Ship equipment degradation state identification method based on multi-source fusion features and series DBN model

Abstract

The invention discloses a ship equipment degradation state identification method based on a multi-source fusion feature and a series DBN model, and belongs to the technical field of health state management, and the method comprises the steps: extracting a center frequency value of each component of a signal feature; carrying out signal feature analysis by adopting an improved multi-scale permutation entropy, and obtaining a stable permutation entropy value according to a scale factor; obtaining a center frequency corresponding to the maximum kurtosis value of the signal, screening a plurality of BIMF component layer-by-layer data sets, and constructing a feature vector set Q through permutation entropy; performing dimension reduction processing on the feature vector set Q to obtain a feature vector set T; training the DBN1 by adopting the feature vector set T, and training the DBN2 by combining the working condition identification result with the feature vector set T to realize ship equipment degradation state identification; and establishing an observer model, obtaining a health baseline of the ship equipment, and evaluating the degradation state of the identified ship equipment by calculating a mahalanobis distance. According to the invention, the degradation state of the key equipment can be accurately and rapidly evaluated.

Description

Technical Field

[0001] The present invention belongs to the technical field of health status management, and more specifically, relates to a method for identifying the degradation status of ship equipment based on multi-source fusion features and a series DBN model. Background Art

[0002] Once a failure or degradation occurs in key rotating equipment such as support bearings and pump groups, it may cause a reduction or loss of system function, affecting the completion of the ship's mission. State degradation modeling can grasp the operating dynamics of the equipment from a healthy state to the current degraded state, and state assessment can provide important guidance for the use and maintenance of the equipment. However, the degradation state of mechanical equipment is often not obtained through direct observation. Therefore, in actual degradation state modeling, the health factor curve of the mechanical equipment is often constructed to characterize the degradation state or degree of degradation of the equipment health level. With the continuous improvement of the technical level of ships, the incompatibility between the existing degradation state assessment methods of key ship equipment and the ship's mission requirements and repair needs has gradually become prominent. Therefore, it is urgent to study a degradation state assessment method for key equipment such as support bearings and pump groups that is more robust, more intelligent, and more anti-interference.

[0003] At present, there are many studies on the assessment of ship degradation status, but the following problems still exist in the equipment degradation status assessment methods: 1) Traditional degradation status methods are gradually unable to meet the requirements of accurate and efficient equipment health management due to excessive reliance on signal processing technology and expert experience, and the accuracy of dealing with complex time series predictions needs to be improved; 2) Although there are many existing status assessment methods, the diagnostic accuracy is not high, and it is difficult to meet the requirements of actual engineering applications; 3) Ship system equipment is complex, the equipment has strong nonlinearity and high complexity, and it is difficult to establish a reliable and accurate mathematical model; 4) The essence of degradation status modeling of mechanical equipment is to construct health factors, but it faces problems such as reliance on a large number of signal processing technologies and expert experience, difficulty in label selection in supervised training, and incomplete feature extraction; 5) There is a strong correlation between different states of the equipment, and they will interfere with each other. One state feature can often correspond to multiple state types, making it difficult to accurately judge the state; 6) The measurable parameters on some equipment are limited, and the monitoring parameters are far less than the number of state categories. Summary of the invention

[0004] In view of the above defects or improvement needs of the prior art, the present invention proposes a method for identifying the degradation state of ship equipment based on multi-source fusion features and a series DBN model to solve the problem of accurate and rapid assessment of the degradation state of key equipment.

[0005] To achieve the above object, the present invention provides a method for identifying degradation status of ship equipment based on multi-source fusion features and a tandem DBN model, comprising:

[0006] Collect vibration, temperature and pressure signals of ship equipment, perform time-frequency analysis on the collected vibration, temperature and pressure signals, and extract signal features;

[0007] Perform variable mode decomposition on the extracted signal features to obtain two-dimensional intrinsic mode function components, and extract the center frequency value of each component;

[0008] The improved multi-scale permutation entropy is used to analyze the signal characteristics after variable mode decomposition, and a stable permutation entropy value is obtained according to the scale factor.

