A ship engine fault diagnosis method based on SCNet

Abstract

The application discloses a ship engine fault diagnosis method based on SCNet, comprising the following steps: S1, obtaining sample data and corresponding fault labels of ship engine fault to obtain a ship engine fault data set; S2, performing noise processing on the sample data in the ship engine fault data set to obtain optimized sample data, and performing preprocessing on the optimized sample data to obtain an optimized fault data set; S3, randomly dividing the optimized fault data set into a training set and a test set according to a preset proportion; S4, constructing an SCNet engine fault diagnosis model based on a convolutional neural network and a self-calibration convolution module; S5, performing model training on the SCNet engine fault diagnosis model according to the training set, and verifying the performance of the trained SCNet engine fault diagnosis model to obtain an optimal SCNet engine fault diagnosis model; and S6, based on the optimal SCNet engine fault diagnosis model, predicting the fault type of the ship engine according to the test set. The method solves the problem that the traditional diesel engine fault diagnosis method mainly depends on the experience of experts, and maintenance personnel usually need to passively identify problems according to their own maintenance experience or existing fault parameter records, and the timeliness and accuracy of the method are often insufficient in the era of unmanned cabins and intelligent shipping.

Description

Technical Field

[0001] This invention relates to the field of marine engine fault diagnosis technology, and in particular to a marine engine fault diagnosis method based on SCNet. Background Technology

[0002] Marine diesel engines, as reciprocating internal combustion engines, are widely used in marine propulsion systems and are the main source of power for ships. However, due to long-term operation in complex working environments, marine diesel engines are prone to failure. Their compact and complex structure, with some moving parts typically bearing significant mechanical and thermal loads, coupled with inherent characteristics such as variable operating conditions, makes the causal relationships of failures intricate and complex. Currently, the traditional methods for fault diagnosis of marine diesel engines mainly include the following:

[0003] 1. Fault diagnosis method based on mechanism model

[0004] This method requires establishing a mechanistic model for the monitored machinery to achieve real-time diagnostics. The key lies in creating a complete and accurate system model of the diesel engine. However, there is often a gap between the model and the actual system, making the modeling process complex and challenging. Furthermore, the mechanistic model may become ineffective as the performance of the diesel engine degrades.

[0005] 2. Fault diagnosis methods based on expert qualitative experience and knowledge

[0006] This method relies on the knowledge and experience of experts for fault diagnosis, primarily including fault tree-based and expert system-based diagnostic methods. This approach requires experts with many years of experience in marine fault diagnosis to perform accurate analysis of marine diesel engine faults. However, over-reliance on expert experience, if errors occur, may affect the accuracy of fault diagnosis.

[0007] 3. Data-driven fault diagnosis methods

[0008] Data-driven intelligent fault diagnosis methods have become an important research direction in mechanical equipment fault diagnosis in recent years. Utilizing technologies such as machine learning, it acquires, transmits, and processes data from condition monitoring systems, mimicking human experts' judgments and decisions regarding equipment condition. This approach avoids, to some extent, the shortcomings of complex modeling and over-reliance on expert experience. Currently, neural networks, support vector machines, and decision trees are the mainstream algorithms employing this method. Among them, deep learning technology, represented by neural networks, has shown significant advantages in learning from large amounts of data and automatically extracting features, making it the mainstream application technology.

[0009] In summary, traditional diesel engine fault diagnosis methods mainly rely on the experience of experts. Maintenance personnel usually need to passively identify problems based on their own maintenance experience or existing fault parameter records. However, in the era of unmanned ship cabins and intelligent shipping, the timeliness and accuracy of this method are often insufficient. Summary of the Invention

[0010] This invention provides a method for diagnosing ship engine faults based on SCNet to overcome the aforementioned technical problems.

