A gearbox fault diagnosis method and system based on cyclic pulse feature decomposition

Abstract

The present invention belongs to the field of fault diagnosis technology, and in particular relates to a gearbox fault diagnosis method and system based on cyclic pulse feature decomposition. The method comprises: constructing a new dictionary matrix, a "cyclic pulse dictionary matrix," to accurately estimate the pulse cycle period contained in a signal formed by mixing multiple periodic pulse signals; using the Goertzel algorithm to obtain the corresponding frequency amplitude according to the order of the energy values ​​of the pulse cycle period; separating and reconstructing the CPCC corresponding to the pulse cycle period, and decomposing the signal into a plurality of CPCCs. The present invention constructs a cyclic pulse dictionary matrix to sparsely represent the pulse cycle period of the periodic pulse signal, accurately estimating the cycle period of the fault pulse feature in the gear vibration signal; and using the Goertzel algorithm to decompose and reconstruct the CPCC in the gear vibration signal according to the pulse cycle period, with the decomposed CPCC containing rich periodic pulse feature information.

Description

Technical Field

[0001] The present invention belongs to, but is not limited to, the technical field of fault diagnosis, and in particular relates to a gearbox fault diagnosis method and system based on cyclic pulse feature decomposition. Background Art

[0002] Gears are key components for transmitting power and torque in mechanical equipment. Failures can cause mechanical equipment to shut down or even lead to safety accidents. Analysis of the principles of gear meshing reveals that when two gears are in meshing, the vibration signal generated by the meshing has periodic characteristics. When a gear fails, it generates periodic shock pulses. These periodic shock pulse signals contain fault information, and these periodic shock pulse signals are often mixed with vibration and noise information generated by other components of the mechanical equipment, causing the periodic fault information to be submerged. Therefore, separating the periodic pulse signal from numerous interference signals is the key to diagnosing gear faults and accurately extracting fault information. Most researchers use different signal decomposition methods to separate the interference information and fault characteristic information in the original signal, extract the fault information from the vibration signal, and obtain the current health status of the gear.

[0003] Empirical mode decomposition (EMD) can adaptively decompose a signal into multiple intrinsic mode functions (IMFs), but EMD can cause modal aliasing during signal decomposition. To overcome this limitation, EEMD was proposed. EEMD adds white noise to the original signal and repeatedly calculates the mean of the IMFs to alleviate modal aliasing. However, modal aliasing still persists when the signal's frequency components are too complex. WT, an algorithm that decomposes signals based on their frequency domain characteristics, has been widely used in signal processing and rotating machinery fault diagnosis. WT decomposes vibration signals into single components with distinct characteristics by selecting appropriate wavelet basis functions, fully exploiting fault signature information hidden in the noise. However, WT has certain limitations. First, manual intervention is required to select the appropriate wavelet basis functions before signal decomposition, making the algorithm non-adaptive. Second, the WT transform essentially involves Fourier transforms with varying window sizes, which still cannot eliminate modal aliasing in the decomposition results. VMD outperforms existing adaptive decomposition methods in fault condition monitoring, exhibiting equivalent filtering characteristics and excellent anti-interference capabilities, and can extract fault signatures from non-stationary data. However, the decomposition performance of VMD is constrained by the decomposition parameters. There is a problem that parameters such as the number of decomposed modes and the decomposition bandwidth need to be manually set, and improper parameter selection will still cause modal confusion in the decomposition results.

[0004] Several classic signal decomposition methods have been selected for analysis above, and they can be categorized into two main groups: one is to decompose the signal from the perspective of the signal's time domain: for example, EMD and EEMD decompose the signal by obtaining the time-scale characteristics of the signal itself; the other is to decompose the signal from the perspective of the signal's frequency domain characteristics: for example, WT decomposes the signal into a series of linear combinations of wavelet functions to analyze the signal's frequency characteristics; VMD decomposes the signal into multiple modal components with different frequency characteristics in a variational manner. However, the characteristic information of gear faults exists in the vibration signal in the form of periodic pulses. By analyzing the decomposition principles and decomposition components of the above decomposition methods, it is not difficult to find that when the original signal is a complex non-sinusoidal waveform, these methods will force it to be decomposed into several single waveform components. The decomposed components cannot represent the periodic pulse characteristics, resulting in the decomposition results lacking physical meaning.

[0005] Although existing decomposition methods can extract gear fault characteristics from vibration signals to a certain extent, the decomposed components still lack the ability to explain the periodic shock pulse characteristics of gear faults. In view of the strong periodicity of the shock pulses generated by gear faults, existing periodic signal separation algorithms are analyzed. Sethares and Staley proposed the periodic transform (PT), which obtains the periodic components in the signal by constructing a "periodic subspace". However, when the periodic components in the signal are complex, the periodic subspace cannot accurately obtain the hidden period in the signal, so PT cannot accurately represent the periodic components in the signal. Vaidyanathan and Pal proposed using the Farey dictionary to sparsely represent the hidden period of the signal and identify the hidden period in the signal. However, when the noise energy value in the signal is too high, it is necessary to introduce an optimal threshold to obtain the periodic information in the signal. The different noise conditions of the signal make it particularly difficult to determine the optimal threshold value. Summary of the Invention

[0006] In response to the problem in the existing technology that the current gear fault diagnosis method has difficulty in accurately extracting the pulse cycle period of complex periodic pulse signals and separating the cyclic pulse feature information, the present invention provides a gearbox fault diagnosis method and system based on cyclic pulse feature decomposition.

[0007] The present invention is implemented as follows: a gearbox fault diagnosis method based on cyclic pulse feature decomposition, the method comprising:

[0008] S1: Construct a new dictionary matrix "cyclic pulse dictionary matrix" to accurately estimate the pulse cycle period contained in the signal composed of multiple periodic pulse signals;

[0009] S2: Use the Goertzel algorithm to obtain the corresponding frequency amplitude according to the energy value of the pulse cycle;

[0010] S3: Separate and reconstruct the CPCC corresponding to the pulse cycle period, and decompose the signal into several CPCCs.

[0011] Furthermore, the S1 specifically includes:

[0012] (1) DFT matrix

[0013] Since the DFT matrix satisfies the Vandermonde matrix condition and has different rows, the DFT matrix is ​​a full-rank matrix; in addition, given a vector Where W P =e -j2π / P , k is an integer and 1≤k≤P, rewrite it as where d i It is a divisor of P and is denoted as d i |P; as a vector in n, The vector has period d i ; Let φ(n) denote the Euler totient function evaluated at n, since k i There is φ(d i ) such values, so φ(d i ) in the form of The signal is exactly equal to d i If you consider For all values ​​of k in 1≤k≤P, each divisor d of P i The generated form is For each d i , obviously φ(d i ) such sequences, because k i There is φ(d i ) values; the total number of such sequences is:

[0014]

[0015] In summary, the P columns of the DFT matrix with dimension P×P can be divided into K categories, each of which is a factor d i |P; in class i there is φ(d i ) columns, can be aggregated into P×φ(d i )matrix Their form is The period is d i ; No column of this DFT matrix has d i |Periods other than P;

[0016] For example, for the 8×8 DFT matrix in equation (5), the first column is a vector with a period of 1, the second column is a vector with a period of 2, the third and fourth columns are vectors with a period of 4, and the rest are vectors with a period of 8;

[0017]

[0018] (2) Construction of Circular Pulse Dictionary Matrix

[0019] According to the periodic characteristics of the DFT matrix above, the P×P DFT matrix D is expressed as:

[0020] D=(c1,c2,…,c P )(6)

[0021] Among them, c P is the vector with period value P in the Pth column of matrix D, c P The dimension is P×1; let m=1,2,…,P-1, and the number of values ​​in m that are prime to the period P is count(■) and is recorded as L P :

