Second-type third-order tensor decomposition method, device and bearing fault detection system

Abstract

The present invention relates to the field of fault diagnosis signal processing technology, and in particular to a second-class third-order tensor decomposition method, device, and bearing fault detection system thereof. The second-class third-order tensor decomposition method comprises: obtaining a vibration acceleration signal, constructing a multi-channel third-order tensor signal X based on the vibration acceleration signal; constructing a computational model for second-class multiplication and transposition of high-order tensors; performing second-class tensor singular value decomposition on the third-order tensor signal X to obtain tensors U, S, and V; and reconstructing the third-order tensor signal X based on the computational model for second-class multiplication and transposition of high-order tensors and tensors U, S, and V to obtain multiple component signals. The second-class third-order tensor decomposition method of the present invention achieves effective signal decomposition and feature extraction to ensure the completeness of rolling bearing status information, facilitates signal analysis and diagnosis, and provides basic support for electromechanical equipment monitoring and remaining life prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis signal processing, and in particular to a second-type third-order tensor decomposition method, device and bearing fault detection system thereof. Background Art

[0002] Rolling bearings are the core components of rotating equipment and the main source of mechanical failures in rotating equipment. Research on rolling bearing fault diagnosis technology has important theoretical research significance and engineering application value for ensuring the safe operation and healthy maintenance of mechanical equipment.

[0003] Vibration analysis is an important method for detecting bearing defects. Currently, the research on bearing fault diagnosis mainly involves signal processing and fault feature extraction.

[0004] Tensors, as higher-order extensions of vectors and matrices, provide a useful representation for existing real-world data with a natural multidimensional structure. Vectors are first-order tensors, and matrices are second-order tensors. Tensor analysis can decompose and denoise signals while preserving the inherent correlations between multiple channel signals, thus offering unique advantages over vector and matrix analysis. Although tensor decomposition methods are highly popular in various research fields, classic tensor decomposition methods such as CP decomposition, Tucker decomposition, and HOSVD are all high-order extensions of matrix singular value decomposition based on n-mode products. They cannot truly implement multi-order tensor multiplication, resulting in inherent problems with tensor decomposition methods, such as the non-unique rank of CP decomposition, the non-pseudo-diagonal core tensor of Tucker decomposition, and poor accuracy. As a promising tool for rolling bearing health monitoring and fault diagnosis, these inherent problems limit signal decomposition and feature extraction, and cannot guarantee the completeness and accuracy of rolling bearing status information, making existing tensor decomposition methods unfavorable for signal analysis and equipment fault diagnosis. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to address some inherent problems in existing tensor decomposition methods, such as limitations in singular value decomposition, an inability to ensure the completeness of rolling bearing state information, and thus a technical problem that is not conducive to signal analysis and diagnosis. The present invention provides a second-class third-order tensor decomposition method that achieves effective signal decomposition and feature extraction, ensures the completeness of rolling bearing state information, facilitates signal analysis and diagnosis, and provides basic support for electromechanical equipment such as rolling bearing state monitoring and remaining life prediction.

[0006] The technical solution adopted by the present invention to solve the technical problem is: a second-type third-order tensor decomposition method, comprising the following steps:

[0007] Step S1: Obtain multi-channel vibration acceleration signals x1, x2, ..., x c ;

[0008] Step S2: According to the vibration acceleration signals x1, x2, ..., x c Constructing multi-channel third-order tensor signals Wherein, I1 is the multi-channel third-order tensor The first order, I2 is the multi-channel third-order tensor The second order and I3 are the multi-channel third-order tensors The third stage;

[0009] Step S3: Construct a computational model for the second-order multiplication and transposition of high-order tensors. Assume that the m-order tensor And n-order tensors A second type of multiplication calculation formula is constructed according to the preset m-order tensor A and the n-order tensor B, and a new tensor C is obtained. The second type of multiplication calculation formula is:

[0010]

[0011] Among them, I m+1-r =J r , r = 1, 2, ..., s, and the new tensor C is a K-order tensor s is the number of all dimensions of the last s orders of the m-order tensor A that are equal to all dimensions of the first s orders of the n-order tensor B, k represents the nth order dimension of the tensor C, I represents the nth order dimension of the tensor A, and J represents the nth order dimension of the tensor B;

[0012] Get the m-order tensor Transpose:

[0013]