[0011] To achieve the above objectives, the technical solution of the present invention is as follows:

[0012] A method for diagnosing marine engine faults based on SCNet includes the following steps:

[0013] S1: Obtain sample data of ship engine failures and corresponding failure labels to obtain a ship engine failure dataset;

[0014] S2: The sample data in the ship engine fault dataset is noise-processed to simulate the real ship engine data acquisition environment and obtain optimized sample data;

[0015] The optimized sample data is then preprocessed to obtain an optimized fault dataset;

[0016] Furthermore, preprocessing includes normalizing the optimized sample data to obtain normalized sample data;

[0017] Perform one-dimensional or two-dimensional tiling on the normalized sample data to obtain the sample tiling data;

[0018] S3: Randomly divide the optimized fault dataset into a training set, a validation set, and a test set according to a preset ratio;

[0019] S4: Construct an SCNet engine fault diagnosis model based on convolutional neural networks and self-calibrating convolutional modules;

[0020] S5: Train the SCNet engine fault diagnosis model based on the training set, and verify the performance of the trained SCNet engine fault diagnosis model to obtain the optimal SCNet engine fault diagnosis model.

[0021] S6: Based on the optimal SCNet engine fault diagnosis model, the test set is used to predict the type of ship engine fault.

[0022] Furthermore, the SCNet engine fault diagnosis model constructed in S4 includes a tiling data input layer, a first feature fusion layer, a second feature fusion layer, a third feature fusion layer, a fourth feature fusion layer, a random dropout layer, a fully connected layer, a softmax layer, and a classification layer connected in sequence.

[0023] The tiling data input layer is used to input the tiling data of the samples in the training set into the first feature fusion layer, and the first feature fusion layer includes a first convolutional layer and a first self-calibrating convolutional module connected in sequence.

[0024] The first convolutional layer is used to extract features from the output of the tiling data input layer to obtain a first extracted feature map; the first self-calibrating convolutional module is used to fuse features from the first extracted feature map to obtain a first optimized feature map.

[0025] The second feature fusion layer includes a second convolutional layer and a second self-calibrating convolutional module connected in sequence. The second convolutional layer is used to extract features from the first optimized feature map to obtain a second extracted feature map. The second self-calibrating convolutional module is used to fuse features from the second extracted feature map to obtain a second optimized feature map.

[0026] The third feature fusion layer includes a third convolutional layer and a third self-calibrating convolutional module connected in sequence. The third convolutional layer is used to extract features from the second optimized feature map to obtain a third extracted feature map. The third self-calibrating convolutional module is used to fuse features from the third extracted feature map to obtain a third optimized feature map.

[0027] The fourth feature fusion layer includes a fourth convolutional layer and a fourth self-calibrating convolutional module connected in sequence. The fourth convolutional layer is used to extract features from the third optimized feature map to obtain a fourth extracted feature map. The fourth self-calibrating convolutional module is used to fuse features from the fourth extracted feature map to obtain a fourth optimized feature map.

[0028] The random dropout layer is used to prevent the network from overfitting by optimizing the fourth optimized feature map, and then transmits the fourth feature map optimized by the random dropout layer to the fully connected layer.

[0029] The fully connected layer is used to transmit the output of the random loss layer to the Softmax layer, and to use the classifier in the Softmax layer to perform feature classification, and then transmit the feature map after feature classification to the classification layer.

[0030] The classification layer is used to output corresponding fault labels based on the feature maps after feature classification.

[0031] Furthermore, the first self-calibrating convolutional module, the second self-calibrating convolutional module, the third self-calibrating convolutional module, and the fourth self-calibrating convolutional module have the same structure, each having an input layer, a first feature branch, a second feature branch, and an output layer.

[0032] The input layer is used to transmit the corresponding input feature maps to the first feature branch and the second feature branch respectively; and the second feature branch includes the first branch path and the second branch path.

[0033] The first feature branch is used to perform a convolution operation on the input feature map to obtain a first convolutional feature map;

[0034] The first branch path is used to perform a convolution operation on the input feature map to obtain a second convolutional feature map. The second branch path is used to perform downsampling, convolution, and upsampling operations on the input feature map in sequence to obtain a global feature map that optimizes the receptive field. The global feature map is then weighted with the input feature map to obtain a weighted feature map. After performing a dot product operation between the weighted feature map and the second convolutional feature map using the sigmoid activation function, a convolution operation is then performed to obtain a self-correcting convolutional feature map.

