

[0070] 1. The generalized wide- and narrow-band sparse decomposition method proposed in the present invention can filter, decompose and remove working conditions of signals more accurately. First, the vibration signal under non-stationary and alternating load is generalized demodulated. Generalized demodulation is an analysis method that can convert a signal with inclined, nonlinear and non-stationary time-frequency distribution into a stationary signal with linear time-frequency distribution and parallel to the time axis. Its advantages are: first, the pre-selected time-frequency ridges in the time-frequency expression of the vibration signal can be converted in the same domain to become a straight line parallel to the time axis; second, when the time-frequency ridges are converted, the time-frequency aggregation is also significantly improved. Secondly, an associative dictionary library composed of two typical broadband signal dictionary libraries of square wave signals and sawtooth wave signals and a narrowband signal dictionary library is established to decompose broadband signals and narrowband signals from noise signals, and searches are performed in the three dictionary libraries to obtain a more accurate sparse solution.

[0071] 2. The present invention adopts the distance evaluation technology method to extract features from the decomposed component signals. After the vibration signal undergoes generalized wide- and narrow-band sparse decomposition, it is necessary to obtain the feature quantities of the decomposed component signals, but not all features can have a positive effect on the fault identification of low-speed and heavy-load bearings. Different feature quantities have different sensitivities to faults, and only feature quantities that are usually called significant features can be selected. In order to improve the accuracy of the classifier, it is necessary to select significant features that match the fault information during calculation and remove irrelevant or redundant features. Therefore, the distance evaluation technology method is used to construct the eigenvalue matrix and screen out significant features to reduce the feature dimension. The basic idea of ​​this method is to use effective factors to select features with small intra-class changes and large inter-class changes, and to normalize the effective factors. The significant features corresponding to the effective factors can better identify different faults.

[0072] 3. In the present invention, significant features are input into the BP neural network for training and testing to realize fault identification of low-speed and heavy-loaded bearings. BP neural network is a training algorithm with wide applicability. A BP neural network with 3 or more layers can approximate any continuous function with any accuracy, and has strong nonlinear mapping ability as well as self-learning, self-organization and self-adaptation capabilities. The application of BP neural network for fault identification of low-speed and heavy-loaded bearings is to utilize its good function approximation ability, judge the output category by comparing the output mode with the standard mode, and thus more accurately classify the fault type corresponding to the experimental data.

[0073] The technical solution of the present invention can not only ensure the safe operation of low-speed and heavy-load bearing equipment, timely discover existing faults and determine the type of faults that have occurred, but also diagnose the early faults of low-speed and heavy-load bearings, thereby providing certain conditions for the predictive maintenance of low-speed and heavy-load bearing equipment. Therefore, taking low-speed and heavy-load bearings as the research object and establishing a low-speed and heavy-load bearing state monitoring and fault identification system is of great and practical significance for ensuring the normal operation of the equipment.

[0074] Second, considering the technical solution as a whole or from the perspective of the product, the technical effects and advantages of the technical solution to be protected by the present invention are described in detail as follows:

[0075] The present invention proposes a method for identifying low-speed and heavy-load bearing faults based on generalized wide-narrowband sparse decomposition. Aiming at the problem that the original signal has broadband, non-stationary and strong noise, the steps of filtering and decomposing the signal are studied in detail, the characteristic quantities of the first three component signals obtained by decomposition are solved and the eigenvalue matrix is ​​constructed, and the distance evaluation technology method is used to reduce the dimension of the eigenvalue matrix, screen out significant features, and input the significant features into the BP neural network for training and testing, thus realizing the identification of low-speed and heavy-load bearing faults.

[0076] The present invention adopts distance evaluation technology to reduce the dimension of the eigenvalue matrix composed of the calculated eigenvalues, which can extract more and more accurate fault information from non-stationary and strongly noise-interfered vibration signals, and screen out more representative significant features, which is convenient for using BP neural network to train and test experimental data, thereby realizing accurate identification of low-speed and heavy-load bearing fault types.