[0014] Step S4: the multi-channel third-order tensor signal Performing a second type of tensor singular value decomposition to obtain tensors U, S, and V, wherein the steps of obtaining the tensors U, S, and V include:

[0015] Based on the expansion rule, the third-order tensor signal Expand to obtain matrix X, perform singular value decomposition on matrix X to obtain matrix U, matrix S and matrix V;

[0016] Based on the folding rule, the matrix U, the matrix S and the matrix V are reconstructed respectively to obtain the tensor The tensor and the tensor

[0017] Step S5: According to the calculation model of the second type of multiplication and transposition of the high-order tensor and the tensor The tensor and the tensor For the third-order tensor signal Reconstruct and obtain multiple component signals.

[0018] Furthermore, specifically, the step S2 uses a stacking method to stack the vibration acceleration signals x1, x2, ..., x c Constructed as a multi-channel third-order tensor signal

[0019] Further, specifically, in step S4, the expansion rule includes: assuming that the third-order tensor Slice the third-order tensor positively, the first slice is a1, the second slice is a2, and the nth slice is an; all slices a1, ..., an are spliced from left to right in the order of the slices, and expanded to form an I×J matrix A. The third-order tensor The expanded formula is as follows:

[0020]

[0021] Further, specifically, in step S4, the folding rule includes: cutting the matrix S into matrices with the number of rows and columns being I1 and I2 respectively; and then folding them into a third-order tensor in the order from left to right and from front to back. The matrix S folding formula is as follows:

[0022]

[0023]

[0024] Fold the matrix V into a fourth-order tensor The folding formula of the matrix V is as follows:

[0025]

[0026] Among them, i=i1+(i2-1)I3, j=i4+(i3-1)I3, i1=i4={1,2,...,Ι3}, i2=i3={1,2,...,Ι2}.

[0027] Further, specifically, in step S5, the multi-channel third-order tensor signal The calculation formula for reconstruction is:

[0028]

[0029] Among them, the tensor The pseudo-diagonal elements of are called singular values.

[0030] Furthermore, specifically, two singular values corresponding to the i-th row = {1, 2, ..., min (I1, I2)} are selected from the two pseudo-diagonals of the slice of the tensor S, and the singular values of other positions except the i-th position are assigned to 0. The reset tensor is recorded as Calculate the component signal corresponding to the i-th singular value as:

[0031]

[0032] Further, specifically, in step S1, the vibration acceleration signals x1, x2, ..., x c The data is collected by M acceleration vibration sensors installed on the bearing, with a collection time of t and the number of data points being N.

[0033] A second-type third-order tensor decomposition device, wherein the second-type third-order tensor decomposition device adopts the second-type third-order tensor decomposition method described above, and the second-type third-order tensor decomposition device includes:

[0034] Acquisition module: Acquire vibration acceleration signals x1, x2, ..., x c ;

[0035] The first building block: according to the vibration acceleration signals x1, x2, ..., x c Constructing multi-channel third-order tensor signals Among them, I1 is a multi-channel third-order tensor The first order, I2 is a multi-channel third-order tensor The second order and I3 are multi-channel third-order tensors The third stage;

[0036] The second building block: building a computational model for high-order tensor second-order multiplication and transposition;

[0037] Singular value decomposition module: performs second-type tensor singular value decomposition on the multi-channel third-order tensor signal X to obtain tensors U, S and V;

[0038] Component signal calculation module: According to the calculation model of the second-order multiplication and transposition of the high-order tensor and the folded tensor U, the tensor S and the tensor V, the third-order tensor signal is calculated. Singular value decomposition is performed to obtain multiple component signals.

[0039] Furthermore, specifically, it further includes: a collection module, the collection module includes a plurality of acceleration vibration sensors, and the plurality of acceleration vibration sensors are installed on the bearing.

[0040] A bearing fault detection system includes: the bearing fault detection system adopts the second type third-order tensor decomposition device as described above.

[0041] The beneficial effect of the present invention is that a second-type third-order tensor decomposition method of the present invention constructs a third-order tensor for the acquired vibration acceleration signal, and performs singular value decomposition on the constructed third-order tensor by constructing a high-order tensor second-type multiplication and transposition calculation model to obtain component signals, thereby achieving effective decomposition and feature extraction of the signal, ensuring the completeness of the rolling bearing state information, facilitating signal analysis and diagnosis, and providing basic support for electromechanical equipment such as rolling bearing state monitoring and remaining life prediction, and the calculated component signals have good accuracy, further improving the rolling bearing state monitoring accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The present invention will be further described below with reference to the accompanying drawings and examples.