[0035] The self-correcting convolutional feature map is concatenated with the first convolutional feature map, and the concatenated output feature map is output through the output layer.

[0036] Furthermore, S5 specifically includes the following steps:

[0037] S51: Use the predicted fault labels as output data, train the SCNet engine fault diagnosis model using the training set, and obtain the trained SCNet engine fault diagnosis model.

[0038] S52: Verify the convergence performance of the trained SCNet engine fault diagnosis model output: that is, determine whether the output of the trained SCNet engine fault diagnosis model converges.

[0039] If the output of the trained SCNet engine fault diagnosis model converges, then the trained SCNet engine fault diagnosis model is confirmed to be the optimal SCNet engine fault diagnosis model.

[0040] Otherwise, the weight parameters of the SCNet engine fault diagnosis model are adaptively adjusted using the backpropagation method, and step S51 is repeated.

[0041] Furthermore, the loss function used in S52 to determine whether the output of the trained SCNet engine fault diagnosis model converges is:

[0042]

[0043] In the formula: G(x) i pred ) represents the predicted fault label obtained after model training; y i true Represents the actual fault label; N represents the number of input samples; i represents the sample data number.

[0044] Furthermore, the expression for the noise reduction process in step S2 is as follows:

[0045] X(i,j)=X0(i,j)+meanX(j)a*(2rand-1)

[0046] In the formula: X(i,j) represents the data after noise processing; X0(i,j) represents the data before noise processing; i represents the sample data number of the ship engine failure; j represents the dimension of the failure feature of the sample data; meanX(j) represents the average value of the j-th dimension failure feature; a represents the noise intensity; (2rand-1) represents a random number with an amplitude of -1 to 1;

[0047] The expression for the normalization process is:

[0048]

[0049] In the formula: X norm X represents the linearly normalized data; X represents the original data; X max With X min These represent the maximum and minimum values ​​of a certain fault characteristic in the original data, respectively.

[0050] Beneficial Effects: This invention provides a ship engine fault diagnosis method based on SCNet. It obtains tiled sample data by sequentially performing noise reduction, normalization, and data tiling on sample data from a ship engine fault dataset. This allows for necessary padding of the sample data to ensure that convolution operations can be performed even at the edges of the data, avoiding the loss of edge information. By improving a standard convolutional neural network (CNN) and introducing a self-calibrating convolutional network module (SCBlock), a ship diesel engine SCNet fault diagnosis model is constructed, achieving automatic diagnosis of thermal faults in ship diesel engines. This model helps crew members quickly and accurately locate faults, reducing reliance on expert experience, and exhibits higher diagnostic accuracy compared to a simple CNN. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a flowchart of the SCNet-based marine engine fault diagnosis method of the present invention;

[0053] Figure 2 This is a schematic diagram of the Wärtsilä 9L34DF engine model built in this embodiment;

[0054] Figure 3 This is a schematic diagram of the SCNet fault diagnosis model for marine diesel engines constructed in this embodiment;

[0055] Figure 4 This is a schematic diagram of the self-calibrating convolution module (SCBlock) in this embodiment;

[0056] Figure 5 This is a flowchart of the SCNet fault diagnosis model training process in this embodiment;

[0057] Figure 6 This is a graph showing the CNN training error in this embodiment;

[0058] Figure 7 This is a schematic diagram of the CNN diagnostic result confusion matrix in this embodiment;

[0059] Figure 8 This is a graph showing the training error of SCNet in this embodiment;

[0060] Figure 9 This is a schematic diagram of the confusion matrix of SCNet diagnostic results in this embodiment. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] This embodiment provides a method for diagnosing marine engine faults based on SCNet, such as... Figure 1 As shown, the specific steps include:

[0063] S1: Obtain sample data of ship engine failures and corresponding failure labels to obtain a ship engine failure dataset;

