Data-driven AUV (Autonomous Underwater Vehicle) actuating mechanism fault detection method

Abstract

The invention discloses an AUV execution mechanism fault detection method based on data driving, and the method comprises the following steps: 1, mapping original nonlinear data to a high-dimensional space through a Gaussian kernel function, and sequentially carrying out the mean centralization and standard processing; 2, performing multi-level decomposition on the standardized data by utilizing deep principal component analysis to generate a plurality of sub-data sets which are not overlapped with each other; step 3, calculating the statistical magnitude of each sub-data set for each sub-data set; and 4, dynamically setting a fault judgment threshold value, and when the statistic exceeds the fault judgment threshold value, determining that an execution mechanism has a fault. According to the method, linear processing of nonlinear data is realized, the multi-level feature extraction capability of deep principal component analysis is combined, and the fault detection precision and reliability are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of AUV actuator fault diagnosis, and in particular to a data-driven AUV actuator fault detection method. Background Art

[0002] Autonomous underwater vehicles (AUVs) are essential equipment for deep-sea exploration and resource development. Their actuators, such as thrusters and elevators, are key components for power output and attitude control. AUVs operate for extended periods in the harsh deep-sea environment of high pressure, low temperatures, and complex ocean currents. Actuators are susceptible to failure due to mechanical wear, environmental interference, or control failure. According to statistics, actuator failures account for 50%-60% of all AUV system failures. Therefore, real-time fault detection technology for AUV actuators is crucial to ensuring the safety and reliability of deep-sea exploration.

[0003] Among existing data-driven fault detection methods, Deep Principal Component Analysis (Deep-PCA) enhances the ability to extract longitudinal features by decomposing data layer by layer. However, it is still a linear method in nature and cannot effectively process the nonlinear dynamic data of AUV systems. Although Kernel Principal Component Analysis (KPCA) maps nonlinear data to a high-dimensional space through a kernel function, which solves the nonlinear limitations of traditional PCA, its single-level feature extraction makes it difficult to capture multidimensional fault information, especially in complex working conditions, and is prone to missing weak fault signals. The main drawbacks of the two existing methods are: Deep-PCA relies on linear assumptions and cannot handle the complex data features caused by strong coupling, nonlinear dynamics, and environmental disturbances in AUV systems; KPCA lacks multi-level decomposition capabilities, insufficient longitudinal mining of fault information, and difficulty extracting effective information from weak fault signals. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides a data-driven AUV actuator fault detection method to solve the technical problems in the existing technology that it is unable to process complex data features caused by strong coupling, nonlinear dynamics and environmental disturbances, and it is difficult to extract effective information from weak fault signals.

[0005] The present invention provides a data-driven AUV actuator fault detection method, comprising the following steps:

[0006] Step 1: Map the original nonlinear data to a high-dimensional space through the Gaussian kernel function, and perform mean centering and standardization processing in turn;

[0007] Step 2: Use deep principal component analysis to perform multi-level decomposition on the standardized data to generate multiple non-overlapping sub-datasets;

[0008] Step 3: Calculate the statistics of each sub-dataset separately;

[0009] Step 4: Dynamically set the fault judgment threshold. When the statistic exceeds the fault judgment threshold, an actuator fault occurs.

[0010] Furthermore, in step 1, the original nonlinear data is mapped to a high-dimensional space through a Gaussian kernel function.

[0011] Furthermore, in step 1, the Gaussian kernel function is:

[0012]

[0013] Where x is the operating data of the actuator; ρ is the constant term, ρ = 10mσ 2 , where m is the sample size and σ is the standard deviation.

[0014] Furthermore, in step 1, the method for performing mean centering processing is:

[0015] The original kernel matrix is ​​replaced by the centralized kernel matrix, where the centralized kernel matrix is:

[0016]

[0017] Where, K is the original kernel matrix.

[0018] Furthermore, in step 2, the specific process of multi-level decomposition is as follows:

[0019] Step 21: Use the processed normalized data as the data to be decomposed;

[0020] Step 22: Obtain the covariance matrix of the data to be decomposed, and perform singular value decomposition on the covariance matrix to obtain the singular value vectors. Decompose the data to be decomposed according to the singular value vectors to obtain several sub-data;

[0021] Step 23: Repeat the decomposition process of step 22 with the sub-data as the data to be decomposed until the decomposition times are reached to obtain the sub-data set:

[0022]

[0023] Where, P j,k is the kth load matrix after the data matrix is ​​decomposed j times; k is the number of principal elements; I is the unit matrix.

