Rolling bearing fault method and system based on Riesz atlas sparse decomposition

Abstract

The invention belongs to the technical field of time-frequency analysis, and particularly relates to a rolling bearing fault method and system based on Riesz atlas sparse decomposition, which utilizes a new time-frequency analysis method-Riesz four-parameter spectrum time-frequency analysis method to obtain four-parameter information of amplitude, frequency, phase and direction of a vibration signal in a time-frequency domain. The time-frequency aggregation and the anti-noise performance of an analysis result are improved; in order to screen out an optimal parameter spectrum and expand the symbol dynamics entropy to a two-dimensional scale, a new entropy calculation method, namely cosine symbol dynamics entropy, is provided; and reconstructing components of the parameter spectrum obtained by entropy screening, and constructing a sparse filter parameter optimization component amplitude, so that a decomposition result has physical significance. The bearing vibration signal under the variable speed working condition and the bearing vibration signal under the constant speed are analyzed, the RSSD method is provided on the basis, and feature extraction and diagnosis of rolling bearing faults under the constant rotating speed working condition and the variable rotating speed working condition are achieved.

Description

Technical Field

[0001] The present invention belongs to but is not limited to the technical field of time-frequency analysis, and in particular relates to a rolling bearing fault detection method and system based on Riesz spectrum sparse decomposition. Background Art

[0002] With the continuous advancement of science and technology, the composition and structure of high-end mechanical equipment are becoming increasingly complex, and the key components of mechanical equipment are facing more extreme service environments and more stringent performance requirements. Among them, the stable and reliable service of rolling bearings is directly related to the safe operation of mechanical equipment. Rolling bearings usually serve in harsh environments and are prone to failure. Therefore, in order to ensure the safe and stable operation of mechanical equipment, it is necessary to collect rolling bearing vibration signals to identify early and weak defects of rolling bearings to reduce unnecessary losses. However, due to the complex structure of mechanical equipment and the changeable working conditions in actual work, the collected vibration signals are mixed with noise and multiple component modes. Therefore, accurately extracting fault information from vibration signals with many interfering component information is the key to realizing bearing fault diagnosis.

[0003] Among the methods for extracting fault information from rolling bearing vibration signals, the adaptive decomposition algorithm for multi-component signals represented by EMD has been widely studied by scholars at home and abroad. A series of improved algorithms have been proposed to address the problems existing in the EMD decomposition process. The integrated empirical mode decomposition EEMD suppresses the mode confusion phenomenon by adding Gaussian white noise to the signal multiple times; the complementary set empirical mode decomposition CEEMD adds positive and negative pairs of Gaussian white noise to eliminate the influence of the introduction of Gaussian white noise; the complete adaptive noise integrated empirical mode decomposition CEEMDAN adaptively adjusts the amplitude and distribution of the noise according to the decomposition results based on EEMD and CEEMD; and the variational mode decomposition VMD determines the frequency center and frequency band of each mode component by iteratively searching for the optimal solution of the variational model, and realizes the adaptive division of the signal in the frequency domain. This type of decomposition method can decompose multi-component signals into different components and has been widely used in the field of mechanical fault diagnosis. However, this method lacks the support of rigorous mathematical theory for whether the defined single component signal has physical meaning.

[0004] Time-frequency analysis can intuitively reflect the frequency composition of the signal and the time-varying characteristics of each frequency component. Among the linear time-frequency methods, short-time Fourier transform (STFT) is a commonly used method for analyzing non-stationary signals. Based on Fourier transform, STFT uses a window function to divide the time series signal into multiple time intervals, and the signal in each time interval is stationary. Wigner-Ville distribution (WVD) has good time-frequency aggregation and high time-frequency resolution when processing single-component signals, but when there are many signal components, serious interference terms will be generated, and the correct fault feature information cannot be extracted. Wavelet transform (WT) realizes different scale decomposition of the signal through wavelet scaling and translation, so that it has high time resolution and low frequency resolution, thereby ensuring good time domain and frequency domain analysis accuracy at the same time. These methods all have their own shortcomings. The time-frequency expression resolution of STFT and WT depends on the selection of window and basis function. Since both are fixed, they are not effective when processing signals containing many interference components. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention provides a rolling bearing fault detection method and system based on Riesz graph sparse decomposition.

[0006] The present invention is implemented as follows: a rolling bearing fault method based on Riesz spectrum sparse decomposition, the method comprising:

[0007] S1: A new time-frequency analysis method, Riesz four-parameter spectrum time-frequency analysis method, is used to obtain the four-parameter information of amplitude, frequency, phase and direction of vibration signals in the time-frequency domain, thereby improving the time-frequency aggregation and anti-noise performance of the analysis results;

[0008] S2: In order to screen out the optimal parameter spectrum, the symbolic dynamics entropy is extended to a two-dimensional scale, and a new entropy value calculation method, cosine symbolic dynamics entropy, is proposed;

[0009] S3: The parameter spectrum screened by entropy is reconstructed into components, and the amplitude of the sparse filter parameter optimization component is constructed to make the decomposition result have physical meaning. On this basis, the RSSD method is proposed.

[0010] Further, the S1 specifically includes:

[0011] (1) Riesz transform

[0012] Traditional time-frequency analysis methods can only obtain the amplitude and frequency information of the signal in the time-frequency domain, which will cause deviations when analyzing the signal. In order to solve this problem, the Riesz transform is innovatively used to analyze the vibration signal. The Riesz transform can obtain the amplitude, frequency, phase and direction information of the signal, thereby more comprehensively characterizing the fault characteristics and improving the time-frequency aggregation and noise resistance of the analysis results.

[0013] For any given one-dimensional real signal f(x) 1D , x∈E, analytical signal f A (x) is defined as the complex function of the signal, where the real part is the signal itself and the imaginary part (conjugate function) is its Hilbert transform;

[0014] f A (x) = f(x) 1D -if H (x) (1)

[0015] Where f H ()=f(x) 1D *h(x) represents the Hilbert transform of the original signal, and the transform kernel is The corresponding frequency response is H(w) = -jsgn(w);

[0016] Similarly, the Riesz transform can generate an analytical signal representation of a two-dimensional signal. For any two-dimensional signal f(x) 2D , x∈R 2 , its Riesz transform f R (·) is defined as:

[0017]

[0018] x = (x1, x2) is the position index vector, y = (y1, y2) is a virtual variable, and |||| is the vector norm. The definition of the Riesz transform can be rewritten in the form of a convolution kernel as

[0019]

[0020] Two-dimensional signal f(x) 2D Riesz transform with two directions

[0021]

[0022] Where x′=(x′1,x′2), p·v· represents the Cauchy principal value integral, and the first and second transform kernel functions of Riesz transform are

[0023]

[0024] f(x) 2D The linear combination with the Riesz transform is f(x) 2D The analytical signal is defined as the monotonic signal f M (x)

[0025]

[0026] Where i and j represent two imaginary units, and {i, j, 1} constitutes R 3 A set of orthogonal bases of space. Thus, f(x) 2D The monotonic signal is usually vectorized by a function or vector field f M : R 2 →R 3 To define,

[0027]

[0028] (2) Riesz four-parameter spectral time-frequency analysis method

[0029] Previous time-frequency transformation methods can only obtain the amplitude and frequency information of the signal in the time-frequency domain. The innovative use of Riesz transform and Log-Gabor filter group to analyze the signal can simultaneously obtain the four-parameter information of the signal in the time-frequency domain: amplitude, frequency, phase and direction. This can more comprehensively characterize the bearing fault characteristics and improve the time-frequency aggregation and anti-noise performance of the analysis results. On this basis, a time-frequency analysis method - Riesz four-parameter spectrum time-frequency analysis method is proposed. This method is divided into the following steps:

[0030] 1) Perform continuous wavelet transform on the original signal. Given a signal x(t), its CWT can be expressed as