[0043] Figure 1 This is a flow chart of the method of embodiment 1 of the present invention.

[0044] Figure 2 This is a time domain waveform diagram of the vibration acceleration signal of the bearing in the horizontal direction in a specific embodiment of the present invention.

[0045] Figure 3 This is a schematic diagram of judging the first component signal after tensor singular value decomposition in a specific embodiment of the present invention.

[0046] Figure 4 It is a structural diagram of embodiment 2 of the present invention.

[0047] In the figure, 300 is an acquisition module; 301 is an acquisition module; 302 is a first construction module; 303 is a second construction module; 304 is a singular value decomposition module; and 305 is a component signal calculation module. DETAILED DESCRIPTION

[0048] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0049] Example 1

[0050] like Figure 1 FIG. 1 is a first embodiment of the present invention, a second-type third-order tensor decomposition method, comprising the following steps:

[0051] Step S1: Obtain vibration acceleration signals x1, x2, ..., x c .

[0052] Vibration acceleration signal x1, x2, ..., xc The data is collected by M acceleration vibration sensors installed on the bearing, with a collection time of t and a number of data points of N. Specifically, acceleration vibration sensors can be installed in the horizontal direction, vertical direction, axial direction or other positions of the bearing system to be diagnosed.

[0053] Take SKF 6308 bearing as an example for calculation. The number of balls in SKF 6308 bearing is N. b =8, shaft speed ω s =42rad / s, an acceleration vibration sensor is installed in the horizontal and vertical directions of the SKF 6308 bearing respectively, the acquisition time is t=2s, the number of data points is N=131072, and the horizontal vibration acceleration signal x1 and the vertical vibration acceleration signal x2 of the bearing to be diagnosed are obtained; based on the obtained vibration acceleration signal x1, a time domain waveform diagram is drawn, as shown Figure 2 shown.

[0054] Step S2: According to the vibration acceleration signals x1, x2, ..., x c Constructing multi-channel third-order tensor signals Among them, I1 is a multi-channel third-order tensor The first order, I2 is a multi-channel third-order tensor The second order and I3 are multi-channel third-order tensors The third order; specifically, the stacking method is used to transform the vibration acceleration signals x1, x2, ..., x c Constructed as a multi-channel third-order tensor signal

[0055] In a specific embodiment of the present invention, the horizontal vibration acceleration signal x1 and the vertical vibration acceleration signal x2 are constructed into a multi-channel third-order tensor signal X using a stacking method. 3078×42×2 , the first-order dimension I1 is 3078, the second-order dimension I2 is 42, and the third-order dimension I3 is 2. It should be noted that the first-order dimension I1 is related to the number of columns, the second-order dimension I2 is related to the number of rows, and the third-order dimension I3 is related to the number of installed acceleration vibration sensors. For example, the third-order tensor signal X 3078×42×2 The vibration acceleration signal x1, with its 131,072 data points N, is truncated into a matrix with 3,078 columns and 42 rows. The excess 1,796 (131,072 - 3,078 * 42) points are discarded. Similarly, the vibration acceleration signal x2 is processed in the same way as the acceleration signal x1: it is also a matrix with 3,078 columns and 42 rows.

[0056] Step S3: Construct a computational model for the second-order multiplication and transposition of high-order tensors. Assume that the m-order tensor And n-order tensors Construct the second-type multiplication calculation formula based on the preset m-order tensor A and n-order tensor B, and obtain the new tensor C. The second-type multiplication calculation formula is:

[0057]

[0058] Among them, the dimension of tensor A is equal to the dimension of tensor B, that is, I m+1-r =J r , r=1,2,…,s, and K-order tensor s is the number of dimensions of the last s orders of the m-order tensor A that are equal to the number of dimensions of the first s orders of the n-order tensor B. For example, when the m-order tensor Only the last dimension and tensor The first-order dimensions of tensor C are equal, then s=1; k represents the nth-order dimension of tensor C, I represents the nth-order dimension of tensor A, and J represents the nth-order dimension of tensor B; k is determined by the dimensions of tensor A and tensor B. Assuming that the dimension of tensor A is 3rd order and the dimension of tensor B is 3rd order, then s=2, 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, k2 is the second-order dimension of tensor C, determined by the third-order dimension J3 of tensor B.