[0009] The fast spectral kurtosis algorithm is used to obtain the center frequency corresponding to the maximum kurtosis value of the signal, and then several BIMF component layer-by-layer data sets are selected according to the center frequency value of each component, and the feature vector set Q is constructed by permutation entropy;

[0010] The feature vector set Q is reduced in dimension by using local tangent space arrangement to remove redundant and noisy features, and the feature vector set T is obtained;

[0011] Build a dual DBN model series framework, optimize the dual DBN model parameters, use feature vector set T to train DBN1 to realize ship working condition identification, and combine the working condition identification results with feature vector set T to train DBN2 to realize ship equipment degradation state identification;

[0012] An observer model is established to obtain the health baseline of ship equipment. The degradation state of ship equipment is evaluated and identified by calculating the Mahalanobis distance between the current degradation state and the health baseline, and an equipment degradation state evaluation report is output.

[0013] In some optional embodiments, the signal characteristics include time-frequency characteristics and frequency domain characteristics, the time-frequency characteristics include mean, absolute mean, peak, peak-to-peak value, root mean square value, root amplitude, standard deviation, waveform factor, peak factor, pulse factor, kurtosis, skewness and peak state; the frequency domain characteristics include average frequency, center frequency, root mean square frequency and root square deviation frequency.

[0014] In some optional implementations, performing variable mode decomposition on the extracted signal features to obtain two-dimensional intrinsic mode function components and extracting the center frequency value of each component includes:

[0015] Perform variable mode decomposition on the extracted signal features and update the modal component U k , component frequency center ω k , Lagrangian operators λ and n, output K two-dimensional intrinsic mode function BIMF components, and extract the center frequency value f of each BIMF component i .

[0016] In some optional implementation schemes, the signal feature analysis is performed using improved multi-scale permutation entropy to obtain a stable permutation entropy value according to a scale factor, including:

[0017] Given a scale factor τ, the signal features after variable mode decomposition are coarse-grained by the scale factor τ;

[0018] Calculate the permutation entropy of the coarse-grained sequence corresponding to each τ, and then calculate the average value to obtain a stable permutation entropy value.

[0019] In some optional implementation schemes, the fast spectral kurtosis algorithm is used to obtain the center frequency corresponding to the maximum kurtosis value of the signal, and then a plurality of BIMF component layer-by-layer data sets are screened according to the center frequency value of each component, and a feature vector set Q is constructed by permutation entropy, including:

[0020] The maximum kurtosis value K of the vibration, temperature and pressure signals of ship equipment is solved by using the fast spectral kurtosis algorithm max The corresponding center frequency f ω and bandwidth B w The interval of

[0021] The center frequency f of BIMF is i and the maximum kurtosis value K max The corresponding center frequency f ω Match and filter out f i Approximate f ω The corresponding BIMFs are defined as a feature vector set Q.

[0022] In some optional embodiments, the dimensionality reduction processing of the feature vector set Q by using the local tangent space arrangement to remove redundant and noise features to obtain the feature vector set T includes:

[0023] Get the local neighborhood matrix of the eigenvector set Q;

[0024] Use the local tangent space permutation algorithm to solve the d-dimensional affine subspace approximation of the points in the local neighborhood matrix, solve the optimal solution, and obtain the local coordinate system;

[0025] The local coordinates are globally arranged by the local coordinate system to obtain the optimal solution of the error minimization formula, and the global coordinates T are constructed as the feature vector set T.

[0026] In some optional implementation schemes, the construction of a dual DBN model series connection framework includes:

[0027] The dual DBN series model framework is built by stacking restricted Boltzmann machines, and the ivy algorithm is used to optimize the parameters of the dual DBN series model. The input of the DBN1 model is the feature vector set T, and the input of the DBN2 model is the output of the DBN1 model and the feature vector set T.

[0028] In some optional implementation schemes, the Ivy algorithm is used to optimize the parameters of the dual DBN series model, and the optimized parameters include the number of hidden layer neurons, learning rate, momentum, and training data input batch size.

[0029] In some optional implementation schemes, the establishment of an observer model, obtaining a health baseline of the ship equipment, evaluating and identifying the degradation state of the ship equipment by calculating the Mahalanobis distance, and outputting an equipment degradation state evaluation report include:

[0030] A deep belief network based on deep learning is established according to the data characteristics and fault types. The network with the best performance is used as a fault classifier after training.

