Fault prediction method and device for ship multi-level fusion model and electronic equipment

Abstract

The present application relates to a kind of ship multi-level fusion model fault prediction method, device and electronic equipment, its method includes: to the original vibration signal data of ship propulsion shafting is preprocessed, obtains enhanced vibration signal data;The multi-dimensional feature extraction of the enhanced vibration signal data is obtained multidimensional feature data;The multidimensional feature data is reduced dimensionality and obtains dimensionality reduction feature data;The dimensionality reduction feature data is input into artificial neural network model and is trained, obtains ship propulsion shafting fault classification model;The dimensionality reduction feature data of the real-time vibration signal of target ship propulsion shafting is input into the ship propulsion shafting fault classification model, obtains the fault type of target ship.The present application carries out multi-level extraction feature to the original vibration signal data of ship propulsion shafting, obtains multiple types of features, so that the recognition of ship fault prediction is high, and prediction accuracy is high.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electrical engineering, and particularly relates to a fault prediction method and device of a ship multi-level fusion model and an electronic device. BACKGROUND

[0002] With the development of science and technology and industrial production, such as large ships and other production and life and national defense and armed essential elements, are being complex, running automation, machine intelligence, etc. high-speed development. As a kind of strong mobility and independent operation equipment, the reliability and stability of the whole machine are put forward more stringent requirements. Although the reliability of the ship power device has been improved, due to the interference of external factors such as weather, sea conditions, and the environment in the cabin is relatively poor, under the action of internal factors such as vibration, noise, temperature and humidity, the ship power device will still have serious failure, so the state monitoring and fault prediction become an important work in the ship operation.

[0003] The ship power device is the core device of the ship, and its running state directly affects the stability of the whole ship navigation. As an important part of the ship power device, the ship propulsion shafting will be affected by various impact forces and periodic excitation forces encountered during the ship navigation. The causes of these forces include: non-standard design of the transmission structure of the shafting, unreasonable installation operation method; a certain end load is generated due to the gravity of the propeller; excitation generated by unstable torque output of the main engine; bending or stress deformation of the shafting due to its own weight, etc. A series of adverse effects may make the state of the ship propulsion shafting more complex and changeable during operation, and abnormal phenomena that may cause the shafting vibration of the ship will inevitably occur. When the amplitude exceeds the gap between the shaft diameter and the bearing, collision and dry friction will occur between them, which is called collision and friction. When collision and friction occur, the working temperature of the shafting rises rapidly, and even causes the shafting failure, which may cause adverse consequences, and seriously affects the navigation performance and safe operation of the ship.

[0004] The current mechanical fault prediction and identification technology is mainly an intelligent fault prediction method based on signal acquisition, feature extraction, fault identification and prediction. This kind of method uses the sensor network distributed in the mechanical system to obtain the effective detection value in the running condition in real time, extracts the characteristic parameters capable of judging the fault occurrence position and fault type of the mechanical system from the detection value, and further uses the parameters to predict the mechanical system fault occurrence in the next period. The fault prediction of the ship propulsion shafting is to use sensors to obtain and collect vibration signals, extract and analyze the vibration signals, and then use the established fault prediction model to identify the fault type. However, the existing fault prediction method of the ship propulsion shafting does not fully consider the harsh running environment of the ship propulsion shafting, the data is not ideal, and the feature level is single, which leads to the problem that the strengthening of the original vibration signal is not paid attention to, and the identification of the mechanical fault prediction is low. SUMMARY

[0005] Therefore, it is necessary to provide a ship multi-level fusion model fault prediction method and device and electronic equipment to solve the problem that the harsh running environment of the ship propulsion shafting is not fully considered, the data is not ideal, the feature level is single, which leads to the problem that the strengthening of the original vibration signal is not paid attention to, and the identification of the mechanical fault prediction is low.

