Ship fault detection method, device and equipment and storage medium

Abstract

According to the ship fault detection method, the ship fault detection device, the ship fault detection equipment and the storage medium provided by the invention, the ship fault data is decomposed by adopting the VMD technology, and a complex signal is decomposed into a plurality of modal components, so that the signal decomposition precision and the result reliability can be effectively improved, and more accurate and representative fault features are extracted. Moreover, the global dependency capturing capability of the Transform model and the excellent classification performance of the SVM classifier are combined, and the ship fault data can be deeply analyzed and accurately classified, so that the accuracy and reliability of fault detection are improved.

Description

Technical Field

[0001] The present invention relates to the field of marine engineering technology, and in particular to a ship fault detection method, device, equipment and storage medium. Background Art

[0002] With the continuous development of ship technology and the increasing complexity of ship systems, ship fault detection and diagnosis have become a key link in ensuring navigation safety and improving ship operation efficiency. Ship failures often affect the normal operation of the ship's power system, power system and other key equipment. Therefore, timely and accurate detection and diagnosis of ship faults are necessary measures to ensure safe navigation of ships.

[0003] At present, traditional ship fault detection methods mostly rely on manual monitoring and judgment based on experience. This method is not only inefficient, but also has a high risk of misdiagnosis or missed diagnosis. With the continuous advancement of information technology, intelligent fault detection methods have gradually become the focus of research. Fault diagnosis methods based on sensor data, especially ship fault detection technologies using advanced algorithms such as machine learning and deep learning, have gradually attracted the attention of academia and industry.

[0004] However, existing machine learning-based fault detection methods still face many challenges, such as the complexity of fault data, the impact of signal noise, and the processing of high-dimensional feature space.

[0005] In summary, the problems existing in the prior art need to be solved urgently. Summary of the invention

[0006] The present invention provides a ship fault detection method, device, equipment and storage medium, which are used to solve the defects in the prior art and realize intelligent detection of ship faults.

[0007] The present invention provides a ship fault detection method, comprising: Obtain the data to be tested; Inputting the data to be detected into a trained fault detection model to obtain the ship fault type; The fault detection model is trained in the following way: Obtain ship fault data; Performing variational mode decomposition on the ship fault data to obtain fault characteristics; The fault characteristics are input into the Transformer-SVM model to perform fault diagnosis training to obtain the fault detection model.

[0008] According to a ship fault detection method provided by the present invention, the step of performing variational modal decomposition on the ship fault data to obtain fault characteristics specifically includes: Performing variational modal decomposition on the ship fault data to obtain modal components; According to the envelope entropy, the modal components are screened to obtain the target modal components; A fault feature is determined according to the target modal component.

[0009] According to a ship fault detection method provided by the present invention, the step of performing variational modal decomposition on the ship fault data to obtain modal components specifically includes: Determine the number of modal components and penalty factors through the GOA algorithm, wherein the number of modal components is the number of modal components obtained by variational modal decomposition, and the penalty factor is used to control the accuracy of variational modal decomposition and the bandwidth of each component; According to the number of modal components and the penalty factor, variational modal decomposition is performed on the ship fault data to obtain modal components.

[0010] According to a ship fault detection method provided by the present invention, the step of determining the number of modal components and the penalty factor by using the GOA algorithm specifically includes: Acquire a candidate solution population, the candidate solution population including a plurality of candidate solutions, the candidate solutions being used to characterize the number of modal components and the penalty factor; Calculating the fitness of each of the candidate solutions; updating the position of the candidate solution according to the random motion model; The candidate solution with the smallest fitness is taken as the optimal solution, where the optimal solution includes the target number of modal components and the target penalty factor.

[0011] According to a ship fault detection method provided by the present invention, the step of determining the fault characteristics according to the target modal components specifically includes: The time domain indicators of the target modal component are collected, and the time domain indicators include mean, variance, peak, kurtosis, effective value, peak factor, pulse factor, waveform factor, and margin factor.

[0012] According to a ship fault detection method provided by the present invention, the step of inputting the fault feature into the Transformer-SVM model, performing fault diagnosis training, and obtaining the fault detection model specifically includes: The Transformer-SVM model includes a Transformer model and an SVM classifier; Capturing the global dependency of the fault features through the Transformer model; The global dependency relationship is input into the SVM classifier for fault diagnosis training to obtain a fault detection model.

[0013] According to a ship fault detection method provided by the present invention, after the step of inputting the data to be detected into a trained fault detection model to obtain the ship fault type, the method further comprises: According to the ship fault type, corresponding warning information is generated and sent to the target terminal.

[0014] The present invention also provides a ship fault detection device, comprising: A data acquisition module, used for acquiring data to be detected; A fault detection module is used to input the data to be detected into a trained fault detection model to obtain the ship fault type; The fault detection model is trained in the following way: Obtain ship fault data; Performing variational mode decomposition on the ship fault data to obtain fault characteristics; The fault characteristics are input into the Transformer-SVM model to perform fault diagnosis training to obtain the fault detection model.

[0015] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any of the above-mentioned ship fault detection methods is implemented.

[0016] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the ship fault detection method as described in any one of the above is implemented.

[0017] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the ship fault detection method as described above is implemented.

[0018] The ship fault detection method, device, equipment and storage medium provided by the present invention can effectively improve the accuracy of signal decomposition and the reliability of results by using VMD technology to decompose ship fault data and decompose complex signals into multiple modal components, thereby extracting more accurate and representative fault features. In addition, by combining the global dependency capture capability of the Transformer model and the excellent classification performance of the SVM classifier, it is possible to conduct in-depth analysis and accurate classification of ship fault data, thereby improving the accuracy and reliability of fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 It is a flow chart of the ship fault detection method provided by the present invention; Figure 2 It is a structural schematic diagram of a ship fault detection device provided by the present invention; Figure 3 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] In order to solve the problems in the prior art, the present invention proposes a ship fault detection method to realize intelligent detection of ship faults. The ship fault detection method is described below. Figure 1 As shown, including but not limited to the following steps: Step 110: Obtain the data to be detected; Step 120: input the data to be detected into the trained fault detection model to obtain the ship fault type; The fault detection model is trained in the following way: Obtain ship fault data; Performing variational mode decomposition on the ship fault data to obtain fault characteristics; The fault characteristics are input into the Transformer-SVM model to perform fault diagnosis training to obtain the fault detection model.

