Asymmetric sparse quaternion hydraulic pump service degradation big data remote restoration method

Abstract

The invention is suitable for the field of fault diagnosis and intelligent operation and maintenance of complex hydraulic equipment, and provides an asymmetric sparse quaternion hydraulic pump service degradation big data remote recovery method. Solving the constructed asymmetric sparse low-rank matrix model by using an alternating direction multiplier method and a fast smooth soft threshold method to obtain an observation signal sparse coefficient matrix of each measurement point; constructing a sparse quaternion digital learning model, and calculating to obtain an observation reconstruction signal of each measuring point by using a quaternion singular value decomposition method; adjusting parameters of the asymmetric sparse low-rank matrix model and the sparse quaternary digital learning model in real time according to the reconstruction error; and detecting whether weak fault information of the reconstructed multi-channel monitoring data is lost or not by using a Hilbert envelope spectrum. The multi-channel degradation data of the hydraulic pump can be reconstructed at a time, important degradation weak detail information can be stored, the reconstruction precision is high, the algorithm operation speed is high, and the industrial online monitoring requirement can be met.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault diagnosis and intelligent operation and maintenance of complex hydraulic equipment, and in particular to a remote restoration method for service degradation big data of an asymmetric sparse quaternion hydraulic pump. Background Art

[0002] At present, technologies such as the Internet of Things and mobile Internet are developing rapidly in the fields of high-end CNC machine tools, robots, new energy vehicles and engineering machinery. With the help of technologies such as the Internet of Things and mobile Internet, real-time monitoring big data of equipment (such as temperature, acceleration, acoustic emission, infrared images) can provide data support for the efficient operation and reliability monitoring of industrial equipment and assembly lines, as well as revealing the causes and evolution laws of their service degradation characteristics.

[0003] However, in the process of remote data transmission and reconstruction, due to external factors such as network failures and terminal storage device failures, traditional data storage and processing methods face huge challenges, such as partial data transmission interruption and data loss. At the same time, in the process of data reconstruction, due to algorithm defects, data may be corrupted and cannot be accurately restored. In the field of complex hydraulic equipment fault diagnosis and intelligent operation and maintenance, traditional data storage and processing methods still have the following bottleneck problems in the process of remote data transmission and reconstruction: (1) Low reconstruction accuracy or large reconstruction error: Due to the complex operating environment of complex hydraulic equipment, the monitoring data collected by sensors is easily affected by uncertain factors such as changes in external working load, ambient temperature and humidity, etc., which may cause large deviations in the reconstructed signal / data, ultimately leading to misjudgment of the state; (2) Important degradation and weak details are easily lost: In the early return operation stage of complex hydraulic equipment and its key parts, the monitored faults (such as blade damage and bearing cracking) have low amplitude, weak fault propagation intensity, and weak energy. In the process of remote data reconstruction, if the data reconstruction algorithm is inappropriate, it is easy to cause the loss of weak fault information and the waveform distortion of the reconstructed signal. (3) Reconstruction takes a long time: Complex hydraulic equipment usually runs continuously, requiring online data reconstruction to be completed in a very short time so that fault diagnosis results can be fed back in time and strategies can be formulated; otherwise, the best time to diagnose the fault may be missed; traditional reconstruction methods (such as compressed sensing methods) take a long time to reconstruct and cannot meet the needs of real-time and rapid response of industrial equipment. Summary of the invention

[0004] In order to overcome the problems of low reconstruction accuracy, easy loss of important degradation and weak detail information in traditional technologies, and long reconstruction time that cannot meet the industrial needs of real-time response and rapid abnormal processing capabilities for online monitoring of hydraulic equipment, the present invention provides a remote restoration method for asymmetric sparse quaternion hydraulic pump service degradation big data. This method can realize online status monitoring and fault diagnosis of complex hydraulic pump equipment, and provide a theoretical basis and technical support for the intelligent operation and maintenance process of hydraulic pump equipment in extreme working conditions.