[0031]

[0032] Where ψ(t) is the morlet wavelet basis, τ,s represents translation and scale transformation;

[0033] 2) Get the Riesz four-parameter spectrum. The Riesz four-parameter spectrum refers to The amplitude spectrum A(τ, s) and phase spectrum constructed by Riesz transform Frequency spectrum f(τ, s) and directional spectrum θ(τ, s); Considering the influence of noise on the four-parameter spectrum, the four-parameter spectrum is filtered using a Log-Gabor filter; For the Log-Gabor filter, it has the advantages of zero DC component and is not limited by bandwidth, and its frequency response function is:

[0034]

[0035] Where ω0 is the center frequency and σ is the scaling factor for adjusting the bandwidth; in order to maintain a fixed shape ratio filter bank, σ is usually r =σ / ω0 remains constant; if observed on a logarithmic frequency scale, the Log-Gabor filter has a Gaussian-like frequency response;

[0036] Since the Log-Gabor filter has no analytical expression in the time domain, the time domain representation is obtained by inverse Fourier transform;

[0037]

[0038]

[0039] Among them, F -1 represents the inverse Fourier transform, express The model, yes The Fourier representation of yes The two Riesz kernels of and They are Derivative with respect to τ, s.

[0040] Further, the S2 specifically includes:

[0041] In order to further improve the computational efficiency and complexity representation accuracy of symbolic dynamics entropy, the two-dimensional matrix is ​​converted into a block matrix using the image block idea, and the cosine similarity between adjacent block matrices is calculated, thereby converting the complexity of calculating the two-dimensional matrix into the complexity of the one-dimensional sequence, which effectively improves the computational efficiency. On this basis, a method for calculating entropy value - cosine symbolic dynamics entropy is proposed, and its steps are as follows:

[0042] Taking A(τ, s) as an example, first divide A(τ, s) into several non-overlapping blocks of the same size, with a block size of 10×10. Convert the block matrix into a vector form to calculate the cosine similarity between adjacent blocks, and obtain a series of cosine similarities; the cosine similarity between two blocks is calculated as follows:

[0043] A(τ,s)={a1,a2,...a N*M} (12)

[0044] D = {d1, d2, ..., d N*M-1}={d(a1-a2), d(a2-a3),..., d(a N*M-1 -a N*M )} (13)

[0045]

[0046] Among them, {a1, a2, ...a N*M} represents the vector form of the block matrix, N = round(length(τ) / 10), M = round(length(s) / 10; Note that the range of cosine similarity d is [-1, 1], which measures the cosine value of the geometric angle; Note that the range of cosine similarity d is [-1, 1]; The defined cosine similarity d is the similarity between two tracks, which measures the cosine value of the geometric angle; Large cosine similarity values ​​indicate that there are similar, predictable, or periodic dynamic changes between the two tracks;

[0047] Divide the D sequence into ε intervals with equal probability. i corresponds to and only corresponds to a unique interval, so the symbol σ i It replaced d i , thus forming a symbol sequence Z{Z k , k = 1, 2, ..., N*M-1}. According to Takens embedding theorem, the subvectors are generated using the embedding dimension m and the time delay μ. Calculate state mode probabilities:

[0048]

[0049] In the formula, The state mode representing the symbol arrangement of each subvector, ε m Indicates the number of symbol arrangement state patterns, type() indicates the mapping relationship from symbol space to pattern space, and |||| indicates the number of elements in the set;

[0050] When the state pattern has been observed When , the symbol that appears after is σ b The probability of (a=1, 2, ..., ε) is the state transition probability:

[0051]

[0052] According to the definition of information entropy, define cosine symbol dynamics entropy CSDE 2D Equal to the sum of state mode probability entropy and state transition probability entropy:

[0053]

[0054] If and only if When CSDE 2D The maximum value is ln(ε m+1 ). 2D Normalization:

[0055] CSDE 2D =CSDE 2D / ln(ε m+1 ) (18)

[0056] CSDE 2D The larger the value, the more random and complex the distribution within the parameter spectrum. 2D The smaller it is, the more regular the distribution of the parameter spectrum is and the less affected by noise. According to statistical analysis, when m = 2, ε = 10, μ = 1, the complexity of the four-parameter spectrum can be accurately quantified and the calculation time is short;

[0057]

[0058] Further, the S3 specifically includes:

[0059] By using Riesz transform and Log-Gabor filter bank to analyze the signal, the obtained four-parameter spectrum is screened using cosine symbol dynamics entropy, and the sparse filter parameters constructed by the reconstructed components are optimized to make the decomposition result have physical meaning. On this basis, a signal adaptive decomposition method based on time-frequency spectrum - RSSD method is proposed. Its main steps are as follows:

[0060] (1) As shown in formulas (8)-(10), the filtered The inverse Fourier transform is used to obtain its time domain representation. As shown in formula (11), the four-parameter spectrum PT (τ, s) is obtained;

[0061] (2) Determine the optimal parameter spectrum, as shown in formulas (12)-(18), and use the cosine signed dynamics entropy to calculate the entropy value CSDE of each parameter spectrum 2D , as shown in formula (20), the optimal parameter spectrum PT is obtained i (τ, s);

[0062] (3) Obtaining local coefficients Given a threshold mean(mean(PT i (τ, s))), further reducing the pseudo local minimum points caused by noise; PT i (τ, s) is divided into several incoherent regions pt like an image i (τ, s), in pt i (τ, s) along s to find the local maximum point max(pt i (τ, s)), considering the applicability, spline interpolation is used to fit the maximum point into the local coefficient

[0063] (4) Reconstruction component r i (t). Perform inverse continuous wavelet transform and reconstruct into component r i(t) Under the premise of ensuring the reconstruction accuracy, the algorithm time consumption is reduced; for the time domain signal r i (t), the reconstruction formula is as follows:

[0064]

[0065]

[0066] in, is the Fourier transform of ψ(t), It means taking the real part.

[0067] (5) Component amplitude sparse optimization only takes the points on the ridge line, which will lose most of the energy. Therefore, let res1(t) = x(t) to construct the following optimization problem P1:

[0068]

[0069] In the optimization problem P1, the optimization parameters are the amplitudes of each component, η||D(res1(t)-r i (t))|| 2 Used to regularize the optimization objective function, D is a differential operator to obtain the sparsest representation of the original signal, and its weight η is usually set to 1;

[0070] Construct the following optimization problem P2, and the optimization parameter is the amplitude coefficient μ:

[0071] P2 Minimize||res1(t)-μr i (t)|| 2 (twenty three)

[0072] The purpose of this step is to obtain r i The optimal amplitude of (t) further ensures the sparsity of the decomposition results;

[0073] (6) Setting local coefficients Length threshold, When the length is too short and the reconstruction is performed, the iteration is terminated and the entire decomposition process is completed.

[0074] Another object of the present invention is to provide a rolling bearing fault detection system based on Riesz spectrum sparse decomposition based on the rolling bearing fault detection method based on Riesz spectrum sparse decomposition, the system specifically comprising:

[0075] The parameter information acquisition module uses a new time-frequency analysis method, the Riesz four-parameter spectrum time-frequency analysis method, to obtain the four-parameter information of the vibration signal in the time-frequency domain: amplitude, frequency, phase, and direction, thereby improving the time-frequency aggregation and anti-noise performance of the analysis results.

[0076] The parameter spectrum screening module is connected with the parameter information acquisition module. In order to screen out the optimal parameter spectrum, the symbolic dynamics entropy is extended to a two-dimensional scale, and a new entropy value calculation method—cosine symbolic dynamics entropy—is proposed.

[0077] The amplitude construction module is connected with the parameter spectrum screening module to reconstruct the parameter spectrum components screened by entropy and construct the amplitude of the sparse filter parameter optimization component, so that the decomposition result has physical meaning. On this basis, the RSSD method is proposed.