[0022] L P =count(gcd(m,P)=1)(7)

[0023] Where gcd(m,P) means obtaining the greatest common divisor of m and P, and reconstructing the DFT matrix into C P :

[0024]

[0025] in It is c P The cyclic shift form, for example, C8 in formula (9) is:

[0026]

[0027] Assume that the input data is x(n), the data length is N, and the matrix C is periodically P Copy and expand to form N×L P Matrix B P , truncate the last C if necessary Pq Matrix C P The first h rows, where q = ceil(N / P), such that N = P × (q-1) + h;

[0028]

[0029] Building from 1 to P max B P Matrix, forming the circulant impulse dictionary matrix A:

[0030]

[0031] Among them, A is an N×L matrix;

[0032] (3) Cyclic pulse period estimation

[0033] If the given signal has a period less than P max If it is a periodic signal, then it must be a linear combination of dictionary columns; this is because if it is a period with period P, then a column with period P as a divisor must be able to span it, so the system of equations in Equation (12) must have a solution y, where x = (x(0), x(1), ..., x(N-1)) T , the equations are:

[0034] x=Ay (12)

[0035] To solve the equations (12), we use the orthogonal matching pursuit algorithm (OMP), let the residual γ0 = x, and the atomic support set is Support matrix A0 = 0, sparsity I = round(P max ×4), the number of iterations t = 1; first find the maximum column inner product index λ t :

[0036]

[0037] In the above formula, a j is the jth column of A, and then the atomic support set Λ is updated t =Λ t-1 ∪λ t , update the support matrix A t =A t-1 ∪a λ , then find x=A t y t The least squares solution of :

[0038]

[0039] Finally, update the residual When t>I, the iteration ends, otherwise t=t+1, and continue to search for the maximum column inner product index λ t ; Let G(P) be the periodic energy value of x(n) in the current period P. G(P) is defined as follows:

[0040]

[0041] The periodic energy value of the signal x(n) ranges from 1 to P max The change is recorded as S(U), where U=1,2,…,P max , S(U) is defined as follows:

[0042] S(U)=(G(1),G(2),…,G(P max ))(16).

[0043] Furthermore, the S2 specifically includes:

[0044] The Goertzel filtering algorithm is used to calculate the kth DFT component of a signal X(n) of length N. The specific expression is shown in formula (17):

[0045]

[0046] By multiplying both sides of the equation After sorting, we get:

[0047]

[0048] The right side of the above equation can be understood as X(n) and h k (n), where u(l) is a unit step function, and l is an integer. The result of the convolution is expressed as Y(n) in the form of convolution, where m is an integer, and we get:

[0049]

[0050] Comparing X(n) with Y(n), we can see that X(k) is the N-times sampled convolution result, as shown in Equation (20):

[0051] X(k)=Y k (N) (20)

[0052] Where k is any non-negative integer, and the transformation equation is obtained by performing Z transformation on the output response:

[0053]

[0054] The above formula can be regarded as First, is a geometric series with a common ratio. For |q|<1, i.e., |z|>1, this sequence tends to converge. Integrating the summation results yields:

[0055]

[0056] The corresponding difference equation is:

[0057]

[0058] where y k (-1) = 0; introduce conjugate factors into both the numerator and denominator of the fraction (22) Construct a filter with second-order conjugate poles and obtain the transfer function of the second-order Goertzel algorithm:

[0059]

[0060] Goertzel filtering can obtain the amplitude information corresponding to a specific frequency. Therefore, the signal pulse cycle period is estimated by combining the cyclic pulse dictionary matrix, and the frequency amplitude of the corresponding period can be extracted through Goertzel filtering.

[0061] Furthermore, the S3 specifically includes:

[0062] According to formula (1), it can be deduced that if x(n) is a mixture of three non-sinusoidal periodic signals, n = 1, 2, ..., N, n∈Z, and their periods are P = {P1, P2, P3}, their exponential sum model is:

[0063]

[0064] The specific decomposition steps are as follows:

[0065] (1) Let r i =x(n), i=1, according to the length of the signal ri and the maximum estimated period P max (P max >P), build from 1 to P max The cyclic pulse dictionary matrix A;

[0066]

[0067] (2) Using the OMP algorithm to solve the signal r i The solution y of the equation composed of and the cyclic pulse dictionary matrix A:

[0068] r i =Ay (27)

[0069] (3) Calculate the value from 1 to P in y max The periodic energy value sequence S(U) is obtained, and the period PM where the maximum energy value of S(U) is located is obtained;

[0070]

[0071] S(U)=(G(1),G(2),…,G(P max )) (29)

[0072] PM=max(S(U)) (30)

[0073] (4) Let S(PM) = 0, and extract the signal r according to the period PM by using the Goertzel filtering algorithm shown in formula (20) i The frequency amplitude sequence X corresponding to the PM in the middle period PM :

[0074] X PM =(Y1(PM),Y2(PM),…,Y PM-1 (PM)) (31)

[0075] Among them, X PM The corresponding frequency is

[0076] (5) For non-frequency F PM Amplitude within index X PM Perform zero-filling interpolation. After zero-filling interpolation, X PM1 The data length is And X PM1 Mirror flip constitutes the bilateral spectrum X PM2 , where X PM2 The length is N, and X PM2 Perform IFFT transformation to restore the cyclic pulse characteristic component CPCCI corresponding to the periodic PM, and realize the reconstruction of the cyclic pulse characteristic component;

[0077] CPCC i =IFFT(X PM2 ,N)(32)

[0078] r i+1 =r i -CPCC i (33)

[0079] (6) If i>count(S(U)>0) or i>P max Then the iteration is terminated, otherwise i=i+1, and the process returns to step (3) and repeats the iteration process.

[0080] Another object of the present invention is to provide a gearbox fault diagnosis system based on cycle pulse feature decomposition based on the gearbox fault diagnosis method based on cycle pulse feature decomposition, wherein the system specifically comprises:

[0081] The matrix construction module is used to construct a new dictionary matrix "cyclic pulse dictionary matrix" to accurately estimate the pulse cycle period contained in the signal composed of multiple periodic pulse signals;

[0082] The frequency amplitude acquisition module is connected to the matrix construction module and uses the Goertzel algorithm to obtain the corresponding frequency amplitude according to the energy value of the pulse cycle;

[0083] The separation and reconstruction module is connected to the frequency amplitude acquisition module to separate and reconstruct the CPCC corresponding to the pulse cycle period and decompose the signal into several CPCCs.

[0084] 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 gearbox fault diagnosis method based on cyclic pulse feature decomposition.

[0085] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the gearbox fault diagnosis method based on cyclic pulse feature decomposition.

[0086] Another object of the present invention is to provide an information data processing terminal, which is used to implement the gearbox fault diagnosis method system based on cyclic pulse feature decomposition.

[0087] 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:

[0088] First, to address the current difficulties in gear fault diagnosis methods in accurately extracting the pulse cycle period of complex periodic pulse signals and separating the cycle pulse characteristic information, this paper proposes a new signal decomposition method, namely the CICD method. This method accurately estimates the pulse cycle period in the vibration signal by constructing a cycle pulse dictionary matrix. The Goertzel algorithm is used to separate the pulse component information in the vibration signal based on the pulse cycle period, and the CPCC of each pulse cycle period is reconstructed, thus realizing the extraction of periodic impact fault information in the gear fault vibration signal. The specific conclusions are as follows:

[0089] (1) A cyclic pulse dictionary matrix is ​​constructed to sparsely represent the pulse cycle period of the periodic pulse signal, and the cycle period of the fault pulse feature in the gear vibration signal is accurately estimated.