[0005] To achieve the above object, the present invention provides the following technical solutions: The remote restoration method of asymmetric sparse quaternion hydraulic pump service degradation big data includes the following steps: Pick up the vibration acceleration signal of each measuring point of the equipment to be tested, and obtain the observation signal of each measuring point; An asymmetric sparse low-rank matrix model is constructed, and the alternating direction multiplier method and the fast smoothing soft threshold method are used to solve the constructed asymmetric sparse low-rank matrix model to obtain the sparse coefficient matrix of the observation signal of each measuring point; Construct a sparse quaternion dictionary learning model, use the quaternion singular value decomposition method and dictionary update layer by layer to obtain the observation reconstruction signal of each measuring point, and realize the reconstruction of multi-channel monitoring data; The original observation signals of each measuring point are compared with the reconstructed multi-channel monitoring data to calculate the reconstruction error. According to the reconstruction error, the parameters of the asymmetric sparse low-rank matrix model and the sparse quaternion digital dictionary learning model are adjusted in real time until the error threshold range is met; The fault characteristic frequency is extracted from the reconstructed multi-channel monitoring data using the Hilbert envelope spectrum to detect whether the weak fault information in the reconstructed multi-channel monitoring data is lost; Based on the online reconstruction data of each channel, remote fault diagnosis and intelligent operation and maintenance are carried out on the equipment under test.

[0006] Furthermore, in the step of picking up the vibration acceleration signal of each measuring point of the equipment to be tested and obtaining the observation signal of each measuring point, the online service data of the equipment to be tested collected by the acceleration sensor is expressed as: Y=AX+NOISE(1); In formula (1), Y is the multi-channel observation data matrix, Y = [y1, y2, y3, …, y n ] T , A is the model degradation matrix, X is the multi-channel sparse coefficient to be estimated, X=[x1,x2,x3,…,x n ] T ,NOISE is the external background noise.

[0007] Furthermore, the step of constructing an asymmetric sparse low-rank matrix model, solving the constructed asymmetric sparse low-rank matrix model using an alternating direction multiplier method and a fast smoothing soft threshold method to obtain a sparse coefficient matrix of the observation signal of each measuring point includes: Using the multi-channel observation data matrix Y, an asymmetric sparse low-rank matrix model is constructed, which is expressed as: (2); In formula (2): F(X) is the cost function; is the Frobenius norm; λ and ξ both represent regularization parameters, where λ>0, ξ>0; represents the singular value penalty function of the smoothing parameterization; represents an asymmetric smoothing parameterized penalty function, the coefficient a is the penalty function waveform control factor, ε is the small data smoothing constant, ε=10 -6 ;M i represents the i-order difference matrix; σ(·) represents the singular value transformation of the signal; i represents the difference order; j represents the j-th row number of the matrix; k represents the k-th column number of the matrix; p represents the number of terms of the smooth parameterized singular value penalty function or the asymmetric smooth parameterized penalty function; m represents the number of rows of the matrix; n represents the number of columns of the matrix; The construction of an asymmetric sparse low-rank matrix model can be transformed into: (3); In formula (3): ; ; D1 represents the first-order difference operator; α represents the penalty function waveform control factor; G1(w) and G2(w) represent the abbreviated form of the formula; Construct an asymmetric sparse low-rank matrix model with constraints, expressed as: (4); In formula (4): ; ; Among them, λ1 represents the regularization parameter; g1(U) and g2(V) represent the abbreviated form of the formula; U and V represent the alternative form of the matrix X; The augmented Lagrangian function model is constructed and expressed as: (5); In formula (5), μ is the Lagrange coefficient; U and V represent the alternative forms of the matrix X; The alternating direction multiplier method and the fast smoothing soft threshold method are used to solve the constructed augmented Lagrangian function model, and the sparse coefficient matrix of the observation signal of each measuring point is obtained, which is expressed as: (6); In formula (6), SFSS (·,·,·,·) is a fast smoothing soft threshold method, specifically: ,in: , ; x represents the preprocessing vector; ξ represents the regularization parameter; μ represents the Lagrange coefficient; ε represents the small data smoothing constant, ε=10 -6 .

