A method for extracting weak fault features of an autonomous underwater vehicle propeller

Abstract

The application provides a weak fault feature extraction method for an autonomous underwater vehicle propeller, and belongs to the technical field of underwater vehicle fault diagnosis, and comprises two parts: fault feature enhancement and feature fusion. First, the application optimizes parameters by judging the Gaussianity of all modes of multi-source state signals and control signals through negative entropy, completes noise reduction, and extracts and enhances fault features based on a modified Bayesian algorithm. Then, the feature signals are divided into multiple time intervals, the faults occurring in each interval are taken as focal elements, all signals except the longitudinal velocity are subjected to first feature fusion, the first fusion result is subjected to second fusion with the feature signal of the longitudinal velocity, the fault features are further enhanced, and the monotonicity between the fault features and the fault degree is presented. The application can provide a basis for subsequent fault detection and identification, and is particularly suitable for state monitoring of autonomous underwater vehicle propellers.

Description

Technical Field

[0001] This invention patent relates to the field of underwater robot fault diagnosis technology, and in particular to a monitoring method applicable to the safety of autonomous underwater robots. Background Technology

[0002] With the increasing scarcity of non-renewable resources on land, the ocean plays an increasingly prominent role in human survival and development. Autonomous Underwater Vehicles (AUVs), as vehicles for ocean development and coastal defense equipment, play an irreplaceable role in political, economic, and military fields. Operating unmanned and untethered in complex marine environments, the safety of AUVs is a crucial research focus in their research and practical application. [1] The thruster is the most important power component in an AUV, and also one of the main sources of AUV failure. Researching thruster fault diagnosis technology is of great research significance and practical value for improving the safety of AUVs.

[0003] Early-stage thruster failures are often minor, resulting in relatively small thrust losses. If these failures can be diagnosed early and timely safety measures (fault tolerance or self-rescue) implemented, more serious malfunctions can be avoided. Therefore, researching methods for diagnosing minor thruster failures in AUVs is currently a hot research topic.

[0004] In the extraction of thruster fault features, broadly speaking, there are data-driven, analytical model-based, and qualitative simulation-based methods. Among these, data-driven methods utilize historical AUV information to extract fault features and have been widely applied in AUV fault diagnosis. A typical research approach in data-driven fault feature extraction involves denoising using wavelet methods, followed by feature extraction using methods such as Black-Scholes (MB), and then fusing fault features from multiple signals using methods such as evidence theory. This approach generally yields good results in extracting fault features when AUV thrusters experience severe faults (thruster output loss greater than 10%).

[0005] This invention studies the fault feature extraction and fusion of weak faults (thruster output loss less than 10%) in AUV thrusters. Many scholars have achieved excellent research results in AUV thruster fault diagnosis technology, but most focus on hard faults and faults with significant output loss, while less attention is paid to weak faults with output loss less than 10% of total output. Compared with strong thruster faults, weak thruster faults are characterized by weak fault features, low signal-to-noise ratio, and fault features hidden within interference features, making them difficult to separate. Based on the aforementioned approach to strong thruster fault feature extraction—namely, wavelet denoising + MB + evidence theory—this invention, when extracting weak fault features, reveals the following two problems: 1) The ratio and difference between fault features and noise features extracted by the wavelet + MB method are both small. 2) The fused fault features do not exhibit a monotonic relationship with the fault severity. Summary of the Invention

[0006] The purpose of this invention is to provide a method for extracting weak fault features of autonomous underwater vehicle (AUV) thrusters based on multi-source state signals and control signals. This invention effectively enhances the difference and ratio between fault feature values ​​and noise feature values ​​of weak faults, while also ensuring a monotonic relationship between fault features and fault severity. This facilitates subsequent fault detection and identification, ensuring the safe and reliable operation of the AUV.

[0007] The objective of this invention is achieved as follows: The steps are as follows:

[0008] Step 1: Initialize the number of modes, center frequencies, and equilibrium parameters of variational mode decomposition (VMD), and use these to perform VMD to obtain the initial mode set;

[0009] Step 2: Obtain the improved optimization evaluation function (MTF) and the entropy values ​​for each mode based on negative entropy;

[0010] Step 3: Find the optimal number of modes and equilibrium parameters. Continuously generate new mode sets through operations such as deletion and merging, and use the improved TF to determine the optimal number of decomposed modes and center frequencies;

[0011] Step 4: Noise reduction and fault feature extraction based on MB. The signal is decomposed using VMD based on the optimal number of modes and center frequency. Residual modes are removed and the signal is reconstructed to obtain a denoised signal. Fault features are then extracted based on MB.