[0006] To solve the above problems, the present application provides a ship multi-level fusion model fault prediction method, comprising:

[0007] The original vibration signal data of the ship propulsion shafting is preprocessed to obtain the strengthened vibration signal data;

[0008] The multi-dimensional feature extraction is performed on the strengthened vibration signal data to obtain multi-dimensional feature data;

[0009] The feature dimension reduction is performed on the multi-dimensional feature data to obtain dimension-reduced feature data;

[0010] The dimension-reduced feature data is input into an artificial neural network model to obtain a ship propulsion shafting fault classification model;

[0011] The dimension-reduced feature data of the real-time vibration signal of the target ship propulsion shafting is input into the ship propulsion shafting fault classification model to obtain the fault type of the target ship.

[0012] In some possible implementation manners, the preprocessing of the original vibration signal data of the ship propulsion shafting to obtain the strengthened vibration signal data comprises:

[0013] The original vibration signal data of the ship propulsion shafting is processed to obtain the first vibration signal data;

[0014] The first original vibration signal data is subjected to sample missing expansion processing based on a generative adversarial network model to obtain second vibration signal data.

[0015] The second vibration signal data is subjected to noise reduction processing to obtain the reinforced vibration signal data.

[0016] In some possible implementation manners, the multi-dimensional feature extraction on the reinforced vibration signal data to obtain multi-dimensional feature data comprises:

[0017] The reinforced vibration signal data is subjected to hierarchical decomposition to obtain multi-level component vibration signal data.

[0018] Frequency domain feature data, scatter entropy feature data and time-frequency feature data are extracted from the multi-level component vibration signal data.

[0019] In some possible implementation manners, the extraction of time-frequency feature data from the multi-level component vibration signal data comprises:

[0020] The multi-level component vibration signal data is subjected to improved variational mode decomposition based on an improved particle swarm algorithm to extract time-frequency feature data.

[0021] In some possible implementation manners, the improved variational mode decomposition of the multi-level component vibration signal data based on the improved particle swarm algorithm to extract time-frequency feature data comprises:

[0022] The penalty factor and the number of signal components in the variational mode decomposition are optimized based on the improved particle swarm algorithm.

[0023] The multi-level component vibration signal data is decomposed by using the optimized penalty factor and the number of signal components to extract time-frequency feature data.

[0024] In some possible implementation manners, the improvement step of the improved particle swarm algorithm comprises: optimizing parameters of the particle swarm algorithm based on a compression factor algorithm, the parameters comprising an inertia weight and a particle velocity.

[0025] In some possible implementation manners, the frequency domain feature data comprises: a mean frequency, a center frequency, a root mean square frequency, a standard deviation frequency, a frequency domain amplitude skewness index and a frequency domain kurtosis; and the time-frequency feature data comprises energy ratio features and energy operator features.

[0026] In some possible implementation manners, the feature dimension reduction on the multi-dimensional feature data to obtain reduced dimension feature data comprises:

[0027] The multi-dimensional feature data is subjected to feature dimension reduction based on a Laplace score algorithm to obtain reduced dimension feature data.

[0028] In another aspect, the present application also provides a fault prediction device of a ship multi-level fusion model, comprising:

[0029] A signal enhancement unit is configured to preprocess original vibration signal data of a ship propulsion shafting to obtain enhanced vibration signal data.

[0030] A multi-dimensional feature unit is configured to perform multi-dimensional feature extraction on the enhanced vibration signal data to obtain multi-dimensional feature data.

[0031] A dimension reduction feature unit is configured to perform feature dimension reduction on the multi-dimensional feature data to obtain dimension-reduced feature data.

[0032] A classification model determination unit is configured to input the dimension-reduced feature data into an artificial neural network model for training to obtain a ship propulsion shafting fault classification model.

[0033] A fault type determination unit is configured to input dimension-reduced feature data of real-time vibration signals of a target ship propulsion shafting into the ship propulsion shafting fault classification model to obtain a fault type of the target ship.

[0034] In another aspect, the present application also provides an electronic device comprising a memory and a processor, wherein,

[0035] The memory is configured to store a program.

[0036] The processor is coupled to the memory and is configured to execute the program stored in the memory to implement steps of the fault prediction method of the ship multi-level fusion model according to any one of the above implementation manners.