[0031] Inputting fault data into a fault classifier to determine the type of system fault;

[0032] The observer is used to obtain the residual of the data to be evaluated, and the features are extracted to form a feature sequence. Then, the distance measurement is carried out to calculate the Mahalanobis distance between the current equipment degradation state and the healthy baseline and normalize it to the health degree. The fault data also needs to be input into the working condition identifier to determine the working condition type.

[0033] The system's equipment degradation status is comprehensively analyzed by health status, fault type and operating condition type, and an evaluation report is output.

[0034] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0035] It can collect multi-source monitoring information of equipment and quickly output detailed and accurate equipment degradation status assessment reports, providing a reliable basis for the subsequent use and maintenance of key equipment;

[0036] Vibration, temperature and pressure multi-source data are closely related to the equipment degradation status. The use of multi-source signals can more accurately and comprehensively reflect the equipment status. The features between data are complementary and have stronger anti-interference ability. The deep features mined can improve the accuracy of equipment degradation status assessment.

[0037] The proposed series dual DBN model can simultaneously realize the identification of ship working conditions and equipment degradation status. In addition, taking the platform working condition identification results as input, the working condition information can be considered in the equipment degradation status identification, which increases the physical meaning and improves the identification accuracy.

[0038] The proposed equipment degradation status assessment method based on health baseline and Mahalanobis distance can realize the equipment degradation status assessment under variable operating conditions and can better meet the actual needs of engineering. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1It is a schematic diagram of the steps of a method for identifying equipment degradation status based on multi-source fusion features and a series DBN model provided by an embodiment of the present invention;

[0040] Figure 2 It is a flowchart of constructing a degradation state identification framework of a series double DBN model of a device degradation state identification method based on multi-source fusion features and a series DBN model provided by an embodiment of the present invention;

[0041] Figure 3 It is a schematic diagram of a DBN model of a method for identifying a device degradation state based on multi-source fusion features and a series DBN model provided by an embodiment of the present invention;

[0042] Figure 4 It is a DBN model training flow chart of a method for identifying equipment degradation states based on multi-source fusion features and a series DBN model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solutions 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. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0044] like Figures 1 to 4 As shown, the ship equipment degradation state identification method based on multi-source fusion features and tandem DBN model of the present invention is as follows: Figure 1 As shown, the following steps are included:

[0045] Step S1: Collect vibration, temperature and pressure signals of ship equipment through acceleration sensors, thermocouples, force sensors and data acquisition systems.

[0046] Step S2: Perform time-frequency analysis on the collected signal to extract signal features, mainly including 13 time-domain features and 4 frequency-domain features;

[0047] Among them, in step S2, time-frequency features are extracted for the obtained multi-source signals, including 13 time domain features, namely, mean, absolute mean, peak, peak-to-peak value, root mean square value, root amplitude, standard deviation, waveform factor, peak factor, impulse factor, kurtosis, skewness and peak state, and 4 frequency domain features, namely, average frequency, center frequency, root mean square frequency and root variance frequency.

[0048] Step S3: Perform variable mode decomposition on the signal to obtain K two-dimensional intrinsic mode function components, and extract the center frequency value f of each component i ;

[0049] In step S3, the signal is subjected to variational mode decomposition, including updating the modal component U in the formula according to the decomposition process. k , component frequency center ω k , Lagrangian operators λ and n, output K two-dimensional intrinsic mode function (bandwidth-limited Inherent mode functions, BIMF) components, and extract the center frequency value f of each component i .

[0050] Step S4: using improved multi-scale permutation entropy to perform feature analysis, and obtaining a stable permutation entropy value according to the scale factor τ;

[0051] Obtaining a stable permutation entropy value is achieved through step S41.

[0052] Step S41: Given a scale factor τ, the original vibration signal sequence after the variable mode decomposition in step S3 is coarse-grained according to the following formula;

[0053]

[0054] Step S42: Calculate the permutation entropy of the coarse-grained sequence corresponding to each τ, and then calculate the average value to obtain a stable permutation entropy value. The permutation entropy is defined by the following formula:

[0055]

[0056] Step S5: Use the fast spectral kurtosis algorithm to obtain the maximum kurtosis value K of the signal max The corresponding center frequency f ω and bandwidth B w In the interval, several BIMF component layer-by-layer data sets are selected according to the center frequency, and the feature vector set Q is constructed by permutation entropy;

[0057] Among them, in terms of fault diagnosis and status assessment, the selection of bandwidth mainly needs to include the third harmonic of the characteristic signal.