[0008] Furthermore, the step of constructing a sparse quaternion dictionary learning model, using the quaternion singular value decomposition method and dictionary update layer by layer to obtain the observation reconstruction signal of each measuring point and realize the reconstruction of multi-channel monitoring data includes: Construct a sparse quaternion dictionary learning model, expressed as: Y=0+y x i+y y i+y z k (7); In formula (7), the training dictionary is: D = D0 + D x i+D y j+D z k,D∈H n×K , the sparse coefficient can be expressed as: X=x=x0+x x i+x y j+x z k,x∈H K ; Among them: D x is the training dictionary of x channel; D y is the training dictionary of the y channel; D z is the training dictionary of the z channel; i, j, k are imaginary units, i 2 =-1,j 2 =-1; k 2 =-1; i 2 =j 2 =k 2 =ijk=-1; The above model can be transformed into: (8); In formula (8): , ; Represents the transpose of the x channel signal; Represents the transpose of the y channel signal; represents the transpose of the z channel signal; at the same time: (9); Among them, y x ,y y ,y z They are three channel signals, which can be expressed by the product of quaternion dictionary and sparse coefficient; According to the initialization dictionary matrix D = {d i , 1≤i≤K}, calculate: (10); Get the multi-channel error matrix: ; Among them, d j represents the jth dictionary atom in the dictionary matrix; d k represents the jth dictionary atom in the dictionary matrix, x k represents the kth element in vector x; x j represents the jth element in vector x; The quaternion singular value decomposition method is used to decompose the multi-channel error matrix and obtain: (11); In formula (11), U is an n×n dimensional matrix spanned by n eigenvectors, Λ is an n×n dimensional matrix with n eigenvalues ​​as diagonals, V is the unitary matrix of U; H represents the Hermitian conjugate transpose of the vector; A vector representing the error matrix; Update the vector dk of the U matrix column by column to dk update , take the first column of matrix U as the updated atom dk update , and update the sparse coefficient corresponding to the atom; Multiply the first column of matrix V with the semi-positive definite diagonal matrix Λ[1,1] as the updated sparse coefficient until the algorithm reaches the number of iterations and outputs the multi-channel reconstructed signal y c =D c x, where D c is the updated dictionary matrix D c ={d i ,1≤i≤K}, c is the number of channels, and K is the number of sampling points.

[0009] Furthermore, the step of comparing the original observation signals of each measuring point with the reconstructed multi-channel monitoring data, calculating the reconstruction error, and adjusting the parameters of the asymmetric sparse low-rank matrix model and the sparse quaternion digital dictionary learning model in real time according to the reconstruction error until the error threshold range is met includes: Calculate the reconstruction error between the original observation signal of each measuring point and the reconstructed multi-channel monitoring data, expressed as: Error=Yyc (12); Where Y is the multi-channel observation data matrix, y c Multi-channel reconstruction signal; Set the error threshold to a constant value of 10 -2 , adjust the parameters of the asymmetric sparse low-rank matrix model and the sparse quaternion dictionary learning model in real time until the error threshold range is met and the iteration is stopped.

[0010] Compared with the prior art, the asymmetric sparse quaternion hydraulic pump service degradation big data remote restoration method of the present invention has the following beneficial effects: first, the method proposed in the present invention is based on the quaternion model, taking into account the spatiotemporal coupling characteristics of the multi-channel degradation data of the hydraulic pump, and can predict multi-channel data at one time instead of predicting channel by channel; Second, compared with the traditional signal reconstruction method, the method proposed by the present invention is not affected by uncertain factors such as changes in external working load, changes in ambient temperature and humidity, and has high reconstruction accuracy; Third, compared with traditional signal reconstruction methods, the method proposed in the present invention can effectively preserve important degradation and weak detail information, has low computational complexity, and has a fast algorithm running speed, which can meet the needs of real-time and rapid response of industrial equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention.

[0012] Figure 1 It is a logic flow chart of the remote restoration method of the asymmetric sparse quaternion hydraulic pump service degradation big data of the present invention; Figure 2 It is a schematic diagram of the implementation process of the asymmetric sparse quaternion hydraulic pump service degradation big data remote restoration method of the present invention; Figure 3 The three-channel vibration acceleration signal of the centrifugal pump impeller damage fault according to the embodiment of the present invention; Figure 4 It is multi-channel sparse coefficient matrix data based on an asymmetric sparse low-rank matrix method according to an embodiment of the present invention; FIG5( a ) is a first channel reconstructed signal (upper) and an amplified signal (lower) thereof obtained according to an embodiment of the present invention; FIG5( b ) is a second channel reconstructed signal (upper) and an amplified signal (lower) thereof obtained according to an embodiment of the present invention; FIG5( c ) is a third channel reconstructed signal (upper) and an amplified signal (lower) thereof obtained according to an embodiment of the present invention; FIG6 (a) is a first channel reconstruction error signal obtained according to an embodiment of the present invention; FIG6( b ) is a second channel reconstruction error signal obtained according to an embodiment of the present invention; FIG6 (c) is a third channel reconstruction error signal obtained according to an embodiment of the present invention; Figure 7 1 and 2 . The original envelope spectrum of the first channel (upper) and the envelope spectrum of the reconstructed signal of the first channel (lower) according to an embodiment of the present invention. DETAILED DESCRIPTION

[0013] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and 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.

