Method for adaptive decomposition of a multichannel signal

Abstract

The present application relates to the technical field of fault diagnosis, and more particularly to a multi-channel signal adaptive decomposition method, comprising the following steps: step S1: obtaining a multi-channel vibration acceleration signal; step S2: constructing a cyclic third-order tensor according to the multi-channel vibration acceleration signal; step S3: performing second-type tensor singular value decomposition on the third-order tensor and reconstructing, and calculating a multi-channel component signal; step S4: repeating steps S2 and S3 to obtain a multi-channel component signal after multiple iterations; and step S5: judging whether a device has failed according to the component signal. The present application provides a multi-channel signal adaptive decomposition method, which has high fault diagnosis accuracy, provides a basis for state monitoring and residual life prediction of rotating machinery devices, avoids missed diagnosis and misdiagnosis during device diagnosis, and improves the safety of device operation.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, and in particular to a multi-channel signal adaptive decomposition method. Background Technology

[0002] With the rapid development of science and technology, mechanical equipment has been widely used in various industries. Rotating machinery, as a typical representative of mechanical equipment, has greatly improved industrial production efficiency and brought great convenience to people's lives. However, due to long-term operation and harsh operating environments, rotating parts and their key components are prone to failure. This can lead to the shutdown of the entire rotating machinery, or even severe economic losses and catastrophic accidents resulting in casualties. Early identification of mechanical faults through fault diagnosis methods is crucial to prevent catastrophic accidents and provide technical assurance for the safe operation of equipment. However, current fault diagnosis technologies are easily affected by the methods used, leading to errors or even missed diagnoses, resulting in low accuracy. Therefore, research on fault diagnosis methods has always been a key focus of scientific research and industry, and is one of the main factors affecting equipment failure health management. Summary of the Invention

[0003] The technical problem to be solved by this invention is: in order to solve the problem that fault diagnosis is easily affected by the diagnostic method in the prior art, resulting in low accuracy of fault diagnosis, this invention provides a multi-channel signal adaptive decomposition method with high fault diagnosis accuracy, providing basic support for condition monitoring and remaining life prediction of rotating machinery equipment, avoiding missed diagnosis and misdiagnosis during equipment diagnosis, and improving the safety of equipment operation.

[0004] The technical solution adopted by this invention to solve its technical problem is: a multi-channel signal adaptive decomposition method, comprising the following steps:

[0005] Step S1: Acquire multi-channel vibration acceleration signals;

[0006] Multiple vibration acceleration sensors are installed on the device to be diagnosed, and data acquisition time is t. n The multi-channel vibration acceleration signals a1(t), ..., a of the device under diagnosis, with N data points. s (t), with a sampling frequency of Fs;

[0007] Step S2: Construct a cyclic third-order tensor based on the multi-channel vibration acceleration signal. Where I1 is a third-order tensor The first order and I2 are third order tensors The second order and I3 are third order tensors The third order;

[0008] Step S3: For the third-order tensor Perform the second type of tensor singular value decomposition and reconstruction to calculate the multi-channel component signals;

[0009] Step S4: Repeat steps S2 and S3 to obtain multi-channel component signals through multiple iterations;

[0010] Step S5: Determine whether the device has malfunctioned based on the component signal;

[0011] Step S2 includes:

[0012] S21: Calculate the multi-channel vibration acceleration signals a1(t), ..., a s The power spectral density of (t) is used to obtain the frequency f corresponding to the maximum spectral peak of each channel. max,1 ,…,f max,s Obtain the maximum spectral peak frequency f in each channel. max :

[0013]

[0014] S22: Based on the maximum spectral peak frequency f max Determine the current multi-channel vibration acceleration signals a1(t), ..., a s The maximum frequency band range in (t) is [f max -Δf, f max +Δf], where Δf is the bandwidth;

[0015] S23: Based on the maximum spectral peak frequency f max Calculate the third-order tensor The second dimension I2

[0016]

[0017] S24: Based on the multi-channel vibration acceleration signals a1(t),...,a s (t) and the second-order dimension I2 construct a cyclic third-order tensor

[0018]

[0019] Where A(:,:,1) represents the first front slice of the third-order tensor A and A(:,:,s) represents the s-th front slice of the third-order tensor A; I1 = N, I3 = the total number of multi-channels s.