[0090] (2) The Goertzel algorithm is used to decompose and reconstruct the CPCC in the gear vibration signal according to the pulse cycle period. The decomposed CPCC contains rich periodic pulse feature information.

[0091] (3) The results of simulation and experimental wave analysis show that the CICD method can accurately extract the fault characteristics of the gear, and the decomposed CPCC envelope spectrum is simple and not affected by other interference components. Compared with the existing signal decomposition methods, the decomposition results of this method are more explanatory of the gear fault characteristics, and the decomposition ability and noise robustness of complex periodic pulse signals are superior.

[0092] Second, as auxiliary evidence for the inventiveness of the claims of the present invention, it is also reflected in the following important aspects:

[0093] (1) The expected benefits and commercial value of the technical solution of the present invention after transformation are:

[0094] (2) The technical solution of the present invention fills the technical gap in the industry at home and abroad:

[0095] The current fault diagnosis method for gear shock pulse extraction provides a gearbox fault diagnosis method and system based on cyclic pulse feature decomposition, which solves the problem of difficulty in obtaining gear shock pulse feature information in vibration signals under noise interference, and fills the technical gap in the method of extracting shock pulse features from vibration signals at home and abroad.

[0096] (3) The technical solution of the present invention solves the technical problems that people have been eager to solve but have never been able to solve successfully:

[0097] The technical solution of the present invention accurately estimates the pulse cycle period in the vibration signal by constructing a cyclic pulse dictionary matrix, uses the Goertzel algorithm to separate the pulse component information in the vibration signal according to the pulse cycle period, and reconstructs the CPCC of each pulse cycle period, thereby realizing the automatic extraction of periodic impact fault information in the gear fault vibration signal and the intelligent diagnosis of gear faults.

[0098] Third, the technical solution of the present invention provides a diagnostic method based on cyclic pulse feature decomposition in the field of gearbox fault diagnosis, which solves key problems in the prior art, mainly manifested in solving the following technical problems and bringing significant technological progress:

[0099] 1. Technical problems solved

[0100] (1) Periodic pulse signals are difficult to separate accurately

[0101] Existing fault diagnosis methods are susceptible to interference from noise and other periodic components when processing complex periodic pulse signals, making it difficult to accurately identify the cycle period of the fault signal. This is especially true when multiple pulse signals are mixed, resulting in low signal decomposition accuracy. This invention constructs a "cyclic pulse dictionary matrix" and leverages its periodic characteristics to effectively decompose the signal. This allows for more accurate estimation of the cycle period of multiple periodic pulse signals, improving the accuracy of fault feature separation.

[0102] (2) Inaccurate extraction of frequency amplitude

[0103] Traditional fault diagnosis methods, due to limitations in the frequency amplitude extraction process, cannot accurately extract key frequency information, affecting the accuracy of fault diagnosis. This invention uses the Goertzel algorithm to analyze the frequencies corresponding to different pulse periods and extract the amplitudes of the corresponding frequencies, effectively overcoming the deviation problem in frequency extraction and ensuring the accuracy of the diagnostic results.

[0104] (3) Unable to effectively handle the separation and reconstruction of multiple periodic signals

[0105] Existing technologies often fail to effectively achieve accurate signal reconstruction when separating multiple periodic signals, resulting in incomplete capture of fault characteristics. This invention utilizes pulse cycle decomposition and reconstruction technology, using cyclic pulse characteristic components (CPCC) to accurately decompose and reconstruct signals. This allows the characteristic information of each pulse cycle to be effectively captured and restored, improving the accuracy and comprehensiveness of fault diagnosis.

[0106] 2. Technological advancement

[0107] (1) The accuracy of fault feature extraction is greatly improved

[0108] By introducing eigendecomposition technology based on a cyclic pulse dictionary matrix, this paper accurately separates complex periodic signals in gearboxes and extracts precise frequency amplitudes using the Goertzel algorithm. This method significantly improves the resolution of fault signals, enabling the capture of subtle fault characteristics that would otherwise be difficult to identify, greatly enhancing the accuracy of gearbox fault diagnosis.

[0109] (2) Improved efficiency of multi-periodic signal decomposition

[0110] This method leverages the periodic nature of the DFT matrix to construct an efficient cyclic pulse dictionary matrix, enabling rapid decomposition and reconstruction of multiple periodic pulse signals. Compared to traditional methods, this method significantly improves the efficiency of processing multi-periodic signals, shortening diagnosis time while ensuring the accuracy and completeness of diagnostic results. It is particularly suitable for online monitoring and real-time fault detection in industrial scenarios.

[0111] (3) Enhanced robustness of diagnostic methods

[0112] By using the Orthogonal Matching Pursuit (OMP) algorithm to find a sparse solution between the signal and the dictionary matrix, the proposed method enhances noise robustness, ensuring accurate fault signal extraction even in the presence of high noise interference. This technological advancement effectively addresses the impact of noise interference on fault diagnosis accuracy in industrial environments, enabling the system to maintain high performance under complex operating conditions.

[0113] (4) Improved automation level of fault diagnosis

[0114] Through a series of algorithmic optimizations, this invention automates the entire process from signal acquisition, cycle estimation, frequency extraction, to fault signature reconstruction, reducing the need for human intervention and improving the automation level of the diagnostic system. This technology is particularly well-suited for online monitoring of complex mechanical systems, enabling real-time automatic detection and alarming of gearbox faults, reducing labor costs and improving operational efficiency.

[0115] (5) Comprehensiveness of signal decomposition

[0116] Traditional fault diagnosis methods have limited processing capabilities for single or small periodic signals. This invention, by constructing a cyclic pulse dictionary matrix, achieves comprehensive decomposition and reconstruction of multi-periodic pulse signals. This technological advancement ensures that no matter how many different periodic signals appear in the gearbox, they can be accurately identified and decomposed, greatly improving the comprehensiveness and applicability of diagnosis.

[0117] (6) Expansion of industrial application scope

[0118] The gearbox fault diagnosis method of this invention is not only applicable to traditional gearboxes but can also be extended to other mechanical equipment fault diagnosis fields containing periodic signals, such as motors, bearings, and transmission systems. This scalability provides fault diagnosis technical support for more industries, improving the maintenance efficiency and reliability of overall industrial systems.

[0119] In summary, the present invention solves the problems of periodic signal separation and frequency extraction in gearbox fault diagnosis by combining cyclic pulse feature decomposition and Goertzel algorithm, and achieves significant technological progress in terms of accuracy, efficiency, robustness and automation level. BRIEF DESCRIPTION OF THE DRAWINGS

[0120] Figure 1 This is a flow chart of a gearbox fault diagnosis method based on cyclic pulse feature decomposition provided by an embodiment of the present invention;

[0121] Figure 2 This is a flow chart of the CICD method provided by an embodiment of the present invention;

[0122] Figure 3 This is a gear impact fault signal simulation provided by an embodiment of the present invention;

[0123] Figure 4 This is a time domain diagram of the impact attenuation simulation signal and the periodic unit pulse simulation signal of the faulty gear provided by the embodiment of the present invention;

[0124] Figure 5 The gear fault vibration shock pulse sequence simulation signal and its envelope spectrum provided by the embodiment of the present invention; (a) simulation signal time domain diagram; (b) simulation signal envelope spectrum;

[0125] Figure 6 Components and envelope spectra of the simulated signal x(t0) decomposed using the CICD method provided by an embodiment of the present invention; (a) the first four CPCC components decomposed and the decomposition residuals; (b) the envelope spectra of the first four CPCC components;

[0126] Figure 7 The components and envelope spectra of the simulated signal x(t) decomposed using the WT decomposition method provided by an embodiment of the present invention; (a) the first four d components and the decomposition residuals; (b) the envelope spectra of the first four d components;