[0012] Step 5: Perform basic probability allocation using the time interval occurrence of the fault as the focal element. For the right main thruster control variable characteristic signal, the lateral thruster control variable fault characteristic signal, and the heading angle fault characteristic signal, perform basic probability allocation calculation using the time interval occurrence of the fault as the "focal element";

[0013] Step 6: Perform fusion based on Dempster's synthesis rules;

[0014] Step 7: Secondary fusion of fault features. The confidence set obtained from the first fusion is fused with the fault feature signal of the velocity signal to obtain the final fault feature result.

[0015] This invention patent also includes the following features:

[0016] 1. Step 2 specifically includes:

[0017] ① Obtain the entropy values ​​of each mode.

[0018] To determine the noise content of each initial mode, this invention introduces negative entropy theory, measuring the noise content by determining the magnitude of the non-Gaussianity of the mode. The theoretical formula for calculating negative entropy is as follows:

[0019] J(y)=H(y guass )-H(y)

[0020] In the formula, y gauss y and y are Gaussian variables with the same covariance. H is the differential entropy, which is calculated as follows:

[0021] H(y)=-∫f(y)logf(y)dy

[0022] Since the calculation of negative entropy depends on factors such as the prior probability distribution of random variables, H in the above formula is unknown. Generally, an approximation strategy is used to obtain an estimate of the negative entropy, and its formula is:

[0023] J(y)≈[E(G(y))-E(G(v))] 2

[0024] In the formula, v is a Gaussian variable with zero mean and the same variance as y; E represents the expectation; G represents a non-quadratic function, and the function G is chosen as G(y) = -exp(-y 2 / 2).

[0025] ② Obtain the improved and optimized evaluation function

[0026] In order to match the evaluation function after introducing negative entropy, this invention improves the optimization evaluation function TF in MVMD based on negative entropy, resulting in an improved optimization evaluation function (MTF). The improved optimization evaluation function (MTF) is as follows:

[0027]

[0028] Where f, frecon, and fres represent the original signal, the reconstructed signal (the sum of all modes), and the residual signal, respectively; β i(i = 1, 2, 3) are the weights of the different components forming the denominator; NE(u) represents negative entropy, where u represents the data sequence; ELR represents the ratio of the energy of the reconstructed signal to that of the original signal, and the larger the value, the less signal loss, and its formula is:

[0029]

[0030] Orth represents the orthogonality between the reconstructed signal and the original signal. The higher the orthogonality value, the stronger the correlation between the two vectors. The formula is:

[0031]

[0032] 2. Step 5 specifically includes:

[0033] The fault characteristic signal X = {x(1), x(2), ..., x(n)} is divided into N2 time intervals from left to right according to a fixed length N, and the fault characteristic vector y of each time interval is obtained. k .

[0034] y k ={x(k),x(k+1),…,x(k+N-1)}

[0035] Where k = 1, 2, 3, ..., N2.

[0036] The "focal element" A is the fault that occurs within the k-th time interval. k Construct a recognition framework Θ = {A1, A2, ..., A n}

[0037] The basic probability allocation function for the fault characteristic signals of the right main thruster control quantity, the lateral thruster control quantity, and the heading angle signal is obtained by performing probability allocation according to the following formula.

[0038]

[0039] In the formula, d(k) is the fault feature vector y of the k-th interval. k The maximum value.

[0040] 3. Step 7 specifically includes:

[0041] The fused confidence set is then fused with the fault characteristic signal of the velocity signal. The fused fault characteristic signal M is obtained according to the following formula. F (t).

[0042] M F (t)=S(t)·ω(t)

[0043]

[0044] Among them, S(t) is the speed signal feature, a i = N1·i; 0 < t < N1·N2, t ∈ N; H(t) is the unit step function.

[0045] Taking the maximum value of the fused fault feature signal M F (t) as the thruster fused fault feature value F, that is, F = maxM F (t).