[0020] Furthermore, specifically, step S3 includes:

[0021] Step S31: For the third-order tensor Perform a second type of tensor singular value decomposition to obtain the left singular tensor U, the core tensor S, and the right singular tensor V, where,

[0022] A=U1S2V T (4);

[0023] Step S32: Perform a Fourier transform on the eigenvectors of the left singular tensor U to obtain the feature set ii = [k, l, ... p] in the maximum frequency band, where k, l, and p represent the nth column of the eigenvectors in U;

[0024] Step S33: Obtain the singular values ​​corresponding to the feature group in the core tensor S, and reconstruct the feature group and the third-order tensor:

[0025]

[0026] Step S34: Reconstruct the third-order tensor using the diagonal averaging method. Reconstructed into multi-channel component signals Obtain the residual signal of the multi-channel vibration acceleration signal

[0027] Further, specifically, in step S31, the step of obtaining the left singular tensor U, the core tensor S, and the right singular tensor V includes:

[0028] For the third-order tensor Expanding the matrix A yields matrix A. Then, performing a second-type tensor singular value decomposition on matrix A yields matrices U, S, and V.

[0029] Step S32: Reconstruct the matrix U, the matrix S, and the matrix V to obtain the left singular tensor U, the core tensor S, and the right singular tensor V.

[0030] Furthermore, specifically, in step S5, the fault characteristic frequency of the device to be diagnosed is obtained, envelope analysis is performed on all the component signals, frequency values ​​close to the fault characteristic frequency of the device to be diagnosed and the harmonics of the frequency values ​​are found, and it is determined whether the device to be diagnosed has a fault.

[0031] Furthermore, specifically, the total number of channels s is equal to the number of vibration acceleration sensors.

[0032] The beneficial effects of this invention are that the multi-channel signal adaptive decomposition method of this invention integrates the multi-channel vibration acceleration signals of the device under test, determines the maximum spectral peak frequency and frequency band range of the multi-channel signals by combining the power spectral density method, and constructs the multi-channel vibration acceleration signals into a cyclic third-order tensor based on the maximum spectral peak frequency. The third-order tensor is decomposed by the second type of tensor singular value decomposition. Based on Fourier transform, the feature group that contributes the most in the current frequency band is selected. The third-order tensor of the current iteration is reconstructed based on the feature group. The iterative third-order tensor is reconstructed into multi-channel component signals by the diagonal averaging method to complete the fault diagnosis analysis of the system under test. It can adaptively enhance the fault characteristics of the channel signals, making them easier to identify, and can also reduce the noise of the channel signals, realizing early fault diagnosis of equipment. It provides basic support for the condition monitoring and remaining life prediction of rolling bearings in electromechanical equipment, avoids missed diagnosis and misdiagnosis during equipment diagnosis, improves the safety of equipment operation, and prevents major accidents caused by common faults such as bearings. It has important practical and engineering value. Attached Figure Description

[0033] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0034] Figure 1 This is a flowchart of the preferred embodiment of the present invention.

[0035] Figure 2 This is a time-domain waveform diagram of the vibration acceleration signal of the device to be diagnosed according to an embodiment of the present invention.

[0036] Figure 3 This is the envelope spectrum of the sixth component signal of the first channel signal in an embodiment of the present invention. Detailed Implementation

[0037] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0038] like Figure 1 The diagram shows the preferred embodiment of the present invention, a multi-channel signal adaptive decomposition method, comprising the following steps:

[0039] Step S1: Acquire multi-channel vibration acceleration signals;

[0040] Multiple vibration acceleration sensors are installed on the device to be diagnosed, and data acquisition time is t. n The multi-channel vibration acceleration signals a1(t), ..., a of the device under diagnosis, with N data points. s (t), with a sampling frequency of Fs.

[0041] In one specific embodiment, three vibration acceleration sensors are used, each installed on the centrifuge to be diagnosed. The three vibration acceleration sensors are respectively arranged in the axial and horizontal directions of the centrifuge, and the data acquisition time is t. n=0.6s, the sampling frequency is set to Fs=25600Hz, the number of data points is N=40000, and the acceleration signals a1(t), a2(t), and a3(t) of three channels are collected.