[0078] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the rolling bearing fault detection method based on Riesz graph sparse decomposition.

[0079] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to execute the steps of the rolling bearing fault detection method based on Riesz graph sparse decomposition.

[0080] Another object of the present invention is to provide an information data processing terminal, which is used to implement the rolling bearing fault system based on Riesz graph sparse decomposition.

[0081] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0082] First, the present invention introduces Riesz transform combined with time-frequency analysis, and proposes a new entropy selection optimal parameter spectrum, introduces sparse ideas to optimize component amplitudes, analyzes bearing vibration signals under variable speed conditions and bearing vibration signals under constant speed, and proposes the RSSD method on this basis, realizing feature extraction and diagnosis of rolling bearing faults under constant speed and variable speed conditions. The specific conclusions are as follows:

[0083] (1) The signal is analyzed by using Riesz transform and Log-Gabor filter bank, and the four-parameter information of the signal in the time-frequency domain, namely amplitude, frequency, phase and direction, is obtained at the same time. On this basis, the Riesz four-parameter spectral time-frequency analysis method is proposed. In addition, the cosine symbolic dynamics entropy is proposed to improve the computational efficiency and complexity representation accuracy of the symbolic dynamics entropy.

[0084] (2) The RSSD method is proposed. This method analyzes the signal by using Riesz transform and Log-Gabor filter group, screens the obtained four-parameter spectrum using cosine signed dynamics entropy, and optimizes the sparse filter parameters constructed by the reconstructed components. This makes the decomposition result have better physical meaning. At the same time, RSSD also has good robustness and time-frequency aggregation.

[0085] (3) Simulation analysis shows that RSSD has better time-frequency aggregation and robustness than EMD and VMD methods. In addition, the analysis results of the vibration signals of rolling bearing faults with variable speed and constant speed show that RSSD can accurately extract the fault characteristic information of rolling bearings under constant speed or variable speed, and combined with envelope spectrum or envelope order spectrum, it can effectively realize the fault diagnosis of rolling bearings under different working conditions.

[0086] Second, the technical solution of the present invention solves a technical problem that people have long been eager to solve but have never been successful: based on the existing rolling bearing fault diagnosis method and the existing technology that may cause modal confusion, fault characteristics of rolling bearings under constant speed and variable speed are extracted to achieve fault diagnosis.

[0087] Third, the present invention solves several key technical problems faced by the prior art in industrial applications through a rolling bearing fault detection method based on Riesz graph sparse decomposition, and has achieved significant technical progress in practical applications. The following are the solutions of the present invention to the prior art problems and its technical progress in the industry:

[0088] 1. Problems with existing technologies

[0089] The time-frequency analysis method has insufficient noise resistance: The traditional time-frequency analysis method can only obtain the amplitude and frequency information of the vibration signal. In the case of large noise or fuzzy signal characteristics, it is often difficult to accurately capture the fault characteristics of the signal, resulting in reduced diagnostic accuracy.

[0090] Signal feature extraction is single and incomplete: Existing methods mostly focus on time domain or frequency domain analysis of signals and cannot simultaneously obtain multi-dimensional information such as phase and direction of the signal, which limits the accuracy of fault detection and diagnostic effect.

[0091] Low computational efficiency of symbolic dynamics entropy: Existing complexity analysis methods based on symbolic dynamics have high computational complexity and are difficult to effectively apply when processing large-scale industrial data, which affects the performance of real-time monitoring and online fault diagnosis of the system.

[0092] 2. Technical advancement of the present invention

[0093] Improved the anti-noise ability and time-frequency aggregation of time-frequency analysis: By introducing the Riesz four-parameter spectrum time-frequency analysis method, the present invention can simultaneously obtain the four-parameter information of the amplitude, frequency, phase and direction of the vibration signal, so that the analysis results have significantly improved in terms of time-frequency aggregation and anti-noise performance. Compared with traditional time-frequency analysis methods, it can capture the fault characteristics in the signal more comprehensively and accurately, especially in a high-noise environment, it can still maintain a high diagnostic accuracy.

[0094] Achieved more comprehensive fault feature extraction: The present invention innovatively applies Riesz transform to rolling bearing fault detection, which can obtain multi-dimensional features of the signal, including phase and direction information, overcoming the deficiency of traditional methods that can only capture amplitude and frequency information in the time-frequency domain. Through these multi-dimensional signal representations, the comprehensiveness of fault detection is improved, and the reliability of fault diagnosis is significantly improved.

[0095] Improved efficiency of entropy calculation: The cosine symbolic dynamics entropy method proposed in this paper greatly improves the calculation efficiency of symbolic dynamics entropy by converting a two-dimensional matrix into a one-dimensional sequence. This method reduces the computational complexity, enabling it to be applied to the processing of large-scale industrial data, thereby supporting real-time monitoring and online fault diagnosis, and greatly improving the feasibility of industrial applications in terms of processing speed.

[0096] Enhanced physical meaning of decomposition results and sparsity of fault signals: Through the signal reconstruction method based on sparse filters, the present invention ensures the physical meaning of fault signal decomposition, can effectively screen out signal components that are highly correlated with fault characteristics, and remove useless noise components, further improving the accuracy and efficiency of diagnosis. The sparse decomposition optimization algorithm makes the rolling bearing fault detection method more accurate and efficient, and is particularly suitable for complex industrial environments.

[0097] 3. Significant technological progress in industrial applications

[0098] Improved the accuracy and reliability of rolling bearing fault diagnosis: By introducing multi-dimensional signal characterization and sparse decomposition technology, rolling bearing fault signals can be extracted efficiently and accurately in complex environments, improving the accuracy of early fault detection and diagnosis and reducing sudden losses caused by equipment failures.

[0099] Adapting to the needs of large-scale industrial data processing: The cosine symbolic dynamics entropy and sparse decomposition optimization algorithm enable the present invention to have higher computational efficiency when processing large-scale vibration data, which can meet the needs of real-time online monitoring in industry and provide effective support for intelligent maintenance.

[0100] Improved predictability of equipment maintenance and saved maintenance costs: This technology has broad application prospects in industrial automation and intelligent manufacturing. By discovering equipment hidden dangers at an earlier stage, it can effectively reduce unplanned equipment downtime, reduce maintenance costs and extend the service life of equipment.

[0101] In summary, the present invention not only solves the key technical problems in the existing rolling bearing fault detection, but also significantly improves the accuracy, real-time and efficiency of fault diagnosis in industrial applications, and has important technical progress and application value in improving the intelligent monitoring and maintenance level of industrial equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0102] Figure 1 It is a flow chart of a rolling bearing fault detection method based on Riesz spectrum sparse decomposition provided by an embodiment of the present invention;

[0103] Figure 2 is a flow chart of RSSD provided by an embodiment of the present invention;

[0104] Figure 3 is a mixed signal and its components provided by an embodiment of the present invention;

[0105] Figure 4 is the RSSD decomposition result provided by the embodiment of the present invention;

[0106] Figure 5 is the EMD decomposition result provided by the embodiment of the present invention;

[0107] Figure 6 is the VMD decomposition result provided by the embodiment of the present invention;

[0108] Figure 7 is the VMD2 decomposition result provided by an embodiment of the present invention;

[0109] Figure 8 The time-frequency spectra of each component provided by the embodiment of the present invention are sequentially the time-frequency spectra of the simulation signal component, the time-frequency spectra of the RSSD component, the time-frequency spectra of the EMD component, and the time-frequency spectra of the VMD component;

[0110] Fig. 9 is a mixed signal and its components provided by an embodiment of the present invention;

[0111] Fig.10 is the RSSD decomposition result provided by the embodiment of the present invention;

[0112] Fig.11 is the EMD decomposition result provided by the embodiment of the present invention;

[0113] Fig.12 is the VMD decomposition result provided by the embodiment of the present invention;