[0014] In view of the problems of low reconstruction accuracy, easy loss of important degradation and weak detail information in existing related technologies, and long reconstruction time that cannot meet the industrial needs of real-time response and rapid abnormal processing capabilities of online monitoring of hydraulic equipment, the present invention proposes an asymmetric sparse quaternion hydraulic pump service degradation big data remote restoration method, which uses Ethernet and sensor fusion network technology to pick up the vibration acceleration signals of equipment measuring points in long-term service; constructs an asymmetric sparse low-rank matrix model, and uses the alternating direction multiplier method and the fast smoothing soft threshold method to solve the constructed asymmetric sparse low-rank matrix model to obtain the sparse coefficient matrix of the observation signal of each measuring point; constructs a sparse quaternion digital dictionary learning model, and uses the quaternion singular value decomposition method to calculate the observation reconstruction signal of each measuring point; according to the reconstruction error, the parameters of the asymmetric sparse low-rank matrix model and the sparse quaternion digital dictionary learning model are adjusted in real time; the Hilbert envelope spectrum is used to detect whether the weak fault information of the reconstructed multi-channel monitoring data is lost; finally, fault diagnosis and intelligent operation and maintenance of the hydraulic pump equipment are performed in the monitoring center.

[0015] The present invention takes into account the spatiotemporal coupling characteristics of multi-channel data of equipment based on the quaternion model, can reconstruct the multi-channel degradation data of the hydraulic pump at one time, and is not affected by uncertain factors such as external working conditions. It can preserve important degradation weak detail information, has high reconstruction accuracy, and has a fast algorithm running speed, which can meet the needs of industrial online monitoring.

[0016] The specific implementation of the asymmetric sparse quaternion hydraulic pump service degradation big data remote restoration method of the present invention is described in detail below in conjunction with specific embodiments.

[0017] Please refer to Figure 1 and Figure 2 The asymmetric sparse quaternion hydraulic pump service degradation big data remote restoration method provided by the present invention comprises the following steps: Step S1: picking up the vibration acceleration signal of each measuring point of the equipment to be tested to obtain the observation signal of each measuring point; In step S1, the present invention uses Ethernet and sensor fusion network technology to pick up the vibration acceleration signal of the long-term service of the measuring point of the equipment to be tested, and the online service data of the equipment to be tested collected by the acceleration sensor is expressed as: Y=AX+NOISE(1); In formula (1), Y is the multi-channel observation data matrix, Y = [y1, y2, y3, …, y n ] T , A is the model degradation matrix, X is the multi-channel sparse coefficient to be estimated, X=[x1,x2,x3,…,x n ] T ,NOISE is the external background noise; Optionally, the equipment to be tested of the present invention is a hydraulic pump equipment and its key important components, such as a pump body, bearings, gears and other components; Step S2: construct an asymmetric sparse low-rank matrix model, and use the alternating direction multiplier method and the fast smoothing soft threshold method to solve the constructed asymmetric sparse low-rank matrix model to obtain the sparse coefficient matrix of the observation signal of each measuring point; Step S3: construct a sparse quaternion dictionary learning model, use the quaternion singular value decomposition method and dictionary update layer by layer to obtain the observation reconstruction signal of each measuring point, and realize the reconstruction of multi-channel monitoring data; Step S4: Compare the original observation signals of each measuring point with the reconstructed multi-channel monitoring data, calculate the reconstruction error, and adjust the parameters of the asymmetric sparse low-rank matrix model and the sparse quaternion digital dictionary learning model in real time according to the reconstruction error until the error threshold range is met; Step S5: extracting fault characteristic frequencies from the reconstructed multi-channel monitoring data using the Hilbert envelope spectrum to detect whether weak fault information in the reconstructed multi-channel monitoring data is lost; Step S6: Based on the online reconstruction data of each channel, remote fault diagnosis and intelligent operation and maintenance are performed on the equipment under test.

