Ship diesel engine fault early warning method and system based on deep learning CNN-BiLSTM

Abstract

The application discloses a kind of based on deep learning CNN-BiLSTM's ship diesel engine fault early warning method and system, whether the ship diesel engine is judged to occur fault and send early warning, to avoid the occurrence of safety accident.The method steps include: S1: obtaining original diesel engine operation data;S2: the original diesel engine operation data obtained is processed to obtain the processed diesel engine operation data;S3: construct CNN-BiLSTM model;S4: CNN-BiLSTM model is trained, and the CNN-BiLSTM model after training is used to predict exhaust temperature value;S5: the exhaust temperature value predicted using CNN-BiLSTM model and the exhaust temperature value actually obtained obtain the monitoring index of diesel engine operation state;S6: determine the reference value of diesel engine in abnormal state, according to the monitoring index of diesel engine operation state and the result of real-time comparison of reference value, whether the diesel engine is judged to occur fault and carries out fault early warning prompt.The system includes data acquisition module, data processing module and fault early warning module, realizes fault early warning function.

Description

Technical Field

[0001] This invention relates to the field of diesel engine fault identification technology, and in particular to a method and system for early warning of marine diesel engine faults based on deep learning CNN-BiLSTM. Background Technology

[0002] Marine diesel engines are important power equipment for ships. Due to the complexity of the structure of marine diesel engines and the complex and variable operating environment, it is necessary to monitor the reliability of marine diesel engine operation in real time.

[0003] Diesel engine exhaust temperature directly affects the performance and lifespan of a diesel engine and is one of the important parameters reflecting its operating status. Therefore, by predicting the exhaust temperature of a ship's diesel engine, it is possible to determine whether the engine has malfunctioned, thereby improving the safety and reliability of the ship.

[0004] Traditional methods for predicting the exhaust temperature of marine diesel engines are mainly based on statistical analysis and empirical models. These methods suffer from low prediction accuracy and poor adaptability, which affect the accuracy of subsequent early warnings of marine diesel engine failures. This can lead to the inability to detect and handle failures in a timely manner, resulting in economic losses in shipping and even threatening the lives and property of ship personnel. Summary of the Invention

[0005] This invention provides a method and system for early warning of marine diesel engine faults based on deep learning CNN-BiLSTM, in order to overcome the technical problem that the low prediction accuracy of traditional marine diesel engine exhaust temperature prediction methods affects the accuracy of early warning of marine diesel engine faults.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] A fault early warning method for marine diesel engines based on deep learning CNN-BiLSTM includes the following steps:

[0008] S1: Obtain raw diesel engine operating data;

[0009] S2: Process the acquired raw diesel engine operating data to obtain processed diesel engine operating data;

[0010] S3: Construct the CNN-BiLSTM model;

[0011] S4: Use the processed diesel engine operating data to train the CNN-BiLSTM model, and use the trained CNN-BiLSTM model to predict the exhaust temperature value.

[0012] S5: Use the exhaust temperature value predicted by the CNN-BiLSTM model and the actual exhaust temperature value to obtain the monitoring index of the diesel engine operating status;

[0013] S6: Determine the reference value for when the diesel engine is in an abnormal state. Based on the real-time comparison of the monitoring index of the diesel engine's operating status with the reference value, determine whether the diesel engine has malfunctioned and issue a fault warning.

[0014] Furthermore, in S1,

[0015] The original operating data of the diesel engine includes exhaust temperature data, high-temperature water-air cooler outlet temperature data, turbocharger speed data, gas inlet pressure data, engine load data, turbocharger outlet exhaust temperature data, high-temperature water cylinder liner inlet pressure data, and high-temperature water cylinder liner outlet temperature data.

[0016] Furthermore, in S2, the steps for processing the acquired raw diesel engine operating data are as follows:

[0017] S21: Normalize the acquired raw diesel engine operating data;

[0018] S22: Perform principal component analysis on the normalized running data to extract feature data;

[0019] S23: Remove outliers from the extracted feature data to obtain the processed diesel engine operating data.