[0077] Third, as auxiliary evidence of the inventiveness of the claims of the present invention, it is also reflected in the following important aspects:

[0078] (1) The technical solution of the present invention provides a new signal decomposition method, the purpose of which is to filter and denoise the vibration signal and more effectively extract fault feature information. It is a supplement to the existing signal decomposition method and fills the technical gap in the fault diagnosis of low-speed and heavy-load bearings under non-stationary and alternating load conditions.

[0079] (2) The technical solution of the present invention can process the vibration signal under low speed and alternating heavy load conditions, accurately extract fault features, and realize the fault identification of low speed and heavy load bearings. It solves the problem of difficulty in fault feature extraction and limited applicability of traditional diagnostic methods, and overcomes the technical difficulties that people have been eager to solve but have never been able to solve. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0081] Figure 1 is a low-speed heavy-load bearing fault identification flow chart provided by an embodiment of the present invention;

[0082] Figure 2 It is an iterative flow chart of the generalized wide-narrowband sparse decomposition method provided by an embodiment of the present invention;

[0083] Figure 3 is a flow chart of a distance assessment technical method provided by an embodiment of the present invention;

[0084] Figure 4 is a BP neural network fault identification flow chart provided by an embodiment of the present invention;

[0085] Figure 5 is a schematic diagram of a fault test bench provided by an embodiment of the present invention;

[0086] Figure 6A vibration signal graph decomposed by a generalized wide-narrowband sparse decomposition method provided by an embodiment of the present invention;

[0087] Figure 7 A vibration signal decomposition diagram using a wide-narrowband sparse decomposition method provided by an embodiment of the present invention;

[0088] Figure 8 A vibration signal diagram decomposed by a variational mode decomposition method provided by an embodiment of the present invention;

[0089] Fig. 9 A vibration signal diagram decomposed by a collective empirical mode decomposition method provided by an embodiment of the present invention;

[0090] Fig.10 It is a significant feature graph of the generalized wide-narrowband sparse decomposition method provided by an embodiment of the present invention;

[0091] Fig.11 It is a significant feature graph of the wide-narrowband sparse decomposition method provided by an embodiment of the present invention;

[0092] Fig.12 is a significant feature graph of the variational mode decomposition method provided by an embodiment of the present invention;

[0093] Fig.13 is a significant feature graph of the collective empirical mode decomposition method provided by an embodiment of the present invention;

[0094] Fig.14 It is a BP neural network recognition graph provided by an embodiment of the present invention;

[0095] Fig.15 is a flow chart of a method for identifying a low-speed and heavy-load bearing fault provided by an embodiment of the present invention;

[0096] Fig.16 It is a structural block diagram of a low-speed and heavy-load bearing fault identification system provided by an embodiment of the present invention;

[0097] In the figure: 1. Sparse decomposition module; 2. Feature extraction module; 3. Fault identification module; 4. Acceleration sensor; 5. Experimental bearing; 6. Load pressurizing device; 7. Coupling; 8. Reducer; 9. 3kW AC motor; 10. Signal acquisition system. DETAILED DESCRIPTION

[0098] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0099] In view of the problems existing in the prior art, the present invention provides a method, system, medium, equipment and terminal for identifying faults of a low-speed and heavy-loaded bearing. The present invention is described in detail below in conjunction with the accompanying drawings.

[0100] 1. Explanatory Examples In order to enable those skilled in the art to fully understand how to implement the present invention, this section provides an illustrative example that expands and describes the technical solution of the claims.

[0101] The present invention proposes a method for identifying low-speed and heavy-load bearing faults based on generalized wide-narrowband sparse decomposition. Aiming at the problem that the original signal has broadband, non-stationary and strong noise, the steps of filtering and decomposing the signal are analyzed in detail, the characteristic quantities of the first three component signals obtained by decomposition are solved and the eigenvalue matrix is ​​constructed, and the distance evaluation technology method is used to reduce the dimension of the eigenvalue matrix, screen out significant features, and input the significant features into the BP neural network for training and testing, thereby realizing the identification of low-speed and heavy-load bearing faults.

[0102] The specific implementation process of low-speed heavy-load bearing fault identification is as follows Figure 1 shown.