[0024] Furthermore, in step 3, the statistics include: squared prediction error statistics and Hotelling squared statistics, and the specific formula is:

[0025]

[0026] Where, P (j+1),(2k-1) is the same as φ j,k The corresponding covariance matrix S j,k Part of the singular value vectors; k is the number of principal elements; Λ j,k is the same as φ j,k The corresponding covariance matrix S j,k The singular value matrix of ; I is the identity matrix.

[0027] Furthermore, in step 4, the fault judgment threshold is dynamically set by a kernel density estimation method.

[0028] Furthermore, in step 4, the fault judgment threshold value set by the kernel density estimation method is:

[0029]

[0030] In the formula, m represents the number of samples; express The i-th sample in the data set; h is the window width; K is the kernel function; N is the SPE j,k and The total number of .

[0031] Furthermore, the window width is:

[0032] h=1.06σm -1 / 5

[0033] Where m represents the number of samples and σ is the standard deviation.

[0034] Beneficial effects of the present invention:

[0035] In this paper, the original nonlinear data is mapped to a high-dimensional linear space by introducing a Gaussian kernel function, which effectively solves the limitation of the existing Deep-PCA method that cannot process nonlinear data.

[0036] In this invention, based on the existing KPCA, Deep-PCA is further used to decompose high-dimensional data step by step to generate multiple non-overlapping sub-datasets. Compared with the existing KPCA method, KDPCA not only realizes the linearization of nonlinear data, but also combines the multi-level feature extraction capabilities of deep principal component analysis, significantly improving the accuracy and reliability of fault detection. The multi-level feature extraction method can more comprehensively mine potential fault information, effectively overcoming the shortcomings of the KPCA method caused by the single-level information feature extraction under weak fault signals.

[0037] This method uses kernel density estimation to adaptively set dynamic thresholds for statistical variables, avoiding the failure of traditional fixed thresholds under complex operating conditions. Compared to existing Deep-PCA and KPCA methods, KDPCA significantly enhances the robustness of fault detection in dynamic environments, enabling more accurate fault identification and reducing false positives and missed detections. It is particularly suitable for real-time fault detection in AUVs operating in highly dynamic deep-sea environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The features and advantages of the present invention will be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the present invention in any way. In the accompanying drawings:

[0039] Figure 1 is a flow chart of a specific embodiment of the present invention;

[0040] Figure 2 is a flow chart of the Deep-PCA principle in a specific embodiment of the present invention;

[0041] Figure 3 This is a flowchart of fault detection in a specific embodiment of the present invention;

[0042] Figure 4 Schematic diagram of AUV coordinates in a specific embodiment of the present invention;

[0043] Figure 5 The AUV thruster T after using the method of the present invention in a specific embodiment of the present invention 2 Statistical graphs;

[0044] Figure 6 : is a graph of the SPE statistics of the AUV propeller after using the method of the present invention in a specific embodiment of the present invention;

[0045] Figure 7 In a specific embodiment of the present invention, the AUV elevator T is used after the method of the present invention is used. 2 Statistical graphs;

[0046] Figure 8 1 is a graph of SPE statistics of the AUV elevator after the method of the present invention is used in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0047] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0048] The present invention will be further described below with reference to specific examples. Those skilled in the art will appreciate that these examples are intended only to illustrate the present invention and are not intended to limit the scope of the present invention, and that modifications to various equivalent forms of the present invention fall within the scope defined by the appended claims.

[0049] like Figure 1-3 As shown, the present invention provides a data-driven AUV actuator fault detection method, comprising the following steps:

[0050] Step 1: Map the original nonlinear data into a high-dimensional space and perform mean centering and standardization processing in sequence;

[0051] Using Gaussian kernel function:

[0052]

[0053] Where x represents the state operation data of the actuator; ρ is a constant term. In order to test the different monitoring performances of ρ, ρ=10mσ 2 It is suitable for monitoring various fault processes, where m is the number of samples and σ is the standard deviation;

[0054] After the Gaussian kernel function is projected, the feature space needs to be mean-centered. Mean centering can be achieved by replacing the original kernel matrix K with the centralized kernel matrix To achieve this, the centralized kernel matrix is:

[0055]

[0056] Where, K is the original kernel matrix.