[0037] The fault prediction method of the ship multi-level fusion model provided by the present application first preprocesses original vibration signal data of a ship propulsion shafting to obtain enhanced vibration signal data, then performs multi-dimensional feature extraction on the enhanced vibration signal data and performs dimension reduction processing to obtain dimension-reduced feature data, further inputs the dimension-reduced feature data into an artificial neural network model for training to obtain a ship propulsion shafting fault classification model, and finally inputs dimension-reduced feature data of real-time vibration signals of a target ship propulsion shafting into the ship propulsion shafting fault classification model to obtain a fault type of the target ship. The present application performs multi-dimensional feature extraction on the preprocessed original vibration signal, fully considers the environment in which the ship propulsion shafting is located, and thus improves the accuracy of ship propulsion shafting fault prediction. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 A method flowchart of an embodiment of the fault prediction method of the ship multi-level fusion model provided by the present application;

[0039] Figure 2 An embodiment structure schematic diagram of a fault prediction device of a ship multi-level fusion model provided by the present application is shown in the figure.

[0040] Figure 3 An embodiment structure schematic diagram of an electronic device provided by the present application is shown in the figure. DETAILED DESCRIPTION

[0041] The preferred embodiments of the present application will be described in detail below with reference to the drawings, wherein the drawings constitute a part of this application and are used to explain the principles of the embodiments of the present application, but are not used to limit the scope of the present application.

[0042] Figure 1 An embodiment flow schematic diagram of a fault prediction method of a ship multi-level fusion model provided by the present application is shown in the figure, and the fault prediction method of the ship multi-level fusion model comprises: Figure 1

[0043] S101, preprocessing original vibration signal data of a ship propulsion shafting to obtain strengthened vibration signal data;

[0044] S102, performing multi-dimensional feature extraction on the strengthened vibration signal data to obtain multi-dimensional feature data;

[0045] S103, performing feature dimension reduction on the multi-dimensional feature data to obtain reduced dimension feature data;

[0046] S104, inputting the reduced dimension feature data into an artificial neural network model for training to obtain a ship propulsion shafting fault classification model;

[0047] S105, inputting reduced dimension feature data of real-time vibration signals of a target ship propulsion shafting into the ship propulsion shafting fault classification model to obtain a fault type of the target ship.

[0048] Compared with the prior art, the embodiment provides a fault prediction method of a ship multi-level fusion model, which firstly preprocesses original vibration signal data of a ship propulsion shafting to obtain strengthened vibration signal data, then performs multi-dimensional feature extraction on the strengthened vibration signal data and performs dimension reduction processing to obtain reduced dimension feature data, further inputs the reduced dimension feature data into an artificial neural network model for training to obtain a ship propulsion shafting fault classification model, and finally inputs reduced dimension feature data of real-time vibration signals of a target ship propulsion shafting into the ship propulsion shafting fault classification model to obtain a fault type of the target ship. The present application performs multi-dimensional feature extraction on the preprocessed original vibration signal, fully considers the environment where the ship propulsion shafting is located, and thus improves the accuracy of ship propulsion shafting fault prediction.

[0049] ​In some embodiments of the present application, in step S101, the original vibration signal data of the ship propulsion shafting is preprocessed to obtain the enhanced vibration signal data, including:

[0050] The original vibration signal data of the ship propulsion shafting is processed for missing values to obtain first vibration signal data;

[0051] The first original vibration signal data is processed for sample missing expansion based on the generative adversarial network model to obtain second vibration signal data;

[0052] The second vibration signal data is processed for noise reduction to obtain the enhanced vibration signal data.

[0053] In specific embodiments of the present application, in step S101, the original vibration signal data of the ship propulsion shafting is preprocessed to obtain the enhanced vibration signal data, and the specific steps are as follows:

[0054] Step one: obtain the original vibration data of the ship propulsion shafting, and the method of obtaining the original vibration data of the ship propulsion shafting is to use multiple acceleration sensors to collect the original vibration signal data of the ship propulsion shafting during operation. Provide data support for subsequent fault prediction.