[0103] 1. Generalized wide-narrowband sparse decomposition method

[0104] 1.1 Generalized Fourier Transform

[0105] Generalized demodulation can convert specific curve components in a multi-component signal with a time-frequency distribution of several curves into a linear straight line parallel to the time axis through a preset phase function. It is more suitable for non-stationary signal analysis under low speed and alternating load. The essence of the generalized demodulation algorithm is the generalized Fourier transform.

[0106] For a real signal x(t), its generalized Fourier transform is defined as:

[0107]

[0108] In the formula, s0(t) is a real-valued function of t. The above formula is actually Doing the standard Fourier transform, we can also transform X G (f) Perform inverse Fourier transform to obtain x(t), that is:

[0109]

[0110] If X G (f) = δ(f-f0), then we have In fact, the generalized Fourier transform transforms a signal with an instantaneous frequency of f(t) = f0 + s′0(t) into a signal with energy concentrated at f(t) = f0. Therefore, for a signal x(t), if its time-frequency distribution is assumed to be f0 + s′0(t), we only need to find a phase function approximating s0(t) to perform generalized demodulation on the original signal, and the demodulation function obtained is The time-frequency distribution of will be a straight line f(t)=f0 parallel to the time axis.

[0111] 1.2 Generalized Wide-Narrowband Sparse Decomposition Method

[0112] 1.2.1 Broadband and Narrowband Joint Dictionary Library

[0113] In order to decompose broadband square wave signals and narrowband sine signals from noise signals, a joint dictionary library consisting of two typical broadband signal dictionaries, square wave signals and sawtooth wave signals, and a narrowband signal dictionary library is first established. The final decomposition result is obtained by searching in the three dictionary libraries. The joint dictionary library is constructed as follows:

[0114] Dic1={A1square(ω1n+θ1,D1)} (3)

[0115] Dic2={A2sawtooth(ω2n+θ2,D2)} (4)

[0116] Dic3={A3(n)cos(ω3n+θ3(n))} (5)

[0117] Among them, θ, ω and A represent the original phase, angular frequency and amplitude in the dictionary library respectively. In Dic3, the maximum angular frequency of A3(n) is restricted to be much smaller than ω3. In addition, θ3(n) is constrained to be a slowly changing function to ensure that the elements in Dic3 are narrowband signals.

[0118] 1.2.2 Iterative process of generalized wide-narrowband sparse decomposition method

[0119] Generalized Wide-Narrowband Sparse Decomposition Method for Solving IMF i The main steps of signal component are as follows, Figure 2 This is an iterative flow chart of the method.

[0120] (1) Set r0(n)=x(n).

[0121] (2) Extract the DC component.

[0122]

[0123] r1(n)=x(n)-IMF0(n) (7)

[0124] in, is the Fourier transform of r0(n), yes The inverse Fourier transform of .

[0125] (3) Extract the generalized demodulated signal.

[0126] According to the characteristics of the original signal, a suitable phase function s0(n) is selected, and the generalized demodulated signal is obtained after generalized Fourier transform of the original signal. X G (f) is the signal after generalized demodulation.

[0127] The iterative process of the generalized wide-narrowband sparse decomposition method is as follows: Figure 2 shown.

[0128] (4) Set i=1 and solve the following optimization problem Q1, which is constructed as follows:

[0129] Q1:Minimize T1(A1,ω1,θ1,D1),T2(A2,ω2,θ2,D2),T3[A3(n),ω3(n),θ3(n)]

[0130]

[0131]

[0132] T j is to adjust the smoothness function to obtain the sparsest decomposition of the original signal, D (2) is a quadratic differential form operator, λ>0 is the weight of residual smoothness, which is usually set to 1.

[0133] (5) Using the artificial chemical reaction optimization algorithm (ACROA) to solve Q1, the optimal solution T j(j＝1,2,3) , and find the minimum value T j , select T j The corresponding optimal solution IMF i (n).

[0134] (6) Update r1(n): r i+1 (n) = r i (n)-IMF i (n).