[0127] Figure 8 The components and envelope spectra of the simulated signal x(t) decomposed using the VMD decomposition method provided by an embodiment of the present invention; (a) the first four IMF components and the decomposition residuals; (b) the envelope spectra of the first four IMF components;

[0128] Figure 9 The components and envelope spectra of the simulated signal x(t) decomposed using the EEMD decomposition method provided by an embodiment of the present invention; (a) the first four IMF components and the decomposition residuals; (b) the envelope spectra of the first four IMF components;

[0129] Figure 10 A steering system gear pump fault test bench provided by an embodiment of the present invention;

[0130] Figure 11 The driven gear of the gear pump provided by the embodiment of the present invention is faulty; (a) broken gear; (b) tooth surface wear;

[0131] Figure 12 The following are the time domain diagrams and envelope spectra of the vibration signal of a broken tooth fault provided by an embodiment of the present invention; (a) the time domain diagram of the vibration signal; (b) the envelope spectra of the vibration signal;

[0132] Figure 13 Components and envelope spectra of a broken tooth fault vibration signal decomposed using the CICD method provided by an embodiment of the present invention; (a) the first four CPCC components and decomposition residuals; (b) the envelope spectra of the first four CPCC components;

[0133] Figure 14 The components and envelope spectra of the broken tooth fault vibration signal decomposed using the WT method provided in an embodiment of the present invention; (a) the first four d components and the decomposition residual; (b) the envelope spectra of the first four d components;

[0134] Figure 15 The components and envelope spectra of the broken tooth fault vibration signal decomposed using the VMD method provided by the embodiment of the present invention; (a) the first four IMF components and the decomposition residuals; (b) the envelope spectra of the first four IMF components;

[0135] Figure 16 The components and envelope spectra of the broken tooth fault vibration signal decomposed using the EEMD method provided in an embodiment of the present invention; (a) the first four IMF components and the decomposition residuals; (b) the envelope spectra of the first four IMF components;

[0136] Figure 17 The following are the time domain diagrams and envelope spectra of the vibration signal of tooth surface wear fault provided by the embodiment of the present invention; (a) time domain diagram of the vibration signal; (b) envelope spectra of the vibration signal;

[0137] Figure 18 The components and envelope spectra of the tooth wear fault vibration signal decomposed using the CICD method provided in an embodiment of the present invention; (a) the first four CPCC components and the decomposition residuals; (b) the envelope spectra of the first four CPCC components;

[0138] Figure 19 The components and envelope spectra of the tooth wear fault vibration signal decomposed using the WT method provided in an embodiment of the present invention; (a) the first four d components and the decomposition residuals; (b) the envelope spectra of the first four d components;

[0139] Figure 20 The components and envelope spectra of the tooth wear fault vibration signal decomposed using the VMD method provided in an embodiment of the present invention; (a) the first four IMF components and the decomposition residuals; (b) the envelope spectra of the first four IMF components;

[0140] Figure 21 The components and envelope spectra of the tooth surface wear fault vibration signal decomposed by the EEMD method provided in an embodiment of the present invention; (a) the first four IMF components decomposed and the decomposition residual; (b) the envelope spectra of the first four IMF components. DETAILED DESCRIPTION

[0141] In order to make the purpose, technical solutions 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 intended to limit the present invention.

[0142] The following are two specific industrial application examples of the gearbox fault diagnosis method based on cyclic pulse feature decomposition according to the present invention:

[0143] Example 1: Wind turbine gearbox fault diagnosis system

[0144] In the field of wind power generation, the gearbox is one of the core components of wind turbines. Its operating environment is complex and the load is highly variable. Gearbox failures can directly affect the power generation efficiency of the wind turbine and even cause the unit to shut down. Since wind turbines are often located in remote areas, regular maintenance and inspections are costly and often delayed. Therefore, an efficient and accurate fault diagnosis method is needed.

[0145] Application scenarios:

[0146] Vibration signal acquisition: The gearbox of the wind turbine generates periodic vibration signals during operation. The present invention continuously collects vibration data of the equipment during operation through a vibration sensor installed on the gearbox.

[0147] Application of cyclic pulse dictionary matrix: Utilizing the cyclic pulse feature decomposition method of the present invention, a cyclic pulse dictionary matrix is ​​constructed to separate and reconstruct the periodic pulse signals of the gearbox, accurately estimate the cycle period of each pulse signal, and extract potential fault features.

[0148] Real-time fault detection and early warning: The Goertzel algorithm is used to quickly extract key frequency amplitudes and perform spectrum analysis to determine whether the gearbox is operating abnormally. If an abnormal signal is detected, the system automatically triggers a fault alarm and sends an early warning to the operation and maintenance team.

[0149] Industrial benefits:

[0150] Improve operation and maintenance efficiency: The present invention can effectively improve the accuracy and real-time performance of gearbox fault diagnosis, reduce the frequency of on-site manual inspections, and reduce maintenance costs.

[0151] Reduce downtime: By real-time monitoring of the gearbox operating status, timely warnings can be issued before a fault occurs, avoiding long equipment downtime due to major faults, thereby improving the power generation efficiency and equipment utilization of wind turbines.

[0152] Example 2: Mine belt conveyor gearbox fault monitoring system

[0153] Belt conveyors are core conveying equipment in mining operations. Their gearboxes often operate in harsh environments such as high loads, dust, and high humidity, making them susceptible to fatigue damage and wear. A gearbox failure can shut down the entire conveyor system, directly impacting ore transportation efficiency. Therefore, early fault diagnosis and real-time monitoring are essential.

[0154] Application scenarios:

[0155] Gearbox vibration signal monitoring: Mining conveyor equipment gearboxes operate constantly, and vibration sensors can be installed at key locations to collect vibration signals during operation. Using the method presented in this invention, these signals are input into a cyclic pulse dictionary matrix for decomposition and analysis.

[0156] Gearbox periodic signal analysis and reconstruction: This invention separates the pulse cycle period in the mixed signal, uses the Goertzel filtering algorithm to extract the fault frequency information of the gearbox, compares these frequencies with the normal operating frequency, and detects abnormal conditions of the gearbox.

[0157] Intelligent fault diagnosis and operation and maintenance decision support: Through the method of the present invention, the system can quickly determine the severity of the gearbox failure in the early stage of mining conveying equipment, and provide maintenance suggestions to the operator to support intelligent maintenance decision-making.

[0158] Industrial benefits:

[0159] Extending equipment life: By promptly detecting minor gearbox faults, the equipment can be prevented from continuing to operate in a serious fault state, thereby reducing wear and extending the service life of the gearbox.

[0160] Improve production efficiency: The present invention can monitor the status of the gearbox in real time, reduce production stagnation caused by sudden failures, ensure the stable operation of mine conveying equipment, and thus improve mine production efficiency.

[0161] These two embodiments demonstrate the practical application of the present invention in the wind power generation and mining industries. Through accurate fault diagnosis methods, real-time monitoring and fault warning of equipment are achieved, which significantly improves the equipment operation efficiency and the economic benefits of the enterprise.

[0162] like Figure 1 、 2 As shown, an embodiment of the present invention provides a gearbox fault diagnosis method based on cyclic pulse feature decomposition, the method comprising:

[0163] S1: Construct a new dictionary matrix "cyclic pulse dictionary matrix" to accurately estimate the pulse cycle period contained in the signal composed of multiple periodic pulse signals;

[0164] S2: Use the Goertzel algorithm to obtain the corresponding frequency amplitude according to the energy value of the pulse cycle;

[0165] S3: Separate and reconstruct the CPCC corresponding to the pulse cycle period, and decompose the signal into several CPCCs.