[0064] Specifically, such as Figure 2 As shown, using the Wärtsilä 9L34DF engine model built with AVL-Boost simulation software as the data source, the simulation model data built with AVL-Boost was compared with the bench test report. The error was within 5%, which met the accuracy requirements and could truly reflect the state of the actual ship engine. Therefore, in the AVL-Boost simulation software, by setting global cases (events), fault sample data of the ship diesel engine under 6 states were collected. The method of using AVL-Boost software to build the Wärtsilä 9L34DF engine model is a well-known existing technology and will not be elaborated here. In the AVL-Boost simulation software, fault data can be obtained by setting global cases (events) in the Case Explorer, i.e., setting fault modes. This example selected six single faults for fault diagnosis experiments, namely six fault states: injection timing lag, increased exhaust valve clearance, reduced single-cylinder fuel supply, low turbocharger efficiency, low air cooler efficiency, and no fuel supply to cylinder 3. The specific states and state numbers of the experimental samples for the six fault states are shown in Table 1. The monitoring points of each diagnostic parameter of the selected marine diesel engine fault diagnosis characteristic parameters are shown in Table 2. There are a total of 13 diagnostic parameters, with 200 monitoring cycles for each state, collecting a total of 5200 sets of 13-dimensional data. The specific marine engine fault dataset obtained is shown in Table 3, with a size of 5200*13.

[0065] Table 1. Diagnostic Status Number and Status

[0066] F1 Fuel injection timing lag F2 Increased exhaust valve clearance F3 Reduced fuel supply per cylinder F4 Inefficient turbocharger F5 Air cooler inefficient F6 Cylinder No. 3 stops fuel supply

[0067] Table 2. Characteristic parameters for fault diagnosis of marine diesel engines

[0068]

[0069]

[0070] Table 3. Ship engine fault dataset

[0071]

[0072] S2: The sample data in the ship engine fault dataset is noise-processed to simulate the real ship engine data acquisition environment and obtain optimized sample data;

[0073] The optimized sample data is then preprocessed to obtain an optimized fault dataset;

[0074] Furthermore, preprocessing includes normalizing the optimized sample data to obtain normalized sample data;

[0075] Normalized sample data is tiled in one or two dimensions to obtain sample tiling data. In convolution operation, data tiling refers to expanding the input data to a size suitable for the convolution kernel. Through data tiling, it can be ensured that the convolution kernel can slide on the input data and perform calculations, thereby obtaining the output feature map.

[0076] Specifically, considering the special operating environment of ships, data acquisition is easily affected by noise, high temperature and vibration. Random noise is added to the sample dataset to simulate the real ship data acquisition environment. In order to avoid the problem of slow data convergence of neural networks due to different data types and value ranges, the collected ship engine operation data is normalized and the diagnostic coefficients are mapped to the [0-1] interval.

[0077] The expression for the noise reduction process is:

[0078] X(i,j)=X0(i,j)+meanX(j)*a*(2rand-1)

[0079] In the formula: X(i,j) represents the data after noise processing; X0(i,j) represents the data before noise processing; i represents the sample data number of the ship engine failure; j represents the dimension of the failure feature of the sample data; meanX(j) represents the average value of the j-th dimension failure feature; a represents the noise intensity; (2rand-1) represents a random number with an amplitude of -1 to 1;

[0080] The expression for the normalization process is:

[0081]

[0082] In the formula: X norm X represents the linearly normalized data; X represents the original data; X max With X min These represent the maximum and minimum values ​​of a certain fault characteristic in the original data, respectively.

[0083] S3: Randomly divide the optimized fault dataset into a training set, a validation set, and a test set according to a preset ratio;

[0084] S4: Construct an SCNet engine fault diagnosis model based on convolutional neural networks and self-calibrating convolutional modules; such as Figure 3As shown, the constructed SCNet engine fault diagnosis model specifically includes a tiling data input layer, a first feature fusion layer, a second feature fusion layer, a third feature fusion layer, a fourth feature fusion layer, a random dropout layer, a fully connected layer, a softmax layer, and a classification layer connected in sequence.