[0018] Furthermore, step S2 provided by the present invention includes the following contents: First, the present invention uses the multi-channel observation data matrix Y to construct an asymmetric sparse low-rank matrix model, which is expressed as: (2); In formula (2): F(X) is the cost function; is the Frobenius norm; λ and ξ both represent regularization parameters, where λ>0, ξ>0; represents the singular value penalty function of the smoothing parameterization; represents an asymmetric smoothing parameterized penalty function, the coefficient a is the penalty function waveform control factor, ε is the small data smoothing constant, ε=10 -6;M i represents the i-order difference matrix; σ(·) represents the singular value transformation of the signal; i represents the difference order; j represents the j-th row number of the matrix; k represents the k-th column number of the matrix; p represents the number of terms of the smooth parameterized singular value penalty function or the asymmetric smooth parameterized penalty function; m represents the number of rows of the matrix; n represents the number of columns of the matrix; Furthermore, the present invention constructs an asymmetric sparse low-rank matrix model as follows: (3); In formula (3): ; ; D1 represents the first-order difference operator, that is ; α represents the penalty function waveform control factor; G1(w) and G2(w) represent the abbreviated form of the formula; Construct an asymmetric sparse low-rank matrix model with constraints, expressed as: (4); In formula (4): ; ; Among them, λ1 represents the regularization parameter; g1(U) and g2(V) represent the abbreviated form of the formula; U and V represent the alternative form of the matrix X, which is used for the subsequent use of the Lagrangian function; The augmented Lagrangian function model is constructed and expressed as: (5); In formula (5), μ is the Lagrange coefficient; U and V represent the alternative forms of matrix X; Z represents an auxiliary matrix; Furthermore, the present invention uses the alternating direction multiplier method and the fast smoothing soft threshold method to solve the constructed augmented Lagrangian function model to obtain the sparse coefficient matrix of the observation signal of each measuring point, which is expressed as: (6); In formula (6), SFSS (·,·,·,·) is a fast smoothing soft threshold method, specifically: , in: , ; where x represents the preprocessing vector; ξ represents the regularization parameter; μ represents the Lagrange coefficient; ε represents the small data smoothing constant, ε=10 -6 .

[0019] Furthermore, the step of constructing a sparse quaternion dictionary learning model, using the quaternion singular value decomposition method and dictionary update layer by layer to obtain the observation reconstruction signal of each measuring point and realize the reconstruction of multi-channel monitoring data includes: Furthermore, the present invention constructs a sparse quaternion dictionary learning model, which is expressed as: Y=0+y x i+y y i+y z k (7); In formula (7), the training dictionary is: D = D0 + D x i+D y j+D z k,D∈H n×K , the sparse coefficient can be expressed as: X=x=x0+x x i+x y j+x z k,x∈H K Among them, D x is the training dictionary of x channel; D y is the training dictionary of the y channel; D z is the training dictionary of the z channel; i, j, k are imaginary units, i 2 =-1,j 2 =-1; k 2 =-1; i 2 =j 2 =k 2 =ijk=-1; The above model can be transformed into: (8); In formula (8): , ; Represents the transpose of the x channel signal; Represents the transpose of the y channel signal; represents the transpose of the z channel signal; at the same time: (9); Among them, y x ,y y ,y z They are three channel signals, which can be expressed by the product of quaternion dictionary and sparse coefficient; According to the initialization dictionary matrix D = {d i , 1≤i≤K}, calculate: (10); Get the multi-channel error matrix: ; Among them, d j represents the jth dictionary atom in the dictionary matrix; dk represents the jth dictionary atom in the dictionary matrix, x k represents the kth element in vector x; x j represents the jth element in vector x; The quaternion singular value decomposition method is used to decompose the multi-channel error matrix and obtain: (11); In formula (11), U is an n×n dimensional matrix spanned by n eigenvectors, Λ is an n×n dimensional matrix with n eigenvalues ​​as diagonals, V is the unitary matrix of U; H represents the Hermitian conjugate transpose of the vector; A vector representing the error matrix; Update the vector dk of the U matrix column by column to dk update , take the first column of matrix U as the updated atom dk update , and update the sparse coefficients corresponding to the atom at the same time; multiply the matrix V as the first column with the semi-positive definite diagonal matrix Λ[1,1] as the updated sparse coefficients until the algorithm reaches the number of iterations and outputs the multi-channel reconstructed signal y c =D c x, where D c is the updated dictionary matrix D c ={d i ,1≤i≤K}, c is the number of channels, and K is the number of sampling points.

[0020] Furthermore, the step of comparing the original observation signals of each measuring point with the reconstructed multi-channel monitoring data, calculating the reconstruction error, and adjusting the parameters of the asymmetric sparse low-rank matrix model and the sparse quaternion digital dictionary learning model in real time according to the reconstruction error until the error threshold range is met includes: Calculate the reconstruction error between the original observation signal of each measuring point and the reconstructed multi-channel monitoring data, expressed as: Error=Yy c (12); Where Y is the multi-channel observation data matrix, y c Multi-channel reconstruction signal; Set the error threshold to a constant value of 10 -2 , adjust the parameters of the asymmetric sparse low-rank matrix model and the sparse quaternion dictionary learning model in real time until the error threshold range is met and the iteration is stopped.