[0135] (7) If the termination condition in equation (10) is satisfied, the iterative process is terminated; otherwise, return to step (4).

[0136]

[0137] 2. Distance evaluation technology for feature extraction

[0138] 2.1 Calculating Eigenvalues

[0139] After the low-speed and heavy-load bearing vibration signal undergoes generalized wide- and narrow-band sparse decomposition, it is necessary to perform feature extraction on the decomposed component signals. Feature extraction requires the selection of appropriate feature parameters for characterization. When a bearing fault occurs, the time-frequency domain distribution and amplitude of the signal are greatly different from those in the normal state, and the corresponding spectrum distribution and amplitude will also change. Therefore, the present invention selects statistical eigenvalues ​​in the time domain and frequency domain to characterize the fault characteristics when performing fault identification of low-speed and heavy-load bearings. The eigenvalue calculation formula is shown in Table 1.

[0140] Table 1 Statistical eigenvalues ​​in time-frequency domain

[0141]

[0142] Note: x(n) (n=1,2,…,N) is the original signal, N is the number of data points; s(h) (h=1,2,…,M) is the amplitude of the original signal spectrum, M is the number of spectral lines; f(h) is the amplitude of the hth spectral line frequency.

[0143] 2.2 Feature Extraction

[0144] After calculating the eigenvalues ​​of the component signals, 22 statistical eigenvalues ​​that characterize the bearing vibration signal are obtained. However, not all eigenvalues ​​can have a positive effect on bearing fault identification. Therefore, the present invention adopts a distance evaluation technology method to construct an eigenvalue matrix and screen out significant features. Assume p i,j,k is the jth statistical parameter of the kth sample in the i-th category, C and N i are the number of categories and samples respectively. j is the effective factor of the jth statistical parameter. The upper and lower bounds of the normalized effective factor are 0 and 1. There is no strict threshold for the selection of the effective factor. The present invention stipulates that the normalized effective factor value is 0.6, and the statistical parameters with an effective factor greater than 0.6 are regarded as significant features. The flow chart of the distance evaluation technical method is as follows: Figure 3 shown.

[0145] Different significant features generally have different amplitude ranges, so the significant features need to be normalized, as shown in the following formula:

[0146]

[0147] where f i,j is the jth significant parameter of the ith data sample, J′ is the number of significant features, and l is the number of samples. is the maximum absolute value of the jth significant parameter of all categories. Formula (11) normalizes all significant parameters.

[0148] The distance evaluation technology method is used to reduce the dimension of the eigenvalue matrix composed of the calculated eigenvalues, which can extract more and more accurate fault information from the non-stationary and strongly noise-interfered vibration signals, and screen out more representative significant features, which is convenient for using BP neural network to train and test experimental data, thereby realizing accurate identification of low-speed and heavy-load bearing fault types.

[0149] 3. BP neural network realizes fault identification of low-speed and heavy-load bearings

[0150] First, the vibration signal of the collected low-speed and heavy-loaded bearing is filtered and decomposed by the proposed generalized wide-narrowband sparse decomposition method. Before extracting the eigenvalue from the signal component, the most representative signal component containing the bearing fault information must be determined. According to the inherent filtering characteristics of the generalized wide-narrowband sparse decomposition method, the signal components covering the high-frequency band are always screened out first in the iterative process. Considering these two factors, the present invention only selects the first three signal components to extract statistical parameter features. Secondly, the distance evaluation technology method is used to extract significant features of the first three component signals after decomposition. That is, the eigenvalues ​​of the decomposed component signals are calculated, and then the significant features that match the fault information are screened out, which is conducive to better identification of different faults. Finally, the significant features are input into the BP neural network for training and testing to realize the fault identification of low-speed and heavy-loaded bearings.

[0151] The flowchart of BP neural network identification of low-speed heavy-load bearing fault is as follows Figure 4 shown.

[0152] 2. Application Examples: In order to prove the creativity and technical value of the technical solution of the present invention, this section provides application examples of the technical solution of the claims on specific products or related technologies.