[0166] Said S1 specifically includes:

[0167] (1) DFT matrix

[0168] Since the DFT matrix satisfies the Vandermonde matrix condition and has different rows, the DFT matrix is ​​a full-rank matrix; in addition, given a vector Where W P =e -j2π / P , k is an integer and 1≤k≤P, rewrite it as where d i It is a divisor of P and is denoted as d i |P; as a vector in n, The vector has period d i ; Let φ(n) denote the Euler totient function evaluated at n, since k i There is φ(d i ) such values, so φ(d i ) in the form of The signal is exactly equal to d i If you consider For all values ​​of k in 1≤k≤P, each divisor d of P i The generated form is For each d i , obviously φ(di ) There are such sequences, because k i There is φ(d i ) values; the total number of such sequences is:

[0169]

[0170] In summary, the P columns of the DFT matrix with dimension P×P can be divided into K categories, each of which is a factor d i |P; in class i there is φ(d i ) columns, can be aggregated into P×φ(d i )matrix Their form is The period is d i ; No column of this DFT matrix has d i |Periods other than P;

[0171] For example, for the 8×8 DFT matrix in equation (5), the first column is a vector with a period of 1, the second column is a vector with a period of 2, the third and fourth columns are vectors with a period of 4, and the rest are vectors with a period of 8;

[0172]

[0173] (2) Construction of Circular Pulse Dictionary Matrix

[0174] According to the periodic characteristics of the DFT matrix above, the P×P DFT matrix D is expressed as:

[0175] D=(c1,c2,…,c P )(6)

[0176] Among them, c P is the vector with period value P in the Pth column of matrix D, c P The dimension is P×1; let m=1,2,…,P-1, and the number of values ​​in m that are prime to the period P is count(■) and is recorded as L P :

[0177] L P =count(gcd(m,P)=1)(7)

[0178] Where gcd(m,P) means obtaining the greatest common divisor of m and P, and reconstructing the DFT matrix into C P :

[0179]

[0180] in It is c P The cyclic shift form, for example, C8 in formula (9) is:

[0181]

[0182] Assume that the input data is x(n), the data length is N, and the matrix C is periodically P Copy and expand to form N×L P Matrix B P , truncate the last C if necessary Pq Matrix C P The first h rows, where q = ceil(N / P), such that N = P × (q-1) + h;

[0183]

[0184] Building from 1 to P max B P Matrix, forming the circulant impulse dictionary matrix A:

[0185]

[0186] Among them, A is an N×L matrix;

[0187] (3) Cyclic pulse period estimation

[0188] If the given signal has a period less than P maxIf it is a periodic signal, then it must be a linear combination of dictionary columns; this is because if it is a period with period P, then a column with period P as a divisor must be able to span it, so the system of equations in Equation (12) must have a solution y, where x = (x(0), x(1), ..., x(N-1)) T , the equations are:

[0189] x=Ay (12)

[0190] Regarding the solution of equation (12), the orthogonal matching pursuit algorithm (OMP) is used, and the residual γ0 = x, and the atomic support set is Support matrix A0 = 0, sparsity I = round(P max ×4), the number of iterations t = 1; first find the maximum column inner product index λ t :

[0191]

[0192] In the above formula, a j is the jth column of A, and then the atomic support set Λ is updated t =Λ t-1 ∪λ t , update the support matrix A t =A t-1 ∪a λ , then find x=A t y t The least squares solution of :

[0193]

[0194] Finally, update the residual When t>I, the iteration ends, otherwise t=t+1, and continue to search for the maximum column inner product index λ t ; Let G(P) be the periodic energy value of x(n) in the current period P. G(P) is defined as follows:

[0195]

[0196] The periodic energy value of the signal x(n) ranges from 1 to P max The change is recorded as S(U), where U=1,2,…,P max , S(U) is defined as follows:

[0197] S(U)=(G(1),G(2),…,G(P max ))(16).

[0198] The S2 specifically includes:

[0199] The Goertzel filtering algorithm is used to calculate the kth DFT component of a signal X(n) of length N. The specific expression is shown in formula (17):

[0200]

[0201] By multiplying both sides of the equation After sorting, we get:

[0202]

[0203] The right side of the above equation can be understood as X(n) and h k (n), where u(l) is a unit step function, and l is an integer. The result of the convolution is expressed as Y(n) in the form of convolution, where m is an integer, and we get:

[0204]

[0205] Comparing X(n) with Y(n), we can see that X(k) is the N-times sampled convolution result, as shown in Equation (20):

[0206] X(k)=Y k (N) (20)

[0207] Where k is any non-negative integer, and the transformation equation is obtained by performing Z transformation on the output response:

[0208]

[0209] The above formula can be regarded as First, is a geometric series with a common ratio. For |q|<1, i.e., |z|>1, this sequence tends to converge. Integrating the summation results yields:

[0210]

[0211] The corresponding difference equation is:

[0212]

[0213] Where yk(-1) = 0; introduce conjugate factors into both the numerator and denominator of the fraction (22) Construct a filter with second-order conjugate poles and obtain the transfer function of the second-order Goertzel algorithm:

[0214]

[0215] Goertzel filtering can obtain the amplitude information corresponding to a specific frequency. Therefore, the signal pulse cycle period is estimated by combining the cyclic pulse dictionary matrix, and the frequency amplitude of the corresponding period can be extracted through Goertzel filtering.

[0216] The S3 specifically includes:

[0217] According to formula (1), it can be deduced that if x(n) is a mixture of three non-sinusoidal periodic signals, n = 1, 2, ..., N, n∈Z, and their periods are P = {P1, P2, P3}, their exponential sum model is:

[0218]

[0219] The specific decomposition steps are as follows:

[0220] (1) Let r i =x(n), i=1, according to the signal r i The length and maximum estimated period P max (P ma x>P), construct from 1 to P max The cyclic pulse dictionary matrix A;

[0221]

[0222] (2) Using the OMP algorithm to solve the signal r i The solution y of the equation composed of and the cyclic pulse dictionary matrix A:

[0223] r i =Ay (27)

[0224] (3) Calculate the value from 1 to P in y max The periodic energy value sequence S(U) is obtained, and the period PM where the maximum energy value of S(U) is located is obtained;

[0225]

[0226] S(U)=(G(1),G(2),…,G(P max )) (29)

[0227] PM=max(S(U)) (30)

[0228] (4) Let S(PM) = 0, and extract the signal r according to the period PM by using the Goertzel filtering algorithm shown in formula (20) i The frequency amplitude sequence X corresponding to the PM in the middle period PM :

[0229] X PM=(Y1(PM),Y2(PM),…,Y PM-1 (PM)) (31)

[0230] Among them, X PM The corresponding frequency is

[0231] (5) For non-frequency F PM Amplitude within index X PM Perform zero-filling interpolation. After zero-filling interpolation, X PM1 The data length is And X PM1 Mirror flip constitutes the bilateral spectrum X PM2 , where X PM2 The length is N, and X PM2 Perform IFFT transformation to restore the cyclic pulse characteristic component CPCC corresponding to the periodic PM i , realizing the reconstruction of the characteristic components of the cyclic pulse;

[0232] CPCCi=IFFT(X PM2 ,N) (32)

[0233] r i+1 =r i -CPCC i (33)

[0234] (6) If i>count(S(U)>0) or i>P max Then the iteration is terminated, otherwise i=i+1, and the process returns to step (3) and repeats the iteration process.