[0085] The tiling data input layer is used to input the tiling data of the samples in the training set into the first feature fusion layer, and the first feature fusion layer includes a first convolutional layer and a first self-calibrating convolutional module connected in sequence.

[0086] The first convolutional layer is used to extract features from the output of the tiling data input layer to obtain a first extracted feature map; the first self-calibrating convolutional module is used to fuse features from the first extracted feature map to obtain a first optimized feature map.

[0087] The second feature fusion layer includes a second convolutional layer and a second self-calibrating convolutional module connected in sequence. The second convolutional layer is used to extract features from the first optimized feature map to obtain a second extracted feature map. The second self-calibrating convolutional module is used to fuse features from the second extracted feature map to obtain a second optimized feature map.

[0088] The third feature fusion layer includes a third convolutional layer and a third self-calibrating convolutional module connected in sequence. The third convolutional layer is used to extract features from the second optimized feature map to obtain a third extracted feature map. The third self-calibrating convolutional module is used to fuse features from the third extracted feature map to obtain a third optimized feature map.

[0089] The fourth feature fusion layer includes a fourth convolutional layer and a fourth self-calibrating convolutional module connected in sequence. The fourth convolutional layer is used to extract features from the third optimized feature map to obtain a fourth extracted feature map. The fourth self-calibrating convolutional module is used to fuse features from the fourth extracted feature map to obtain a fourth optimized feature map.

[0090] The random dropout layer is used to prevent the network from overfitting by optimizing the fourth optimized feature map, and then transmits the fourth feature map optimized by the random dropout layer to the fully connected layer.

[0091] The fully connected layer is used to transmit the output of the random loss layer to the Softmax layer, and to use the classifier in the Softmax layer to perform feature classification, and then transmit the feature map after feature classification to the classification layer.

[0092] The classification layer is used to output corresponding fault labels based on the feature maps after feature classification.

[0093] Preferably, the first self-calibrating convolutional module, the second self-calibrating convolutional module, the third self-calibrating convolutional module, and the fourth self-calibrating convolutional module have the same structure, and each has an input layer, a first feature branch, a second feature branch, and an output layer.

[0094] The input layer is used to transmit the corresponding input feature maps to the first feature branch and the second feature branch respectively; and the second feature branch includes the first branch path and the second branch path.

[0095] The first feature branch is used to perform a convolution operation on the input feature map to obtain a first convolutional feature map;

[0096] The first branch path is used to perform a convolution operation on the input feature map to obtain a second convolutional feature map. The second branch path is used to perform downsampling, convolution, and upsampling operations on the input feature map in sequence to obtain a global feature map that optimizes the receptive field. The global feature map is then weighted with the input feature map to obtain a weighted feature map. After performing a dot product operation between the weighted feature map and the second convolutional feature map using the sigmoid activation function, a convolution operation is then performed to obtain a self-correcting convolutional feature map.

[0097] The self-correcting convolutional feature map is concatenated with the first convolutional feature map, and the concatenated output feature map is output through the output layer.