[0153] The technical solution of the present invention takes the low-speed and heavy-load bearings of steel plant cranes as the research object, and uses the proposed technical solution to realize the fault identification of low-speed and heavy-load bearings. The vibration signal is collected by installing the acceleration sensor on the low-speed and heavy-load bearings used in the experiment. In view of the problem that the original signal has broadband, non-stationary and strong noise, a generalized wide-narrowband sparse decomposition method is creatively proposed, and the vibration signal is filtered, denoised and de-conditioned by the method. The characteristic quantities of the component signals are calculated to construct the eigenvalue matrix. The distance evaluation technology method can select the features with small intra-class changes and large inter-class changes, reduce the dimension of the eigenvalue matrix, and realize the screening of significant features. Finally, the selected significant features are input into the BP neural network for training and testing. The test results show that the technical solution of the present invention can achieve high accuracy for the fault identification of low-speed and heavy-load bearings. It shows that the technical solution of the present invention has certain engineering practical significance and high technical value.

[0154] 3. Evidence of the effects of the embodiments. The embodiments of the present invention have achieved some positive effects during the development or use process, and indeed have great advantages over the prior art. The following content is described in conjunction with the data, charts, etc. of the test process.

[0155] Taking the vibration signal collected by the bearing fault test bench in the laboratory as an example, the specific implementation case analysis and effect verification are carried out.

[0156] 1. Specific implementation case analysis

[0157] 1.1 Experimental setup and data collection

[0158] The present invention takes the low-speed and heavy-load bearings of cranes on the production line of a steel company as the research object and performs fault identification on them.

[0159] The analysis data comes from the low-speed heavy-load bearing vibration signal collected in the laboratory. Figure 5 , which consists of an acceleration sensor 4, an experimental bearing 5, a load pressurizing device 6, a coupling 7, a reduction box 8, a 3kW AC motor 9, and a signal acquisition system 10.

[0160] The support shaft is driven by a motor, and the speed is controlled by an AC drive. The speed condition can be set by adjusting the frequency converter indication. Two low-speed heavy-load bearings are installed on the support shaft, with the healthy bearing on the left and the experimental bearing on the right, which are used to set different fault types of the bearings. The load state of the experimental bearing is set by pressing the pressure handle, and an acceleration sensor is placed on the housing of the experimental bearing to collect vibration signals.

[0161] During the experiment, the sampling frequency was set to 1kHz, and the sampling time for each set of experimental data was 10 minutes. Before the experiment began, the acceleration sensor was installed on the experimental bearing seat, and the load pressure on the bearing was set to 25MPa. Since the low-speed and heavy-load bearings of the steel plant's cranes mainly work at 38r / min, the bearing speed was treated the same in the experiment. Before each set of data was collected, the experimental bearings with different fault states that were set in advance were replaced, and finally a set of data was taken under normal state, inner ring fault, and outer ring fault, a total of three sets of experimental data, for experimental analysis. The relevant parameters of the low-speed and heavy-load bearings of the crane are shown in Table 2.

[0162] Table 2 Experimental bearing parameters

[0163]

[0164] 1.2 Fault identification of low-speed and heavy-loaded bearings of cranes

[0165] The vibration signals of the experimental bearings in three states, normal, inner ring fault and outer ring fault, are selected for fault identification of low-speed and heavy-loaded crane bearings. First, the vibration signals of the experimental bearings in three healthy states under the same speed and load conditions are intercepted at equal time, and the vibration signals of each state are divided into 30 groups. The experimental data set is shown in Table 3, that is, the data set in Table 3 contains three states of the experimental bearings, and each state contains 30 samples.

[0166] Table 3 Experimental dataset

[0167]

[0168] The specific steps for identifying the low-speed and heavy-load bearing fault of a crane are as follows:

[0169] (1) The vibration signals of the experimental bearings under three conditions are decomposed to obtain a series of signal components.

[0170] (2) Take the first three IMFs components to calculate all the eigenvalues ​​(the 22 time-frequency domain statistical parameters in Table 1) and use them as the eigenvalue vector. Therefore, each group of signals has a total of 66 eigenvalues.