[0057] The specific process of standardization is as follows:

[0058] The high-dimensional data set φ is obtained by Gaussian kernel function mapping in step 1 m×m , calculate the mean and variance of each row in preparation for standardization:

[0059]

[0060] Where j is the jth row of the high-dimensional dataset φ; z j (i) is the i-th number in the j-th row;

[0061] Normalize each row of the high-dimensional dataset φ:

[0062]

[0063] Where, is the data obtained after the j-th row data is standardized;

[0064] By standardization, we get a data set φ with a mean of 0 and a variance of 1:

[0065]

[0066] Step 2: Use Deep Principal Component Analysis (Deep-PCA) to perform multi-level decomposition on the standardized data to generate multiple non-overlapping sub-datasets. The specific process of multi-level decomposition is as follows:

[0067] Step 21: Calculate the covariance matrix S of the standardized data set φ and perform singular value decomposition on S:

[0068]

[0069] Where m represents the number of samples; 0,1 ∈R m×m , Λ 0,1 =diag(λ 0,1 ,...λ 0,m ), λ 0,m represents the mth singular value of S; P1∈R m×m is the singular value vector of S. According to the number of main elements, P1 can be decomposed into P 1，1 and P 1，2 , that is, P1=[P 1,1 ,P 1，2 ];

[0070] Step 22: Decompose φ into two parts, namely φ 1,1 With φ 1,2 :

[0071] φ=φ 1,1 +φ 1,2

[0072]

[0073] Where I is the identity matrix;

[0074] Step 23: Decompose the φ 1,1 and φ 1,2 As the new φ, calculate φ separately 1,1 and φ 1,2 The covariance matrix S 1,1 With S 1,2 , and the covariance matrix S 1,1 With S 1,2 Perform singular value decomposition and transform φ 1,1 and φ 1,2 It is divided into two parts:

[0075] φ 1,1 =φ 2,1 +φ2,2

[0076] φ 1,2 =φ 2,3 +φ 2,4

[0077] φ=φ 2,1 +φ 2,2 +φ 2,3 +φ 2,4

[0078] Repeat the above steps and decompose the data φ j times to get 2 j non-overlapping subsets, then the k-th dataset in layer j is:

[0079]

[0080] Where, P j,k is the kth load matrix after the data matrix is ​​decomposed j times; k is the number of principal elements.

[0081] Step 3: Calculate the statistics of each sub-dataset separately;

[0082] Statistics include: squared prediction error statistic SPE and Hotelling squared statistic T 2 , the specific formula is:

[0083]

[0084] Where, SPE j,k and For the dataset φ j,k Two statistics of (j+1),(2k-1) is the same as φ j,k The corresponding covariance matrix S j,k Part of the singular value vectors; k is the number of principal elements; Λ j,k is the same as φ j,k The corresponding covariance matrix S j,k The singular value matrix of ; I is the identity matrix.

[0085] Step 4: Dynamically set the fault judgment threshold through the kernel density estimation method. When the statistic exceeds the fault judgment threshold, an actuator failure occurs.

[0086] Among them, the fault judgment threshold set by the kernel density estimation method is:

[0087]

[0088] In the formula, m represents the number of samples; express The i-th sample in the data set; K is the kernel function; N is SPE j,k and The total number of; h is the window width, the optimal window bandwidth h=1.06σm -1 / 5 , where m represents the number of samples and σ is the standard deviation.

[0089] The method of the present invention is simulated and verified below:

[0090] Step A1: Figure 4 As shown in the figure, a Simulink model is established for the Remus AUV. The AUV model is composed of hydrostatics, hydrodynamic lift and drag, and added mass. The external forces and moments caused by propeller and lift fin control inputs are defined in terms of the airframe's system parameters. The equations with coefficients and the nonlinear equations describing the rigid-body dynamics of the airframe are determined in the reference, enabling realistic simulation of the AUV's inherent nonlinear motion.

[0091] Step A2: Connect components and set initial parameters according to the modules given in the reference so that the simulation process can simulate the actual movement of the AUV in the deep sea.