[0023] Step 110: Obtaining the data to be tested In this embodiment, the data to be detected refers to data from various sensors and power systems of the ship, which generally include but are not limited to vibration signals, temperature data, current and voltage waveforms, etc. In order to ensure the comprehensiveness and accuracy of the detection data, signals can be collected in real time from various key equipment and systems of the ship (such as the main engine, generator, thruster, steering gear, etc.).

[0024] Signal acquisition: The raw data required for fault diagnosis is collected through sensors installed on the ship, such as acceleration sensors, temperature sensors, voltage sensors, current sensors, etc.

[0025] Data preprocessing: Perform preprocessing operations such as filtering and denoising on the collected raw data to improve data quality, remove possible noise, and standardize the signal to make it suitable for subsequent analysis.

[0026] Step 120: Input the data to be detected into the trained fault detection model to obtain the ship fault type After completing the data preprocessing, the next step is to input the data to be detected into the trained fault detection model to obtain the type of ship fault. This step is carried out in the following two parts: Data input: The pre-processed ship data is passed to the pre-trained fault detection model. At this point, the model has the ability to identify different fault modes through input data.

[0027] Fault diagnosis output: Based on the input signal data, the model outputs the prediction results of the ship fault type after analysis and reasoning. The output results may include the specific type and severity of the fault or related diagnostic information.

[0028] Fault detection model training process The training process of the fault detection model is carried out through the following steps, which describes in detail how to train through the combination of variational mode decomposition (VMD), GOA optimization algorithm and Transformer-SVM model.

[0029] 1. Obtain ship fault data In order to train the fault detection model, a large amount of ship fault data is first needed. This data should include various operation data of the ship in normal working state and when various faults occur. By learning from these data, the model can identify different fault types.

[0030] Data source: Various types of ship fault data can be obtained through long-term accumulation or experimental simulation. The data contains multi-dimensional signals in different states before and after the fault occurs.

[0031] Data labeling: To ensure the accuracy of the training process, the collected data should be labeled to indicate the time and type of fault, to provide support for the subsequent training process.

[0032] 2. Perform variational modal decomposition on ship fault data to obtain fault characteristics Variational mode decomposition (VMD) is an advanced signal processing method that can decompose complex ship fault data into multiple modal components, thus facilitating subsequent feature extraction. The specific process is as follows: Modal decomposition: The collected ship fault data is input into the VMD algorithm. VMD decomposes the complex signal into multiple modal components with different frequencies and bandwidths in an adaptive manner. Each modal component represents a specific frequency band characteristic of the signal.

[0033] Feature extraction: By analyzing the various modal components obtained by decomposition, the features related to the fault are extracted. These features may include time domain features (such as mean, variance, peak, kurtosis, etc.), frequency domain features (such as frequency, bandwidth, etc.) and other statistical characteristics.

[0034] 3. Input fault features into the Transformer-SVM model for fault diagnosis training The core step of training the model is to input the fault features into the Transformer-SVM model for fault classification training. The specific process is as follows: Transformer model training: Feature modeling: The Transformer model is able to process time series data with long-term and short-term dependencies, and captures the global dependencies in ship failure data through the self-attention mechanism. This mechanism enables the model to find potential failure modes in time series data.

[0035] Encoder and decoder: The fault features are encoded through the Transformer encoder, global information is extracted, and passed to the decoder for subsequent processing.

[0036] SVM classifier training: Feature input: The feature vector output by the Transformer model is passed as input to the SVM classifier. The SVM constructs a maximum margin hyperplane, performs fault classification training, and generates a classification model.

[0037] Training process: The SVM classifier uses the input fault feature vector to train a classification model that can match the fault feature with the fault type. The model has strong generalization ability and can identify a variety of ship fault types.

[0038] Through the above steps, the trained fault detection model will have the ability to identify different types of ship faults and can be used for the prediction and diagnosis of ship faults in practical applications.

[0039] The ship fault detection method provided by the present invention combines VMD, GOA optimization algorithm, Transformer model and SVM classifier. The ship fault data is decomposed by VMD, and the VMD parameters are optimized by GOA algorithm, so that the signal decomposition is more accurate; by utilizing the global dependency capture ability of the Transformer model and the excellent classification ability of the SVM classifier, the present invention realizes the accurate identification of the ship fault type. This method can not only effectively improve the accuracy and real-time performance of ship fault detection, but also realize automated and intelligent fault diagnosis, providing reliable technical support for the safe operation of ships.

[0040] As a further optional embodiment, the step of performing variational mode decomposition on the ship fault data to obtain fault characteristics specifically includes: Performing variational modal decomposition on the ship fault data to obtain modal components; According to the envelope entropy, the modal components are screened to obtain the target modal components; A fault feature is determined according to the target modal component.

[0041] In a further implementation of the present invention, a modal component screening method based on envelope entropy is provided for processing ship fault data to improve the accuracy and effectiveness of fault characteristics. This step includes the following sub-steps: 1. Perform variational modal decomposition on the ship fault data to obtain modal components In this step, the ship fault data is first processed by the variational mode decomposition (VMD) algorithm. VMD adaptively decomposes the signal into multiple modal components, each of which represents a specific frequency band of the signal. The specific steps are as follows: Data input: Input the ship fault data into the VMD algorithm. The data may be vibration signals, current and voltage waveforms, or other types of monitoring signals.

[0042] Modal decomposition: VMD algorithm decomposes the original signal into multiple modal components with different frequencies and bandwidths through an iterative optimization process. These modal components each represent different spectral characteristics in the signal and can reflect important changes in the ship fault signal.

[0043] 2. According to the envelope entropy, the modal components are screened to obtain the target modal components Envelope entropy is an indicator used to quantify signal complexity and irregularity, and can effectively identify important components related to faults. In this step, the envelope entropy of each modal component is calculated to screen out the target modal component that best represents the fault characteristics. The specific process is as follows: Envelope calculation: For each modal component, calculate its envelope signal. The envelope signal can be obtained through Hilbert transform, which reflects the amplitude change of the modal component.