[0059] In addition, when K=k1 represents the first-order dimension of tensor C, when the order of tensor A is 1, it corresponds to vector multiplication, and when the order of tensor A is 2, it corresponds to matrix multiplication. The second-order multiplication of high-order tensors can be degenerated into standard vector and matrix multiplication;

[0060] Get m-order tensor Transpose:

[0061]

[0062] Step S4: performing a second-type tensor singular value decomposition on the multi-channel third-order tensor signal X to obtain tensors U, S, and V. The steps of obtaining the tensors U, S, and V include:

[0063] Based on the expansion rule, the third-order tensor signal Expand to obtain matrix X, perform singular value decomposition on matrix X to obtain matrix U, matrix S and matrix V; Specifically, based on the expansion rule, the third-order tensor signal X 3078×42×2 Expand and obtain the matrix X 3078×84 ; Then X 3078×84 Perform singular value decomposition to obtain the matrix U 3078×3078 , matrix S 3078×84 and matrix V 84×84 .

[0064] Based on the folding rule, the matrix U, the matrix S and the matrix V are reconstructed respectively to obtain the tensor Tensor and tensors Specifically, the folding rule is used to transform the matrix U 3078×3078 、S 3078×84 , and V 84×84 Folded into tensor U, tensor S, and tensor V.

[0065] It should be noted that in step S4, the expansion rule includes: assuming that the third-order tensor Slice the third-order tensor positively, the first slice is a1, the second slice is a2, and the nth slice is an; all slices a1, ..., an are spliced from left to right in the order of slicing, and expanded to form an I×J matrix A, the third-order tensor The expanded formula is as follows:

[0066]

[0067] In a specific embodiment, the multi-channel third-order tensor signal is directly The matrix X is formed by the expansion rule.

[0068] In step S4, the folding rule includes: cutting the matrix S into matrices with the number of rows and columns being I1 and I2 respectively; and then folding them into a third-order tensor in the order from left to right and from front to back. The matrix S folding formula is as follows:

[0069]

[0070] Among them, i=i1, j=i3+(i2-1)Ι3, i1={1,2,...,Ι1}, i2={1,2,...,Ι2}, i3={1,2,...Ι3};

[0071] Fold the matrix V into a fourth-order tensor The folding formula of matrix V is as follows:

[0072]

[0073] Among them, i=i1+(i2-1)I3, j=i4+(i3-1)I3, i1=i4={1,2,...,Ι3}, i2=i3={1,2,...,Ι2}.

[0074] It should be noted that since the matrix U is still a matrix after reconstruction, there is no need to fold the matrix U here.

[0075] Step S5: According to the second-order multiplication of the high-order tensor in step S3 and the calculation model of the transposition and the folded tensor Tensor and tensors For third-order tensor signals Reconstruct and obtain multiple component signals.

[0076] Among them, the third-order tensor signal The calculation formula for reconstruction is:

[0077]

[0078] Among them, the tensor The pseudo-diagonal elements of are called singular values.

[0079] Select the two singular values corresponding to the i-th row = {1, 2, ..., min (I1, I2)} in the two pseudo-diagonals of the slice of tensor S, and assign the singular values of other positions except the i-th position to 0. The reset tensor is recorded as Calculate the component signal corresponding to the i-th singular value as:

[0080]

[0081] The component signal is selected by formula 7, and all the calculated component signals are analyzed and diagnosed to detect whether there is a fault characteristic frequency in the component signal. The characteristics of the component signal can also be diagnosed. For example, if the component signal has a double impact feature, the fault width can be diagnosed. For example, by detecting whether there is a fault characteristic frequency in the component signal, the component signal corresponding to the first singular value is calculated by formula (7). The component signal is as follows Figure 3 As shown, analyze the component signal corresponding to the first singular value Six impact features with a period of To can be observed. When the component signal To corresponding to the first singular value is the inverse of the bearing fault characteristic frequency, it can be determined that the bearing has a fault.