[0098] In this embodiment, the SCBlock module is as follows: Figure 4 As shown, this is a self-calibrating convolutional module that integrates multiple convolutional kernels while considering contextual relationships. It can expand the receptive field of input features, providing richer features for downstream tasks. Compared to conventional convolutional operations and attention mechanisms, it can achieve better results. In this embodiment, the pooling layers in the convolutional neural network are replaced with SCBlock modules (Self-Calibration) to construct the SCNet engine fault diagnosis model. The working process of the self-calibrating convolutional module is divided into two parts: a conventional convolutional module and an adaptive calibration module, as shown... Figure 4As shown. First, the input feature map X can be the output feature map of any convolutional layer in the CNN. Then, the input feature map is subjected to both ordinary convolution (K1) and self-calibration operations (K2, K3, K4). For the ordinary convolution operation, the input feature map X is convolved with a fixed-size kernel to obtain the output feature map Y2. This operation can preserve the spatial context as much as possible. In order to further expand the receptive field of the neural network, the input feature X is also input into the self-calibration convolution module: path K2 first performs a downsampling operation (r×Down) on the input, followed by convolution and upsampling operations (r×Up) to obtain a larger receptive field, improve the network's feature extraction and feature fusion capabilities, and is accompanied by residual operations to avoid gradient explosion caused by too many network layers. Then, the upsampled output is summed element-wise with the input feature map X, and then multiplied by the feature map obtained by the convolution operation of path K3 (element-wise summation) after passing through the sigmoid activation function. Multiplication) is used, and then through the K4 path, higher-level features are further extracted to obtain the final self-calibrating convolutional feature map Y1. Finally, the ordinary convolutional feature map of the K1 channel is concatenated with the final self-calibrating convolutional feature map to obtain the output feature map Y. After the features are extracted by the self-calibrating convolutional module, a fully connected layer is connected to combine and abstract the features extracted by the self-calibrating convolutional module to generate a representative feature vector. Finally, a softmax function is connected to achieve the classification task. At the same time, in order to prevent overfitting, a random loss layer is added to the structure. F1, F2, F3 and F4 are just letter representations of the feature maps (that is, F1 and F3 are the input feature map X; F2 is the feature map after the downsampling operation; F4 is the feature map after the dot product operation) and Identity is the input feature map X. The parameters of the convolutional neural network are shown in Table 4.

[0099] Table 4. Parameters of Convolutional Neural Networks

[0100]

[0101] S5: Train the SCNet engine fault diagnosis model based on the training set, and verify the performance of the trained SCNet engine fault diagnosis model through the validation set to obtain the optimal SCNet engine fault diagnosis model.

[0102] like Figure 5 As shown, the specific steps include:

[0103] S51: Use the predicted fault labels as output data, train the SCNet engine fault diagnosis model using the training set, and obtain the trained SCNet engine fault diagnosis model.

[0104] S52: Determine whether the output of the trained SCNet engine fault diagnosis model has converged;

[0105] If the output of the trained SCNet engine fault diagnosis model converges, then the trained SCNet engine fault diagnosis model is confirmed to be the optimal SCNet engine fault diagnosis model.

[0106] Otherwise, the weight parameters of the SCNet engine fault diagnosis model are adaptively adjusted using the backpropagation method, and step S51 is repeated.

[0107] Specifically, the loss function used to determine whether the output of the trained SCNet engine fault diagnosis model converges is:

[0108]

[0109] In the formula: G(x) i pred ) represents the predicted fault label obtained after model training; y i true Represents the actual fault label; N represents the number of input samples; i represents the sample data number.

[0110] In this embodiment, the pre-processed dataset Excel file is read from an Excel file using a Python script file. As shown in Table 3, there are 5200 sets of 13-dimensional data. The dataset is randomly divided into 0.8 sets for training and 0.2 sets for testing according to a preset ratio. During the SCNet neural network training process, the solver is set to Adam with the following parameters: 200 iterations, 8 batches, L2 regularization coefficient of 0.0001, and cross-entropy loss function. Training ends when the loss function is minimized or the maximum number of iterations is reached.

[0111] S6: Based on the optimal SCNet engine fault diagnosis model, the test set is used to predict the type of ship engine fault.

[0112] Experimental and Result Analysis in this Example

[0113] (1) Analysis of CNN model diagnostic results

[0114] After building the CNN model, the processed 4160 sets of data were input into the model for training. The resulting training error curve is shown in the figure below. Figure 6As shown, the model terminates training after reaching the maximum number of iterations. After model training, 1040 test samples are input into the CNN diagnostic model for diagnosis, and the results are as follows. Figure 7 As shown, the model's average diagnostic accuracy is 88%. The diagnostic results indicate that the constructed CNN diagnostic model can achieve fault diagnosis to a certain extent, but the results are not ideal and still need optimization.