[0235] Gear impact type fault vibration mechanism:

[0236] When a gear pair experiences an impact fault, due to the short duration of gear meshing, the gear impact fault can be represented by a periodic impact pulse sequence, where the interval between periodic pulses is the gear rotational frequency. Analysis of the gear meshing principle shows that gear meshing is divided into two processes: initial meshing and disengagement. When a gear defect occurs, the meshing point generates a gradually increasing and then gradually attenuated impact pulse from entering to disengaging the defect position. The single impact attenuation signal of a gear impact fault can be represented by Equation (1):

[0237]

[0238] Where T is the time range of the shock attenuation wavelet, ω d =2πf z , f zis the center frequency of the shock attenuation wavelet, and ξ is the attenuation exponent of the shock attenuation wavelet. According to the convolution characteristics of the signal, a periodic shock pulse sequence can be constructed by convolving a single shock attenuation signal with a periodic unit pulse signal. The expression of the periodic unit pulse signal is:

[0239]

[0240] Where k i is the number of unit pulse signals, i=1,2,…,Z,P i is the unit pulse period interval. Convolution of the signals in equations (1) and (2) can be expressed as:

[0241] x i (t) = x a (t)*δ i (t) (3)

[0242] According to the periodic shock pulse sequence constructed by formula (1-3), the sampling frequency fs of the signal is set to 1000 Hz, and the shock attenuation wavelet x a (t) The time range T is set to 5s, the center frequency f z The frequency is set to 300Hz, the k1 of the periodic unit pulse signal δ1(t) is set to 17, and the period interval P1 is set to 283. a (t), periodic unit pulse simulation signal δ1(t), periodic shock pulse sequence x1(t) Figure 1 shown.

[0243] 1. Specific application fields or related products of the present invention.

[0244] An embodiment of the present invention provides a gearbox fault diagnosis system based on cyclic pulse feature decomposition based on the gearbox fault diagnosis method based on cyclic pulse feature decomposition, and the system specifically includes:

[0245] The matrix construction module is used to construct a new dictionary matrix "cyclic pulse dictionary matrix" to accurately estimate the pulse cycle period contained in the signal composed of multiple periodic pulse signals;

[0246] The frequency amplitude acquisition module is connected to the matrix construction module and uses the Goertzel algorithm to obtain the corresponding frequency amplitude according to the energy value of the pulse cycle;

[0247] The separation and reconstruction module is connected to the frequency amplitude acquisition module to separate and reconstruct the CPCC corresponding to the pulse cycle period and decompose the signal into several CPCCs.

[0248] An embodiment of the present invention provides a computer device, which includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor performs the steps of the gearbox fault diagnosis method based on cyclic pulse feature decomposition.

[0249] 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 gearbox fault diagnosis method based on cyclic pulse feature decomposition.

[0250] An embodiment of the present invention provides an information data processing terminal, which is used to implement the gearbox fault diagnosis method system based on cyclic pulse feature decomposition.

[0251] 2. Relevant evidence of the technical effects obtained by the embodiments of the present invention.

[0252] 1. Simulation signal analysis

[0253] According to the vibration characteristics of gear impact fault, the simulation signal of gear fault periodic impact vibration is constructed by formula (1-3), the signal sampling frequency fs = 1000 Hz, the impact wavelet time range T = 5s, and the impact attenuation wavelet center frequency f z =300Hz, number of unit pulse signals k1=17, k2=28, k3=54, period interval of unit pulse signal xb1(t) P1=283, P2=177, P3=91, impact attenuation simulation signal x of faulty gear a (t) and periodic unit pulse simulation signals δ1(t), δ2(t), δ3(t) as Figure 4 shown.

[0254] Gear fault vibration shock pulse sequence simulation signal and its envelope spectrum Figure 5 As shown, Figure 5 (a) x1(t) = x a (t)*δ1(t),x2(t)=0.7(x a (t)*δ2(t)), x3(t)=0.5(x a (t)*δ3(t)), x(t)=x1(t)+x2(t)+x3(t)+e(n), where e(n) is -8dB Gaussian white noise. The three constructed signals are periodic shock pulse signals with different fault degrees. Envelope spectrum analysis is performed on x1(t), x29t), x3(t), and x(t), and the envelope spectrum is as follows: Figure 5 (b) shown. Figure 5In (b), we can clearly see that the first harmonic frequencies of x1(t), x2(t), and x3(t) are 3.6 Hz, 5.6 Hz, and 11 Hz, respectively. Due to the interference of Gaussian noise, the first harmonic frequencies of x1(t), x2(t), and x3(t) cannot be discerned in the envelope spectrum of x(t). Therefore, directly performing envelope analysis on the simulation signal cannot extract the fault frequency of the constructed gear fault simulation signal.

[0255] In order to verify the superiority of the CICD method, WT, EEMD and VMD are compared with the proposed method CICD, and the simulation signal x(t) is decomposed respectively. The first four components of the decomposition results and their envelope spectra are shown as follows: Figure 6-Figure 9 shown.

[0256] Figure 6 (a) is the time domain diagram decomposed by CICD method. Figure 6 (a) The decomposition results clearly show that the three gear fault vibration shock pulse sequences constructed are successfully separated from x(t). Figure 6 (b) is the envelope spectrum of the cyclic pulse characteristic component obtained by the CICD method. The first harmonic frequencies in the envelope spectra of the CPCC1, CPCC2, and CPCC3 components are 3.6 Hz, 5.6 Hz, and 11 Hz, respectively, which are consistent with the first harmonic frequencies of the constructed simulation signals x1(t), x2(t), and x3(t). In addition, the frequency doubling information of the first harmonic in each component is obvious. Therefore, the CICD method can effectively separate the gear fault simulation signal component from x(t).

[0257] Figure 7 (a) and Figure 8 (a) The components decomposed by WT method and VMD method respectively. The WT decomposition wavelet selects the 8th order Daubechies wavelet, and the decomposition level is 10. The decomposition results of the first 4 levels are taken to draw the envelope spectrum. The envelope spectrum Figure 7 As shown in (b), VMD takes the first four decomposition components for envelope analysis, and its envelope spectrum is as follows Figure 8 As shown in (b), Figure 7 (b) Figure 8 The frequencies with more prominent amplitudes in (b) only contain the first harmonic frequency of the simulation signal x3(t). The first harmonic frequencies of x1(t) and x2(t) cannot be extracted. Therefore, the WT method and VMD method cannot completely separate the fault frequency contained in x(t). Figure 9 The components and their envelope spectra decomposed by EEMD method are as follows: Figure 9(b) only contains the first harmonic frequency of the simulation signal x2(t), while the first harmonic frequencies of x1(t) and x3(t) cannot be separated from the noise. Therefore, the EEMD method cannot completely separate the fault frequency contained in x(t).

[0258] Through the verification and analysis results of the above gear fault simulation signal decomposition using four methods, it can be seen that the CICD decomposition method can separate the components of different fault degrees and effectively extract the fault impact component of the periodic component. Therefore, the CICD decomposition method proposed in this paper is suitable for gear fault diagnosis of different degrees.

[0259] 4. Test Signal Analysis

[0260] In order to further illustrate the advantages of the CICD method in gear fault diagnosis, this section will use the CICD method to diagnose the vibration data of the steering system gear pump fault test bench. Figure 9 As shown in the figure, the test bench consists of a pressure sensor, electromagnetic reversing valve, acceleration sensor, gear pump, test gear, servo motor, stop valve, hydraulic cylinder, oil tank, and pressure gauge. Fault types include tooth surface wear, partial tooth breakage, and missing teeth. The number of teeth on the driving and driven wheels of the gear pump is 10. During the test, the acceleration sensor is installed at the oil inlet, oil outlet, and top of the gear pump. The specific installation direction is as follows: Figure 10 As shown in the mark, ①Pressure sensor ②Electromagnetic reversing valve ③Acceleration sensor 1 ④Acceleration sensor 2 ⑤Gear pump ⑥Acceleration sensor 3 ⑦Servo motor ⑧Stop valve ⑨Hydraulic cylinder ⑩Oil tank Pressure gauge.