[0114] Fig.13 is the VMD2 decomposition result provided by an embodiment of the present invention;

[0115] Fig.14 The time-frequency spectra of each component provided by the embodiment of the present invention are, in order, the time-frequency spectra of the simulation signal component, the time-frequency spectra of the RSSD component, the time-frequency spectra of the EMD component, and the time-frequency spectra of the VMD component;

[0116] Fig.15 An energy recovery variable speed bearing fault test bench provided by an embodiment of the present invention;

[0117] Fig.16 It is an EDM 6206 bearing fault part provided by an embodiment of the present invention;

[0118] Fig.17 The bearing speed signal and vibration signal with outer ring fault provided by the embodiment of the present invention;

[0119] Fig.18 The RSSD decomposition and angular domain resampling of components and envelope order spectrum provided by the embodiment of the present invention;

[0120] Fig.19 The EMD decomposition and angular domain resampling of components and envelope order spectrum provided by the embodiment of the present invention;

[0121] Fig. 20 The VMD decomposition and angular domain resampling of components and envelope order spectrum provided by the embodiment of the present invention;

[0122] Fig.21 is a time domain diagram of a constant speed bearing outer ring fault signal provided by an embodiment of the present invention;

[0123] Fig. 22 is the envelope spectrum of the constant speed bearing outer ring fault signal provided by the embodiment of the present invention;

[0124] Fig.23 is a fault vibration signal component obtained by using RSSD provided in an embodiment of the present invention;

[0125] Fig.24 is a fault vibration signal component obtained by using EMD provided in an embodiment of the present invention;

[0126] Fig.25 is a fault vibration signal component obtained by using VMD provided in an embodiment of the present invention;

[0127] Fig.26 The embodiment of the present invention provides an envelope spectrum of components of a faulty bearing signal decomposed using RSSD, EMD and VMD methods. DETAILED DESCRIPTION

[0128] 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.

[0129] like Figure 1 As shown, a rolling bearing fault detection method based on Riesz graph sparse decomposition includes:

[0130] S1: A new time-frequency analysis method, Riesz four-parameter spectrum time-frequency analysis method, is used to obtain the four-parameter information of amplitude, frequency, phase and direction of vibration signals in the time-frequency domain, thereby improving the time-frequency aggregation and anti-noise performance of the analysis results;

[0131] S2: In order to screen out the optimal parameter spectrum, the symbolic dynamics entropy is extended to a two-dimensional scale, and a new entropy value calculation method, cosine symbolic dynamics entropy, is proposed;

[0132] S3: The parameter spectrum screened by entropy is reconstructed into components, and the amplitude of the sparse filter parameter optimization component is constructed to make the decomposition result have physical meaning. On this basis, the RSSD method is proposed.

[0133] The embodiment of the present invention proposes a rolling bearing fault detection method based on Riesz spectrum sparse decomposition. First, in step S1, the Riesz four-parameter spectrum time-frequency analysis method is adopted. This method simultaneously analyzes the four parameter information of amplitude, frequency, phase and direction of the vibration signal, so that the analysis results in the time-frequency domain have higher time-frequency aggregation, and can still effectively improve the noise resistance performance when there is large noise in the signal. Compared with traditional time-frequency analysis technology, this method can more accurately capture the characteristic changes of vibration signals and provide a more reliable data basis for subsequent fault identification.

[0134] Then, in step S2, symbolic dynamics entropy is used as a tool to screen the optimal parameter spectrum. Symbolic dynamics entropy is a quantitative index that reflects the complexity of the signal. In order to more effectively screen out the parameter spectrum that meets the rolling bearing fault characteristics, this paper proposes the concept of cosine symbolic dynamics entropy and expands it to a two-dimensional scale. This new entropy value calculation method can more accurately capture the complex changes and nonlinear characteristics of the signal, thereby improving the accuracy and reliability of fault feature extraction.

[0135] In step S3, the optimal parameter spectrum selected by entropy calculation is reconstructed to form a sparse decomposition model for rolling bearing fault detection. The reconstructed components are further optimized by sparse filters to make the decomposition results have clearer physical meanings, that is, the characteristics of the fault signal are more significant and easy to identify. The sparse decomposition model can effectively eliminate useless information and retain key components related to the fault, thereby improving the detection accuracy.

[0136] Finally, based on the above, the RSSD (Riesz Spectral Sparse Decomposition) method is proposed. The RSSD method combines time-frequency analysis, symbolic dynamics entropy and sparse decomposition technology, and has high precision, strong noise resistance and good physical meaning. This method can efficiently extract fault features from rolling bearing vibration signals, is suitable for equipment fault detection and health monitoring in complex industrial environments, and has important application value.

[0137] The S1 specifically includes:

[0138] (1) Riesz transform

[0139] Traditional time-frequency analysis methods can only obtain the amplitude and frequency information of the signal in the time-frequency domain, which will cause deviations when analyzing the signal. In order to solve this problem, the Riesz transform is innovatively used to analyze the vibration signal. The Riesz transform can obtain the amplitude, frequency, phase and direction information of the signal, thereby more comprehensively characterizing the fault characteristics and improving the time-frequency aggregation and noise resistance of the analysis results.

[0140] For any given one-dimensional real signal f(x) 1D , x∈R, analytical signal f A (x) is defined as the complex function of the signal, where the real part is the signal itself and the imaginary part (conjugate function) is its Hilbert transform;

[0141] f A (x) = f(x) 1D -if H (x) (1)

[0142] Where f H ()=f(x) 1D *h(x) represents the Hilbert transform of the original signal, and the transform kernel is The corresponding frequency response is H(w) = -jsgn(w);

[0143] Similarly, the Riesz transform can generate an analytical signal representation of a two-dimensional signal. For any two-dimensional signal f(x) 2D , x∈R 2 , its Riesz transform f R (·) is defined as:

[0144]

[0145] x = (x1, x2) is the position index vector, y = (y1, y2) is a virtual variable, and |||| is the vector norm. The definition of the Riesz transform can be rewritten in the form of a convolution kernel as

[0146]

[0147] The two-dimensional signal f(x)2D has Riesz transforms in two directions

[0148]

[0149] Where x′=(x′1,x′2), p·v· represents the Cauchy principal value integral, and the first and second transform kernel functions of Riesz transform are

[0150]

[0151] f(x) 2D The linear combination with the Riesz transform is f(x) 2D The analytical signal is defined as the monotonic signal f M (x)

[0152]

[0153] Where i and j represent two imaginary units, and {i, j, 1} constitutes R 3 A set of orthogonal bases of space. Thus, f(x) 2D The monotonic signal is usually vectorized by a function or vector field f M : R 2 →R 3 To define,

[0154]

[0155] (2) Riesz four-parameter spectral time-frequency analysis method

[0156] Previous time-frequency transformation methods can only obtain the amplitude and frequency information of the signal in the time-frequency domain. The innovative use of Riesz transform and Log-Gabor filter group to analyze the signal can simultaneously obtain the four-parameter information of the signal in the time-frequency domain: amplitude, frequency, phase and direction. This can more comprehensively characterize the bearing fault characteristics and improve the time-frequency aggregation and anti-noise performance of the analysis results. On this basis, a time-frequency analysis method - Riesz four-parameter spectrum time-frequency analysis method is proposed. This method is divided into the following steps:

[0157] 1) Perform continuous wavelet transform on the original signal. Given a signal x(t), its CWT can be expressed as

[0158]

[0159] Where ψ(t) is the morlet wavelet basis, τ,s represents translation and scale transformation;

[0160] 2) Get the Riesz four-parameter spectrum. The Riesz four-parameter spectrum refers to The amplitude spectrum A(τ, s) and phase spectrum constructed by Riesz transform Frequency spectrum f(τ, s) and directional spectrum θ(τ, s); Considering the influence of noise on the four-parameter spectrum, the four-parameter spectrum is filtered using a Log-Gabor filter; For the Log-Gabor filter, it has the advantages of zero DC component and is not limited by bandwidth, and its frequency response function is:

[0161]

[0162] Where ω0 is the center frequency and σ is the scaling factor for adjusting the bandwidth; in order to maintain a fixed shape ratio filter bank, σ is usually r =σ / ω0 remains constant; if observed on a logarithmic frequency scale, the Log-Gabor filter has a Gaussian-like frequency response;

[0163] Since the Log-Gabor filter has no analytical expression in the time domain, the time domain representation is obtained by inverse Fourier transform;

[0164]

[0165]

[0166] Among them, F -1 represents the inverse Fourier transform, express The model, yes The Fourier representation of yes The two Riesz kernels of and They are Derivative with respect to τ, s.