[0115] (2) Analysis of SCNet model diagnostic results

[0116] By inserting the SCBlock module into the constructed CNN model, and inputting the processed 4160 sets of data into the model for training, the resulting training error curve is shown in the figure below. Figure 8 As shown, the model terminates training after reaching the maximum number of iterations. 1040 test samples were input into the SCNet diagnostic model for diagnosis, and the diagnostic results are as follows. Figure 9 As shown, the average accuracy of the model is 97%. The CNN optimized by SCBlock has higher diagnostic accuracy than the standalone CNN diagnostic model, and has higher recognition accuracy. It is effective for fault diagnosis of ship engines. Table 5 shows a comparison of the model diagnostic results of the CNN model and the SCNet model.

[0117] Table 5. Comparison of Model Diagnostic Results

[0118] CNN diagnostic model 1040 10.9％ 200 89.10％ SCNet diagnostic model 1040 2.17％ 200 97.83％

[0119] The present invention provides a marine engine fault diagnosis method based on SCNet, which has the following advantages.

[0120] 1. High applicability: The method of adding noise to simulate the real ship data acquisition environment improves the realism. Data normalization eliminates the influence of different dimensions and orders of magnitude of various diagnostic parameters on the diagnostic results.

[0121] 2. High efficiency and accuracy: By using convolutional neural networks (CNN) for feature extraction and representation learning, the information in the data is fully explored, reducing the need for manual features and improving the efficiency and accuracy of diagnosis.

[0122] 3. Strong adaptability: By introducing a self-calibrating convolution module (SCBlock), the relationship between subtasks can be enhanced. Through self-calibration operations, the correlation between long-range spatial domains and channels can be adaptively constructed, helping CNN to generate more discriminative feature representations. Because it has richer information, it can adapt to different fault modes and working conditions, improving the accuracy and stability of diagnosis.

[0123] 4. Good real-time performance: By adopting online data processing and real-time diagnosis, thermal faults can be detected and dealt with in a timely manner, reducing diesel engine downtime and maintenance costs.

[0124] In summary, the SCNet-based fault diagnosis method for marine diesel engines can automatically diagnose thermal faults in marine diesel engines, helping crew members to quickly and accurately locate the fault location, reducing reliance on expert experience, and achieving higher diagnostic accuracy compared to standalone CNNs. It can be widely promoted in the field of marine engineering.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for diagnosing marine engine faults based on SCNet, characterized in that, Specifically, the following steps are included: S1: Obtain sample data of ship engine failures and corresponding failure labels to obtain a ship engine failure dataset; S2: The sample data in the ship engine fault dataset is noise-processed to simulate the real ship engine data acquisition environment and obtain optimized sample data; The optimized sample data is then preprocessed to obtain an optimized fault dataset; Furthermore, preprocessing includes normalizing the optimized sample data to obtain normalized sample data; Perform one-dimensional or two-dimensional tiling on the normalized sample data to obtain the sample tiling data; S3: Randomly divide the optimized fault dataset into a training set and a test set according to a preset ratio; S4: Construct an SCNet engine fault diagnosis model based on a convolutional neural network and a self-calibrating convolutional module; the constructed SCNet engine fault diagnosis model includes a tiling data input layer, a first feature fusion layer, a second feature fusion layer, a third feature fusion layer, a fourth feature fusion layer, a random dropout layer, a fully connected layer, a softmax layer, and a classification layer connected in sequence. The tiling data input layer is used to input the tiling data of the samples in the training set into the first feature fusion layer, and the first feature fusion layer includes a first convolutional layer and a first self-calibrating convolutional module connected in sequence. The first convolutional layer is used to extract features from the output of the tiling data input layer to obtain a first extracted feature map; the first self-calibrating convolutional module is used to fuse features from the first extracted feature map to obtain a first optimized feature map. The second feature fusion layer includes a second convolutional layer and a second self-calibrating convolutional module connected in sequence. The second convolutional layer is used to extract features from the first optimized feature map to obtain a second extracted feature map. The second self-calibrating convolutional module is used to fuse features from the second extracted feature map to obtain a second optimized feature map. The third feature fusion layer includes a third convolutional layer and a third self-calibrating convolutional module connected in sequence. The third convolutional layer is used to extract features from the second optimized feature map to obtain a third extracted feature map. The third self-calibrating convolutional module is used to fuse features from the third extracted feature map to obtain a third optimized feature map. The fourth feature fusion layer includes a fourth convolutional layer and a fourth self-calibrating convolutional module connected in sequence. The fourth convolutional layer is used to extract features from the third optimized feature map to obtain a fourth extracted feature map. The fourth self-calibrating convolutional module is used to fuse features from the fourth extracted feature map to obtain a fourth optimized feature map. The random dropout layer is used to perform optimization based on the fourth optimized feature map to prevent network overfitting, and then transmits the fourth optimized feature map after optimization by the random dropout layer to the fully connected layer. The fully connected layer is used to transmit the output of the random dropout layer to the Softmax layer, and to perform feature classification using the classifier in the Softmax layer, and then transmit the feature map after feature classification to the classification layer. The classification layer is used to output the corresponding fault label based on the feature map after feature classification. S5: Train the SCNet engine fault diagnosis model based on the training set, and verify the performance of the trained SCNet engine fault diagnosis model to obtain the optimal SCNet engine fault diagnosis model. S6: Based on the optimal SCNet engine fault diagnosis model, the test set is used to predict the type of ship engine fault.