[0044] Envelope entropy calculation: For the envelope signal of each modal component, calculate its envelope entropy. The larger the envelope entropy, the more complex the dynamic characteristics of the component in the time series, which may be related to the fault.

[0045] Modal component screening: According to the size of the envelope entropy, the modal components with higher envelope entropy are screened out. These components usually contain more fault information and can help diagnose ship faults more accurately.

[0046] 3. Determine fault characteristics based on target modal components Once the target modal components are screened, useful fault features can be extracted from them, which can be used for fault diagnosis. The specific steps include: Time domain feature extraction: Perform time domain analysis on the target modal components to extract common time domain features, such as mean, variance, peak, kurtosis, etc. These features can reflect the volatility and trend changes of the signal and are an important basis for fault diagnosis.

[0047] Frequency domain feature extraction: Perform frequency domain analysis on the target modal components to extract features such as frequency characteristics and energy spectrum. These features help identify frequency characteristics unique to different fault modes.

[0048] Feature fusion: Combine time domain and frequency domain features to construct a comprehensive fault feature vector to provide sufficient information for subsequent fault diagnosis.

[0049] Through variational modal decomposition and envelope entropy screening, this embodiment provides an efficient fault feature extraction method. This method can accurately extract representative modal components from complex ship fault data, and screen out features with high fault correlation through envelope entropy, thereby providing effective support for ship fault diagnosis. This embodiment improves the accuracy and reliability of fault feature extraction, and helps to improve the accuracy of the overall fault detection system.

[0050] In the VMD algorithm, the decomposed modal component is defined as an AM-FM signal. Assuming that the original signal can be decomposed into k modal components, the expression of the kth modal component is:

[0051] In the formula, the phase is a non-decreasing function, and ; represents the envelope function, and and instantaneous frequency Compared to It changes slowly.

[0052] The bandwidth of each modal component can be estimated using the Casson criterion:

[0053] In the formula, represents the maximum deviation of the instantaneous frequency from the center, represents the deviation rate of instantaneous frequency, Represents the highest frequency of the envelope function. Under the constraint that the sum of the components is equal to the input signal, the sum of the estimated bandwidths of the components is minimized, and then after a series of transformations, the following constrained variational model is constructed:

[0054] In the formula, represents the modal components obtained by decomposition; Indicates the center frequency corresponding to the modal component; Indicates that the modal components are obtained by Hilbert transform The one-sided spectrum of represents the original input signal; Represents the sum of modal components. The VMD method adaptively decomposes the signal by searching for the optimal solution of the above constrained variational model, gradually updates the center frequency and bandwidth of each component during the iterative solution, and finally adaptively divides the modal components according to the frequency domain characteristics of the signal itself. However, when solving this model, it is necessary to introduce a quadratic penalty factor α and a Lagrangian operator λ(t) to convert the above constrained variational problem into the following unconstrained variational problem:

[0055] GOA (Gazelle Optimization Algorithm) is used to optimize VMD parameters, and the minimum envelope entropy is used as the fitness function. The modal component corresponding to the index value of the minimum envelope entropy is extracted, and its 9 indicators are calculated. The 9 time domain indicators of this optimal modal component are collected: mean, variance, peak, kurtosis, effective value, peak factor, pulse factor, waveform factor, and margin factor. In this way, the characteristic vector is extracted, and the characteristic vector can reflect the fault status of the signal. Optimize VMD parameters, the fitness function is the minimum envelope entropy It can be expressed as:

[0056] The minimum value of envelope entropy is used as the fitness function. Envelope entropy represents the sparse characteristics of the original signal. When there is more noise and less feature information in the IMF, the envelope entropy value is larger, otherwise, the envelope entropy value is smaller.

[0057] As a further optional embodiment, the step of performing variational modal decomposition on the ship fault data to obtain modal components specifically includes: Determine the number of modal components and penalty factors through the GOA algorithm, wherein the number of modal components is the number of modal components obtained by variational modal decomposition, and the penalty factor is used to control the accuracy of variational modal decomposition and the bandwidth of each component; According to the number of modal components and the penalty factor, variational modal decomposition is performed on the ship fault data to obtain modal components.

[0058] In a further embodiment of the present invention, the step of performing variational mode decomposition on ship fault data specifically includes the following sub-steps: This embodiment optimizes the parameters of variational mode decomposition (VMD) by introducing GOA (Grey Wolf Optimization Algorithm) to improve the accuracy and robustness of fault detection.

[0059] 1. Determine the number of modal components and penalty factors through the GOA algorithm In the process of variational modal decomposition, the number of modal components and the penalty factor are key parameters that have a direct impact on the performance of VMD. Specifically, the number of modal components determines the number of modal components obtained by decomposition, while the penalty factor controls the accuracy of decomposition and the bandwidth of each modal component. To optimize these parameters, this embodiment uses the GOA algorithm to automatically determine the optimal values ​​of these two parameters.

[0060] The specific process is as follows: Obtaining a population of candidate solutions: The GOA algorithm first initializes a population of candidate solutions, each of which represents a combination of the number of modal components and a penalty factor. Each candidate solution consists of two parts: Number of modal components (k): the number of modal components that we want to obtain after VMD decomposition.

[0061] Penalty factor (α): used to control the accuracy of VMD decomposition and affect the bandwidth of each modal component.

[0062] By setting a reasonable search range, each individual in the candidate solution population contains a combination of the number of modal components and a penalty factor.

[0063] Calculate the fitness value: For each candidate solution in the population, calculate its fitness value. The fitness function is usually evaluated based on the reconstruction error of the VMD decomposition result. The candidate solution with a smaller reconstruction error indicates that the selected combination of the number of modal components and the penalty factor can effectively extract the main information from the original signal.

[0064] The fitness function can be calculated as:

[0065] Among them, ReconstructionError(k,α) is the error of reconstructing the signal after VMD decomposition.

[0066] Update the position of the candidate solution: Update the position of the candidate solution according to the update rules in the GOA algorithm. GOA simulates the hunting process of gray wolves, in which the selection and position adjustment of prey are determined according to the fitness and position of the current prey. The algorithm optimizes the position of the candidate solution through multiple iterations to find the optimal combination of the number of modal components and penalty factors.