[0042] Step S2: Construct a cyclic third-order tensor based on the multi-channel vibration acceleration signal. Where I1 is a third-order tensor The first order and I2 are third order tensors The second order and I3 are third order tensors The third order;

[0043] Step S2 includes:

[0044] S21: Calculate the multi-channel vibration acceleration signals a1(t), ..., a s The power spectral density of (t) is used to obtain the frequency f corresponding to the maximum spectral peak of each channel. max,1 ,…,f max,s Obtain the maximum spectral peak frequency f in each channel. max :

[0045]

[0046] S22: Based on the maximum spectral peak frequency f max Determine the current multi-channel vibration acceleration signals a1(t), ..., a s The maximum frequency band range in (t) is [f max -Δf, f max +Δf], where Δf is the bandwidth;

[0047] S23: Based on the maximum spectral peak frequency f max Calculate the third-order tensor The second dimension I2

[0048]

[0049] S24: Based on the multi-channel vibration acceleration signals a1(t),…,a s (t) and the second-order dimension I2 construct a cyclic third-order tensor

[0050]

[0051] Where A(:,:,1) represents the first front slice of the third-order tensor A and A(:,:,s) represents the s-th front slice of the third-order tensor A; A i1,i2,i3 I1 represents any element of the multi-channel third-order tensor signal A; I3 = N, I3 = the total number of multi-channels s, and the total number of multi-channels s is equal to the number of vibration acceleration sensors.

[0052] In one specific embodiment, the maximum spectral peak frequency f of the three acquired acceleration signals a1(t), a2(t), and a3(t) is calculated according to the steps described above. max And the maximum frequency band range, and construct it as a multi-channel third-order tensor. Where I1 = 40,000, I2 = (0.8~1.2)25600 / f max , I3 = 3.

[0053] The maximum spectral peak frequency is calculated by the power spectral density method to construct a third-order tensor signal. Compared with the tensors constructed by periodically truncating the signal in the prior art, the present invention can easily construct a third-order tensor signal without empirical values. The construction speed of the third-order tensor signal is fast and the constructed third-order tensor signal has no interference frequencies, which improves the timeliness and accuracy of fault diagnosis.

[0054] Step S3: For the third-order tensor Perform the second type of tensor singular value decomposition and reconstruction to calculate the multi-channel component signals;

[0055] Step S3 includes:

[0056] Step S31: For the third-order tensor Perform a second type of tensor singular value decomposition to obtain the left singular tensor U, the core tensor S, and the right singular tensor V, where,

[0057] A=U1S2V T (4);

[0058] In step S31, the steps of obtaining the left singular tensor U, the core tensor S, and the right singular tensor V include:

[0059] Step S311: For the third-order tensor Expanding the matrix A yields matrix A. Performing a second-type tensor singular value decomposition on matrix A yields matrices U, S, and V.

[0060] Step S312: Reconstruct matrices U, S, and V to obtain the left singular tensor U, the core tensor S, and the right singular tensor V;

[0061] Step S313: Based on the formulas for calculating the second type of multiplication of higher-order tensors, the formula for calculating the transpose, and the left singular tensor U, the core tensor S, and the right singular tensor V, obtain the third-order tensor A = U1S2V. T :

[0062] It should be noted that: assuming an m-order tensor and nth order tensor Based on the preset m-order tensor A and n-order tensor B, construct the second type of multiplication calculation formula and obtain a new tensor C. The second type of multiplication calculation formula is as follows:

[0063]