[0055] Step two: process the obtained original vibration data of the ship propulsion shafting for missing values to obtain first original vibration signal data, which can strengthen non-ideal original vibration signal data, solve the problems of partial data missing, insufficient samples, and large noise, and provide support for subsequent feature extraction and fault prediction. The missing value processing method is hot card filling method. Find an object most similar to it in the complete data, fill the current value with the most similar value, specifically using fixed parameters, and only referring to one KNN.

[0056] Step three: the first original vibration signal data in step two is processed for sample missing expansion, and the expansion processing method is to use the generative adversarial network UGAN to ensure the diversity of generated samples.

[0057] The energy objective function of the discriminator D is:

[0058] f D =D(x)+[m-D(G(z,y))] + =||Dec(Enc(x))-x||+[m-||Dec(Enc(G(z,y)))-G(z,y)||] +

[0059] Where [·] + =max(0,·), Enc represents encoding operation, and Dec represents decoding operation.

[0060] The loss function of the generator G is:

[0061] L G =||D(G(y))||=||Dec(Enc(G(z)))-G(z)||

[0062] Step four: the second vibration signal data is denoised to obtain the reinforced vibration signal data, and the denoising method is singular value decomposition.

[0063] (1) Based on the phase space reconstruction theory, the time series signal is reconstructed into a matrix A:

[0064]

[0065] Wherein, L is selected as half of the signal length. And the time series signal is subjected to Fourier transform, so as to determine the number of main frequencies n.

[0066] (2) Singular value decomposition is performed on the reconstructed matrix A, 2n is taken as the order of effective rank, and other singular values are set to 0, so as to obtain a new reconstructed matrix B.

[0067] (3) The corresponding elements in B are added and averaged to obtain the denoised reinforced vibration signal data.

[0068] In some embodiments of the present application, in step S102, the multi-dimensional feature extraction of the reinforced vibration signal data obtains multi-dimensional feature data, including:

[0069] The reinforced vibration signal data is hierarchically decomposed to obtain multi-level component vibration signal data;

[0070] Frequency domain feature data, scatter entropy feature data and time-frequency feature data are extracted from the multi-level component vibration signal data.

[0071] In some embodiments of the present application, the time-frequency feature data is extracted from the multi-level component vibration signal data, including:

[0072] The multi-level component vibration signal data is subjected to improved variational mode decomposition based on an improved particle swarm algorithm to extract time-frequency feature data.

[0073] In some embodiments of the present application, the time-frequency feature data is extracted from the multi-level component vibration signal data based on the improved particle swarm algorithm, including:

[0074] The number of signal components and the penalty factor in the variational mode decomposition are optimized based on the improved particle swarm algorithm.

[0075] The multi-level component vibration signal data is decomposed by the optimized penalty factor and the number of signal components, and time-frequency feature data is extracted.

[0076] In some embodiments of the present application, the improvement step of the improved particle swarm algorithm comprises:

[0077] The parameters of the particle swarm algorithm are optimized based on the compression factor algorithm, including the inertia weight and the particle velocity. In some embodiments of the present application, the frequency domain feature data comprises the mean frequency, the center frequency, the root mean square frequency, the standard deviation frequency, the frequency domain amplitude skewness index and the frequency domain frequency kurtosis, and the time-frequency feature data comprises the energy ratio feature and the energy operator feature.

[0078] In specific embodiments of the present application, in step S102, the multi-dimensional feature extraction is performed on the reinforced vibration signal data to obtain multi-dimensional feature data, and the specific steps are as follows:

[0079] Step four: performing hierarchical decomposition on the reinforced vibration signal data to obtain multi-level component reinforced vibration signal data; obtaining the multi-level component reinforced vibration signal data, i.e. the ship multi-level fusion model, and the specific steps are as follows:

[0080] Step four one: defining an average operator Q0and a difference operator Q1as follows:

[0081]

[0082] In the formula, 2 n-1 is the length of the operator, n is a positive integer, Q0(x) and Q1(x) represent the low-frequency component and the high-frequency component of the original time series in the first layer decomposition, respectively.