[0020] Furthermore, in S4,

[0021] S41: Divide the processed diesel engine operating data into a training set and input the training set into the CNN-BiLSTM model for training;

[0022] S42: During training, mean absolute error (MAE), mean squared error (MSE), and mean absolute percentage error (MAPE) are used as metrics to evaluate the CNN-BiLSTM model. The optimal parameters are selected by controlling variables to obtain the optimal CNN-BiLSTM model.

[0023] Furthermore, in S5, the exhaust temperature values ​​predicted by the CNN-BiLSTM model and the actual obtained exhaust temperature values ​​are combined with Mahalanobis distance, and a mapping method is used to construct the monitoring index function MF for ship exhaust temperature, specifically including:

[0024] S51: Calculate the Mahalanobis distance between the exhaust temperature value predicted by the CNN-BiLSTM model and the actual obtained exhaust temperature value. The calculation formula is as follows:

[0025]

[0026] In the formula D MH x is the calculated value of Mahalanobis distance. i y is the column vector of the matrix of actual obtained exhaust temperature values.j S is the column vector of the predicted exhaust temperature value matrix. -1 For x i With y j The inverse matrix of the covariance;

[0027] S52: Construct the monitoring index function MF, and use the Mahalanobis distance D MH As input, obtain the monitoring index of the diesel engine's operating status;

[0028]

[0029] In the formula, α is the adjustment factor, and D is the Mahalanobis distance. MH The domain of MF is (0, +∞), and the range of MF is (0, 1).

[0030]

[0031] In the formula The mean of Mahalanobis distance, MF Q The confidence coefficient for normal exhaust temperature.

[0032] Furthermore, in S6, a reference value for determining when the diesel engine is in an abnormal state is established, and a warning value MF is set using the 3σ criterion. w and threshold MF f Alert value MF w The threshold value MF serves as a reference for potential abnormal operation of the diesel engine. f The reference values ​​for diesel engine malfunctions are as follows:

[0033] S61: Determine the warning value K for Mahalanobis distance 2σ Threshold K of Mahalanobis distance 3σ The calculation formula is as follows:

[0034]

[0035]

[0036] In the formula , respectively, are the mean and standard deviation of the Mahalanobis distance;

[0037] S62: According to K 2σ and K 3σ The warning value MF of the monitoring index was calculated. w and threshold MF f ;

[0038]

[0039]

[0040] S63: The monitoring index and warning value MF of the diesel engine operating status. w and threshold MF f For comparison, if the monitored index exceeds the warning value MF w This indicates a possible diesel engine malfunction. If the monitored index exceeds the threshold MF, it will indicate that the diesel engine may be malfunctioning. f It will then issue a warning indicating that the diesel engine has malfunctioned.

[0041] A fault early warning system for marine diesel engines based on deep learning CNN-BiLSTM includes a data acquisition module, a data processing module, and a fault early warning module. The data acquisition module is electrically connected to the data processing module, and the data processing module is electrically connected to the fault early warning module.

[0042] The data acquisition module is used to collect diesel engine operating data in real time and transmit it to the data processing module;

[0043] The data processing module is used to receive diesel engine operating data and transmit the processed diesel engine operating data to the fault warning module.

[0044] The fault early warning module is used to acquire the monitoring index of the diesel engine's operating status and compare the monitoring index with the set warning value MF. w and threshold MF f Real-time comparison will be performed; if the monitored index exceeds the warning value MF... w The fault warning module then issues a warning that the diesel engine may malfunction; if the monitored index exceeds the threshold MF... f The fault warning module then issues a warning that the diesel engine has malfunctioned.

[0045] Beneficial effects: This invention combines convolutional neural network (CNN) with bidirectional long short-term memory network (BiLSTM) to extract the spatiotemporal features of marine diesel engine exhaust temperature more comprehensively, solving the problem of low prediction accuracy of diesel engine exhaust temperature. This enables accurate judgment of whether marine diesel engines have malfunctioned. By monitoring the operating status of marine diesel engines in real time and issuing early warnings, measures can be taken in advance to avoid marine diesel engine malfunctions and safety accidents. Attached Figure Description

[0046] 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.