[0058] In this step, feature screening needs to be performed, which specifically includes steps S51-S53.

[0059] Step S51: Use the fast spectral kurtosis algorithm to solve the maximum kurtosis value K of the measured equipment vibration, temperature and pressure signals max The corresponding center frequency f ω and bandwidth B w The kurtosis value of all frequency bands is solved according to the following formula.

[0060]

[0061] During this process, the model outputs the splash category and location information. The location information includes the center coordinates and its pixel width and height. At the same time, the pixel area difference of the same splash in adjacent images can also be obtained to prepare for determining whether it is the same splash.

[0062] Step S52: Set the center frequency f of BIMF i , and the maximum kurtosis value K of the signal max The corresponding center frequency f ω Match and filter out f i Approximate f ω The corresponding BIMFs are defined as a set Q.

[0063] Step S6: Using local tangent space arrangement to perform dimensionality reduction processing on the vector set Q, remove redundant and noise features, and obtain a feature vector set T;

[0064] Among them, in step S6, the neighborhood of the sample points of the data set can be used as the calculation basis in the original high-dimensional space to form a reasonable neighborhood matrix, and this matrix is ​​used to calculate the matrix tangent space that can contain most of the set characteristics of the high-dimensional data. Feature screening is required, which specifically includes steps S61-S63:

[0065] Step S61: Obtain the local neighborhood matrix of the feature vector set T, about the sample point x i , x i ∈R m , i = 1, 2, ..., n, define the neighborhood matrix as X i =[x i1 ,x i2 ,...,x ik ], is the point x i and x i It consists of k-1 nearby neighboring points;

[0066] Step S62: local linear fitting, using the local tangent space arrangement algorithm to solve the points in the d-dimensional affine subspace approximation neighborhood matrix, solve the optimal solution, and obtain the local coordinate system;

[0067] Furthermore, LTSA is used to solve the d-dimensional affine subspace approximation X i The points in are as follows:

[0068]

[0069] Q is an m×d-order orthogonal matrix, and the local coordinate system is defined as θ=[θ1, θ2, ..., θ k ] Define the neighborhood matrix X i of is the neighborhood center matrix The singular value decomposition of Q i ∈Rm×n ,∑i∈R m×k , V i ∈R k×k . Solve the optimal solution and obtain the local coordinate system as follows:

[0070] Step S63: Globally arrange the local coordinates to obtain the optimal solution of the error minimization formula and construct the global coordinates T.

[0071] Among them, θ obtained by the LTSA algorithm in the low-dimensional feature space contains most of the local geometric structures, and the global coordinate τ ij conform to: The optimal solution of the error minimization formula is obtained as follows: According to the calculated eigenvector corresponding to the minimum d+1th eigenvalue of the permutation matrix θ: u1,u2,…,u d+1 Finally, we get the optimal solution of E(T) and construct the global coordinates T = [u1,u2,…,u d+1 ].

[0072] Step S7: Build a dual DBN model series framework, use the Ivy algorithm to optimize model parameters, first train DBN1 with data set T to realize ship working condition recognition, and combine the working condition recognition results with data set T to train DBN2 to realize ship equipment degradation state recognition;

[0073] In step S7, Figure 2 and Figure 3 As shown, a dual DBN series model framework is built by stacking restricted Boltzmann machines, and the ivy algorithm is used to optimize the model parameters. The input of the DBN1 model is the data set T, and the input of the DBN2 model is the output of the DBN1 model and the data set T. After training and testing, the DBN1 and DBN2 models are used for ship operating condition identification and ship equipment degradation state identification.