[0083] Step four two: in order to describe the hierarchical analysis of the signal, when j=0 or 1, the matrix form of the kth layer operator X is defined as follows:

[0084]

[0085] Step four three: in order to obtain the hierarchical components X k,e of each layer in the hierarchical decomposition process, the above-defined operators need to be repeatedly used, and a one-dimensional vector [γ1,γ2,…,γ k ] and an integer value are also defined, where {γ p ,p=1,2,…,k}∈{0,1} represents the average or difference operator of the pth layer. Accordingly, the hierarchical component of the e th node of the kth layer can be represented as:

[0086]

[0087] That is, the ship multi-level fusion model is obtained.

[0088] Step five: 6 frequency domain features and scatter entropy features are extracted from the multi-level component enhanced vibration signal data obtained in step four; time-frequency domain features are extracted from the multi-level component enhanced vibration signal data obtained in step four, 5 energy ratios of effective components are obtained as time-frequency features by improving the variational mode decomposition on the multi-level component enhanced vibration signal data, in addition, an energy operator is extracted from the vibration signal, which is also a time-frequency domain feature. The calculation expression of the energy operator is:

[0089]

[0090] In the formula, r(n) represents the Teager energy difference signal, r(n)=x 2 (n)-x(n-1)x(n+1), represents the mean value of r(n).

[0091] In step five, the scatter entropy feature acquisition process is:

[0092] Step five one: the time series X=[X j ,j=1,2,…,N] is mapped to y through a normal distribution function formula, y={y j}, y j ∈(0,1):

[0093] Where u represents expectation, and σ 2 represents variance.

[0094] Step five two: y is mapped to the range of [1,2,…,c] by linear transformation:

[0095] R is the rounding function, and c is the number of categories.

[0096] Step five three: the embedding vector

[0097] m is the embedding dimension, and d is the time delay.

[0098] Step five four: calculate the scatter pattern:

[0099] If , then The corresponding scatter pattern is πv0v1]v m-1 ,v=1,2,],c. Each pattern is composed of c data, and each value has m values, so there are c m corresponding scatter patterns.

[0100] Step five: find the probability of each spread pattern:

[0101] where represents the embedding vector maps to the number of spread patterns.

[0102] Step five six: define the spread entropy value of the original signal X by the definition of Shannon entropy:

[0103]

[0104] In step five, by improving the vibration signal data of multi-level components, the improved variation modal decomposition is carried out, the fitness function is constructed for the two key parameters of modal number and penalty factor, and the improved particle swarm optimization algorithm is used to get the best parameter combination under different signals. The improved particle swarm optimization algorithm is improved in inertia weight and particle swarm speed, as follows:

[0105] Firstly, the most important inertia weight ω in PSO algorithm is optimized and selected by using adaptive inertia weight strategy, because ω is closely related to the number of iterations and the fitness of particles. Through adaptive method, the inertia weight can be automatically adjusted, and then the algorithm converges to the optimal solution quickly.

[0106] In the adaptive inertia weight strategy, if the minimum value of the objective function is solved, the smaller the fitness is, the closer to the optimal solution it is, and a smaller weight is needed at this time to facilitate local search; otherwise, a larger weight should be taken to improve the global search ability of particles. The adjustment method is as follows:

[0107]

[0108] where:

[0109] 1)ω min and ω max are the minimum and maximum values of the inertia coefficient given in advance, ω min = 0.4, ω max = 0.9;

[0110] 2) That is, the average value of the whole population fitness calculated at the kth iteration;

[0111] 3) That is, the minimum fitness in the whole population at the kth iteration.

[0112] (2) Compression factor method

[0113] The shrinkage factor is used to optimize the PSO model, which can ensure that the particle swarm algorithm has strong convergence when solving, and weaken the limitation of the speed size. The particle velocity calculation expression is:

[0114]

[0115] The particle velocity is obtained, so as to determine the two parameters of the mode number K and the penalty factor alpha.