[0171] (3) Normalize the eigenvalue vector to obtain the normalized eigenvalue vector α′ j (j=1,2,…,66). The distance evaluation technique is used to select feature values, and the features with normalized effective factor values ​​greater than 0.6 are selected as significant features.

[0172] (4) The BP neural network is used to train and test the significant features to achieve the identification and classification of the low-speed and heavy-load bearing fault types of cranes.

[0173] In order to compare and verify the effectiveness of the method proposed in this invention, four methods, namely generalized wide-narrowband sparse decomposition, wide-narrowband sparse decomposition, variational mode decomposition and ensemble empirical mode decomposition, are compared respectively. The distance evaluation technical method is used to extract the fault features of the experimental signal of the low-speed and heavy-load bearing of the crane, and the BP neural network is used for fault identification.

[0174] Firstly, four methods were used to decompose a total of 90 groups of experimental signals under three states to extract fault feature information. Figure 6 , Figure 7 , Figure 8 and Fig. 9 The decomposition results of a randomly selected group of samples from the third type of outer ring fault data using four methods are given. From the decomposition diagram, it can be seen that the components decomposed by the generalized wide-narrowband sparse decomposition method are relatively stable, and the decomposition results are more accurate. Among the 30 groups of data in each of the three health states, 20 groups of data are randomly selected for training, and the remaining 10 groups of data are used as test samples, that is, a total of 60 groups of training samples and 30 groups of test samples. Secondly, the distance evaluation technology method is combined to select significant features for the eigenvalues ​​required in Table 1. The significant feature screening results are shown in Fig.10 , Fig.11 , Fig.12 and Fig.13 As shown (significant features are marked with circles). It can be seen that the ensemble empirical mode decomposition method combined with the distance evaluation technology finally obtained 9 significant features, the variational mode decomposition method combined with the distance evaluation technology finally obtained 15 significant features, the wide-narrow band sparse decomposition method combined with the distance evaluation technology finally obtained 24 significant features, and the generalized wide-narrow band sparse decomposition method combined with the distance evaluation technology finally obtained 32 significant features, indicating that the generalized wide-narrow band sparse decomposition method can extract more fault feature information. Finally, the significant features obtained by the four methods combined with the distance evaluation technology were input into the BP neural network for training and testing to realize the fault identification of the low-speed and heavy-load bearings of the crane. The recognition results of the BP neural network for the test samples are shown in Fig.14 As shown, it can be seen that in the recognition results of the test samples obtained by the generalized wide-narrowband sparse decomposition method combined with the distance evaluation technology and the BP neural network, only one case of misclassification occurred in state 2, while other test samples were accurately classified, with a total accuracy of 96.67%. However, the other three methods combined with the BP neural network produced more errors in the final recognition results. Table 4 shows the specific classification results obtained by the four decomposition methods combined with the BP neural network. The example shows that the method proposed by the present invention combined with the BP neural network can more accurately and effectively identify the fault type of the low-speed heavy-load bearing of the crane.

[0175] The present invention takes low-speed and heavy-loaded bearings as the analysis object and proposes a generalized wide-narrow-band sparse decomposition method. By decomposing the vibration signal and combining the distance evaluation technology method to screen the significant features of the eigenvalues, the BP neural network is used to achieve accurate identification of the fault types of low-speed and heavy-loaded bearings. The generalized wide-narrow-band sparse decomposition method has certain advantages in suppressing modal aliasing and improving time-frequency resolution. Moreover, when decomposing non-stationary signals under strong noise interference, the decomposition accuracy can be maintained at a high level, and the effective components can be accurately extracted. Experimental analysis shows that the generalized wide-narrow-band sparse decomposition method combined with the distance evaluation technology method and the use of the BP neural network have shown good performance in identifying different fault types of low-speed and heavy-loaded bearings, and can achieve a high recognition accuracy.

[0176] Table 4 Recognition accuracy of four decomposition methods combined with BP neural network

[0177]

[0178] like Fig.15 As shown, the low-speed heavy-load bearing fault identification method provided by the embodiment of the present invention includes the following steps:

[0179] S101, proposed a generalized wide-narrowband sparse decomposition method;

[0180] S102, extracting features using distance evaluation technology;

[0181] S103, using BP neural network to realize fault identification of low-speed heavy-load bearings.