[0167] The S2 specifically includes:

[0168] In order to further improve the computational efficiency and complexity representation accuracy of symbolic dynamics entropy, the two-dimensional matrix is ​​converted into a block matrix using the image block idea, and the cosine similarity between adjacent block matrices is calculated, thereby converting the complexity of calculating the two-dimensional matrix into the complexity of the one-dimensional sequence, which effectively improves the computational efficiency. On this basis, a method for calculating entropy value - cosine symbolic dynamics entropy is proposed, and its steps are as follows:

[0169] Taking A(τ, s) as an example, first divide A(τ, s) into several non-overlapping blocks of the same size, with a block size of 10×10. Convert the block matrix into a vector form to calculate the cosine similarity between adjacent blocks, and obtain a series of cosine similarities; the cosine similarity between two blocks is calculated as follows:

[0170] A(τ,s)={a1,a2,...a N*M} (12)

[0171] D = {d1, d2, ..., d N*M-1}={d(a1-a2), d(a2-a3),..., d(a N*M-1 -a N+M )} (13)

[0172]

[0173] Among them, {a1, a2, ...a N*M} represents the vector form of the block matrix, N = round(length(τ) / 10), M = round(length(s) / 10; Note that the range of cosine similarity d is [-1, 1], which measures the cosine value of the geometric angle; Note that the range of cosine similarity d is [-1, 1]; The defined cosine similarity d is the similarity between two tracks, which measures the cosine value of the geometric angle; Large cosine similarity values ​​indicate that there are similar, predictable, or periodic dynamic changes between the two tracks;

[0174] Divide the D sequence into ε intervals with equal probability. i corresponds to and only corresponds to a unique interval, so the symbol σ i It replaced d i , thus forming a symbol sequence Z{Z k , k=1,2,...,N*M-1}. According to Takens embedding theorem, subvectors are generated using embedding dimension m and time delay μ. Calculate state mode probabilities:

[0175]

[0176] In the formula, The state mode representing the symbol arrangement of each subvector, ε m Indicates the number of symbol arrangement state patterns, type() indicates the mapping relationship from symbol space to pattern space, and |||| indicates the number of elements in the set;

[0177] When the state pattern has been observed When , the symbol that appears after is σ bThe probability of (a=1, 2, ..., ε) is the state transition probability:

[0178]

[0179] According to the definition of information entropy, define cosine symbol dynamics entropy CSDE 2D Equal to the sum of state mode probability entropy and state transition probability entropy:

[0180]

[0181] If and only if When CSDE 2D The maximum value is ln(ε m+1 ). 2D Normalization:

[0182] CSDE 2D =CSDE 2D / ln(ε m+1 ) (18)

[0183] CSDE 2D The larger the value, the more random and complex the distribution within the parameter spectrum. 2D The smaller it is, the more regular the distribution of the parameter spectrum is and the less affected by noise. According to statistical analysis, when m = 2, ε = 10, μ = 1, the complexity of the four-parameter spectrum can be accurately quantified and the calculation time is short;

[0184]

[0185] The S3 specifically includes:

[0186] By using Riesz transform and Log-Gabor filter bank to analyze the signal, the obtained four-parameter spectrum is screened using cosine symbol dynamics entropy, and the sparse filter parameters constructed by the reconstructed components are optimized to make the decomposition result have physical meaning. On this basis, a signal adaptive decomposition method based on time-frequency spectrum - RSSD method is proposed. Its main steps are as follows:

[0187] (1) As shown in formulas (8)-(10), the filtered The inverse Fourier transform is used to obtain its time domain representation. As shown in formula (11), the four-parameter spectrum PT (τ, s) is obtained;

[0188] (2) Determine the optimal parameter spectrum, as shown in formulas (12)-(18), and use the cosine signed dynamics entropy to calculate the entropy value CSDE of each parameter spectrum 2D , as shown in formula (20), the optimal parameter spectrum PT is obtainedi (τ, s);

[0189] (3) Obtaining local coefficients Given a threshold mean(mean(PT i (τ, s))), further reducing the pseudo local minimum points caused by noise; PT i (τ, s) is divided into several incoherent regions pt like an image i (τ, s), in pt i (τ, s) along s to find the local maximum point max(pt i (τ, s)), considering the applicability, spline interpolation is used to fit the maximum point into the local coefficient

[0190] (4) Reconstruction component r i (t). Perform inverse continuous wavelet transform and reconstruct into component r i (t) Under the premise of ensuring the reconstruction accuracy, the algorithm time consumption is reduced; for the time domain signal r i (t), the reconstruction formula is as follows:

[0191]

[0192]

[0193] in, is the Fourier transform of ψ(t), It means taking the real part.

[0194] (5) Component amplitude sparse optimization only takes the points on the ridge line, which will lose most of the energy. Therefore, let res1(t) = x(t) to construct the following optimization problem P1:

[0195]

[0196] In the optimization problem P1, the optimization parameters are the amplitudes of each component, η||D(res1(t)-r i (t))|| 2 Used to regularize the optimization objective function, D is a differential operator to obtain the sparsest representation of the original signal, and its weight η is usually set to 1;

[0197] Construct the following optimization problem P2, and the optimization parameter is the amplitude coefficient μ:

[0198] P2 Minimize||res1(t)-μr i (t)|| 2 (twenty three)

[0199] The purpose of this step is to obtain r i The optimal amplitude of (t) further ensures the sparsity of the decomposition results;

[0200] (6) Setting local coefficients Length threshold, When the length is too short and the reconstruction is performed, the iteration is terminated and the entire decomposition process is completed.

[0201] 1. Specific application fields or related products of the present invention.

[0202] The embodiment of the present invention provides a rolling bearing fault detection system based on Riesz spectrum sparse decomposition based on the rolling bearing fault detection method based on Riesz spectrum sparse decomposition, and the system specifically includes:

[0203] The parameter information acquisition module uses a new time-frequency analysis method - Riesz four-parameter spectrum time-frequency analysis method to obtain the four-parameter information of amplitude, frequency, phase and direction of the vibration signal in the time-frequency domain, thereby improving the time-frequency aggregation and anti-noise performance of the analysis results;

[0204] The parameter spectrum screening module is connected with the parameter information acquisition module. In order to screen out the optimal parameter spectrum, the symbolic dynamics entropy is extended to a two-dimensional scale, and a new entropy value calculation method - cosine symbolic dynamics entropy is proposed;

[0205] The amplitude construction module is connected with the parameter spectrum screening module to reconstruct the parameter spectrum components screened by entropy and construct the amplitude of the sparse filter parameter optimization component, so that the decomposition result has physical meaning. On this basis, the RSSD method is proposed.

[0206] An embodiment of the present invention provides a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the rolling bearing fault detection method based on Riesz graph sparse decomposition.

[0207] An embodiment of the present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the rolling bearing fault detection method based on Riesz graph sparse decomposition.

[0208] An embodiment of the present invention provides an information data processing terminal, which is used to implement the rolling bearing fault system based on Riesz graph sparse decomposition.

[0209] 2. Relevant evidence of the technical effects obtained by the embodiments of the present invention.