[0116] By using the adaptive inertia weight strategy and the compression factor method, the problems of low search efficiency and easy falling into local optimal solution caused by the fixed value of the key parameter in the particle swarm algorithm can be solved.

[0117] In some embodiments of the present application, in step S103, the multi-dimensional feature data is reduced in dimension to obtain reduced dimension feature data, comprising:

[0118] The multi-dimensional feature data is reduced in dimension based on the Laplace score algorithm to obtain reduced dimension feature data.

[0119] In specific embodiments of the present application, the first step: 13 features are obtained from step five, the 13 features include 6 frequency domain features, a dispersion entropy feature, and 6 time-frequency features, wherein the 6 time-frequency features include 5 energy ratio features and 1 energy operator feature, and the 13 features are normalized;

[0120] The second step: although the above feature set can reflect different information of the vibration signal and the system fault, the sensitivity of different fault feature parameters is different, and there is a part of irrelevant or redundant phenomenon. Before the feature information is input into the classifier, a screening and dimension reduction method is needed to obtain reduced dimension feature data, that is, sensitive features, so as to improve the classification performance and avoid the curse of dimensionality. The feature selection algorithm selects the Laplace score dimension reduction method, and the specific evaluation process is as follows:

[0121] Input the training sample feature matrix Where m is the sample number, n is the feature dimension, f ri is the rth feature of the ith sample.

[0122] The third step is to construct a neighbor graph G, wherein G has m sample points, x i corresponding to the ith node. If x i and x j are "neighbors", the two points are connected, and x i and x j are defined as neighbor nodes; otherwise, the two points are not connected.

[0123] The weight matrix S is definedij for:

[0124]

[0125] Where t is a suitable constant.

[0126] Step 4: Define L r For the r-th feature f r Laplace score, L r The calculation method is as follows:

[0127]

[0128] In the formula, f r =[f r1 ,f r2 ,…f rm ] T , D=diag(SI), I=[1,1,…1] T L = DS, where matrix L is the Laplace matrix of the neighborhood graph G.

[0129] in:

[0130]

[0131] Where var(f) r f is the r-th feature r The variance, standardized f r We can obtain:

[0132]

[0133] Furthermore, the Laplacian score of the r-th feature can be calculated: σ = 0.1

[0134]

[0135] L r Sort the data in ascending order and output L sequentially. r Molecules (f) ri -f rj The smaller the value of f, the smaller the feature differences within the sample. The r-th feature f r The larger the variance, the greater the feature differences between samples, and the higher the separability. Therefore, the Laplace score L corresponding to the feature... r The smaller the value, the higher its importance.

[0136] In step S104, the dimensionality-reduced feature data is input into an artificial neural network model for training to obtain a ship propulsion shaft system fault classification model. The artificial neural network model is a probabilistic neural network (PNN) model, using a Gaussian function as the activation function, with an input vector dimension of 10. The number of neurons in the pattern layer varies depending on samples randomly drawn from all data, the number of neurons in the summation layer is 10, and the smoothing factor is 0.2.

[0137] To better implement the fault prediction method of a multi-level fusion model for ships according to the embodiments of the present invention, based on the fault prediction method of a multi-level fusion model for ships, correspondingly, as follows: Figure 2 As shown, this embodiment of the invention also provides a fault prediction device for a multi-level fusion model of a ship. A fault prediction device 200 for a multi-level fusion model of a ship includes:

[0138] The signal enhancement unit 201 is used to preprocess the original vibration signal data of the ship's propulsion shaft system to obtain enhanced vibration signal data.

[0139] Multidimensional feature unit 202 is used to extract multidimensional features from the enhanced vibration signal data to obtain multidimensional feature data;

[0140] The dimensionality reduction feature unit 203 is used to perform feature dimensionality reduction on the multidimensional feature data to obtain dimensionality-reduced feature data.

[0141] The fault classification determination unit 204 is used to input the dimensionality reduction feature data into the artificial neural network model for training to obtain a ship propulsion shaft system fault classification model.