[0082] A second-type third-order tensor decomposition method of the present invention constructs a third-order tensor of the acquired vibration acceleration signal, and performs singular value decomposition on the constructed third-order tensor by constructing a second-type multiplication and transposition calculation model of a high-order tensor to obtain component signals, thereby achieving effective decomposition and feature extraction of the signal, ensuring the completeness of the rolling bearing state information, facilitating signal analysis and diagnosis, and providing basic support for electromechanical equipment such as rolling bearing state monitoring and remaining life prediction. The calculated component signals are accurate, further improving the rolling bearing state monitoring accuracy.

[0083] Example 2

[0084] like Figure 2FIG. 1 is a second embodiment of the present invention. Based on the same inventive concept as the second-type third-order tensor decomposition method in the first embodiment, the present invention further provides a second-type third-order tensor decomposition device, including:

[0085] Acquisition module 301: Acquisition of vibration acceleration signals x1, x2, ..., x c ;

[0086] The first building block 302: according to the vibration acceleration signals x1, x2, ..., x c Constructing multi-channel third-order tensor signals Among them, I1 is a multi-channel third-order tensor The first order, I2 is a multi-channel third-order tensor The second order and I3 are multi-channel third-order tensors The third stage;

[0087] Second construction module 303: constructing a computational model for high-order tensor second-type multiplication and transposition;

[0088] Singular value decomposition module 304: multi-channel third-order tensor signal Perform the second type of tensor singular value decomposition to obtain tensors U, S and V. Specifically, based on the expansion rule, the third-order tensor signal Expand to obtain matrix X, perform singular value decomposition on matrix X to obtain matrix U, matrix S and matrix V; reconstruct matrix U, matrix S and matrix V based on the folding rule to obtain tensors U, tensor S and tensor V;

[0089] Component signal calculation module 305: According to the calculation model of the second-order multiplication and transposition of the high-order tensor and the folded tensors U, S and V, the third-order tensor signal is calculated. Reconstruct and obtain multiple component signals.

[0090] The acquisition module 300 includes a plurality of acceleration vibration sensors, and the plurality of acceleration vibration sensors are installed on the bearing.

[0091] A second-class third-order tensor decomposition device realizes signal processing and application diagnosis by working in various functional modules. The implementation is simple and the division of labor is clear, which further improves the signal processing speed and has good practicality.

[0092] The foregoing Figure 1The various variations and specific examples of a second-type third-order tensor decomposition method in Example 1 are also applicable to a second-type third-order tensor decomposition device in this embodiment. Through the above detailed description of a second-type third-order tensor decomposition method, those skilled in the art can clearly know the implementation method of a second-type third-order tensor decomposition device in this embodiment, and it can produce the same technical effect as the second-type third-order tensor decomposition method. Therefore, for the sake of brevity of the specification, it will not be described in detail here.

[0093] Example 3

[0094] Based on the same inventive concept as the second-type third-order tensor decomposition device in the aforementioned embodiment 2, the present invention also provides a bearing fault detection system, including: the bearing fault detection system adopts the second-type third-order tensor decomposition device as above.

[0095] With the above-described preferred embodiments of the present invention as a guide, and with reference to the above description, relevant personnel are fully capable of making various changes and modifications without departing from the technical scope of this invention. The technical scope of this invention is not limited to the contents of the specification and must be determined according to the scope of the claims.

Claims

1. A second-type third-order tensor decomposition method, characterized in that: The steps include: Step S1: Obtain multi-channel vibration acceleration signals x1, x2, ..., x c ; Step S2: According to the vibration acceleration signals x1, x2, ..., x c Constructing multi-channel third-order tensor signals Wherein, I1 is the multi-channel third-order tensor The first order, I2 is the multi-channel third-order tensor The second order and I3 are the multi-channel third-order tensors The third stage; Step S3: Construct a computational model for the second-order multiplication and transposition of high-order tensors. Assume that the m-order tensor And n-order tensors A second type of multiplication calculation formula is constructed according to the preset m-order tensor A and the n-order tensor B, and a new tensor C is obtained. The second type of multiplication calculation formula is: Among them, I m+1-r =J r , r = 1, 2, ..., s, and the new tensor C is a K-order tensor s is the number of all dimensions of the last s orders of the m-order tensor A that are equal to all dimensions of the first s orders of the n-order tensor B, k represents the nth order dimension of the tensor C, I represents the nth order dimension of the tensor A, and J represents the nth order dimension of the tensor B; Get the m-order tensor Transpose: Step S4: the multi-channel third-order tensor signal Performing a second type of tensor singular value decomposition to obtain tensors U, S, and V, wherein the steps of obtaining the tensors U, S, and V include: Based on the expansion rule, the third-order tensor signal Expand to obtain matrix X, perform singular value decomposition on matrix X to obtain matrix U, matrix S and matrix V; Based on the folding rule, the matrix U, the matrix S and the matrix V are reconstructed respectively to obtain the tensor The tensor and the tensor Step S5: According to the calculation model of the second type of multiplication and transposition of the high-order tensor and the tensor The tensor and the tensor For the third-order tensor signal Reconstruct and obtain multiple component signals.