[0067] Select the optimal solution: After multiple rounds of iterations, the algorithm considers the most adaptable candidate solution as the optimal solution. The number of modal components and penalty factor corresponding to the optimal solution are the optimal parameters required for VMD decomposition.

[0068] 2. Perform variational modal decomposition on ship fault data based on the number of modal components and penalty factors Once the optimal number of modal components and penalty factors are determined by the GOA algorithm, these parameters can be applied to the variational modal decomposition step of the ship fault data. The specific process is as follows: Set the parameters of VMD: The modal component number k and penalty factor α obtained by GOA algorithm optimization are used as input parameters of VMD algorithm.

[0069] Perform variational mode decomposition: Use variational mode decomposition (VMD) to process ship fault data. The VMD algorithm decomposes the original signal into several modes (or components) with different frequency components, and each modal component contains different frequency band characteristics of the signal.

[0070] The goal of VMD is to minimize the bandwidth of each modal component through an iterative optimization process while retaining the main characteristics of the signal as much as possible.

[0071] Output modal components: VMD algorithm will decompose multiple modal components according to the optimal parameters, each modal component reflects the characteristics of ship fault data in a specific frequency band. These modal components will be further analyzed and screened as fault features in subsequent steps.

[0072] 3. Advantages of variational mode decomposition By optimizing the number of modal components and penalty factors through the GOA algorithm, the performance of VMD in ship fault data processing can be significantly improved. The main advantages include: Improved accuracy: The optimized parameters can more accurately control the bandwidth of each mode, making the modal decomposition more accurate and better able to extract fault features from the signal.

[0073] Enhanced robustness: Through the selection of a reasonable number of modal components and penalty factors, the VMD algorithm can better adapt to the complex noise and uncertainty in ship fault data, thereby improving the robustness of fault diagnosis.

[0074] Reduce computational complexity: Through the optimization of the GOA algorithm, the optimal decomposition parameters can be found in a shorter time, avoiding the tedious process of manual parameter adjustment and saving computing resources and time costs.

[0075] As a further optional embodiment, the step of determining the number of modal components and the penalty factor by using the GOA algorithm specifically includes: Acquire a candidate solution population, the candidate solution population including a plurality of candidate solutions, the candidate solutions being used to characterize the number of modal components and the penalty factor; Calculating the fitness of each of the candidate solutions; updating the position of the candidate solution according to the random motion model; The candidate solution with the smallest fitness is taken as the optimal solution, where the optimal solution includes the target number of modal components and the target penalty factor.

[0076] In a further implementation of the present invention, an optimization process is provided for the step of determining the number of modal components and the penalty factor by the GOA algorithm to ensure that the variational modal decomposition (VMD) can obtain the optimal modal decomposition effect. The step specifically includes the following sub-steps: 1. Obtain candidate solution population In GOA (Gazelle Optimization Algorithm), the candidate solution population is the basis of the optimization process. The candidate solution population consists of multiple individuals, each of which represents a possible solution, including the number of modal components and penalty factors. These candidate solutions will be used to search for the optimal number of modal components and penalty factors, thereby optimizing the VMD algorithm.

[0077] Initialize candidate solution population: By randomly generating an initial candidate solution population, each candidate solution contains a modal component number (k) and a penalty factor (α). These candidate solutions will form a position matrix, and the position of each candidate solution in the population represents a possible combination of the modal component number and the penalty factor.

[0078] 2. Calculate the fitness of each candidate solution The fitness of each candidate solution is evaluated by calculating the quality of the modal decomposition result corresponding to the solution. In this embodiment, the fitness function uses envelope entropy to measure the effectiveness of the modal decomposition result.

[0079] Fitness calculation: For each candidate solution, based on the number of modal components and penalty factors it represents, the ship fault data is decomposed by VMD, and the envelope entropy of the decomposed modal components is calculated. The smaller the envelope entropy, the better the quality of the modal decomposition, and therefore the higher the fitness value.

[0080] Fitness sorting: After the fitness calculation of all candidate solutions is completed, these candidate solutions are sorted by fitness, and the candidate solution with the smallest fitness is preferentially selected for optimization.

[0081] 3. Update the position of the candidate solution according to the random motion model The GOA algorithm imitates the behavior of gazelles in nature to update the position of candidate solutions and uses a random motion model to update candidate solutions. In each iteration, the candidate solutions will perform local search and global exploration based on the position of the current optimal solution (most adaptive solution).

[0082] Random motion model: Based on Levy flight and Brownian motion, the GOA algorithm simulates the random motion of candidate solutions in the search space. Levy flight helps to jump to new areas globally, while Brownian motion is used to finely search the neighborhood of the current optimal solution.

[0083] Position update: In each iteration, the candidate solution updates its position based on its current position, speed, fitness and relative position of the optimal solution. The updated candidate solution will be used for the next round of fitness calculation.

[0084] 4. Take the candidate solution with the smallest fitness as the optimal solution After multiple iterations, all candidate solutions will be updated to new positions. After continuous optimization, a candidate solution with the minimum fitness will eventually be found. The number of modal components and penalty factor corresponding to this solution are the optimal solution.

[0085] Optimal solution selection: After multiple iterations and updates, the candidate solution with the smallest fitness is finally selected as the optimal solution. The number of modal components (k) and penalty factor (α) corresponding to the optimal solution are the optimal parameter configurations in the VMD algorithm.

[0086] Parameter determination: The number of modal components and penalty factors corresponding to the optimal solution are used as input parameters in the VMD algorithm for further signal decomposition to obtain high-quality fault features.

[0087] Specifically, the optimization process is as follows: (1) Initialization GOA is a population-based optimization algorithm that uses randomly initialized fault features (X) as search agents, which represent different diagnostic parameters and feature selection strategies. The search agent is defined as an nd matrix of candidate solutions, as defined in Eq. GOA uses the upper bound constraint (UB) and the lower bound constraint (LB) of the problem to randomly define the range of values ​​that the population vector can take.