[0142] The fault type determination unit 205 is used to input the dimensionality reduction feature data of the real-time vibration signal of the target ship's propulsion shaft system into the ship propulsion shaft system fault classification model to obtain the fault type of the target ship.

[0143] The fault prediction device 200 for a ship propulsion shafting provided in the above embodiments can realize the technical solution described in the above embodiment of the fault prediction method for a multi-level fusion model of a ship. The specific implementation principle of each module or unit can be found in the corresponding content in the above embodiment of the fault prediction method for a multi-level fusion model of a ship, which will not be repeated here.

[0144] like Figure 3 As shown, the present invention also provides an electronic device 300. The electronic device 300 includes a processor 301, a memory 302, and a display 303. Figure 3 Only some components of the electronic device 300 are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0145] In some embodiments, processor 301 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 302 or process data, such as a fault prediction method for a multi-level fusion model of a ship in this invention.

[0146] In some embodiments, processor 301 may be a single server or a group of servers. The server group may be centralized or distributed. In some embodiments, processor 301 may be local or remote. In some embodiments, processor 301 may be implemented on a cloud platform. In one embodiment, the cloud platform may include a private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, intranet, multi-cloud, etc., or any combination thereof.

[0147] In some embodiments, memory 302 may be an internal storage unit of electronic device 300, such as a hard disk or memory of electronic device 300. In other embodiments, memory 302 may also be an external storage device of electronic device 300, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on electronic device 300.

[0148] Furthermore, the memory 302 may include both internal storage units of the electronic device 300 and external storage devices. The memory 302 is used to store application software and various types of data installed on the electronic device 300.

[0149] In some embodiments, display 303 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 303 is used to display information from electronic device 300 and to display a visual user interface. Components 301-303 of electronic device 300 communicate with each other via a system bus.

[0150] In one embodiment, when processor 301 executes a fault prediction program for a multi-level fusion model of a ship stored in memory 302, the following steps can be implemented:

[0151] The raw vibration signal data of the ship's propulsion shaft system is preprocessed to obtain enhanced vibration signal data;

[0152] Multidimensional feature data is obtained by performing multidimensional feature extraction on the enhanced vibration signal data.

[0153] The multidimensional feature data is subjected to feature dimensionality reduction to obtain dimensionality-reduced feature data;

[0154] The reduced-dimensionality feature data is input into an artificial neural network model for training to obtain a ship propulsion shaft system fault classification model.

[0155] The dimensionality-reduced feature data of the real-time vibration signal of the target ship's propulsion shaft system is input into the ship propulsion shaft system fault classification model to obtain the fault type of the target ship.

[0156] It should be understood that when the processor 301 executes a fault prediction program for a multi-level fusion model of a ship in the memory 302, in addition to the functions mentioned above, it can also perform other functions, as can be found in the description of the corresponding method embodiments above.

[0157] Furthermore, this embodiment of the invention does not specifically limit the type of electronic device 300 mentioned. Electronic device 300 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the invention, electronic device 300 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0158] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0159] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting faults in a ship's propulsion shafting system, characterized in that, include: The raw vibration signal data of the ship's propulsion shaft system is preprocessed to obtain enhanced vibration signal data; Multidimensional feature data is obtained by performing multidimensional feature extraction on the enhanced vibration signal data. The multidimensional feature data is subjected to feature dimensionality reduction to obtain dimensionality-reduced feature data; The reduced-dimensionality feature data is input into an artificial neural network model for training to obtain a ship propulsion shaft system fault classification model. The dimensionality-reduced feature data of the real-time vibration signal of the target ship's propulsion shaft system is input into the ship propulsion shaft system fault classification model to obtain the fault type of the target ship; The step of extracting multidimensional feature data from the enhanced vibration signal data includes: The enhanced vibration signal data is decomposed hierarchically to obtain vibration signal data with multi-level components; Frequency domain feature data, scatter entropy feature data, and time-frequency feature data are extracted from the vibration signal data of the multi-level components. The extraction of time-frequency feature data from the multi-level component vibration signal data includes: An improved particle swarm optimization algorithm is used to perform improved variational mode decomposition on the vibration signal data of the multi-level components to extract time-frequency feature data. The improved particle swarm optimization algorithm is used to perform improved variational mode decomposition on the multi-level component vibration signal data to extract time-frequency feature data, including: The penalty factor and the number of signal components in the variational mode decomposition are optimized based on an improved particle swarm optimization algorithm. The vibration signal data of the multi-level components is decomposed by optimizing the penalty factor and the number of signal components, and time-frequency feature data is extracted. The improved particle swarm optimization algorithm includes the following steps: The parameters of the particle swarm optimization algorithm are optimized based on the compression factor algorithm, and the parameters include inertia weight and particle velocity.