[0047] Figure 1This is a flowchart of the marine diesel engine fault early warning method based on deep learning CNN-BiLSTM in this invention;

[0048] Figure 2 This is a block diagram of the marine diesel engine fault early warning system based on deep learning CNN-BiLSTM in this invention;

[0049] Figure 3 This is a graph showing the contribution rate and cumulative contribution rate of each principal component in this invention;

[0050] Figure 4 This is a framework diagram of the Convolutional Neural Network (CNN) in this invention;

[0051] Figure 5 This is a framework diagram of the Bidirectional Long Short-Term Memory (BiLSTM) network in this invention;

[0052] Figure 6 The image shows the prediction results of the CNN-BiLSTM model in this invention.

[0053] Figure 7 This is a monitoring index diagram for fault prediction in this invention. Detailed Implementation

[0054] 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.

[0055] Current research on fault early warning for marine diesel engines is generally divided into state prediction type and state classification type. This invention uses the state prediction method, which uses the real-time exhaust temperature value of the diesel engine to predict the exhaust temperature value at a future time. Then, it uses mathematical statistics to analyze the deviation between the predicted exhaust temperature value and the actual exhaust temperature value to set the early warning value for fault warning. The reliability of the method of this invention is verified through experiments.

[0056] This invention provides a method for early warning of marine diesel engine faults based on deep learning CNN-BiLSTM, such as... Figure 1 As shown, the process includes the following steps: S1: Obtain the raw diesel engine operating data, some of which are shown in Table 1.

[0057] Table 1:

[0058] 1 469.8 88 9420 7.47 19.1 384.6 3.34 87.9 … … … … … … … … 2000 463.6 87.8 9220 7.49 17.9 385.6 3.33 87.8 ... ... ... ... ... ... ... ... 4000 472.5 88.1 9500 7.47 19.3 387.1 3.39 88 ... ... ... ... ... ... ... ... 6000 471.8 87.9 9580 7.47 19.3 387 3.35 87.7 ... ... ... ... ... ... ... ...

[0059] In this embodiment, the original diesel engine operating data is the actual operating data of the 6L34DF dual-fuel generator diesel engine of an LNG ship. In order to reduce the amount of calculation, improve the program running speed, and reduce the impact of engine room noise, the operating data is sampled at intervals of 6 seconds, with a total of 7200 monitoring points selected. Eight operating parameters of this type of diesel engine are selected, i.e., 7200*8. The collected data is the data of the diesel engine of the ship operating under stable conditions.

[0060] Specifically, the selected operating parameter data for the eight diesel engines included exhaust temperature data T. P The unit is ℃; the high-temperature water-air cooler outlet temperature data T1 is in ℃; the turbocharger speed data N is in r / min; the gas inlet pressure data P1 is in MPa; the engine load data L is in KW; the turbocharger outlet exhaust temperature data T2 is in ℃; the high-temperature water cylinder liner inlet pressure data P2 is in MPa; and the high-temperature water cylinder liner outlet temperature data T3 is in ℃.

[0061] Among these parameters, exhaust temperature reflects the ship's combustion efficiency and engine performance; excessively high exhaust temperature may indicate incomplete combustion or a problem with the diesel engine. High-temperature water-air cooler outlet temperature reflects whether the engine's cooling water circulation is normal and whether the engine load is excessive. Turbocharger speed reflects the main engine load. Gas inlet pressure reflects the working status of the gas supply system and the stability of gas quality. Engine load reflects the ship's operating status and performance. Turbocharger outlet exhaust temperature reflects the combustion efficiency of the ship's main engine and the working status of the turbocharger. High-temperature water cylinder liner inlet pressure reflects the pressure status of the ship's cooling system. High-temperature water cylinder liner outlet temperature reflects the engine's thermal load and the working efficiency of the cooling system.