[0064] In this case, the dimension of tensor A is equal to the dimension of tensor B, i.e., I m+1-r =J r r = 1, 2, ..., s, and K-order tensor Let s be the number of dimensions of the m-th order tensor A that are equal to the number of dimensions of the n-th order tensor B that are equal to the number of dimensions of the m-th order tensor B that are equal to the number of dimensions of the n-th order tensor B. For example, when the m-th order tensor... Only the final dimension and tensor are present. When the first-order dimensions of tensors are equal, s = 1; k represents the order dimension of tensor C, I represents the order dimension of tensor A, and J represents the order dimension of tensor B; k is determined by the dimensions of tensor A and tensor B. For example, if tensor A and tensor B have a dimension of 3, then s = 2, and the dimensions of k include k1 and k2. k1 is the first-order dimension of tensor C, determined by the first-order dimension I1 of tensor A, and k2 is the second-order dimension of tensor C, determined by the third-order dimension J3 of tensor B. Furthermore, when k = k1 represents the first-order dimension of tensor C, if the order of tensor A is 1, it corresponds to vector multiplication; if the order of tensor A is 2, it corresponds to matrix multiplication. Second-type multiplication of higher-order tensors can degenerate into standard vector and matrix multiplication.

[0065] Obtain m-order tensor Transpose:

[0066]

[0067] That is, the third-order tensor A = U1S2V T The calculation is performed using formulas (5) and (6) above. By using the second type of tensor singular value decomposition, component signals are obtained, achieving effective decomposition and feature extraction of the signal, ensuring the completeness of the status information detection of the equipment to be diagnosed, facilitating signal analysis and diagnosis, and further improving the accuracy of fault diagnosis.

[0068] Step S32: Perform Fourier transform on the eigenvectors of the left singular tensor U to obtain the feature set ii = [k, l, ... p] in the maximum frequency band, where k, l, p and other parameters represent which column of the eigenvectors in U. If the 1st, 3rd and 4th column eigenvectors all belong to the maximum frequency band, then k = 1, l = 3 and p = 4.

[0069] Step S33: Obtain the singular values ​​corresponding to the feature set ii = [k, l, ... p] in the core tensor S, and perform analysis on the feature set and the third-order tensor A = U1S2V. T Refactor:

[0070]

[0071] Step S34: Use the diagonal averaging method to reconstruct the third-order tensor Reconstructed into multi-channel component signals Obtain the residual signal of multi-channel vibration acceleration signal

[0072] Step S4: Repeat steps S2 and S3 to obtain the multi-channel component signals and the final residual signal after multiple iterations; in one embodiment, the iteration termination condition can be set according to the standard mean square deviation threshold th of the residual signal. If the standard mean square deviation of the residual signal is greater than the threshold th, the iteration continues, that is, steps S2 and S3 continue to run; if the standard mean square deviation of the residual signal is less than the threshold th, the iteration terminates and all multi-channel component signals are output.

[0073] Step S5: Determine whether the device has malfunctioned based on the component signals; specifically, obtain the fault characteristic frequency of the device to be diagnosed, perform envelope analysis on all component signals, find the frequency values ​​and harmonics of the frequency values ​​that are close to the fault characteristic frequency of the device to be diagnosed, and determine whether the device to be diagnosed has a fault. Figure 3 The figure shows the envelope spectrum of the sixth component signal of the first channel signal in a specific embodiment. From the figure, frequency values ​​and their harmonics that are close to the fault characteristic frequency of the device to be diagnosed can be found. The frequency value that is close to the fault characteristic frequency of the device to be diagnosed is f. o Its multiplication factor is 2f o 3f o When the frequency value f of the fault characteristics of the device to be diagnosed... o When the frequency is approximately equal to the fault characteristic frequency of the equipment to be diagnosed, it can be determined that the centrifuge has a fault; otherwise, the centrifuge has no fault.

[0074] This invention discloses a multi-channel signal adaptive decomposition method that integrates multi-channel vibration acceleration signals from the device under test. It uses the power spectral density method to determine the maximum spectral peak frequency and bandwidth of the multi-channel signals, and constructs a cyclic third-order tensor based on the maximum spectral peak frequency. The third-order tensor is decomposed by the second type of tensor singular value decomposition. Based on Fourier transform, the feature group that contributes the most in the current frequency band is selected. The third-order tensor of the current iteration is reconstructed based on the feature group. The iterative third-order tensor is reconstructed into multi-channel component signals by the diagonal averaging method to complete the fault diagnosis analysis of the system under test. It can adaptively enhance the fault characteristics of the channel signals, making them easier to identify, and can also reduce the noise of the channel signals, realizing early fault diagnosis of equipment. It provides basic support for the condition monitoring and remaining life prediction of rolling bearings in electromechanical equipment, avoids missed diagnosis and misdiagnosis during equipment diagnosis, improves the safety of equipment operation, and prevents major accidents caused by common faults such as bearings. It has important practical and engineering value.