[0210] 1. Simulation analysis

[0211] In order to demonstrate the superiority and accuracy of the RSSD method in mixed signal decomposition, a detailed comparison is made between its decomposition capabilities and those of the EMD and VMD methods. First, the following simulation model s(t) is constructed. s(t) consists of an AM / FM component s1(t), an oscillating decay component s2(t) and a variable frequency component s3(t). A 5dB intermittent Gaussian white noise nt is added to the signal. The time domain of the simulation signal and each component is shown in Figure 3 shown.

[0212]

[0213] The simulation signal is decomposed by RSSD, EMD and VMD methods respectively. The decomposition results are shown in Figure 4 , Figure 5 and Figure 6 .Depend on Figure 4 It can be seen that RSSD successfully separates the signal component from the intermittent noise, and the phase of the decomposition result coincides with the phase of the component in the original signal, and the amplitude of the decomposition component is almost consistent with the amplitude of the component in the original signal. Figure 5 From the EMD decomposition results, there is still noise in the IMF component and the modes are confused between the components, which proves that the decomposition ability of EMD is weaker than that of RSSD. Figure 6 We can see that VMD successfully separates the noise, and IMF1 seems to have no noise at all. Therefore, IMF1 is decomposed by VMD again, and the decomposition result is as follows: Figure 7 shown.

[0214] from Figure 7 It can be seen that VMD2 separates the signal components. The separation effect of AM and FM components is better. However, from IMF1, IMF3 and IMF4, there is a phenomenon of mode confusion between the variable frequency component and the oscillation attenuation component during the decomposition process, resulting in the problem of incomplete decomposition. The time domain signal of IMF4 contains not only variable frequency components but also oscillation attenuation components. Therefore, it is preliminarily determined that compared with EMD and VMD, RSSD decomposition is indeed more accurate in processing intermittent noise.

[0215] The time spectrum of the simulated signal and the components obtained by each decomposition method is plotted, such as Figure 8 The time-frequency spectrum depicts the instantaneous frequency and instantaneous amplitude of each component in the form of a three-dimensional grayscale graph. The amplitude is represented by grayscale, and the brighter the point, the larger the amplitude. Figure 8It can be seen that the low-frequency energy of the EMD decomposition components is mixed together; the 100Hz component is seriously affected by the intermittent noise, and blanks appear in 0.1-0.3s and 0.7-0.8. The time-frequency spectrum of the variable frequency components in RSSD decomposition and VMD2 decomposition is almost the same as that of the simulation signal components; while the 100Hz component is slightly interfered by the intermittent noise, and there are some fluctuations in 0.1-0.3s and 0.7-0.8, which is generally consistent with the simulation signal components; in the 30Hz component, due to the mode confusion of VMD2 decomposition, the amplitude of VMD2 components is lower than that of the simulation signal components in 0.1-0.4s, which is consistent with the results of VMD decomposition in the previous paragraph. Therefore, it is temporarily determined that RSSD does have certain advantages compared with EMD and VMD.

[0216] Quantification of the components of RSSD, VMD and EMD decomposition, the correlation coefficient and the correlation between the original signal r j , the orthogonality index IO and the calculation time T compare the orthogonality and calculation speed of the method. As can be seen from Table 1, the correlation coefficient of RSSD is slightly larger than VMD and larger than EMD. This shows that the components of RSSD have the best decomposition performance with the original signal. And RSSD has the smallest orthogonality index. The smaller IO is, the better the orthogonality is. It is worth noting that the decomposition speed of RSSD is lower than EMD, but faster than VMD.

[0217] Table 1 Evaluation indicators of decomposition results

[0218]

[0219] To further illustrate the superiority of RSSD in signal processing, the following simulation model s(t) is created. s(t) consists of an oscillating attenuation component s1(t), a variable frequency component s2(t) with a gradually increasing amplitude, and an amplitude-frequency modulated component s3(t). In order to better simulate the actual situation, 5dB Gaussian white noise nt is added to the signal. The time domain of the simulation signal and each component is shown in the figure below.

[0220]

[0221] EMD, VMD and RSSD are still used for decomposition and comparison. The decomposition diagram is as follows Figure 10-12 As shown. Fig.10 It can be seen that the components of RSSD decomposition are consistent with the components of the simulation signal in terms of amplitude and frequency, separating the signal components from the noise. Fig.11 As for the EMD decomposition results, the EMD components still have noise and need further processing. Fig.12 and Figure 6 Similarly, VMD separates the signal component, that is, IMF1 contains the signal component.

[0222] right Fig.12 The IMF1 component in is decomposed again by VMD, and the decomposition result is as follows Fig.13 .from Fig.13 It can be seen that s1(t), s2(t) and s3(t) all have modal confusion in the decomposition process, where s1(t) and English s2(t) interfere with each other because of their similar frequencies. Fig.13 IMF2 and IMF4 in s3(t) can be reflected, and s3(t) is also slightly affected and divided into IMF1 and IMF3. In summary, RSSD has stronger decomposition ability than EMD and VMD.

[0223] The time spectrum of the simulated signal and the components obtained by each decomposition method is plotted, such as Fig.14 .from Fig.14 It can be seen that the components decomposed by EMD are covered by noise. The AM / FM signals in the VMD2 component are affected, and the energy nearby is relatively divergent; the oscillation attenuation and frequency conversion components interfere with each other, causing the oscillation attenuation component points to be relatively dim, and the latter part of the frequency conversion component disappears directly and the energy diverges. It can be seen from the time-frequency spectrum of the RSSD component that the time-frequency spectrum of the RSSD component is almost the same as that of the simulation signal component. Therefore, RSSD does have a better decomposition effect than EMD and VMD.

[0224] Quantification of the components of RSSD, VMD and EMD decomposition, the correlation coefficient and the correlation between the original signal r i , the orthogonality index IO and the calculation time T compare the orthogonality and calculation speed of the method. As can be seen from Table 2, the correlation coefficient of RSSD is greater than that of EMD. It is worth mentioning that the three correlation coefficients of VMD r i , r3 is the AM / FM component, which has a correlation coefficient close to RSSD, and the other two coefficients are smaller than RSSD, which is consistent with the above results. This shows that the RSSD component has the best decomposition performance with the original signal. RSSD also has the smallest orthogonality index. The smaller the IO, the better the orthogonality. It is worth noting that the decomposition speed of RSSD is lower than that of EMD, but faster than VMD.

[0225] Table 2 Evaluation indicators of decomposition results

[0226]

[0227] 2. Application of RSSD method in rolling bearing diagnosis

[0228] 2.1 Application of RSSD method in variable speed bearing diagnosis

[0229] In order to verify the feasibility of RSSD in practical applications, this method is used in actual variable speed bearing fault diagnosis. Fig.15 As shown. Use electric spark cutting to cut the outer ring of the bearing, such as Fig.16 (c) The test bench is mainly composed of a generator, a torque sensor, a speed sensor, an acceleration sensor, a coupling, a test bearing, a Siemens PLC and a data acquisition device. The test bearing is an SKF 6206 deep groove ball bearing, the fault depth is set to 0.2mm, and the sampling frequency is 1024Hz.

[0230] The characteristic order K of the outer ring fault is obtained by the calculation formula of the characteristic frequency of the outer ring fault, which is calculated to be 3.568. First, the speed change of the bearing is analyzed, such as Fig.17 As shown in the figure, it increases from 0 to 810 r / min within 0 to 14 s. From the vibration signal time domain diagram, it can be seen that as the speed increases, the amplitude also increases, which proves that the vibration signal is related to the speed signal.