[0182] like Fig.16 As shown, the low-speed heavy-load bearing fault identification system provided by the embodiment of the present invention includes:

[0183] Sparse decomposition module 1, used to propose a generalized wide-narrowband sparse decomposition method;

[0184] Feature extraction module 2, used for extracting features using distance evaluation technology;

[0185] The fault identification module 3 is used to implement the fault identification of the low-speed heavy-load bearing by using the BP neural network.

[0186] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0187] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the field of the present invention within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for identifying low-speed and heavy-load bearing faults, characterized in that: The low-speed heavy-load bearing fault identification method comprises: The signal is filtered and decomposed, the characteristic quantities of the first three component signals obtained by decomposition are solved and the eigenvalue matrix is ​​constructed; the distance evaluation technology is used to reduce the dimension of the eigenvalue matrix and screen out significant features; the significant features are input into the BP neural network for training and testing to realize the fault identification of low-speed and heavy-load bearings; The low-speed heavy-load bearing fault identification method also includes: The generalized wide-narrowband sparse decomposition method is used to filter and decompose the collected vibration signals of low-speed and heavy-load bearings. The most representative signal components containing bearing fault information are determined before extracting eigenvalues ​​from the signal components. According to the inherent filtering characteristics of the generalized wide-narrowband sparse decomposition method, the first three signal components are selected to extract statistical parameter features. The distance evaluation technology method is used to calculate the eigenvalues ​​of the first three component signals after decomposition, and then the significant features matching the fault information are screened out. The significant features are input into the BP neural network for training and testing to realize the fault identification of low-speed and heavy-load bearings. The low-speed heavy-load bearing fault identification method comprises the following steps: Step 1, propose a generalized wide-narrowband sparse decomposition method; Step 2: Use distance evaluation technology to extract features; Step 3: Use BP neural network to realize fault identification of low-speed and heavy-load bearings; The generalized wide-narrowband sparse decomposition method in step 1 includes generalized Fourier transform; Generalized demodulation converts the specific curve components in the multi-component signal whose time-frequency distribution is a number of curves into a linear straight line parallel to the time axis through a preset phase function. It is used for non-stationary signal analysis under low speed and alternating load. The essence of the generalized demodulation algorithm is the generalized Fourier transform. For a signal x(t), the generalized Fourier transform is defined as: In the formula, s0(t) is a real-valued function of t. The generalized Fourier transform is essentially Perform a standard Fourier transform on X G (f) Perform inverse Fourier transform to obtain x(t): Let X G (f) = δ(f-f0), then The generalized Fourier transform is to transform a signal with an instantaneous frequency of f(t) = f0 + s′0(t) into a signal with energy concentrated at f(t) = f0; for a signal x(t), if the time-frequency distribution is f0 + s′0(t), then find a phase function approximating s0(t) to perform generalized demodulation on the original signal, and the demodulation function obtained is The time-frequency distribution of is a straight line parallel to the time axis f(t) = f0; The generalized wide-narrowband sparse decomposition method in step 1 also includes: (1) Broadband and narrowband joint dictionary library A joint dictionary consisting of two typical wideband signal dictionary libraries, square wave signal and sawtooth wave signal, and a narrowband signal dictionary library is established to decompose the wideband square wave signal and the narrowband sine signal from the noise signal; the final decomposition result is obtained by searching in the three dictionary libraries; the joint dictionary library is constructed as follows: Dic1={A1square(ω1n+θ1,D1)}; Dic2={A2sawtooth(ω2n+θ2,D2)}; Dic3={A3(n)cos(ω3n+θ3(n))}; Among them, θ, ω and A represent the original phase, angular frequency and amplitude in the dictionary library respectively; in Dic3, the maximum angular frequency of A3(n) is limited to be less than ω3, and θ3(n) is constrained to be a slowly changing function to ensure that the elements in Dic3 are narrowband signals; (2) Iterative process of generalized wide-narrowband sparse decomposition method Solving IMF using generalized wide-narrowband sparse decomposition method i Signal components include: 1) Set r0(n) = x(n); 2) Extract the DC component; r1(n)=x(n)-IMF0(n); in, is the Fourier transform of r0(n), yes The inverse Fourier transform of 3) Extracting generalized demodulated signal; According to the characteristics of the original signal, a suitable phase function s0(n) is selected, and the generalized Fourier transform of the original signal is performed to obtain a generalized demodulated signal; X G (f) is the signal after generalized demodulation; 4) Set i=1 and solve the following optimization problem Q1, which is constructed as follows: Among them, T j is to adjust the smoothness function to obtain the sparsest decomposition of the original signal; D (2) is the quadratic differential form operator; λ>0 is the weight of residual smoothness, which is set to 1; 5) Use the artificial chemical reaction optimization algorithm ACROA to solve Q1 and obtain the optimal solution T j(j＝1,2,3) , and find the minimum value T j , select T j The corresponding optimal solution IMF i (n); 6) Update r1(n): r i+1 (n) = r i (n)-IMF i (n); 7) If the following termination condition is met, the iteration process is terminated; otherwise, return to step 4); 2. The low-speed heavy-load bearing fault identification method according to claim 1, characterized in that: The distance evaluation technology in step 2 for feature extraction includes: (1) Calculate eigenvalues After the vibration signal of the low-speed and heavy-loaded bearing is subjected to generalized wide-narrow-band sparse decomposition, the decomposed component signal is subjected to feature extraction. Feature extraction requires the selection of appropriate feature parameters for characterization; when performing fault identification of the low-speed and heavy-loaded bearing, the statistical feature values ​​in the time domain and frequency domain are selected to characterize the fault characteristics; (2) Feature extraction After calculating the eigenvalues ​​of the component signals, 22 statistical eigenvalues ​​that characterize the bearing vibration signals are obtained, and the eigenvalue matrix is ​​constructed using the distance evaluation technology method to screen out significant features; let p i,j,k is the jth statistical parameter of the kth sample in the i-th category, C and N i are the number of categories and samples respectively; j is the effective factor of the jth statistical parameter; the upper and lower bounds of the normalized effective factor are 0 and 1, the normalized effective factor value is 0.6, and the statistical parameters with effective factors greater than 0.6 are regarded as significant features; The significant features are normalized as shown below: Among them, f i,j is the jth significant parameter of the i-th data sample, J′ is the number of significant features, and l is the number of samples; is the maximum absolute value of the j-th significant parameter of all categories.