[0074] like Figure 4 As shown, the ivy algorithm in this step simulates the growth pattern of ivy plants by coordinating orderly population growth and the diffusion and evolution of ivy plants; the growth rate of ivy plants is modeled through differential equations and data-intensive experimental processes, and the algorithm uses the knowledge of nearby ivy plants to determine the growth direction and self-improve by selecting the nearest and most important neighbors. It mainly includes steps S71-S75:

[0075] Step S71: random initialization;

[0076] Step S72: Coordinated and orderly species growth, assuming that the growth rate of the ivy plant is G v is a function of time given by a differential equation;

[0077] Step S73: Establish an equation based on the sunlight growth law;

[0078] Step S74: Propagation and evolution, after the stage in which species i roams globally through the search space to the nearest and most important neighbor ii, there is a stage in which member Ii tries to directly follow the best member IBest of the entire population and search for a better optimal solution around member IBest;

[0079] Step S75: Survivor selection, simulating the two alternating stages in the life of ivy, climbing and expansion; the following decision method is used, when the objective function value f(Ii) of member Ii is less than β*f(Ibest), β=(2+rand) / 2. The ivy tree begins to expand the width of branches and leaves; otherwise, the ivy will grow upward and climb.

[0080] To optimize the DBN model, it is necessary to determine the number of hidden layers and the number of input and output neurons of the DBN model, and use the Ivy algorithm to optimize the DBN model parameters. The optimized parameters include the number of hidden layer neurons, learning rate, momentum, and training data input batch size.

[0081] Step S8: Establish an observer model, obtain the health baseline of the equipment, evaluate the equipment degradation state by calculating the Mahalanobis distance, and output an equipment degradation state evaluation report.

[0082] In this step, a health baseline and Mahalanobis space are constructed, which specifically includes steps S81-S84.

[0083] Step S81: Establish a deep belief network based on deep learning according to data characteristics and fault types, and use the network with the best performance as a fault classifier after training;

[0084] Step S82: input the collected fault data into a fault classifier to determine the system fault type;

[0085] Step S83: Use the observer to obtain the residual of the data to be evaluated, and extract the features to form a feature sequence, then carry out distance measurement, calculate the Mahalanobis distance between the current state and the health baseline and normalize it to the health degree (α CV ), the fault data also needs to be input into the working condition identifier to determine the working condition type;

[0086] The state identified in step S7 is a qualitative state assessment, and the state assessment in step S8 is mainly a quantitative state assessment.

[0087] Step S84: Comprehensively analyze the system's equipment degradation status by health, fault type, and operating condition type, and output an evaluation report.

[0088] The calculation method of Mahalanobis distance is as follows:

[0089] 1) Calculate the mean of each vector as follows:

[0090]

[0091] 2) Calculate the standard deviation of each vector as follows:

[0092]

[0093] 3) Orthogonalize the eigenvectors to obtain Z ij , and find its transposed matrix Z T , as shown below:

[0094]

[0095] 4) Calculate the correlation matrix A of the orthogonal matrix, where each element

[0096]

[0097] 5) The Mahalanobis distance is:

[0098] d MD,j =Z T A -1 Z

[0099] The larger the distance, the farther the feature point to be evaluated is from the healthy baseline, and the more serious the system performance degradation is; the smaller the Mahalanobis distance, the closer the feature point to be evaluated is from the healthy baseline, and the better the system performance is. In order to more intuitively describe the performance of the system, the Mahalanobis distance is normalized to the health degree, as shown below:

[0100] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, and two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0101] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying degradation status of ship equipment based on multi-source fusion features and tandem DBN model, characterized in that: include: Collect vibration, temperature and pressure signals of ship equipment, perform time-frequency analysis on the collected vibration, temperature and pressure signals, and extract signal features; Perform variable mode decomposition on the extracted signal features to obtain two-dimensional intrinsic mode function components, and extract the center frequency value of each component; The improved multi-scale permutation entropy is used to analyze the signal characteristics after variable mode decomposition, and a stable permutation entropy value is obtained according to the scale factor. The fast spectral kurtosis algorithm is used to obtain the center frequency corresponding to the maximum kurtosis value of the signal, and then several BIMF component layer-by-layer data sets are selected according to the center frequency value of each component, and the feature vector set Q is constructed by permutation entropy; The feature vector set Q is reduced in dimension by using local tangent space arrangement to remove redundant and noisy features, and the feature vector set T is obtained; Build a dual DBN model series framework, optimize the dual DBN model parameters, use feature vector set T to train DBN1 to realize ship working condition identification, and combine the working condition identification results with feature vector set T to train DBN2 to realize ship equipment degradation state identification; An observer model is established to obtain the health baseline of ship equipment. The degradation state of ship equipment is evaluated and identified by calculating the Mahalanobis distance between the current degradation state and the health baseline, and an equipment degradation state evaluation report is output.