[0092] Step A3: After the simulation is completed, the data is exported in the data inspector and initialized. The processed data is then subjected to fault detection based on the KDPCA algorithm.

[0093] Figure 5 、 6 This is a KDPCA-based AUV thruster T2 and SPE statistic diagram; the statistic is more sensitive to abnormal changes after a fault occurs, and the detection effect is more stable. In addition, the SPE statistic is more robust than the T2 statistic and can more accurately characterize the fault state. By comparison, the detection sensitivity of SPE is higher than that of T 2 Higher sensitivity.

[0094] Figure 7 、 8 The statistical diagram of AUV elevator T2 and SPE based on KDPCA; SPE and T 2 The change trends are similar, indicating that the statistical SPE and T 2 Contains similar fault information. It is not difficult to see from the figure that KDPCA can quickly detect anomalies when a fault occurs and maintain high detection stability during the duration of the fault.

[0095] Figure 5-8 Describes the statistical diagram of AUV actuator fault detection, considering T 2 The traditional method only relies on one of the residuals exceeding the threshold to generate a jump to determine the fault, while the method of the present invention considers both statistics at the same time. 2 It has better sensitivity than SPE.

[0096] Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention. Such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A data-driven AUV actuator fault detection method, characterized in that: The steps include: Step 1: Map the original nonlinear data to a high-dimensional space through the Gaussian kernel function, and perform mean centering and standardization processing in turn; Step 2: Use deep principal component analysis to perform multi-level decomposition on the standardized data to generate multiple non-overlapping sub-datasets; Step 3: Calculate the statistics of each sub-dataset separately; Step 4: Dynamically set the fault judgment threshold. When the statistic exceeds the fault judgment threshold, an actuator fault occurs.

2. The data-driven AUV actuator fault detection method according to claim 1, characterized in that: In step 1, the original nonlinear data is mapped to a high-dimensional space through a Gaussian kernel function.

3. The data-driven AUV actuator fault detection method according to claim 2, characterized in that: In step 1, the Gaussian kernel function is: Where x is the operating data of the actuator; ρ is the constant term, ρ = 10mσ 2 , where m is the sample size and σ is the standard deviation.

4. The data-driven AUV actuator fault detection method according to claim 1, characterized in that: In step 1, the method for performing mean centering is: The original kernel matrix is ​​replaced by the centralized kernel matrix, where the centralized kernel matrix is: Where, K is the original kernel matrix.

5. The data-driven AUV actuator fault detection method according to claim 1, characterized in that: In step 2, the specific process of multi-level decomposition is: Step 21: Use the processed normalized data as the data to be decomposed; Step 22: Obtain the covariance matrix of the data to be decomposed, and perform singular value decomposition on the covariance matrix to obtain the singular value vectors. Decompose the data to be decomposed according to the singular value vectors to obtain several sub-data; Step 23: Repeat the decomposition process of step 22 with the sub-data as the data to be decomposed until the decomposition times are reached to obtain the sub-data set: Where, P j,k is the kth load matrix after the data matrix is ​​decomposed j times; k is the number of principal elements; I is the identity matrix.

6. The data-driven AUV actuator fault detection method according to claim 1 or 5, characterized in that: In step 3, the statistics include: squared prediction error statistics and Hotelling squared statistics, and the specific formula is: Where, P (j+1),(2k-1) is the same as φ j,k The corresponding covariance matrix S j,k Part of the singular value vectors; k is the number of principal elements; Λ j,k is the same as φ j,k The corresponding covariance matrix S j,k The singular value matrix of ; I is the identity matrix.

7. The data-driven AUV actuator fault detection method according to claim 1, characterized in that: In step 4, the fault judgment threshold is dynamically set by using a kernel density estimation method.

8. The data-driven AUV actuator fault detection method according to claim 7, characterized in that: In step 4, the fault judgment threshold set by the kernel density estimation method is: In the formula, m represents the number of samples; express The i-th sample in the data set; h is the window width; K is the kernel function; N is the SPE j,k and The total number of .

9. The data-driven AUV actuator fault detection method according to claim 8, characterized in that: The window width is: h=1.06σm -1 / 5 Where m represents the number of samples and σ is the standard deviation.

Images