2. The method for predicting faults in a ship propulsion shafting system according to claim 1, characterized in that, The raw vibration signal data of the ship's propulsion shafting is preprocessed to obtain enhanced vibration signal data, including: The first vibration signal data is obtained by processing the missing values ​​of the original vibration signal data of the ship's propulsion shaft system; The second vibration signal data is obtained by augmenting the first vibration signal data with missing samples based on a generative adversarial network model. The enhanced vibration signal data is obtained by performing noise reduction processing on the second vibration signal data.

3. The method for predicting faults in a ship propulsion shafting system according to claim 1, characterized in that, The frequency domain feature data includes: mean frequency, center frequency, root mean square frequency, standard deviation frequency, frequency domain amplitude skewness index, and frequency domain frequency kurtosis; the time-frequency feature data includes energy ratio feature and energy operator feature.

4. The method for predicting faults in a ship propulsion shafting system according to claim 1, characterized in that, The step of performing dimensionality reduction on the multidimensional feature data to obtain dimensionality-reduced feature data includes: The multidimensional feature data is reduced in dimensionality using the Laplace score algorithm to obtain dimensionality-reduced feature data.

5. A fault prediction device for a ship's propulsion shafting system, comprising: The signal enhancement unit is used to preprocess the original vibration signal data of the ship's propulsion shaft system to obtain enhanced vibration signal data. A multi-dimensional feature unit is used to extract multi-dimensional features from the enhanced vibration signal data to obtain multi-dimensional feature data. A dimensionality reduction feature unit is used to perform dimensionality reduction on the multidimensional feature data to obtain dimensionality-reduced feature data. The classification model determination unit is used to input the dimensionality reduction feature data into the artificial neural network model for training to obtain a ship propulsion shaft system fault classification model. The fault type determination unit is used to input the dimensionality reduction feature data of the real-time vibration signal of the target ship's propulsion shaft system into the ship propulsion shaft system fault classification model to obtain the fault type of the target ship. The step of extracting multidimensional feature data from the enhanced vibration signal data includes: The enhanced vibration signal data is decomposed hierarchically to obtain vibration signal data with multi-level components; Frequency domain feature data, scatter entropy feature data, and time-frequency feature data are extracted from the vibration signal data of the multi-level components. The extraction of time-frequency feature data from the multi-level component vibration signal data includes: An improved particle swarm optimization algorithm is used to perform improved variational mode decomposition on the vibration signal data of the multi-level components to extract time-frequency feature data. The improved particle swarm optimization algorithm is used to perform improved variational mode decomposition on the multi-level component vibration signal data to extract time-frequency feature data, including: The penalty factor and the number of signal components in the variational mode decomposition are optimized based on an improved particle swarm optimization algorithm. The vibration signal data of the multi-level components is decomposed by optimizing the penalty factor and the number of signal components, and time-frequency feature data is extracted. The improved particle swarm optimization algorithm includes the following steps: The parameters of the particle swarm optimization algorithm are optimized based on the compression factor algorithm, and the parameters include inertia weight and particle velocity.

6. An electronic device, characterized in that, Including memory and processor, among which, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the fault prediction method for a ship propulsion shafting system as described in any one of claims 1 to 4.

Images