[0046] Compared with the prior art, the beneficial effects of the present invention are as follows: There is a problem that the difference between the weak fault features of the thruster obtained based on the traditional wavelet + MB method and the noise features is small and difficult to separate. Compared with the wavelet + MB method, the ratio between the fault feature value extracted by the MVMD + MB method and the noise feature increases, but the increase amplitude is small and still cannot meet the requirements of subsequent fault diagnosis. To address the above problems, the present invention specifically proposes a fault feature enhancement method based on improved MVMD and modified Bayesian classification algorithm (MB). This method uses negative entropy to replace the permutation entropy in MVMD, judges the Gaussianity of the mode through negative entropy, judges the noise content degree of the mode according to the size of the Gaussianity of the mode, and continuously deletes the modes with small negative entropy to achieve the purpose of noise reduction, so as to further enhance the difference and ratio between the weak fault feature value and the noise feature value. It is necessary to perform fusion processing on multiple feature extractions to satisfy the monotonic relationship between the fault feature and the fault degree, and at the same time further expand the signal-to-noise ratio. The fault feature obtained after feature fusion based on the evidence theory does not show a monotonic relationship with the fault degree. Although the monotonic mapping between the fault feature and the fault degree under the strong fault of the thruster can be achieved through the evidence theory method based on the energy of the wave peak region, the effect on weak faults is still not good. To address the above problems, the present invention patent proposes a fault feature fusion method based on the combination of focal element and secondary fusion. This method divides the feature signal into multiple time intervals according to a fixed length. In any time interval, taking the occurrence of a fault in this time interval as the "focal element", dividing the maximum value of the fault feature in this time interval by the sum of the maximum values of the fault features in all time intervals of the signal, obtaining the probability of the thruster fault occurrence reflected by each signal such as the longitudinal speed in this time interval, and then obtaining the probability of the thruster fault occurrence at each moment, that is, the basic probability assignment function. Then, based on the Dempster rule, fault feature fusion is performed. Next, when performing feature fusion based on the evidence theory, the longitudinal speed signal does not participate in the fusion, and the first fusion is performed on the probabilities of other signals based on the evidence theory; then, the result of the first fusion is fused with the fault feature of the longitudinal speed signal for the second time. Through the above steps, the monotonic mapping relationship can be presented between the fused thruster fault feature and the fault degree, so as to perform subsequent fault detection and identification work and ensure the safe and reliable operation of the AUV. Description of the Drawings

[0047] Figure 1 This is a flowchart of the fault feature extraction method of this invention patent.

[0048] Figure 2 The results are the longitudinal velocity signal fault feature extraction results of the various methods in this invention patent.

[0049] Figure 3 This is a table showing the differences and ratios between fault features and noise features obtained by various methods in this invention patent.

[0050] Figure 4 The fault characteristic table of AUV thrusters obtained by various fusion methods of this invention patent.

[0051] Figure 5-7 These are all mapping diagrams showing the relationship between fused fault feature values ​​and fault severity for each method in this invention patent, wherein: Figure 5 It is a mapping relationship diagram based on the fusion method of traditional evidence theory. Figure 6 It is a mapping diagram based on the evidence theory of peak region energy. Figure 7 The fusion method of this invention has a mapping relationship diagram. Detailed Implementation

[0052] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0053] Figure 1 This is a flowchart of the AUV fault feature extraction method of this invention patent. Figure 1 The specific implementation steps of the method for extracting weak fault features of autonomous underwater robot thrusters based on multi-source state signals and control signals are as follows:

[0054] Step 1: Initialize the number of modes, center frequencies, and equilibrium parameters for Variational Mode Decomposition (VMD), and use these to perform VMD to obtain an initial set of modes. This step is the same as the traditional MVMD method.

[0055] Step 2: Obtain the improved optimization evaluation function (MTF) and the entropy values ​​of each mode based on negative entropy.

[0056] In this step, this patent uses the negative entropy method to obtain the entropy value of each mode, and obtains the improved optimization evaluation function (MTF) based on the negative entropy. The specific implementation includes the following two aspects: ① obtaining the entropy value of each mode; ② obtaining the improved optimization evaluation function.

[0057] ① Obtain the entropy values ​​of each mode.