[0062] S2: Process the acquired raw diesel engine operating data to obtain processed diesel engine operating data;

[0063] In a specific embodiment, the steps for processing the acquired raw diesel engine operating data are as follows:

[0064] S21: To avoid the problem of slow convergence speed of neural network training due to differences in the dimensions and orders of magnitude of various operating data, the acquired raw diesel engine operating data is normalized. The normalization formula is as follows:

[0065]

[0066] In the formula, x norm The data is normalized, and x represents the original diesel engine operating data. max x min These are the maximum and minimum values ​​in the original diesel engine operating data, respectively.

[0067] S22: Perform principal component analysis on the normalized running data to extract feature data;

[0068] Specifically, Python code was used to perform dimensionality reduction on the normalized data of eight operating parameters, retaining key information and extracting the main features from the data. The formulas for calculating the principal component scores are as follows:

[0069]

[0070]

[0071] In the formula: F i For the score of the i-th principal component, X j For runtime parameters, W ij The weights of each variable in the principal components, θ i These are the coefficients corresponding to each variable in the component matrix. By taking the square root of the eigenvalues ​​corresponding to the principal components, principal component analysis was performed on the normalized data of the eight operating parameters to obtain the correlation coefficient matrix shown in Table 2, the variance contribution rate of each component shown in Table 3, and the principal component coefficient matrix shown in Table 4.

[0072] Table 2:

[0073] <![CDATA[T P ]]> 1.000 0.608 0.920 -0.043 0.943 0.389 -0.007 0.553 <![CDATA[T1]]> 0.608 1.000 0.512 0.002 0.547 0.310 0.000 0.615 N 0.920 0.512 1.000 -0.026 0.950 0.464 0.006 0.377 <![CDATA[P1]]> -0.043 0.002 -0.026 1.000 -0.035 0.011 0.016 -0.020 L 0.943 0.547 0.950 -0.035 1.000 0.369 0.001 0.459 <![CDATA[T2]]> 0.389 0.310 0.464 0.011 0.369 1.000 0.009 0.237 <![CDATA[P2]]> -0.007 0.000 0.006 0.016 0.001 0.009 1.000 -0.013 <![CDATA[T3]]> 0.553 0.615 0.377 -0.020 0.459 0.237 -0.013 1.000

[0074] Table 3:

[0075] 1 3.873048 48.4131 48.4131 2 1.019800 12.7475 61.1606 3 0.986344 12.3293 73.4899 4 0.894839 11.8612 85.3511 5 0.757531 9.4691 94.8202

[0076] Table 4:

[0077] <![CDATA[T P ]]> 0.246275 -0.024683 -0.027972 -0.098546 -0.263373 <![CDATA[T1]]> 0.190904 0.017818 0.089242 0.503729 0.174280 N 0.236658 0.010854 -0.053009 -0.348286 -0.235904 <![CDATA[P1]]> -0.009272 0.716062 0.680180 -0.049845 -0.150673 L 0.240172 -0.012937 -0.046322 -0.215064 -0.344556 <![CDATA[T2]]> 0.136887 0.113231 0.021580 -0.415051 0.992566 <![CDATA[P2]]> -0.000351 0.672891 -0.726202 0.140762 -0.014527 <![CDATA[T3]]> 0.169072 -0.031991 0.098301 0.701387 0.194603

[0078] Principal component analysis was performed on the normalized data of eight parameters. First, the correlation coefficients between the eight parameters were calculated. The closer the absolute value is to 1, the stronger the linear relationship between the variables; the closer the absolute value is to 0, the weaker the linear relationship. Generally, an absolute value ≥ 0.8 is considered highly correlated, 0.5-0.8 moderately correlated, 0.3-0.5 lowly correlated, and an absolute value < 0.3 indicates extremely weak correlation, which can be considered as no correlation. In this embodiment, variables with an absolute correlation coefficient ≥ 0.5 related to the exhaust temperature of marine diesel engines were used as input parameters. Based on the KMO and Bartlett tests, the KMO value was 0.786 and the Sig value was 0.000, indicating a high degree of correlation among the characteristic factors, meeting the requirements of principal component analysis. Principal component analysis was then performed to extract principal components with eigenvalues ​​greater than 0.5 and a variance contribution rate of approximately 95%. Figure 3As shown, the cumulative variance contribution rate of the first five principal components reached 94.145%, indicating that these five principal components can represent 94.145% of the information from the original eight operating parameter data, and can represent the selected normalized diesel engine operating data. Therefore, only the first five feature data are needed for the research on marine diesel engine exhaust temperature prediction, reducing data redundancy and improving the diagnostic efficiency of the model.