3. A low-speed and heavy-load bearing fault identification system using the low-speed and heavy-load bearing fault identification method according to any one of claims 1 to 2, characterized in that: The low-speed heavy-load bearing fault identification system comprises: Sparse decomposition module, used to propose a generalized wide- and narrow-band sparse decomposition method; A feature extraction module, used for extracting features using distance evaluation technology; The fault identification module is used to realize the fault identification of low-speed and heavy-load bearings by using BP neural network.

4. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the low-speed heavy-load bearing fault identification method according to any one of claims 1 to 2, including the following steps: The signal is filtered and decomposed, the characteristic quantities of the first three component signals obtained by decomposition are solved and the eigenvalue matrix is ​​constructed; the distance evaluation technology is used to reduce the dimension of the eigenvalue matrix and screen out significant features; The significant features are input into the BP neural network for training and testing to realize the fault identification of low-speed and heavy-load bearings.

5. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the low-speed heavy-load bearing fault identification method according to any one of claims 1 to 2, comprising the following steps: The signal is filtered and decomposed, and the characteristic quantities of the first three component signals obtained by decomposition are solved and the eigenvalue matrix is ​​constructed; the distance evaluation technology is used to reduce the dimension of the eigenvalue matrix and screen out the significant features; the significant features are input into the BP neural network for training and testing to realize the low-speed and heavy-load bearing fault identification.

6. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the low-speed and heavy-load bearing fault identification system as described in claim 3.

Images