[0058] To determine the noise content of each initial mode, this invention introduces negative entropy theory, measuring the noise content by determining the magnitude of the non-Gaussianity of the mode. The theoretical formula for calculating negative entropy is as follows:

[0059] J(y)=H(yguass H(y) (1)

[0060] In the formula, y gauss y and y are Gaussian variables with the same covariance. H is the differential entropy, which is calculated as follows:

[0061] H)y)=-∫f(y)logf(y)dy (2)

[0062] Since the calculation of negative entropy depends on factors such as the prior probability distribution of random variables, H in the above formula is unknown. Generally, an approximation strategy is used to obtain an estimate of the negative entropy, and its formula is:

[0063] J(y)≈[E(G(y))-E(G(v))] 2 (3)

[0064] In the formula, v is a Gaussian variable with zero mean and the same variance as y; E represents the expectation; G represents a non-quadratic function, and the function G is chosen as G(y) = -exp(-y 2 / 2).

[0065] ② Obtain the improved and optimized evaluation function

[0066] In order to match the evaluation function after introducing negative entropy, this invention improves the optimization evaluation function TF in MVMD based on negative entropy, resulting in an improved optimization evaluation function (MTF). The improved optimization evaluation function (MTF) is as follows:

[0067]

[0068] Where f, frecon, and fres represent the original signal, the reconstructed signal (the sum of all modes), and the residual signal, respectively; β i (i = 1, 2, 3) are the weights of the different components forming the denominator; NE(u) represents negative entropy, where u represents the data sequence; ELR represents the ratio of the energy of the reconstructed signal to that of the original signal, and the larger the value, the less signal loss, and its formula is:

[0069]

[0070] Orth represents the orthogonality between the reconstructed signal and the original signal. The higher the orthogonality value, the stronger the correlation between the two vectors. The formula is:

[0071]

[0072] Step 3: Find the optimal number of modes and equilibrium parameters. New mode sets are continuously generated through deletion and merging operations. The optimal number of decomposed modes and center frequencies are then determined using an improved TF (True-Fractal) algorithm. This step is the same as the traditional MVMD method.

[0073] Step 4: Noise reduction and fault feature extraction based on MB. The signal is decomposed using VMD based on the optimal mode number and center frequency. Residual modes are removed and the signal is reconstructed to obtain a denoised signal. Fault features are then extracted based on MB.

[0074] Step 5: Perform basic probability allocation using the time interval occurrence of the fault as the focal element. For the right main thruster control variable characteristic signal, the lateral thruster control variable fault characteristic signal, and the heading angle fault characteristic signal, perform basic probability allocation calculation using the time interval occurrence of the fault as the "focal element";

[0075] Traditional evidence-based methods use the occurrence of a fault at a specific moment (or beat) as the "focal element" for feature fusion, which fails to yield fused features among the maximum probabilities of fault occurrence for multiple signals. This patent uses a specific time interval as the "focal element" for feature fusion, selecting an appropriate time interval length to ensure that the maximum values ​​of fault feature signals from multiple signals occur within the same time interval. In this patent, the feature signals are divided into multiple time intervals of fixed length. Within any time interval, the occurrence of a fault within that interval is taken as the "focal element." The maximum value of the fault feature in that time interval is divided by the sum of the maximum values ​​of the fault features across all time intervals of that signal to obtain the probability of thruster fault occurrence reflected by each signal, such as the longitudinal velocity, within that time interval. This yields the thruster fault occurrence probability for each time interval. Finally, the fault fusion features are calculated based on Dempster's rule. The specific process is as follows:

[0076] The fault characteristic signal X = {x(1), x(2), ..., x(n)} is divided into N2 time intervals from left to right according to a fixed length N, and the fault characteristic vector y of each time interval is obtained. k .

[0077] y k ={x(k),x(k+1),…,x(k+N-1)} (7)

[0078] Where k = 1, 2, 3, ..., N2.

[0079] The "focal element" A is the fault that occurs within the k-th time interval. k Construct a recognition framework Θ = {A1, A2, ..., A n}

[0080] The basic probability allocation function for the fault characteristic signals of the right main thruster control quantity, the lateral thruster control quantity, and the heading angle signal is obtained by performing probability allocation according to the following formula.

[0081]

[0082] In the formula, d(k) is the fault feature vector y of the k-th interval.k The maximum value.

[0083] Step 6: Perform fusion based on Dempster's synthesis rules.

[0084] Under the same identification framework Θ, there exist n sets of evidence E1, E2, ..., E n The corresponding basic probability assignment function is m1,m2,…,m n The focal elements are A1, A2, ..., A n For The basic probability assignment function for n sets of evidence on Θ is m1,m2,…,m n The synthesis rules for Dempster are as follows:

[0085]

[0086] in,

[0087]

[0088] Step 7: Secondary fusion of fault features. The confidence set obtained from the first fusion is fused with the fault feature signal of the velocity signal to obtain the final fault feature result.