[0261] The effectiveness of the CICD algorithm is verified by the vibration data of the gear tooth breakage fault and tooth surface wear fault test. The gear pump driven gear tooth breakage fault part and tooth surface wear fault part are as follows: Figure 11 As shown in the figure, envelope analysis is first performed on the original signal to determine whether the data is not directly diagnostic data. Then, CICD is compared with WT, VMD, and EEMD decomposition methods. The vibration signals of the two faults are decomposed using the four decomposition methods respectively. Envelope analysis is performed on the components decomposed by the four methods to verify whether the four methods can effectively separate the fault characteristic frequencies masked by noise in the vibration signal.

[0262] (1) Gear tooth breakage

[0263] During the test, a broken tooth fault was set on the driven wheel of the gear pump. The sampling frequency of the acquisition card was set to 4096Hz, the speed of the gear pump was set to 1200rmp, and the characteristic frequency of the broken tooth fault of the driven gear was 20Hz. Figure 12 (a) is the time domain diagram of the vibration signal of the broken tooth fault. Figure 12 (b) is the envelope spectrum, and the red dotted line in the envelope spectrum is the frequency double of the fault characteristic frequency f g1 and the double frequency f g2 ,It can be seen from the envelope spectrum that the fault characteristic frequency of the first and second harmonics are masked by noise. Therefore, it is impossible to extract the fault characteristic frequency by directly performing envelope spectrum analysis on the original vibration signal. The CICD algorithm will be used to diagnose the gear tooth breakage fault, and WT, VMD, and EEMD will be used for experimental comparison. The decomposition results and envelope spectra of the four decomposition methods are shown in the figure. Figure 13-16 shown.

[0264] Figure 13 The results of decomposing the broken tooth fault vibration signal using the CICD algorithm and the envelope spectrum of the first four components are shown in the figure. Figure 13 From the envelope spectrum in (b), we can see that the first harmonic frequency in the envelope spectrum of CPCC1 is close to the fault characteristic frequency f g1 The fault characteristic frequency is extracted accurately. The first harmonic frequency in the envelope spectrum of CPCC2 is consistent with f g2 The first harmonic frequency in the envelope spectrum of CPCC3 and CPCC4 is close to f g1 near.

[0265] Figure 14 The components obtained by using WT to decompose the vibration signal of broken tooth fault and the envelope spectrum of the first four components, the envelope spectrum Figure 13 The frequency at which the amplitudes of the d1 and d4 components in (b) are more prominent is close to f g1 , the frequency at which the d4 component amplitude is more prominent is close to f g2 , but it cannot accurately extract the fault characteristic frequency and is easily affected by the surrounding interference information. Figure 15 and Figure 16 They are the components obtained by decomposing the broken tooth fault diagnosis signal using VMD and EEMD and the envelope spectrum of the first four components, Figure 15 The frequencies and f at which the envelope spectrum amplitudes of the IMF2 components in (b) are more prominent g1 near, Figure 16 The frequencies and f at which the envelope spectrum amplitudes of the IMF4 components in (b) are more prominent g1 It is similar to the envelope spectrum analysis result of WT decomposition, but it does not accurately extract the fault characteristic frequency and is easily affected by the surrounding interference information.

[0266] (2) Gear tooth surface wear failure

[0267] In the test, a tooth surface wear fault was set on the driven wheel of the gear pump. The sampling frequency of the acquisition card was set to 4096Hz, the speed of the gear pump was set to 800rmp, and the fault characteristic frequency of the driven gear tooth surface wear was 13Hz. Figure 17 (a) is the time domain diagram of the vibration signal of tooth surface wear fault, Figure 17 (b) is the envelope spectrum of the fault vibration signal. The red dashed lines in the envelope spectrum are the double frequency f of the fault characteristic frequency. g1 and the double frequency f g2 ,It can be seen from the envelope spectrum that the ,first and second harmonics of the fault characteristic frequencies are ,massaged by noise. Therefore, directly performing envelope spectrum analysis on ,the original vibration signal cannot extract the fault characteristic frequencies.

[0268] Figures 18-21 The vibration signal of tooth surface wear fault is decomposed by CICD method to obtain components and the envelope spectrum of the first four components, by WT method to obtain components and the envelope spectrum of the first four components, by VMD method to obtain components and the envelope spectrum of the first four components, and by EEMD method to obtain components and the envelope spectrum of the first four components.

[0269] from Figure 18 As can be seen from (b), in the envelope spectra of the first four CPCC components decomposed by the CICD method, the first harmonic frequency of CPCC1 is 13 Hz, which is consistent with the fault characteristic frequency f g1 The first harmonic frequency of CPCC2 is 12Hz, which is consistent with the fault characteristic frequency f g1 The first harmonic frequency of CPCC3 is 23Hz, which is close to the fault characteristic frequency f g2 The first harmonic frequency of CPCC4 is 26Hz, which is close to the fault characteristic frequency f g2 Therefore, the CICD method can be used to decompose and perform envelope analysis to accurately extract the fault characteristic frequency of the tooth surface wear fault vibration signal.

[0270] The envelope spectrum of the first four components of the tooth wear fault vibration signal is obtained by WT decomposition. Figure 19 As shown in (b), among the envelope spectra of the first four components, only the envelope spectrum of the d4 component has a more prominent amplitude close to f g1 The fault characteristic frequency of the tooth surface wear fault is not accurately extracted from the vibration signal. The VMD is used to decompose the tooth surface wear fault vibration signal to obtain the envelope spectrum of the first four components as shown in the figure. Figure 20 As shown in (b), the frequencies with more prominent amplitudes in the envelope spectra of IMF1 and IMF3 components in the envelope spectra of the first four components are related to f g1The frequency with the most prominent amplitude in the envelope spectrum of the IMF2 component is close to f g2 However, the fault characteristic frequency f is not accurately extracted. g2 EEMD is used to decompose the vibration signal of tooth surface wear fault to obtain the envelope spectrum of the first four components as follows: Figure 21 As shown in (b), among the envelope spectra of the first four components, only the envelope spectrum corresponding to IMF3 has a more prominent amplitude close to f g1 The fault characteristic frequency of the tooth surface wear fault is not accurately extracted from the vibration signal, just like the WT transformation.

[0271] The comparison results of the above two experiments show that the proposed CICD decomposition method can accurately separate the gear fault characteristic frequency from the interference information, and the fault diagnosis effect is better than the other three methods. In addition, the separated component envelope spectrum is very simple and is not affected by other interference information. It is an effective gear fault diagnosis method.

[0272] 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 portion can be implemented using dedicated logic; the software portion can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those skilled in the art will appreciate that the above-mentioned devices and methods can be implemented using computer-executable instructions and / or contained in processor control code, for example, 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. The devices and 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, 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, or can be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0273] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.