[0231]

[0232] In order to verify the effectiveness of RSSD decomposition of experimental signals, it is compared with EMD and VMD methods. Envelope order spectrum analysis is combined to perform variable speed bearing outer ring fault diagnosis. Each method is used to decompose the vibration signal, and the decomposition result takes the first four components; the first component with a higher center frequency is selected for angular domain resampling to obtain a relatively stable component; the final angular domain signal is subjected to envelope Fourier transform to obtain the envelope order spectrum and analyze it. The decomposition results are shown in Figure 1. Fig.18 (a) Fig.19 (a) and Fig. 20 As shown in (a), the angular domain resampled signal is Fig.18 (b) Fig.19 (b) and Fig. 20 As shown in (b), the envelope order spectrum is Fig.18 (c) Fig.19 (c) and Fig. 20 (c) as shown.

[0233] Through Fig.19 From the analysis, it can be seen that the envelope order spectrum obtained by EMD decomposition of the experimental signal and then angular domain resampling cannot find the characteristic order corresponding to the outer ring fault of the variable speed bearing, so it is impossible to diagnose the outer ring fault. Fig. 20In the envelope order spectrum of , a higher peak appears at 3.5603, which is close to the characteristic order corresponding to the variable speed bearing outer ring fault bearing, but only one characteristic order appears, so the outer ring fault diagnosis result is not particularly obvious. The envelope order spectrum of RSSD clearly shows multiple higher peaks, three of which are 3.5658, 7.1373 and 10.7087, which are close to 1 times, 2 times and 3 times the characteristic order of the variable speed bearing outer ring fault, and the amplitude is higher than the characteristic order amplitude of VMD, and the extracted fault characteristics are more obvious. The experimental results show that the RSSD method proposed in the present invention can effectively extract the fault characteristics of variable speed signals. At the same time, compared with EMD and VMD, RSSD is more superior in processing variable speed signals, and can effectively extract fault characteristics and realize effective diagnosis of variable speed bearing outer ring faults.

[0234] 2.2 Application of RSSD method in constant speed bearing diagnosis

[0235] In order to further verify the effectiveness of RSSD method in rolling bearing fault diagnosis, the RSSD method is introduced to Fig.15 The outer ring fault data of the test bench shown in the figure is decomposed. Fig.16 As shown in (c), the sampling rate is 1024 Hz, the rotation frequency f0 is set to 13.5 Hz, and the fault characteristic frequency f is calculated by formula (26): r 48.168Hz. The envelope spectrum analysis of the time domain data is performed, and the time domain diagram and envelope spectrum of the bearing fault signal are shown below.

[0236] Neither the time domain diagram nor the envelope spectrum can observe the fault characteristics of the outer ring of the bearing. Therefore, RSSD, EMD and VMD are used to decompose the vibration signal respectively, and the components obtained are as follows Figure 23-25 .

[0237] The components of each method are used for subsequent analysis, and the component with the richest fault information is taken for envelope spectrum analysis. The envelope spectrum obtained is as follows: Fig.26 As shown in the figure. The transfer frequency is prominent in the envelope spectrum of the RSSD component, and the fault characteristic frequency is accurately extracted. Therefore, RSSD can accurately diagnose the outer ring fault of the energy recovery bearing test bench. At the same time, it can be seen that there are more prominent transfer frequencies and frequency doublings in the envelope spectrum of the VMD component, and the fault characteristic frequency is weak. Therefore, it is impossible to diagnose the outer ring fault of the bearing using VMD. The fault characteristic frequency in the EMD envelope spectrum is not very obvious, and it is difficult to effectively diagnose the bearing.

[0238] 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.

[0239] 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 technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A rolling bearing fault detection system based on Riesz graph sparse decomposition, characterized in that the system include: The time-frequency analysis module uses the Riesz four-parameter spectrum time-frequency analysis method to obtain the amplitude, frequency, phase and direction of the vibration signal; The entropy value calculation module characterizes the complexity of the signal based on the cosine symbol dynamics entropy calculation method to screen out the optimal parameter spectrum; The sparse decomposition module reconstructs the components of the selected parameter spectrum and optimizes the sparse filter parameters; The control module realizes adaptive decomposition and fault diagnosis of rolling bearing fault signals through RSSD method.

2. The rolling bearing fault detection system based on Riesz graph sparse decomposition according to claim 1, characterized in that: The time-frequency analysis module further comprises: Riesz transform unit, to obtain the amplitude, frequency, phase and direction information of the signal in the time-frequency domain; The Log-Gabor filter bank is used to filter the signal in the Riesz four-parameter spectrum to improve the signal's noise resistance and analysis accuracy.

3. The rolling bearing fault detection system based on Riesz graph sparse decomposition according to claim 1, characterized in that: The entropy value calculation module includes: A cosine similarity calculation unit, used for calculating the cosine similarity between adjacent blocks of a block matrix in a two-dimensional symbolic dynamics entropy; The symbol sequence generation unit performs sequence segmentation and symbolization on the signal based on the symbol dynamics entropy, generates a state pattern of symbol arrangement, and calculates the state transition probability.

4. The rolling bearing fault detection system based on Riesz spectrum sparse decomposition according to claim 1, characterized in that: The sparse decomposition module is implemented by the following steps: The vibration signal is reconstructed using a four-parameter spectrum; The signal components are reconstructed using the inverse continuous wavelet transform method; The component amplitudes are sparsely optimized and the sparsest representation of the component amplitudes for solving the optimization problem is constructed to ensure the sparsity and physical meaning of the decomposition results.

5. A rolling bearing fault detection method based on Riesz graph sparse decomposition, characterized in that: The method includes: S1: A new time-frequency analysis method, Riesz four-parameter spectrum time-frequency analysis method, is used to obtain the four-parameter information of amplitude, frequency, phase and direction of vibration signals in the time-frequency domain, thereby improving the time-frequency aggregation and anti-noise performance of the analysis results; S2: In order to screen out the optimal parameter spectrum, the symbolic dynamics entropy is extended to a two-dimensional scale, and a new entropy value calculation method, cosine symbolic dynamics entropy, is proposed; S3: The parameter spectrum screened by entropy is reconstructed into components, and the amplitude of the sparse filter parameter optimization component is constructed to make the decomposition result have physical meaning. On this basis, the RSSD method is proposed.