[0075] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A method of adaptive decomposition of a multichannel signal, characterized by, The method comprises the following steps: Step S1: acquiring a multi-channel vibration acceleration signal; A plurality of vibration acceleration sensors are arranged on the device to be diagnosed, and a plurality of channel vibration acceleration signals a1(t), a2(t), a3(t) of the device to be diagnosed with a collection time length t, a data point number N, and a sampling frequency Fs are acquired n . s (t), a sampling frequency Fs are acquired Step S2: constructing a cyclic third-order tensor from the multi-channel vibration acceleration signal wherein I1 is a first order of a third-order tensor I2 is a second order of a third-order tensor and I3 is a third order of a third-order tensor ​ Step S3: Perform a second type of tensor singular value decomposition on the third order tensor and reconstruct to compute the multi-channel component signals; Step S4: repeating the step S2 and the step S3 to acquire a plurality of iterations of multi-channel component signals; Step S5: judging whether the device is faulty according to the component signals; In the step S2, the following steps are included: S21: calculating the power spectral density of the multi-channel vibration acceleration signal a1(t),...,a s (t) to obtain the frequency f max,1 ,…,f max,s corresponding to the maximum spectral peak of each channel max : f max = max(f max,1 ,…,f max,s ) (1) S22: determining a maximum spectral peak frequency f max a1(t),...,a s (t) in the current multi-channel vibration acceleration signal a1(t),...,a max (t) is in the range [f max +Δf], where Δf is a bandwidth. S23: calculating the maximum spectral peak frequency f max , a second order dimension I2 of the third order tensor ​ I2 = (0.8 ~ 1.2) F s / f max (2); S24: constructing a cyclic third-order tensor T3 from the multichannel vibration acceleration signals a1(t),...,aM(t) and the second-order dimension I2 s (t) and the second-order dimension I2 Wherein, A(:,:,1) represents the first frontal slice of the third-order tensor A, and A(:,:,s) represents the s-th frontal slice of the third-order tensor A; I1=N, I3=the total number of channels s.

2. The method for multichannel signal adaptive decomposition of claim 1, wherein, In the step S3, the following steps are included: Step S31: Perform the second type of tensor singular value decomposition on the third order tensor to obtain the left singular tensor U, the core tensor S and the right singular tensor V, wherein, A = U1S2V T (4); Step S32: performing Fourier transform on the eigenvectors of the left singular tensor U to acquire a characteristic group ii=[k,l,…p] in the maximum frequency band range, wherein k, l, p represent the column number of the eigenvectors in U; Step S33: acquiring the singular values corresponding to the characteristic group in the core tensor S, and reconstructing the characteristic group and the third-order tensor: Step S34: Reconstructing the third-order tensor using diagonal averaging method Reconstructing into multi-channel component signals Obtaining residual signals of the multi-channel vibration acceleration signals 3. The method for multichannel signal adaptive decomposition of claim 2, wherein, In the step S31, the steps of acquiring the left singular tensor U, the core tensor S and the right singular tensor V include: unfolding the third order tensor to obtain a matrix A, performing a second type tensor singular value decomposition on the matrix A to obtain a matrix U, a matrix S and a matrix V; Step S32: reconstructing the matrix U, the matrix S and the matrix V to acquire the left singular tensor U, the core tensor S and the right singular tensor V.

4. The method for multi-channel signal adaptive decomposition of claim 1, wherein, In the step S5, the fault characteristic frequency of the device to be diagnosed is acquired, envelope analysis is performed on all the component signals, the frequency value close to the fault characteristic frequency of the device to be diagnosed and the multiple of the frequency value are found out, and it is judged whether the device to be diagnosed is faulty.

5. The method for multi-channel signal adaptive decomposition of claim 1, wherein, The total number of channels s is equal to the number of vibration acceleration sensors.

Images