[0089] Experimental studies revealed that the longitudinal velocity signal plays a dominant role in the fault feature fusion process for thruster fault diagnosis. Further analysis showed that when a weak fault occurs in the thruster, the maximum value of the fault feature in the longitudinal velocity signal becomes smaller after basic probability allocation during feature fusion based on evidence theory. This weakens the dominant role of the longitudinal velocity signal after fusion, potentially leading to a non-monotonic relationship between the fused fault features and the fault severity.

[0090] Therefore, in this patent, when performing feature fusion based on evidence theory, the longitudinal velocity fault feature signal is not included in the fusion; instead, the basic probability assignment functions of other signals are used for the first fusion based on evidence theory. Then, the result of the first fusion is fused a second time with the longitudinal velocity fault feature signal. The specific process is as follows:

[0091] The fused confidence set is then fused with the fault characteristic signal of the velocity signal. The fused fault characteristic signal M is obtained according to the following formula. F (t).

[0092] M F (t)=S(t)·ω(t) (11)

[0093]

[0094] Where S(t) represents the velocity signal characteristic, a i= N1·i; 0 < t < N1·N2, t ∈ N; H(t) is the unit step function.

[0095] Taking the maximum value of the fused fault feature signal M F (t) as the thruster fused fault feature value F, i.e., F = max M F (t).

[0096] Application case:

[0097] A comparative experiment was carried out on the weak fault feature extraction method of the thruster of the autonomous underwater vehicle proposed in this patent based on multi-source state signals and control signals to verify the effectiveness of this patent in fault feature extraction. Among them, the wavelet + MB method, the MVMD + MB method, the traditional evidence theory fusion method, and the evidence theory fusion method based on the energy of the wave peak region were used for comparative experiment verification respectively.

[0098] The data used in the experimental verification process of this patent comes from the beaver II AUV experimental carrier independently developed by this research team. In the pool test, the target speed of the AUV is 0.3 m / s. Starting from a standstill, after reaching the target speed, it makes a steady straight-line motion, and the control frequency is 5 Hz, that is, the control period is 0.2 s. The fault soft simulation method is used to simulate the left main thruster force loss fault. Starting from the 250th beat, the left main thruster has a force loss fault until the end of the experiment. Fault experiments with 5%, 8%, 10%, 20%, 30%, and 40% left main thruster force loss were carried out respectively, and the longitudinal speed signal, right main thruster control amount, side thruster control amount, and heading angle signal of the AUV were obtained.

[0099] Taking the 5% left main thruster force loss fault as an example, the results of the fault feature extraction of each method in the longitudinal speed signal are shown as Figure 2 shown. Under the faults of 5%, 8%, 10%, 20%, 30%, and 40% left main thruster force loss, the results of the difference and ratio between the fault features and the noise features obtained by each method are shown in the table of Figure 3 . From the above experimental results and analysis, it is reflected that in the case of weak faults of the AUV thruster, that is, when the output force loss degree is 5%, 8%, 10%, compared with the MVMD + MB method and the wavelet + MB method, the difference and ratio between the fault feature value and the noise feature value of this patent are significantly enhanced; in addition, this patent enhances the difference and ratio between the fault feature value and the noise feature value under strong faults of the AUV thruster, verifying the effectiveness of this patent in enhancing the difference and ratio between the fault feature value and the noise feature value.

[0100] The AUV thruster fault features obtained by different fusion methods are shown in the table of Figure 4 . For the convenience of analyzing problems, according to Figure 4Mapping relationships between fused fault eigenvalues ​​and fault severity are plotted for fusion methods based on traditional evidence theory, fusion methods based on peak region energy evidence theory, and the fusion method of this invention. Figure 5-7 As shown. From Figure 5-7 The results show that the fusion fault feature value obtained by this patent has a monotonic relationship with the fault degree. Regardless of the fault degree, a fusion fault feature value corresponds to a unique fault degree value, which reflects that this patent can solve the problem that the fusion fault feature value and the fault degree do not have a monotonic relationship. Figure 4 and Figure 5-7 The results show that the above experimental results and analysis reflect the effectiveness of this patent in enhancing the monotonicity of fusion fault characteristic values ​​and fault degree.