[0021] In summary, the method proposed in the present invention is based on the quaternion model, takes into account the spatiotemporal coupling characteristics of multi-channel degradation data of the hydraulic pump, and can predict multi-channel data at one time instead of channel by channel; compared with the traditional signal reconstruction method, the method proposed in the present invention is not affected by uncertain factors such as changes in external working load, changes in ambient temperature and humidity, and has high reconstruction accuracy; in addition, compared with the traditional signal reconstruction method, the method proposed in the present invention can effectively preserve important degradation and weak detail information, has low computational complexity, and has a fast algorithm running speed, which can meet the needs of real-time and rapid response of industrial equipment.

[0022] In the asymmetric sparse quaternion hydraulic pump service degradation big data remote restoration method described in this embodiment, the experimental data of the present invention comes from the centrifugal pump fault test bench data set of Anhui Agricultural University. The selected equipment to be tested is a horizontal centrifugal pump with a hydraulic pump model of 25CWZ-5A220V. The inherent parameters include a flow rate of 3m 3 / h, efficiency is 23.6%, head is 20m, distribution power is 1.1kw, speed is 2950r / min, suction height is 5m, etc.

[0023] The NI-1062DAQ acquisition chassis and acceleration sensor are used to collect multi-channel centrifugal pump impeller damage signals. The sampling frequency is fs=1024. Figure 3 It is the three-channel vibration acceleration signal of the centrifugal pump impeller damage fault, the sampling point is 101376, and the sampling duration is 99s.

[0024] To illustrate the visualization and resolution of the proposed method, 512 sampling points are selected and the time is 0.5s. Figure 3 The first three channels are selected.

[0025] based on Figure 3 The three-channel vibration acceleration signal of the centrifugal pump impeller damage fault is used to construct an asymmetric sparse low-rank matrix model. The alternating direction multiplier method and the fast smoothing soft threshold method are used to solve the constructed asymmetric sparse low-rank matrix model to obtain the sparse coefficient matrix of the observation signal of each measuring point. The parameters of the asymmetric sparse low-rank matrix model are set as follows: regularization parameter λ=0.01, regularization parameter ξ=0.01, smoothing constant ɛ=10 -6 , the Lagrange coefficient μ=0.01, the number of iterations is 1000, and the evaluation noise deviation of the three channel signals is obtained according to Sigma=median(abs(y1-median(y1))) / 0.6745 to obtain Sigma=25.5871; finally, the multi-channel sparse coefficient matrix data of the asymmetric sparse low-rank matrix method is obtained, as shown Figure 4 shown.

[0026] The sparse quaternion dictionary learning model constructed by quaternion singular value decomposition method and dictionary layer-by-layer update is used to finally obtain the observation reconstruction signal of each measuring point. The result is shown in Figure 5, where: FIG5( a ) is a first channel reconstructed signal (upper) and an amplified signal (lower) thereof obtained according to an embodiment of the present invention; FIG5( b ) is a second channel reconstructed signal (upper) and an amplified signal (lower) thereof obtained according to an embodiment of the present invention; Figure 5 (c) is a third channel reconstructed signal (upper) and its amplified signal (lower) obtained in an embodiment of the present invention; It can be seen that the reconstructed three-channel signals have the same waveform trend as the original three-channel signals, the waveform amplitude fluctuations are consistent, and the matching is high.

[0027] The error threshold range is set to [-0.5, 0.5], and the reconstruction error between the reconstructed multi-channel monitoring data and the original observation signal of each measuring point is calculated. The results are shown in Figure 6, where: FIG6 (a) is a first channel reconstruction error signal obtained according to an embodiment of the present invention; FIG6( b ) is a second channel reconstruction error signal obtained according to an embodiment of the present invention; FIG6 (c) is a third channel reconstruction error signal obtained according to an embodiment of the present invention; From the reconstruction errors of the three channels, it can be clearly seen that the reconstruction error range obtained by the method proposed in the present invention basically fluctuates in the range of [-0.5, 0.5]. The reconstruction error range fluctuates slightly, the reconstruction accuracy is high, and the error threshold range requirement is met.