2. The method according to claim 1, characterized in that The signal characteristics include time-frequency characteristics and frequency domain characteristics. The time-frequency characteristics include mean, absolute mean, peak, peak-to-peak value, root mean square value, root amplitude, standard deviation, waveform factor, peak factor, pulse factor, kurtosis, skewness and peak state; the frequency domain characteristics include average frequency, center frequency, root mean square frequency and root square deviation frequency.

3. The method according to claim 1 or 2, characterized in that: The step of performing variable mode decomposition on the extracted signal features to obtain two-dimensional intrinsic mode function components and extracting the center frequency value of each component includes: Perform variable mode decomposition on the extracted signal features and update the modal component U k , component frequency center ω k , Lagrangian operators λ and n, output K two-dimensional intrinsic mode function BIMF components, and extract the center frequency value f of each BIMF component i .

4. The method according to claim 3, characterized in that The improved multi-scale permutation entropy is used to perform signal feature analysis, and a stable permutation entropy value is obtained according to the scale factor, including: Given a scale factor τ, the signal features after variable mode decomposition are coarse-grained by the scale factor τ; Calculate the permutation entropy of the coarse-grained sequence corresponding to each τ, and then calculate the average value to obtain a stable permutation entropy value.

5. The method according to claim 4, characterized in that The fast spectral kurtosis algorithm is used to obtain the center frequency corresponding to the maximum kurtosis value of the signal, and then several BIMF component layer-by-layer data sets are selected according to the center frequency value of each component, and a feature vector set Q is constructed by permutation entropy, including: The maximum kurtosis value K of the vibration, temperature and pressure signals of ship equipment is solved by using the fast spectral kurtosis algorithm max The corresponding center frequency f ω and bandwidth B w The interval of The center frequency f of BIMF is i and the maximum kurtosis value K max The corresponding center frequency f ω Match and filter out f i Approximate f ω The corresponding BIMFs are defined as a feature vector set Q.

6. The method according to claim 5, characterized in that The local tangent space arrangement is used to perform dimensionality reduction processing on the feature vector set Q, remove redundant and noise features, and obtain the feature vector set T, including: Get the local neighborhood matrix of the eigenvector set Q; Use the local tangent space permutation algorithm to solve the d-dimensional affine subspace approximation of the points in the local neighborhood matrix, solve the optimal solution, and obtain the local coordinate system; The local coordinates are globally arranged by the local coordinate system to obtain the optimal solution of the error minimization formula, and the global coordinates T are constructed as the feature vector set T.

7. The method according to claim 6, characterized in that The dual DBN model series connection framework is constructed, including: The dual DBN series model framework is built by stacking restricted Boltzmann machines, and the ivy algorithm is used to optimize the parameters of the dual DBN series model. The input of the DBN1 model is the feature vector set T, and the input of the DBN2 model is the output of the DBN1 model and the feature vector set T.

8. The method according to claim 7, characterized in that The Ivy algorithm is used to optimize the parameters of the dual DBN series model. The optimized parameters include the number of hidden layer neurons, learning rate, momentum, and training data input batch size.

9. The method according to claim 8, characterized in that The observer model is established to obtain the health baseline of the ship equipment, the degradation state of the ship equipment is identified by calculating the Mahalanobis distance evaluation, and the equipment degradation state evaluation report is output, including: A deep belief network based on deep learning is established according to the data characteristics and fault types. The network with the best performance is used as a fault classifier after training. Inputting fault data into a fault classifier to determine the type of system fault; The observer is used to obtain the residual of the data to be evaluated, and the features are extracted to form a feature sequence. Then, the distance measurement is carried out to calculate the Mahalanobis distance between the current equipment degradation state and the healthy baseline and normalize it to the health degree. The fault data also needs to be input into the working condition identifier to determine the working condition type. The system's equipment degradation status is comprehensively analyzed by health status, fault type and operating condition type, and an evaluation report is output.

Images