[0101] In summary, this invention relates to a method for extracting weak fault features of autonomous underwater robot thrusters based on multi-source state signals and control signals. This method belongs to the field of underwater robot fault diagnosis technology and comprises two parts: fault feature enhancement and feature fusion. First, this patent uses negative entropy to determine the Gaussianity of all modes of the multi-source state signals and control signals, thereby optimizing parameters and achieving noise reduction. Then, fault features are extracted and enhanced based on a modified Bayesian algorithm. Next, the feature signals are divided into multiple time intervals, with the occurrence of a fault in each interval as the focal element. A first feature fusion is performed on all signals except for the longitudinal velocity. The result of the first fusion is then fused a second time with the feature signal of the longitudinal velocity to further enhance the fault features, while simultaneously ensuring a monotonic relationship between the fault features and the fault severity. This patent provides a foundation for subsequent fault detection and identification, and is particularly suitable for state monitoring of autonomous underwater robot thrusters.

Claims

1. A method for extracting weak fault features of an autonomous underwater robot thruster, characterized in that, The steps are as follows: Step 1: Initialize the number of modes, center frequency, and equilibrium parameters of variational mode decomposition (VMD), and use these to perform VMD to obtain the initial mode set; Step 2: Obtain the improved optimization evaluation function MTF and the entropy values ​​of each mode based on negative entropy; The theoretical formula for calculating the negative entropy J is: J(y)=H(y guass )-H(y) Among them, y guass Let y and y be Gaussian variables with the same covariance; H(·) is the differential entropy, H(y)=-∫f(y)logf(y)dy, and an approximation strategy is used to obtain the estimate of the negative entropy: J(y)≈[E(G(y))-E(G(v))] 2 Where v is a Gaussian variable with zero mean and the same variance as y; E(·) represents the expectation; G(·) represents a non-quadratic function, G(y) = -exp(-y 2 / 2); Based on the improvement of the optimization evaluation function TF in MVMD by negative entropy, the improved optimization evaluation function MTF is obtained as follows: Where f, frecon, and fres represent the original signal, reconstructed signal, and residual signal, respectively; β1, β2, and β3 are the weights of the different components constituting the denominator; NE(u) represents negative entropy, where u represents the data sequence; ELR represents the ratio of the energy of the reconstructed signal frecon to that of the original signal f, with a larger value indicating less signal loss. Orth represents the orthogonality between the reconstructed signal frecon and the original signal f. The higher the orthogonality value, the stronger the correlation between the two vectors. Step 3: Find the optimal number of modes and equilibrium parameters. Continuously generate new mode sets through deletion and merging operations. Use the improved optimization evaluation function MTF to obtain the optimal number of decomposed modes and center frequencies. Step 4: Denoise reduction and fault feature extraction based on MB. The signal is decomposed by VMD based on the optimal number of modes and center frequency, residual modes are removed and the signal is reconstructed to obtain a denoised signal. Then, fault feature extraction is performed based on MB. Step 5: Perform basic probability allocation calculations for the fault characteristic signals of the right main thruster control, the side thruster control, and the heading angle, with the occurrence of faults in the time interval as the "focal element"; The fault characteristic signal X = {x(1), x(2), ..., x(n)} is divided into N2 time intervals from left to right according to a fixed length N, and the fault characteristic vector y of each time interval is obtained. k ; y k ={x(k),x(k+1),...,x(k+N-1)} Where k = 1, 2, 3, ..., N2; The "focal element" A is the fault that occurs within the k-th time interval. k Build a recognition framework The basic probability allocation function m(A) of the fault characteristic signal is obtained by performing probability allocation according to the following formula. k ) Where d(k) is the fault feature vector y of the k-th interval. k The maximum value; Step 6: Perform fusion based on Dempster's synthesis rules; When performing feature fusion, the longitudinal velocity fault feature signal is not included in the fusion, and the basic probability assignment function of other signals is used for the first fusion based on evidence theory; Step 7: Secondary fusion of fault features. The confidence set obtained from the first fusion is fused with the fault feature signal of the velocity signal to obtain the final fault feature result. M F (t)=S(t)·ω(t) Where S(t) represents the velocity signal characteristic; α i =N1·i, 0<t<N1·N2; H(t) is the unit step function; Using fused fault characteristic signal M F The maximum value of (t) is taken as the characteristic value F of the thruster fusion fault, that is, F = maxM F (t).

Images