[0079] S23: Remove outliers from the extracted feature data to obtain the processed diesel engine operating data.

[0080] Specifically, in this embodiment, Mahalanobis distance is used to filter out outliers, i.e., those abnormal values ​​that differ significantly from other data points. Mahalanobis distances are established between exhaust temperature and high-temperature water-air cooler outlet temperature, turbocharger speed, engine load, and high-temperature water cylinder liner outlet temperature. Then, a chi-square test is used to filter out outliers in exhaust temperature. The overall degrees of freedom of the extracted feature data is 4, and the significance level for determining the Mahalanobis distance is 0.005. According to the chi-square test, the chi-square degrees of freedom are calculated to be 14.86026. Therefore, points with a Mahalanobis distance greater than 14.86026 are outliers. By using Mahalanobis distance to filter outliers, 80 outliers are removed from the extracted feature data, eliminating inaccurate and untrue data collected due to the special working environment of ships, and improving the accuracy of fault prediction. Some of the data is shown in Table 5.

[0081] Table 5:

[0082] 1 0.439024 0.5 0.341772 0.470588 0.571429 ... ... ... ... ... 1500 0.158537 0.375 0.113924 0.176471 0.428571 ... ... ... ... ... 3000 0.213415 0.25 0.126582 0.176471 0.285714 ... ... ... ... ... 4500 0.091463 0.25 0.164557 0.088235 0.142857

[0083] S3: A CNN-BiLSTM model was constructed using the PyCharm platform and Python 3.6 environment, relying on the PyThorch deep learning framework. This model includes a Convolutional Neural Network (CNN) framework and a Bidirectional Long Short-Term Memory (BiLSTM) framework. A one-dimensional convolutional neural network is used to mine time-series data to extract local features, thereby improving the accuracy and reliability of the prediction model. The BiLSTM, composed of a forward LSTM and a backward LSTM, can simultaneously utilize information from historical and future moments to predict the current diesel engine operating state. Through forward and backward propagation, the BiLSTM can more comprehensively capture bidirectional semantic dependencies, thus better understanding time-series data. Therefore, the BiLSTM exhibits better feature representation, stronger modeling capabilities, and better predictive expression in the problem of marine diesel engine exhaust temperature prediction. Figure 4 and Figure 5 As shown, Figure 5In the diagram, w1 represents the weight coefficients from the input layer to the preceding LSTM; w2 represents the weight coefficients between units in the forward LSTM; w3 represents the weight coefficients from the input layer to the backward LSTM; w5 represents the weight coefficients between units in the backward LSTM; w4 represents the weight coefficients from the forward LSTM to the output layer; w6 represents the weight coefficients from the backward LSTM to the output layer; and h represents the weight coefficients. ft For the hidden state of the forward LSTM, h bt For backward LSTM hidden layers, h t This is the output value after passing through the BiLSTM model.

[0084] S4: Use the processed diesel engine operating data to train the CNN-BiLSTM model, and use the trained CNN-BiLSTM model to predict the exhaust temperature value.

[0085] In a specific embodiment, S4 includes,

[0086] S41: Divide the first 70% of the processed diesel engine operating data into a training set and 70%-100% into a test set. The data in the training set and the data in the test set are continuous in time. Input the training set into the CNN-BiLSTM model for training.

[0087] The specific training process involves extracting the spatial features of the temporal sequence from the input data through convolutional layer 1. The output spatial feature data undergoes dimensionality reduction processing via a max pooling layer. The dimensionality-reduced spatial feature data is then input into the hidden state h of the feedforward LSTM. ft and backward LSTM hidden layer h bt Then, the hidden state h of the forward LSTM is... ft and backward LSTM hidden layer h bt Connected together, they are fed into the output layer for prediction; the BiLSTM model outputs h. t The output layer uses a fully connected layer 1 to connect the output h of the BiLSTM. t The final output is the trained CNN-BiLSTM model data, which incorporates the non-linear activation function ReLU.