2. The second type of third-order tensor decomposition method according to claim 1, characterized in that: The step S2 uses a stacking method to stack the vibration acceleration signals x1, x2, ..., x c Constructed as a multi-channel third-order tensor signal 3. The second type of third-order tensor decomposition method according to claim 1, characterized in that: In step S4, the expansion rule includes: assuming that the third-order tensor Slice the third-order tensor positively, the first slice is a1, the second slice is a2, and the nth slice is an; all slices a1, ..., an are spliced from left to right in the order of slicing, and expanded to form an I×J matrix A. The third-order tensor The expanded formula is as follows:

4. The second type of third-order tensor decomposition method according to claim 1, characterized in that: In step S4, the folding rule includes: cutting the matrix S into matrices with the number of rows and columns being I1 and I2 respectively; and then folding them into a third-order tensor in the order from left to right and from front to back. The matrix S folding formula is as follows: Among them, i=i1, j=i3+(i2-1)Ι3, i1={1,2,...,Ι1}, i2={1,2,...,Ι2}, i3={1,2,...Ι3}; Fold the matrix V into a fourth-order tensor The folding formula of the matrix V is as follows: Among them, i=i1+(i2-1)I3, j=i4+(i3-1)I3, i1=i4={1,2,...,Ι3}, i2=i3={1,2,...,Ι2}.

5. The second type of third-order tensor decomposition method according to claim 1, characterized in that: In step S5, the multi-channel third-order tensor signal The calculation formula for reconstruction is: Among them, the tensor The pseudo-diagonal elements of are called singular values.

6. The second type of third-order tensor decomposition method according to claim 5, characterized in that: Select the two singular values corresponding to the i-th row = {1, 2, ..., min (I1, I2)} in the two pseudo-diagonals of the slice of the tensor S, assign the singular values of other positions except the i-th position to 0, and the reset tensor is recorded as Calculate the component signal corresponding to the i-th singular value as:

7. The second type of third-order tensor decomposition method according to claim 1, characterized in that: In step S1, the vibration acceleration signals x1, x2, ..., x c The data is collected by M acceleration vibration sensors installed on the bearing, with a collection time of t and the number of data points being N.

8. A second type third-order tensor decomposition device, characterized in that: The second type of third-order tensor decomposition device adopts the second type of third-order tensor decomposition method according to any one of claims 1 to 7, and the second type of third-order tensor decomposition device includes: Acquisition module (301): Acquisition of vibration acceleration signals x1, x2, ..., x c ; The first building block (302): according to the vibration acceleration signals x1, x2, ..., x c Constructing multi-channel third-order tensor signals Wherein, I1 is the multi-channel third-order tensor The first order, I2 is the multi-channel third-order tensor The second order and I3 are the multi-channel third-order tensors The third stage; The second construction module (303): constructs a computational model for high-order tensor second-order multiplication and transposition; Singular value decomposition module (304): for the multi-channel third-order tensor signal Perform the second type of tensor singular value decomposition to obtain tensors U, S, and V; Component signal calculation module (305): performs the third-order tensor signal calculation based on the second-order multiplication and transposition calculation model of the high-order tensor and the folded tensor U, the tensor S and the tensor V. Singular value decomposition is performed to obtain multiple component signals.

9. The second type third-order tensor decomposition device according to claim 8, characterized in that: Also includes: A collection module (300) includes a plurality of acceleration vibration sensors, and the plurality of acceleration vibration sensors are mounted on a bearing.

10. A bearing fault detection system, characterized in that: include: The bearing fault detection system adopts the second type third-order tensor decomposition device as described in claim 8 or 9.

Images