[0088]

[0089] Where X is the position vector matrix of the candidate population; j is the randomly generated vector position of the i-th population in the j-th dimension, n is the number of fault features, and d is the search space (dimension) defined by the optimization problem.

[0090]

[0091] Where rand is a random number, UB j and LB j are the upper and lower bounds of the problem, respectively. In each iteration, each xi;j generates a candidate solution. The minimum solution is considered to be the best solution obtained so far.

[0092] The optimal solution currently obtained is named as the most diagnostically significant fault feature, and a matrix is ​​constructed as follows. This matrix is ​​used to search and find the fault feature in the next step.

[0093]

[0094] in The position vector representing the top fault feature.

[0095] (2) Brownian motion A random motion whose displacements follow a normal (Gaussian) probability distribution function with specific mean and unit variance I = 0 and r2 = 1. At a point x, the standard Brownian motion is defined as follows.

[0096]

[0097] (3) Le´vy flight Le´vy flight performs a random walk using the Le´vy distribution (power-law tail) given by the following equation.

[0098]

[0099] In the formula, is the flight distance, represents the power law exponent. The following expression represents the Le´vy stable process as an integral.

[0100]

[0101] in is the distribution index that controls everything about the motion, and c is the unit of scale. This work uses an algorithm proposed to produce stable Levy motion. The algorithm uses a range of 0.3-1.99 , defined as shown in Eq.

[0102]

[0103] in ; and The definitions are as follows: and

[0104]

[0105] (4) Development This stage assumes that the gazelle is grazing quietly when no potential fault features are identified, or when a predator is stalking the gazelle. When stalked by a predator, the gazelle's movement becomes more directional and controlled. This can be likened to a local search in the fault feature selection process, which is refined and optimized using the previously found high-quality feature combinations. In this stage, the Brownian motion with uniform controllable step size features is used to effectively cover the neighborhood of the domain. The mathematical model of this behavior is shown in Eq.

[0106]

[0107] in is the solution of the next iteration, gazelle is the solution of the current iteration, s is the predation speed of the gazelle, is a vector containing random numbers representing Brownian motion, and R is a vector of uniform random numbers [0,1].

[0108] 5) Search The exploratory behavior that begins when a gazelle detects a predator can be likened to an initial fault feature search. The gazelle responds to danger by flicking its tail, stamping its feet, or standing on all four feet 2 meters in the air; this is simulated by scaling the height of 2 meters to a number between 0 and 1, and by converting these behaviors into digital parameters (e.g., between 0 and 1) to represent different positions in the search space. This algorithmic stage uses Le’vy flight; this process involves small steps and occasional long jumps. This approach improves coverage of the search space and increases the likelihood of finding a global optimal solution. Once a predator is detected, the gazelle flees and the predator chases. The mathematical model of the gazelle’s behavior once it detects a predator is shown in Eq.

[0109]

[0110] Where S is the maximum speed that the gazelle can reach and RL represents a random number vector based on Levy distribution. The mathematical model of the predator's behavior of chasing the gazelle is shown in Eq.

[0111]

[0112] In a study on Mongolian gazelles, the authors also assumed that, although gazelles are not endangered, they have an annual survival rate of 0.66, meaning that predators can only succeed if it is 0.34. PSRs are predator success rates and affect the gazelle's ability to escape, which means that the algorithm avoids getting stuck in a local minimum. The PSR effect is modeled as follows:

[0113] Where U represents a binary vector, which is constructed by generating a random number r in [0,1]. R1 and r2 are random indices of the gazelle matrix.

[0114] By optimizing the modal component number and penalty factor of VMD through the GOA algorithm, this embodiment can effectively improve the accuracy and performance of VMD signal decomposition. By introducing the optimization algorithm, the most suitable decomposition parameters can be automatically selected, so that the ship fault detection system can more accurately extract fault features, thereby improving the accuracy and reliability of fault diagnosis. This optimization method enables the entire system to process and analyze more robustly when facing complex ship fault data.

[0115] As a further optional embodiment, the step of determining the fault feature according to the target modal component specifically includes: The time domain indicators of the target modal component are collected, and the time domain indicators include mean, variance, peak, kurtosis, effective value, peak factor, pulse factor, waveform factor, and margin factor.

[0116] In a further implementation of the present invention, a specific method for extracting fault features is provided for the target modal components obtained by VMD (variational modal decomposition). By collecting the time domain indicators of the target modal components, the fault state can be effectively reflected, thereby improving the precision and accuracy of ship fault detection. This step specifically includes the following sub-steps: 1. Collect time domain indicators of target modal components After the variational mode decomposition (VMD) processing, we get a series of modal components. The target modal components obtained by screening are used as the basis for subsequent fault feature extraction. In this step, it is first necessary to perform time domain analysis on the target modal components and calculate a set of key time domain indicators that can effectively characterize the characteristics of the fault signal. Time domain indicators include but are not limited to the following: Mean: Calculates the mean of the target modal component to describe the average level of the signal. The mean reflects the overall deviation of the signal.

[0117] Variance: Calculates the variance of the target modal component to measure the volatility or dispersion of the signal. A larger variance indicates a more drastic signal fluctuation, which may be related to a fault.

[0118] Peak value: Calculates the peak value of the target modal component, which indicates the maximum amplitude in the signal. Peak value is often used to detect sudden changes in the signal, especially abnormal fluctuations when a fault occurs.

[0119] Kurtosis: Calculates the kurtosis of the target modal component, reflecting the sharpness of the signal. Higher kurtosis values ​​usually indicate the presence of fast-changing events in the signal, which may be related to faults or shocks.

[0120] Effective value (RMS value): Calculate the effective value (root mean square value) of the target modal component, which is a measure of signal strength and reflects the energy of the signal.

[0121] Crest Factor: Calculates the crest factor of the target modal component, which represents the ratio of the peak value to the RMS value of the signal. A larger crest factor may indicate a more significant anomaly in the signal.

[0122] Pulse Factor: Calculates the pulse factor of the target modal component, which is used to measure the strength of the pulse-like features in the signal. Signals with large pulse factors usually appear as sudden events when a fault occurs.