[0028] The Hilbert envelope spectrum is used to extract the fault characteristic frequency of the reconstructed multi-channel monitoring data to detect whether the weak fault information in the reconstructed multi-channel monitoring data is lost. Figure 7 The original envelope spectrum of the first channel of an embodiment of the present invention (above) and the envelope spectrum of the reconstructed signal of the first channel (below) show that the reconstructed signal obtained by the method proposed in the present invention is demodulated by Hilbert envelope, and the fault characteristic frequency of 96 Hz, amplitude 2012641, and frequency 190 Hz, amplitude 8.11699 are basically consistent with the original signal, which is demodulated by Hilbert envelope, and the fault characteristic frequency of 96 Hz, amplitude 2012585, and frequency 190 Hz, amplitude 8.11566 are basically consistent, and the weak fault signal is well preserved without distortion.

[0029] Finally, the restoration method of the embodiment of the present invention can perform remote fault diagnosis and intelligent operation and maintenance of the centrifugal pump and its impeller components based on the online reconstruction data of each channel.

[0030] In combination with the above embodiments, it can be seen that the restoration method proposed in the present invention can predict multi-channel data at one time, is not affected by uncertain factors such as changes in external working conditions and load, changes in ambient temperature and humidity, has high reconstruction accuracy, and can effectively preserve important degradation and weak detail information, and has good industrial application value.

[0031] Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and the embodiments. It can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily realized. Therefore, without departing from the general concept defined by the claims and equivalent scope, the present invention is not limited to the specific details and the illustrations shown and described here.

Claims

1. Asymmetric sparse quaternion hydraulic pump service degradation big data remote restoration method, characterized by: The following steps are involved: Pick up the vibration acceleration signal of each measuring point of the equipment to be tested, and obtain the observation signal of each measuring point; An asymmetric sparse low-rank matrix model is constructed, and the alternating direction multiplier method and the fast smoothing soft threshold method are used to solve the constructed asymmetric sparse low-rank matrix model to obtain the sparse coefficient matrix of the observation signal of each measuring point; Construct a sparse quaternion dictionary learning model, use the quaternion singular value decomposition method and dictionary update layer by layer to obtain the observation reconstruction signal of each measuring point, and realize the reconstruction of multi-channel monitoring data; The original observation signals of each measuring point are compared with the reconstructed multi-channel monitoring data to calculate the reconstruction error. According to the reconstruction error, the parameters of the asymmetric sparse low-rank matrix model and the sparse quaternion digital dictionary learning model are adjusted in real time until the error threshold range is met; The fault characteristic frequency is extracted from the reconstructed multi-channel monitoring data using the Hilbert envelope spectrum to detect whether the weak fault information in the reconstructed multi-channel monitoring data is lost; Based on the online reconstruction data of each channel, remote fault diagnosis and intelligent operation and maintenance are carried out on the equipment under test.

2. The asymmetric sparse quaternion hydraulic pump service degradation big data remote restoration method according to claim 1 is characterized in that: In the step of picking up the vibration acceleration signal of each measuring point of the equipment to be tested and obtaining the observation signal of each measuring point, the online service data of the equipment to be tested collected by the acceleration sensor is expressed as: Y=AX+NOISE(1); In formula (1), Y is the multi-channel observation data matrix, Y = [y1, y2, y3, …, y n ] T , A is the model degradation matrix, X is the multi-channel sparse coefficient to be estimated, X=[x1,x2,x3,…,x n ] T ,NOISE is the external background noise.

3. The asymmetric sparse quaternion hydraulic pump service degradation big data remote restoration method according to claim 2 is characterized in that: The step of constructing an asymmetric sparse low-rank matrix model, solving the constructed asymmetric sparse low-rank matrix model using an alternating direction multiplier method and a fast smoothing soft threshold method to obtain a sparse coefficient matrix of an observation signal at each measuring point includes: Using the multi-channel observation data matrix Y, an asymmetric sparse low-rank matrix model is constructed, which is expressed as: （2）； In formula (2): F(X) is the cost function; is the Frobenius norm; λ and ξ both represent regularization parameters, where λ>0, ξ>0; represents the singular value penalty function of the smooth parameterization; represents an asymmetric smoothing parameterized penalty function, the coefficient a is the penalty function waveform control factor, ε is the small data smoothing constant, ε=10 -6 ;M i represents the i-order difference matrix; σ(·) represents the singular value transformation of the signal; i represents the difference order; j represents the j-th row number of the matrix; k represents the k-th column number of the matrix; p represents the number of terms of the smooth parameterized singular value penalty function or the asymmetric smooth parameterized penalty function; m represents the number of rows of the matrix; n represents the number of columns of the matrix; The construction of an asymmetric sparse low-rank matrix model can be transformed into: （3）； In formula (3): ； ； D1 represents the first-order difference operator; α represents the penalty function waveform control factor; G1(w) and G2(w) represent the abbreviated form of the formula; Construct an asymmetric sparse low-rank matrix model with constraints, expressed as: （4）； In formula (4): ； ； Among them, λ1 represents the regularization parameter; g1(U) and g2(V) represent the abbreviated form of the formula; U and V represent the alternative form of the matrix X; The augmented Lagrangian function model is constructed and expressed as: （5）； In formula (5), μ is the Lagrange coefficient; U and V represent the alternative forms of matrix X; Z represents an auxiliary matrix; The alternating direction multiplier method and the fast smoothing soft threshold method are used to solve the constructed augmented Lagrangian function model, and the sparse coefficient matrix of the observation signal of each measuring point is obtained, which is expressed as: （6）； In formula (6), SFSS (·,·,·,·) is a fast smoothing soft threshold method, specifically: ,in: , ; x represents the preprocessing vector; ξ represents the regularization parameter; μ represents the Lagrange coefficient; ε represents the small data smoothing constant, ε=10 -6 .