2. The method for diagnosing marine engine faults based on SCNet according to claim 1, characterized in that, The first self-calibrating convolution module, the second self-calibrating convolution module, the third self-calibrating convolution module, and the fourth self-calibrating convolution module have the same structure, each having an input layer, a first feature branch, a second feature branch, and an output layer; The input layer is used to transmit the corresponding input feature maps to the first feature branch and the second feature branch respectively; and the second feature branch includes the first branch path and the second branch path. The first feature branch is used to perform a convolution operation on the input feature map to obtain a first convolutional feature map; The first branch path is used to perform a convolution operation on the input feature map to obtain a second convolutional feature map. The second branch path is used to perform downsampling, convolution, and upsampling operations on the input feature map in sequence to obtain a global feature map that optimizes the receptive field. The global feature map is then weighted with the input feature map to obtain a weighted feature map. After multiplying the weighted feature map with the second convolutional feature map using the sigmoid activation function, a convolution operation is then performed to obtain a self-calibrated convolutional feature map. The self-calibrated convolutional feature map is concatenated with the first convolutional feature map, and the concatenated output feature map is output through the output layer.

3. The method for diagnosing marine engine faults based on SCNet according to claim 2, characterized in that, S5 specifically includes the following steps: S51: Use the predicted fault labels as output data, train the SCNet engine fault diagnosis model using the training set, and obtain the trained SCNet engine fault diagnosis model. S52: Verify the convergence performance of the trained SCNet engine fault diagnosis model output: that is, determine whether the output of the trained SCNet engine fault diagnosis model converges. If the output of the trained SCNet engine fault diagnosis model converges, then the trained SCNet engine fault diagnosis model is confirmed to be the optimal SCNet engine fault diagnosis model. Otherwise, the weight parameters of the SCNet engine fault diagnosis model are adaptively adjusted using the backpropagation method, and step S51 is repeated.

4. The method for diagnosing marine engine faults based on SCNet according to claim 3, characterized in that, The loss function used in S52 to determine whether the output of the trained SCNet engine fault diagnosis model converges is: In the formula: This represents the predicted fault label obtained after model training. Indicates the actual fault label; Indicates the number of input samples; Indicates the sample data number.

5. The method for diagnosing marine engine faults based on SCNet according to claim 1, characterized in that, The expression for the noise reduction process in step S2 is: In the formula: This represents the data before noise processing; The sample data number indicates a ship engine malfunction; Dimensions representing the fault characteristics of the sample data; Indicates the first The average value of the fault characteristics; Indicates noise intensity; This represents a random number with an amplitude between -1 and 1. The expression for the normalization process is: In the formula: This represents the data after linear normalization; Represents the original data; and These represent the maximum and minimum values ​​of a certain fault characteristic in the original data, respectively.

Images