[0123] Form Factor: Calculates the form factor of the target modal component to measure the complexity of the signal waveform. A larger form factor may indicate that the signal contains more high-frequency components, which may be related to the fault state of the device.

[0124] Margin Factor: Calculates the margin factor of the target modal component, reflecting the stability and abnormal response of the signal. A higher margin factor may indicate a higher abnormal strength of the signal.

[0125] 2. Calculate and process time domain indicators For each target modal component, by calculating the above time domain indicators, a set of values ​​reflecting the signal characteristics can be obtained. Each time domain indicator can provide information on the performance of the signal in different aspects, such as volatility, impact, stability, etc. These characteristics can help further identify and classify the type of ship faults.

[0126] Data storage and processing: The calculated time domain indicators will be collected and stored in the feature database. Each set of time domain indicators corresponds to a specific target modal component, which is then associated with the relevant characteristics of the ship fault.

[0127] Fault mode association: Through the correlation analysis between time domain indicators and fault modes, effective fault features can be further extracted for subsequent fault diagnosis and classification.

[0128] 3. Determine the fault characteristics The time domain index set obtained through the above steps is used as the key feature of ship fault diagnosis. These time domain indicators can reveal the intrinsic characteristics of the target modal components and reflect the possible abnormalities or fault modes in the signal. Through the comprehensive analysis of time domain indicators, the characteristics of different fault types can be determined, thus providing a reliable basis for fault classification and diagnosis.

[0129] Feature selection: In the subsequent fault diagnosis process, the most representative time domain indicators can be selected as the final fault features according to the nature and characteristics of the fault. This process can be further optimized using feature selection algorithms to screen out the time domain indicators that best reflect the fault status.

[0130] Classification and identification: Finally, the extracted fault features will be passed as input to the classifier (such as SVM, neural network, etc.) for further fault identification and classification.

[0131] By collecting the time domain indicators of the target modal components and further analyzing and processing these indicators, the features related to ship faults can be effectively extracted. These time domain indicators include mean, variance, peak, kurtosis, etc. They reflect the different characteristics of the signal and help to accurately identify the type of ship faults. In the fault detection process, through the comprehensive analysis of these time domain characteristics, more accurate fault diagnosis can be achieved, providing strong support for ship maintenance and management.

[0132] As a further optional embodiment, the step of inputting the fault feature into the Transformer-SVM model to perform fault diagnosis training to obtain the fault detection model specifically includes: The Transformer-SVM model includes a Transformer model and an SVM classifier; Capturing the global dependency of the fault features through the Transformer model; The global dependency relationship is input into the SVM classifier for fault diagnosis training to obtain a fault detection model.

[0133] In a further implementation of the present invention, the step of inputting the fault features in the ship fault detection method into the Transformer-SVM model for fault diagnosis training specifically includes the following sub-steps. This step combines the powerful characteristics of the Transformer model and the superior performance of the SVM classifier, and obtains the final fault detection model by capturing global dependencies and performing efficient classification training. The following is a detailed description: 1. Transformer-SVM model construction The Transformer-SVM model consists of two main modules: the Transformer model and the SVM classifier. This model structure can effectively process and analyze the input fault features and perform fault diagnosis.

[0134] Transformer model: Transformer is a neural network model based on the self-attention mechanism, which is widely used in the processing of sequence data. In ship fault detection, the Transformer model is mainly used to capture the global dependencies between fault features. Through the self-attention mechanism, Transformer can effectively capture the long-term dependencies and complex relationships in the fault features, thereby improving the model's expressiveness and prediction accuracy.

[0135] SVM classifier: Support vector machine (SVM) is a classic supervised learning algorithm, especially suitable for small sample learning and nonlinear classification problems. In the present invention, the SVM classifier is used to classify and train fault features based on the global dependency extracted by the Transformer model. By learning different types of fault data, the SVM classifier can efficiently identify the fault type of the ship.

[0136] 2. Capturing global dependencies of fault features After the fault features are input into the Transformer model, the Transformer processes the input features through the self-attention mechanism to capture the global dependencies between the features. Specifically, the Transformer model performs the following operations: Self-attention mechanism: Transformer can automatically focus on the most relevant feature information by calculating the attention weights between input features. This enables the model to dynamically adjust the degree of attention to different features, thereby mining possible long-distance dependencies between features.

[0137] Positional encoding: Transformer uses positional encoding to supplement the order information of input data, which is particularly important for time series data. In ship fault data, fault features often have a temporal or spatial order. Positional encoding helps the model understand this time series information, thereby improving the ability to identify fault patterns.

[0138] Multi-head attention mechanism: Through the multi-head attention mechanism, Transformer can learn different feature relationships in parallel, thereby mining dependency patterns in data in multiple dimensions. This allows the model to more comprehensively capture the complex laws in fault features.

[0139] 3. Input global dependencies into the SVM classifier After the Transformer model processes the fault features and extracts the global dependencies, the output is passed as input data to the SVM classifier. The SVM classifier is trained for fault diagnosis through the following steps: Feature mapping: SVM maps input data to a higher-dimensional feature space through a kernel function. In a high-dimensional space, the classification boundaries of data are often more obvious, making it easier to distinguish different types of fault modes.

[0140] Classification training: By learning from the labeled data, the SVM classifier establishes a decision boundary that can distinguish different fault types. The training goal of the SVM is to find a hyperplane that maximizes the classification interval, thereby ensuring the accuracy and robustness of the classification.

[0141] Fault type prediction: Once trained, the SVM classifier is able to perform real-time fault diagnosis based on the input fault features and their global dependencies, and output the possible fault types of the ship.

[0142] 4. Generation of Fault Detection Model After two-stage processing by Transformer and SVM, the final fault detection model can not only capture the global dependency of fault features, but also has strong classification capabilities. The model can quickly and accurately identify the fault type based on new ship fault data, and provide support for subsequent fault warnings, maintenance decisions, etc.

[0143] Model optimization: During the training process, the model can be optimized through cross-validation, grid search and other techniques to adjust the hyperparameters in the Transformer and SVM models to improve the accuracy and efficiency of fault detection.

[0144] Real-time detection: The trained fault detection model can be deployed in the ship system to perform fault diagnosis on real-time data. As more fault data and feedback are input, the model can be continuously optimized to adapt to different ship fault modes.