[0088] S42: During training, the mean absolute error (MAE), mean squared error (MSE), and mean absolute percentage error (MAPE) are used as metrics to evaluate the CNN-BiLSTM model, as shown in Table 6.

[0089] Table 6:

[0090] MAE 0.27996633 MSE 0.16509529 MSPE 0.00059659

[0091] The optimal parameters, as shown in Table 7, were selected using the controlled variable method, resulting in the optimal CNN-BiLSTM model.

[0092] Table 7:

[0093] cnn_layers Convolutional layer 1 pooling_layer Pooling layer Max Pool num_layers LSTM stacking layers 1 input_size Input feature dimension 5 hidden_size Hidden layer state dimension 64 seq_len Predicted step size 5 batch_size Sample training batch 100 epochs Number of training rounds 1000 learning_rate Learning rate 0.0005 loss loss function MSE loss optimizer Optimizer Adam

[0094] This embodiment uses Dropout technology to optimize the model and prevent overfitting of the CNN-BiLSTM model during training. Dropout technology reduces the dependency between nodes by randomly discarding nodes during training, thereby avoiding overfitting. In this embodiment, the Dropout value is selected as 0.5.

[0095] The data from the training set is input into the trained CNN-BiLSTM model, and the predicted exhaust temperature values ​​are compared with the actual exhaust temperature values ​​in the test set. The comparison results are as follows: Figure 6 As shown in the figure, the predicted exhaust temperature value is basically consistent with the actual exhaust temperature value in the test set, with a large number of overlaps. Moreover, the data fluctuation range is consistent with the normal operating range of the diesel engine, indicating that the CNN-BiLSTM model established in this invention has high prediction accuracy and precision, and the prediction results can meet the requirements of practical applications.

[0096] S5: The exhaust temperature value predicted by the CNN-BiLSTM model and the actual obtained exhaust temperature value are combined with Mahalanobis distance, and a monitoring index function MF for ship exhaust temperature is constructed using a mapping method. In a specific embodiment, S5 includes:

[0097] S51: Calculate the Mahalanobis distance between the exhaust temperature value predicted by the CNN-BiLSTM model and the actual obtained exhaust temperature value. The calculation formula is as follows:

[0098]

[0099] In the formula D MH x is the calculated value of Mahalanobis distance. i y is the column vector of the matrix of actual obtained exhaust temperature values. j S is the column vector of the predicted exhaust temperature value matrix. -1 For x i With y j The inverse matrix of the covariance;

[0100] S52: Due to the high uncertainty in the magnitude of the Mahalanobis distance, a function mapping method is used to fix it within a certain range. By constructing the monitoring index function MF, the Mahalanobis distance D is... MH The domain of definition is (0, +∞), and the range of MF is (0, 1). The Mahalanobis distance D... MH The input is mapped to the interval [0, 1] to obtain the monitoring index of the diesel engine's operating status, thereby accurately and intuitively monitoring the diesel engine's operating status;

[0101]

[0102] In the formula, α is the adjustment factor;

[0103]

[0104] In the formula The mean of Mahalanobis distance, MF Q The confidence coefficient for normal exhaust temperature;

[0105] S6: Based on the monitoring index MF, a reference value for when the diesel engine is in an abnormal state is determined, which can more intuitively and effectively monitor the operating status of the diesel engine. Therefore, a warning value and a threshold value for exhaust temperature are set. The warning value is a reference value for the diesel engine to work abnormally, and the threshold value is a reference value for diesel engine failure. Based on the real-time comparison of the monitoring index of the diesel engine operating status with the reference value, it is determined whether the diesel engine has failed and a fault warning is issued.