6. The rolling bearing fault detection method based on Riesz graph sparse decomposition according to claim 1, characterized in that: The S1 specifically includes: (1) Riesz transform Traditional time-frequency analysis methods can only obtain the amplitude and frequency information of the signal in the time-frequency domain, which will cause deviations when analyzing the signal. In order to solve this problem, the Riesz transform is innovatively used to analyze the vibration signal. The Riesz transform can obtain the amplitude, frequency, phase and direction information of the signal, thereby more comprehensively characterizing the fault characteristics and improving the time-frequency aggregation and noise resistance of the analysis results. For any given one-dimensional real signal f(x) 1D , x∈R, analytical signal f A (x) is defined as the complex function of the signal, where the real part is the signal itself and the imaginary part (conjugate function) is its Hilbert transform; f A (x)=f(x) 1D -if H (x) (1) Where f H ( )=f(x) 1D *h(x) represents the Hilbert transform of the original signal, and the transform kernel is The corresponding frequency response is H(w) = -jsgn(w). Similarly, the Riesz transform can generate an analytical signal representation of a two-dimensional signal. For any two-dimensional signal f(x) 2D , x∈R 2 , its Riesz transform f R (·) is defined as: x = (x1, x2) is the position index vector, y = (y1, y2) is a virtual variable, and || || is the vector norm. The definition of the Riesz transform can be rewritten in the form of a convolution kernel as Two-dimensional signal f(x) 2D Riesz transform with two directions Where x′=(x′1,x′2), pv represents the Cauchy principal value integral, and the first and second transform kernel functions of Riesz transform are f(x) 2D The linear combination with the Riesz transform is f(x) 2D The analytical signal is defined as the monotonic signal f M (x) Where i and j represent two imaginary units, and {i, j, 1} constitutes R 3 A set of orthogonal bases of space. Thus, f(x) 2D The monotonic signal is usually vectorized by a function or vector field f M : R 2 →R 3 To define, (2) Riesz four-parameter spectral time-frequency analysis method The previous time-frequency transformation method can only obtain the amplitude and frequency information of the signal in the time-frequency domain. The innovative use of Riesz transform and Log-Gabor filter group to analyze the signal can simultaneously obtain the four-parameter information of the signal in the time-frequency domain: amplitude, frequency, phase and direction. This can more comprehensively characterize the bearing fault characteristics and improve the time-frequency aggregation and anti-noise performance of the analysis results. On this basis, a time-frequency analysis method, the Riesz four-parameter spectrum time-frequency analysis method, is proposed. This method is divided into the following steps: 1) Perform continuous wavelet transform on the original signal. Given a signal x(t), its CWT can be expressed as Where ψ(t) is the morlet wavelet basis, τ,s represents translation and scale transformation; 2) Get the Riesz four-parameter spectrum. The Riesz four-parameter spectrum refers to The amplitude spectrum A(τ, s) and phase spectrum constructed by Riesz transform Frequency spectrum f(τ, s) and directional spectrum θ(τ, s); Considering the influence of noise on the four-parameter spectrum, the four-parameter spectrum is filtered using a Log-Gabor filter; For the Log-Gabor filter, it has the advantages of zero DC component and is not limited by bandwidth, and its frequency response function is: Where ω0 is the center frequency and σ is the scaling factor for adjusting the bandwidth; in order to maintain a fixed shape ratio filter bank, σ is usually r =σ / ω0 remains constant; if observed on a logarithmic frequency scale, the Log-Gabor filter has a Gaussian-like frequency response; Since the Log-Gabor filter has no analytical expression in the time domain, the time domain representation is obtained by inverse Fourier transform; Among them, F -1 represents the inverse Fourier transform, express The model, yes The Fourier representation of yes The two Riesz kernels of and They are Derivative with respect to τ, s.

7. The rolling bearing fault detection method based on Riesz graph sparse decomposition as claimed in claim 1, characterized in that: The S2 specifically includes: In order to further improve the computational efficiency and complexity representation accuracy of symbolic dynamics entropy, the two-dimensional matrix is ​​converted into a block matrix using the image block idea, and the cosine similarity between adjacent block matrices is calculated, thereby converting the complexity of calculating the two-dimensional matrix into the complexity of the one-dimensional sequence, which effectively improves the computational efficiency. On this basis, a method for calculating entropy value - cosine symbolic dynamics entropy is proposed, and its steps are as follows: Taking A(τ, s) as an example, first divide A(τ, s) into several non-overlapping blocks of the same size, with a block size of 10×10. Convert the block matrix into a vector form to calculate the cosine similarity between adjacent blocks, and obtain a series of cosine similarities; the cosine similarity between two blocks is calculated as follows: A(τ,s)={a1,a2,...a N*M } (12) <h2 style=";text-align:left;direction:ltr">D = {d1, d2,..., d<h2 style=";text-align:left;direction:ltr"> N*M-1 <h2 style=";text-align:left;direction:ltr">}={d(a1-a2),d(a2-a3),...,d(a<h2 style=";text-align:left;direction:ltr"> N*M-1 <h2 style=";text-align:left;direction:ltr"> -a<h2 style=";text-align:left;direction:ltr"> N*M <h2 style=";text-align:left;direction:ltr"> )} (13) Among them, {a1, a2, ...a N*M } represents the vector form of the block matrix, N = round(length(τ) / 10), M = round(length(s) / 10; the range of cosine similarity d is [-1, 1], which measures the cosine value of the geometric angle; note that the range of cosine similarity d is [-1, 1]; the defined cosine similarity d is the similarity between two tracks, which measures the cosine value of the geometric angle; larger cosine similarity values ​​indicate similar, predictable, or periodic dynamic changes between the two tracks; Divide the D sequence into ε intervals with equal probability. i corresponds to and only corresponds to a unique interval, so the symbol σ i It replaced d i , thus forming a symbol sequence Z{Z k , k=1,2,...,N*M-1}. According to Takens embedding theorem, subvectors are generated using embedding dimension m and time delay μ. Calculate state mode probabilities: In the formula, The state mode representing the symbol arrangement of each subvector, ε m Indicates the number of symbol arrangement state patterns, type() indicates the mapping relationship from symbol space to pattern space, and || || indicates the number of elements in the set; When the state mode has been observed When , the symbol that appears after is σ b The probability of (a=1, 2, ..., ε) is the state transition probability: According to the definition of information entropy, define cosine symbol dynamics entropy CSDE 2D Equal to the sum of state mode probability entropy and state transition probability entropy: If and only if When CSDE 2D The maximum value is ln(ε m+1 ). 2D Normalization: CSDE 2D =CSDE 2D / ln(e m+1 ) (18) CSDE 2D The larger the value, the more random and complex the distribution within the parameter spectrum. 2D The smaller it is, the more regular the distribution of the parameter spectrum is and the less affected by noise. According to statistical analysis, when m = 2, ε = 10, μ = 1, the complexity of the four-parameter spectrum can be accurately quantified and the calculation time is short; 8. The rolling bearing fault detection method based on Riesz graph sparse decomposition as claimed in claim 1, characterized in that: The S3 specifically includes: By using Riesz transform and Log-Gabor filter bank to analyze the signal, the obtained four-parameter spectrum is screened using cosine symbol dynamics entropy, and the sparse filter parameters constructed by the reconstructed components are optimized to make the decomposition result have physical meaning. On this basis, a signal adaptive decomposition method based on time-frequency spectrum - RSSD method is proposed, and its main steps are as follows: (1) As shown in formulas (8)-(10), the filtered The inverse Fourier transform is used to obtain its time domain representation. As shown in formula (11), the four-parameter spectrum PT (τ, s) is obtained; (2) Determine the optimal parameter spectrum, as shown in formulas (12)-(18), and use the cosine signed dynamics entropy to calculate the entropy value CSDE of each parameter spectrum 2D , as shown in formula (20), the optimal parameter spectrum PT is obtained i (τ, s); (3) Obtaining local coefficients Given a threshold mean(mean(PT i (τ, s))), further reducing the pseudo local minimum points caused by noise; PT i (τ, s) is divided into several incoherent regions pt like an image i (τ, s), at pt i (τ, s) along s to find the local maximum point max(pt i (τ, s)), considering the applicability, spline interpolation is used to fit the maximum point into the local coefficient (4) Reconstruction component r i (t). Perform inverse continuous wavelet transform and reconstruct into component r i (t) Under the premise of ensuring the reconstruction accuracy, the algorithm time consumption is reduced; for the time domain signal r i (t), the reconstruction formula is as follows: in, is the Fourier transform of ψ(t), It means taking the real part. (5) Component amplitude sparse optimization only takes the points on the ridge line, which will lose most of the energy. Therefore, let res1(t) = x(t) to construct the following optimization problem P1: In the optimization problem P1, the optimization parameters are the amplitudes of each component, η||D(res1(t)-r i (t))|| 2 Used to regularize the optimization objective function, D is a differential operator to obtain the sparsest representation of the original signal, and its weight η is usually set to 1; Construct the following optimization problem P2, and the optimization parameter is the amplitude coefficient μ: P2 Minimize||res1(t)-μr i (t)|| 2 (23) The purpose of this step is to obtain r i The optimal amplitude of (t) further ensures the sparsity of the decomposition results; (6) Setting local coefficients Length threshold, When the length is too short and the reconstruction is performed, the iteration is terminated and the entire decomposition process is completed.

9. A computer-readable storage medium storing a computer program, wherein the computer program is processed When executed by the processor, the processor executes the steps of the rolling bearing fault method based on Riesz graph sparse decomposition as described in any one of claims 5 to 8.

10. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the rolling bearing fault system based on Riesz graph sparse decomposition as described in claim 1.

Images