[0145] Specifically, the Transformer-SVM model is used to implement fault classification. The Transformer encoder is used to mine the complex relationships between data features and the long-term and short-term dependencies in the time series. In the final classification stage of the model, the traditional softmax classifier is replaced with the SVM classifier to further improve the classification accuracy of the model. The extracted features are divided into training sets and test sets (3:1) and sent to the Transformer-SVM model for training. Using Transformer to extract features and SVM classification can improve the accuracy of fault diagnosis.

[0146] Transformer is a model with an encoder-decoder structure. This model abandons the loop structure and relies on the Attention mechanism to control the global dependency between input and output. At each step of decoding, the model uses the previously generated symbol as an additional input to generate the next symbol, that is, the model is autoregressive. Each layer of the Encoder has two operations: Self-Attention and Feed Forward; each layer of the Decoder has three operations: Self-Attention, Encoder-Decoder Attention, and Feed Forward. Both Self-Attention and Encoder-Decoder Attention are Multi-Head Attention mechanisms.

[0147] The Attention function is the core function of the Transformer model, and its calculation process is as shown in formula (17):

[0148] In the formula, , , , , Each is the output of the self-attention mechanism, is a learnable linear transformation matrix.

[0149] Assume that the training set , contains N samples in total, among which , is the K-dimensional feature vector, represents the class label. Here we assume that the training set data has M classes. The goal is to find the decision function Predict new data categories. The training set data is assumed to have M categories. The essence of the SVM method is to construct a binary SVM between every two categories. For the i-th and j-th categories, a binary SVM is trained, which is a quadratic programming problem in the field of operations research. The mathematical model is as follows:

[0150] By inputting fault features into the Transformer-SVM model, combining the Transformer model to capture global dependencies and the SVM classifier for fault diagnosis training, the accuracy and efficiency of ship fault detection can be effectively improved. The Transformer model establishes complex associations between features, and the SVM classifier performs efficient classification based on these global dependencies, ultimately generating a high-precision fault detection model, providing strong technical support for ship maintenance and management.

[0151] As a further optional embodiment, after the step of inputting the to-be-detected data into a trained fault detection model to obtain the ship fault type, the method further includes: According to the ship fault type, corresponding warning information is generated and sent to the target terminal.

[0152] In a further implementation of the present invention, the last step in the ship fault detection method, i.e., the step of generating warning information according to the ship fault type output by the fault detection model and sending it to the target terminal, specifically includes the following sub-steps. This embodiment provides an automated warning and remote monitoring function for the fault detection system, ensuring timely response and decision support in ship operation.

[0153] 1. Get the ship fault type In the steps of the present invention, the data to be detected has been processed by the trained fault detection model, and the fault type of the ship is obtained. The fault type generally includes the following possibilities: Mechanical failure: such as ship's engine failure, propeller damage, etc.

[0154] Electrical failure: such as power system failure, sensor failure, etc.

[0155] Structural failures: such as hull cracks, water stains in the cabin, etc.

[0156] Operational failures: such as automatic control system problems, navigation equipment failures, etc.

[0157] This step analyzes the input real-time monitoring data and uses the prediction results of the fault detection model to classify the current fault condition of the ship into a specific type, and transmits the results to the subsequent modules of the system for processing.

[0158] 2. Generate early warning information based on fault type Once the fault type of the ship is determined, the system will then generate corresponding warning information based on the fault type. The specific process includes the following: Warning level judgment: According to different types of faults and their severity, warning information will be divided into different levels, such as: Red alert: major failure, such as a failure that may cause the ship to stop operating or cause an accident.

[0159] Orange warning: A more serious fault that needs to be repaired as soon as possible, but will not immediately affect the safety of the ship.

[0160] Yellow warning: Minor fault, the ship can continue to operate, but needs to be repaired as soon as possible.

[0161] In addition, the warning level can be dynamically adjusted based on factors such as the impact scope of the fault, potential risks, and the ship's operating environment.

[0162] Warning information content: The content of the warning information should include the following basic elements: Fault Type: Accurately describe the fault type, such as "Abnormal engine temperature" or "Navigation system failure".

[0163] Fault location: The location where the fault occurred, helping maintenance personnel quickly locate the problem area.

[0164] Fault severity level: Warning level (such as red, orange, yellow).

[0165] Repair suggestions: For different fault types, the system can provide specific repair suggestions or operation instructions.

[0166] Estimated impact: Assess the potential impact of the fault on ship operations, such as reduced navigation speed, ship suspension, route adjustment, etc.

[0167] For example, if "engine temperature is too high" is detected, the warning message may be: "Red warning, it is recommended to stop immediately and check the engine cooling system to avoid engine damage." 3. Send warning information to the target terminal After the warning information is generated, the system sends it to the target terminal so that relevant personnel can receive the warning in time and take countermeasures. The target terminal can be one or more of the following: Crew control terminal: such as a console or operator computer on board a vessel, where crew members can view fault conditions in real time and take necessary actions immediately.

[0168] Ship management system: such as cloud platform and background monitoring system, which can remotely receive ship fault data and warning information, conduct global monitoring and timely dispatch response.

[0169] Shipowner or manager terminal: The shipowner or ship manager's mobile device, desktop system, etc., facilitates remote monitoring of ship failures and assigns technicians to handle them when necessary.

[0170] Maintenance service center: When the fault is more complex or serious, the warning information can be sent directly to the remote maintenance service center, where professional technicians will provide support and guide on-site operations.

[0171] The system determines the most suitable communication method (such as SMS, email, push notification, APP message, etc.) and transmission channel according to the fault type and warning level to ensure that the warning information can reach the target terminal accurately and in a timely manner.

[0172] 4. Subsequent processing of warning information Response mechanism: The target terminal that receives the warning information can initiate the corresponding response mechanism according to the warning level. For example, when a red warning is received, the ship may be forced to stop or adjust the waterway; while a yellow warning may trigger regular inspections or arrange for the crew to conduct further diagnosis.

[0173] Maintenance scheduling: For more serious faults, early warning information can trigger the maintenance scheduling process, automatically generate maintenance task orders and assign them to designated maintenance teams.