[0106] Specifically, the 3σ criterion is used to set the warning value MF. w and threshold MF f The specific steps are as follows:

[0107] S61: Determine the warning value K for Mahalanobis distance 2σ Threshold K of Mahalanobis distance 3σ The calculation formula is as follows:

[0108]

[0109]

[0110] In the formula are the mean and standard deviation of the Mahalanobis distance, respectively;

[0111] S62: According to K 2σ and K 3σ The warning value MF of the monitoring index was calculated. w and threshold MF f ;

[0112]

[0113]

[0114] S63: The monitoring index and warning value MF of the diesel engine operating status. w and threshold MF f For comparison, if the monitored index exceeds the warning value MF w This indicates a possible diesel engine malfunction. If the monitored index exceeds the threshold MF, it will indicate that the diesel engine may be malfunctioning. fIt will then issue a warning indicating that the diesel engine has malfunctioned.

[0115] Specifically, a method of manually constructing a dataset is used to simulate the state data when a marine diesel engine malfunctions. This embodiment simulates a high-temperature water-air cooler with blocked water pipes, causing the air cooler to be unable to effectively dissipate heat, resulting in a decrease in the air cooler's cooling effect, and subsequently triggering a malfunction state in which the exhaust temperature of the marine diesel engine abnormally rises. Figure 7 As shown, before the fault was introduced, the monitoring index was consistently higher than the monitoring threshold, indicating that the diesel engine was in a relatively stable operating state. It fluctuated at certain times, but the fluctuations remained within the threshold. After a brief fluctuation, it automatically returned to the warning value, thus only indicating a possible fault in the diesel engine. After the fault was introduced, the monitoring index fluctuated for a period before experiencing a sharp drop. When the monitoring index remained below the threshold, a warning was issued indicating that the diesel engine had already failed. Therefore, the marine diesel engine fault warning method proposed in this embodiment can meet the requirements of real-time monitoring and evaluation of marine diesel engine exhaust temperature, while also issuing fault warnings for abnormal exhaust temperature in the future. It has high accuracy and sensitivity, and can basically meet the needs of modern marine diesel engine exhaust temperature fault warning.

[0116] like Figure 2 As shown, a marine diesel engine fault early warning system based on deep learning CNN-BiLSTM includes a data acquisition module, a data processing module, and a fault early warning module. The data acquisition module is electrically connected to the data processing module, and the data processing module is electrically connected to the fault early warning module.

[0117] The data acquisition module is used to collect diesel engine operating data in real time and transmit it to the data processing module;

[0118] The data processing module is used to receive diesel engine operating data and transmit the processed diesel engine operating data to the fault warning module.

[0119] The fault early warning module is used to acquire the monitoring index of the diesel engine's operating status and compare the monitoring index with the set warning value MF. w and threshold MF f Real-time comparison will be performed; if the monitored index exceeds the warning value MF... w The fault warning module then issues a warning that the diesel engine may malfunction; if the monitored index exceeds the threshold MF... f The fault warning module then issues a warning that the diesel engine has malfunctioned.