Claims

1. A gearbox fault diagnosis method based on cyclic pulse feature decomposition, characterized in that: The method includes: S1: Construct a new dictionary matrix "cyclic pulse dictionary matrix" to accurately estimate the pulse cycle period contained in the signal composed of multiple periodic pulse signals; S2: Use the Goertzel algorithm to obtain the corresponding frequency amplitude according to the energy value of the pulse cycle; S3: Separate and reconstruct the CPCC corresponding to the pulse cycle period and decompose the signal into several CPCCs; Said S1 specifically includes: (1) DFT matrix Since the DFT matrix satisfies the Vandermonde matrix condition and has different rows, the DFT matrix is ​​a full-rank matrix; in addition, given a vector Where W P =e -j2π / P , k is an integer and 1≤k≤P, rewrite it as where d i It is a divisor of P and is denoted as d i |P; as a vector in n, The vector has period d i ; Let φ(n) denote the Eulertotient function evaluated at n, since k i There is φ(d i ) such values, so φ(d i ) in the form of The signal is exactly equal to d i If you consider For all values ​​of k in 1≤k≤P, each divisor d of P i The generated form is For each d i , obviously φ(d i ) such sequences, because k i There is φ(d i ) values; the total number of such sequences is: In summary, the P columns of the DFT matrix with dimension P×P can be divided into K categories, each of which is a factor d i |P; in class i there is φ(d i ) columns, can be aggregated into P×φ(d i )matrix Their form is The period is d i ; No column of this DFT matrix has d i |Periods other than P; For example, for the 8×8 DFT matrix in equation (5), the first column is a vector with a period of 1, the second column is a vector with a period of 2, the third and fourth columns are vectors with a period of 4, and the rest are vectors with a period of 8; (2) Construction of Circular Pulse Dictionary Matrix According to the periodic characteristics of the DFT matrix above, the P×P DFT matrix D is expressed as: D=(c1,c2,…,c P )(6) Among them, c P is the vector with period value P in the Pth column of matrix D, c P The dimension is P×1; let m=1,2,…,P-1, and the number of values ​​in m that are prime to the period P is count(■) and is recorded as L P : L P =count(gcd(m,P)=1) (7) Where gcd(m,P) means obtaining the greatest common divisor of m and P, and reconstructing the DFT matrix into C P : in It is c P The cyclic shift form, for example, C8 in formula (9) is: Assume that the input data is x(n), the data length is N, and the matrix C is periodically P Copy and expand to form N×L P Matrix B P , truncate the last C if necessary Pq Matrix C P The first h rows, where q = ceil(N / P), such that N = P × (q-1) + h; Building from 1 to P max B P Matrix, forming the circulant impulse dictionary matrix A: Among them, A is an N×L matrix; (3) Cyclic pulse period estimation If the given signal has a period less than P max If it is a periodic signal, then it must be a linear combination of dictionary columns; this is because if it is a period with period P, then a column with period P as a divisor must be able to span it, so the system of equations in Equation (12) must have a solution y, where x = (x(0), x(1), ..., x(N-1)) T , the equations are: x=Ay (12) To solve the equations (12), we use the orthogonal matching pursuit algorithm (OMP), let the residual γ0 = x, and the atomic support set is Support matrix A0 = 0, sparsity I = round(P max ×4), the number of iterations t = 1; first find the maximum column inner product index λ t : In the above formula, a j is the jth column of A, and then the atomic support set Λ is updated t =Λ t-1 ∪λ t , update the support matrix A t =A t-1 ∪a λ , then find x=A t y t The least squares solution of : Finally, update the residual When t>I, the iteration ends, otherwise t=t+1, and continue to search for the maximum column inner product index λ t ; Let G(P) be the periodic energy value of x(n) in the current period P. G(P) is defined as follows: The periodic energy value of the signal x(n) ranges from 1 to P max The change is recorded as S(U), where U=1,2,…,P max , S(U) is defined as follows: S(U)=(G(1),G(2),…,G(P max )) (16)。 2. The gearbox fault diagnosis method based on cyclic pulse feature decomposition according to claim 1 is characterized in that: The S2 specifically includes: The Goertzel filtering algorithm is used to calculate the kth DFT component of a signal X(n) of length N. The specific expression is shown in formula (17): By multiplying both sides of the equation After sorting, we get: The right side of the above equation can be understood as X(n) and h k (n), where u(l) is a unit step function, and l is an integer. The result of the convolution is expressed as Y(n) in the form of convolution, where m is an integer, and we get: Comparing X(n) with Y(n), we can see that X(k) is the N-times sampled convolution result, as shown in Equation (20): X(k)=Y k (N) (20) Where k is any non-negative integer, and the transformation equation is obtained by performing Z transformation on the output response: The above formula can be regarded as First, is a geometric series with a common ratio. For |q|<1, i.e., |z|>1, this sequence tends to converge. Integrating the summation results yields: The corresponding difference equation is: where y k (-1) = 0; introduce conjugate factors into both the numerator and denominator of the fraction (22) Construct a filter with second-order conjugate poles and obtain the transfer function of the second-order Goertzel algorithm: Goertzel filtering can obtain the amplitude information corresponding to a specific frequency. Therefore, the signal pulse cycle period is estimated by combining the cyclic pulse dictionary matrix, and the frequency amplitude of the corresponding period can be extracted through Goertzel filtering.

3. The gearbox fault diagnosis method based on cyclic pulse feature decomposition according to claim 1 is characterized in that: The S3 specifically includes: According to formula (1), it can be deduced that if x(n) is a mixture of three non-sinusoidal periodic signals, n = 1, 2, ..., N, n∈Z, and their periods are P = {P1, P2, P3}, their exponential sum model is: The specific decomposition steps are as follows: (1) Let r i =x(n), i=1, according to the signal r i The length and maximum estimated period P max (P max >P), build from 1 to P max The cyclic pulse dictionary matrix A; (2) Using the OMP algorithm to solve the signal r i The solution y of the equation composed of and the cyclic pulse dictionary matrix A: r i =Yes (27) (3) Calculate the value from 1 to P in y max The periodic energy value sequence S(U) is obtained, and the period PM where the maximum energy value of S(U) is located is obtained; S(U)=(G(1),G(2),…,G(P max )) (29) PM=max(S(U)) (30) (4) Let S(PM) = 0, and extract the signal r according to the period PM by using the Goertzel filtering algorithm shown in formula (20) i The frequency amplitude sequence X corresponding to the PM in the middle period PM : X PM =(Y1(PM),Y2(PM),…,Y PM-1 (PM)) (31) Among them, X PM The corresponding frequency is (5) For non-frequency F PM Amplitude within index X PM Perform zero-filling interpolation. After zero-filling interpolation, X PM1 The data length is And X PM1 Mirror flip constitutes the bilateral spectrum X PM2 , where X PM2 The length is N, and X PM2 Perform IFFT transformation to restore the cyclic pulse characteristic component CPCC corresponding to the periodic PM i , realizing the reconstruction of the characteristic components of the cyclic pulse; CPCC i =IFFT(X PM2 ,N)(32) r i+1 =r i -CPCC i (33) (6) If i>count(S(U)>0) or i>P max Then the iteration is terminated, otherwise i=i+1, and the process returns to step (3) and repeats the iteration process.

4. A gearbox fault diagnosis system based on the gearbox fault diagnosis method based on cycle pulse feature decomposition according to any one of claims 1 to 3, characterized in that: The system specifically includes: The matrix construction module is used to construct a new dictionary matrix "cyclic pulse dictionary matrix" to accurately estimate the pulse cycle period contained in the signal composed of multiple periodic pulse signals; The frequency amplitude acquisition module is connected to the matrix construction module and uses the Goertzel algorithm to obtain the corresponding frequency amplitude according to the energy value of the pulse cycle; The separation and reconstruction module is connected to the frequency amplitude acquisition module to separate and reconstruct the CPCC corresponding to the pulse cycle period and decompose the signal into several CPCCs.

5. 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 performs the steps of the gearbox fault diagnosis method based on cyclic pulse feature decomposition as described in any one of claims 1 to 3.

6. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the gearbox fault diagnosis method based on cyclic pulse feature decomposition according to any one of claims 1 to 3.

7. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the gearbox fault diagnosis system based on cyclic pulse feature decomposition as claimed in claim 4.

Images