[0174] Real-time feedback: The system can provide feedback on the processing status to relevant personnel based on the progress of fault handling, ensuring the transparency of ship maintenance and operation management.

[0175] By adding the function of generating and sending early warning information to the ship fault detection system, the present invention can promptly inform relevant personnel when a fault occurs and take appropriate measures to deal with potential safety risks. This real-time early warning and remote monitoring function can greatly improve the efficiency of ship fault response, reduce the risks in ship operation, and ensure that the ship can quickly intervene and handle the fault when it occurs, thereby improving the safety and reliability of the ship.

[0176] The ship fault detection device provided by the present invention is described below. Figure 2 As shown, the ship fault detection device described below and the ship fault detection method described above can refer to each other.

[0177] A ship fault detection device, comprising: The data acquisition module 210 is used to acquire the data to be detected; A fault detection module 220 is used to input the data to be detected into a trained fault detection model to obtain a ship fault type; The fault detection model is trained in the following way: Obtain ship fault data; Performing variational mode decomposition on the ship fault data to obtain fault characteristics; The fault characteristics are input into the Transformer-SVM model to perform fault diagnosis training to obtain the fault detection model.

[0178] Figure 3 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 3 As shown, the electronic device may include: a processor 310, a communications interface 320, a memory 330 and a communication bus 340, wherein the processor 310, the communications interface 320 and the memory 330 communicate with each other through the communication bus 340. The processor 310 may call the logic instructions in the memory 330 to execute the ship fault detection method, which includes: Obtain the data to be tested; Inputting the data to be detected into a trained fault detection model to obtain the ship fault type; The fault detection model is trained in the following way: Obtain ship fault data; Performing variational mode decomposition on the ship fault data to obtain fault characteristics; The fault characteristics are input into the Transformer-SVM model to perform fault diagnosis training to obtain the fault detection model.

[0179] In addition, the logic instructions in the above-mentioned memory 330 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0180] On the other hand, the present invention further provides a computer program product, the computer program product comprising a computer program, the computer program can be stored in a non-transitory computer-readable storage medium, when the computer program is executed by a processor, the computer can execute the ship fault detection method provided by the above methods, the method comprising: Obtain the data to be tested; Inputting the data to be detected into a trained fault detection model to obtain the ship fault type; The fault detection model is trained in the following way: Obtain ship fault data; Performing variational mode decomposition on the ship fault data to obtain fault characteristics; The fault characteristics are input into the Transformer-SVM model to perform fault diagnosis training to obtain the fault detection model.

[0181] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the ship fault detection method provided by the above methods is implemented, and the method comprises: Obtain the data to be tested; Inputting the data to be detected into a trained fault detection model to obtain the ship fault type; The fault detection model is trained in the following way: Obtain ship fault data; Performing variational mode decomposition on the ship fault data to obtain fault characteristics; The fault characteristics are input into the Transformer-SVM model to perform fault diagnosis training to obtain the fault detection model.

[0182] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0183] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0184] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A ship fault detection method, characterized in that: include: Obtain the data to be tested; Inputting the data to be detected into a trained fault detection model to obtain the ship fault type; The fault detection model is trained in the following way: Obtain ship fault data; Performing variational mode decomposition on the ship fault data to obtain fault characteristics; The fault characteristics are input into the Transformer-SVM model to perform fault diagnosis training to obtain the fault detection model.

2. The ship fault detection method according to claim 1, characterized in that: The step of performing variational modal decomposition on the ship fault data to obtain fault characteristics specifically includes: Performing variational modal decomposition on the ship fault data to obtain modal components; According to the envelope entropy, the modal components are screened to obtain the target modal components; A fault feature is determined according to the target modal component.

3. The ship fault detection method according to claim 2, characterized in that: The step of performing variational modal decomposition on the ship fault data to obtain modal components specifically includes: Determine the number of modal components and penalty factors through the GOA algorithm, wherein the number of modal components is the number of modal components obtained by variational modal decomposition, and the penalty factor is used to control the accuracy of variational modal decomposition and the bandwidth of each component; According to the number of modal components and the penalty factor, variational modal decomposition is performed on the ship fault data to obtain modal components.

4. The ship fault detection method according to claim 3, characterized in that: The step of determining the number of modal components and the penalty factor by the GOA algorithm specifically includes: Acquire a candidate solution population, wherein the candidate solution population includes a plurality of candidate solutions, and the candidate solutions are used to characterize the number of modal components and the penalty factor; Calculating the fitness of each of the candidate solutions; updating the position of the candidate solution according to the random motion model; The candidate solution with the smallest fitness is taken as the optimal solution, where the optimal solution includes the target number of modal components and the target penalty factor.

5. The ship fault detection method according to claim 4, characterized in that: The step of determining the fault characteristics according to the target modal component specifically includes: The time domain indicators of the target modal component are collected, and the time domain indicators include mean, variance, peak, kurtosis, effective value, peak factor, pulse factor, waveform factor, and margin factor.

6. The ship fault detection method according to claim 1, characterized in that: The step of inputting the fault features into the Transformer-SVM model, performing fault diagnosis training, and obtaining the fault detection model specifically includes: The Transformer-SVM model includes a Transformer model and an SVM classifier; Capturing the global dependency of the fault features through the Transformer model; The global dependency relationship is input into the SVM classifier for fault diagnosis training to obtain a fault detection model.

7. The ship fault detection method according to claim 1, characterized in that: After the step of inputting the data to be detected into the trained fault detection model to obtain the ship fault type, the method further includes: According to the ship fault type, corresponding warning information is generated and sent to the target terminal.

8. A ship fault detection device, characterized in that: include: A data acquisition module, used for acquiring data to be detected; A fault detection module is used to input the data to be detected into a trained fault detection model to obtain the ship fault type; The fault detection model is trained in the following way: Obtain ship fault data; Performing variational mode decomposition on the ship fault data to obtain fault characteristics; The fault characteristics are input into the Transformer-SVM model to perform fault diagnosis training to obtain the fault detection model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the ship fault detection method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the ship fault detection method according to any one of claims 1 to 7 is implemented.

Images