[0120] 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 therein. Such 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 early warning of marine diesel engine faults based on deep learning CNN-BiLSTM, characterized in that, Includes the following steps: S1: Obtain raw diesel engine operating data; the raw operating data of the diesel engine includes exhaust temperature data, high-temperature water-air cooler outlet temperature data, turbocharger speed data, gas inlet pressure data, engine load data, turbocharger outlet exhaust temperature data, high-temperature water cylinder liner inlet pressure data, and high-temperature water cylinder liner outlet temperature data. S2: Process the acquired raw diesel engine operating data to obtain processed diesel engine operating data; S3: Construct the CNN-BiLSTM model; S4: Use the processed diesel engine operating data to train the CNN-BiLSTM model, and use the trained CNN-BiLSTM model to predict the exhaust temperature value. Training the CNN-BiLSTM model includes: The spatial features of the temporal sequence of the input data are extracted by a one-dimensional convolutional neural network. The output spatial feature data is then processed by a max pooling layer for dimensionality reduction. The spatial feature data after dimensionality reduction is then input into the hidden state of the forward LSTM and the hidden state of the backward LSTM, respectively. The hidden state of the forward LSTM and the hidden state of the backward LSTM are then connected together and fed into the output layer for prediction. The output layer uses a fully connected layer to introduce the output of BiLSTM into the non-linear activation function ReLU, and finally outputs the trained CNN-BiLSTM model data. S5: Use the exhaust temperature value predicted by the CNN-BiLSTM model and the actual exhaust temperature value to obtain the monitoring index of the diesel engine operating status; In S5, the exhaust temperature values ​​predicted by the CNN-BiLSTM model and the actual obtained exhaust temperature values ​​are combined with Mahalanobis distance, and a monitoring index function MF for ship exhaust temperature is constructed using a mapping method. Specifically, this includes: S51: Calculate the Mahalanobis distance between the exhaust temperature value predicted by the CNN-BiLSTM model and the actual obtained exhaust temperature value. The calculation formula is as follows: In the formula The value is the Mahalanobis distance. This is the column vector of the matrix of actual obtained exhaust temperature values. This is the column vector of the matrix of predicted exhaust temperature values. for The inverse matrix of the covariance; S52: Construct the monitoring index function MF, and use Mahalanobis distance As input, obtain the monitoring index of the diesel engine's operating status; ； In the formula Mahalanobis distance is used as an adjustment factor. The domain of MF is (0,+∞), and the range of MF is (0,1). ； In the formula The mean of Mahalanobis distance, The confidence coefficient for normal exhaust temperature; S6: Determine the reference value for when the diesel engine is in an abnormal state. Based on the real-time comparison of the monitoring index of the diesel engine's operating status with the reference value, determine whether the diesel engine has malfunctioned and provide a fault warning. In S6, a reference value for determining the abnormal state of the diesel engine is established, and a warning value is set using the 3σ criterion. and threshold Alert value Threshold values ​​are reference values ​​for potential abnormal operation of diesel engines. The reference values ​​are for diesel engine malfunctions. The specific steps are as follows: S61: Determine the warning value for Mahalanobis distance Threshold of Mahalanobis distance The calculation formula is as follows: ； ； In the formula , are the mean and standard deviation of the Mahalanobis distance, respectively; S62: According to and The warning value of the monitoring index was calculated. and threshold ; ； ； S63: Monitoring indices and warning values ​​for diesel engine operating status. and threshold By comparison, if the monitored index exceeds the warning value... This indicates a possible diesel engine malfunction; if the monitored index exceeds the threshold... It will then issue a warning indicating that the diesel engine has malfunctioned.

2. The marine diesel engine fault early warning method based on deep learning CNN-BiLSTM according to claim 1, characterized in that, In S2, the steps for processing the acquired raw diesel engine operating data are as follows: S21: Normalize the acquired raw diesel engine operating data; S22: Perform principal component analysis on the normalized running data to extract feature data; S23: Remove outliers from the extracted feature data to obtain the processed diesel engine operating data.

3. The marine diesel engine fault early warning method based on deep learning CNN-BiLSTM according to claim 1, characterized in that, In S4, S41: Divide the processed diesel engine operating data into a training set and input the training set into the CNN-BiLSTM model for training; S42: During training, mean absolute error (MAE), mean squared error (MSE), and mean absolute percentage error (MAPE) are used as metrics to evaluate the CNN-BiLSTM model. The optimal parameters are selected by controlling variables to obtain the optimal CNN-BiLSTM model.

4. A marine diesel engine fault early warning system based on the deep learning CNN-BiLSTM-based fault early warning method for marine diesel engines according to any one of claims 1-3, characterized in that, It includes a data acquisition module, a data processing module, and a fault early warning module. The data acquisition module is electrically connected to the data processing module, and the data processing module is electrically connected to the fault early warning module. The data acquisition module is used to collect diesel engine operating data in real time and transmit it to the data processing module; The data processing module is used to receive diesel engine operating data and transmit the processed diesel engine operating data to the fault warning module. The fault early warning module is used to acquire the monitoring index of the diesel engine's operating status and compare the monitoring index with the set warning value. and threshold Real-time comparisons are performed; if the monitored index exceeds the warning value... The fault warning module then issues a warning that the diesel engine may malfunction; If the monitoring index exceeds the threshold The fault warning module then issues a warning that the diesel engine has malfunctioned.

Images