4. The asymmetric sparse quaternion hydraulic pump service degradation big data remote restoration method according to claim 3 is characterized in that: The steps of constructing a sparse quaternion dictionary learning model, using a quaternion singular value decomposition method and dictionary update layer by layer to obtain an observation reconstruction signal of each measuring point, and realizing reconstruction of multi-channel monitoring data include: Construct a sparse quaternion dictionary learning model, expressed as: Y=0+y x i+y y i+y z k (7); In formula (7), the quaternion training dictionary is: D = D0 + D x i+D y j+D z k,D∈H n×K , the sparse coefficient can be expressed as: X=x=x0+x x i+x y j+x z k,x∈H K ; Among them, D x is the training dictionary of x channel; D y is the training dictionary of the y channel; D z is the training dictionary of the z channel; i, j, k are imaginary units, i 2 =-1,j 2 =-1; k 2 =-1; i 2 =j 2 =k 2 =ijk=-1; The above model can be transformed into: (8); In formula (8): , ; Represents the transpose of the x channel signal; Represents the transpose of the y channel signal; represents the transpose of the z channel signal; at the same time: （9）； Among them, y x ,y y ,y z They are three channel signals, which can be expressed by the product of quaternion dictionary and sparse coefficient; According to the initialization dictionary matrix D = {d i , 1≤i≤K}, calculate: （10）； Get the multi-channel error matrix: ; Among them, d j represents the jth dictionary atom in the dictionary matrix; d k represents the jth dictionary atom in the dictionary matrix, x k represents the kth element in vector x; x j represents the jth element in vector x; The quaternion singular value decomposition method is used to decompose the multi-channel error matrix and obtain: （11）； In formula (11), U is an n×n dimensional matrix spanned by n eigenvectors, Λ is an n×n dimensional matrix with n eigenvalues ​​as diagonals, V is the unitary matrix of U; H represents the Hermitian conjugate transpose of the vector; A vector representing the error matrix; Update the vector dk of the U matrix column by column to dk update , take the first column of matrix U as the updated atom dk update , and update the sparse coefficient corresponding to the atom; Multiply the first column of matrix V with the semi-positive definite diagonal matrix Λ[1,1] as the updated sparse coefficient until the algorithm reaches the number of iterations and outputs the multi-channel reconstructed signal y c =D c x, where D c is the updated dictionary matrix, D c ={d i ,1≤i≤K}, c is the number of channels, and K is the number of sampling points.

5. The asymmetric sparse quaternion hydraulic pump service degradation big data remote restoration method according to claim 4 is characterized in that: The step of comparing the original observation signals of each measuring point with the reconstructed multi-channel monitoring data, calculating the reconstruction error, and adjusting the parameters of the asymmetric sparse low-rank matrix model and the sparse quaternion digital dictionary learning model in real time according to the reconstruction error until the error threshold range is met includes: Calculate the reconstruction error between the original observation signal of each measuring point and the reconstructed multi-channel monitoring data, expressed as: Error=Yy c (12) Where Y is the multi-channel observation data matrix, y c Multi-channel reconstruction signal; Set the error threshold to a constant value of 10 -2 , adjust the parameters of the asymmetric sparse low-rank matrix model and the sparse quaternion dictionary learning model in real time until the error threshold range is met and the iteration is stopped.

Images