Ship engine fault diagnosis method, ship engine model training method, ship engine fault diagnosis equipment and storage medium

Abstract

The embodiment of the invention provides a ship engine fault diagnosis method, a model training method, equipment and a storage medium, and belongs to the technical field of equipment fault detection. According to the method, real data and virtual data of a ship engine under different working conditions are acquired to correspondingly obtain a real data set and a virtual data set, and the real data set and the virtual data set are input into a fusion network for data fusion to obtain network fusion data; and inputting the network fusion data into a fault prediction network for fault prediction to obtain a fault prediction result, updating parameters of the fusion network and the fault prediction network according to the fault prediction result and a fault real result, and determining a ship engine fault diagnosis model according to the trained fault prediction network. According to the invention, the accuracy of ship engine fault diagnosis can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of equipment fault detection, and particularly to a method for diagnosing faults of a marine engine, a method for training a model, a device, and a storage medium. Background Art

[0002] A marine diesel engine is a heat engine that provides power by burning diesel oil and is the main power source for marine navigation. Its reliability directly affects the navigation safety of the entire ship. The diesel engine has a complex composition structure, a high degree of correlation between systems and components, and the operation of the equipment is affected by multiple factors such as the environment and humans. It is very difficult to detect when a fault occurs.

[0003] In recent years, fault diagnosis methods have emerged and developed continuously, and artificial intelligence technology has continuously appeared in the field of fault diagnosis, gradually forming an intelligent diagnosis technology based on vibration signals and centered on artificial intelligence technology. However, in real industrial scenarios, it is very difficult to collect a large amount of fault data, which will lead to problems such as overfitting and poor generalization ability of the fault diagnosis model established from small sample data. In related technologies, the virtual-real fusion technology can construct a virtual simulation model of a marine diesel engine, and by simulating the operation conditions of the marine diesel engine under different working conditions, generate signals of different fault types under different working conditions, so as to achieve the purpose of data enhancement. However, since the virtual simulation data comes from the virtual simulation model of the marine engine, the reliability of the virtual simulation data depends on the accuracy of the virtual simulation model. When the reliability of the virtual simulation data is low, it will lead to low accuracy of the constructed marine engine fault diagnosis model, thus resulting in low accuracy of marine engine fault diagnosis. Summary of the Invention

[0004] The main purpose of the embodiments of the present application is to propose a method for diagnosing faults of a marine engine, a method for training a model, a device, and a storage medium, aiming to improve the accuracy of marine engine fault diagnosis.

[0005] To achieve the above object, on the one hand, an embodiment of the present application proposes a method for training a marine engine fault diagnosis model, including the following steps:

[0006] Obtain real data and virtual data of the marine engine under different working conditions, and correspondingly obtain a real data set and a virtual data set;

[0007] Input the real data set and the virtual data set into a fusion network for data fusion to obtain network fusion data;

[0008] Input the network fusion data into a fault prediction network for fault prediction to obtain a fault prediction result;

[0009] Update the parameters of the fusion network and the fault prediction network according to the fault prediction result and the true fault result, and determine the ship engine fault diagnosis model according to the trained fault prediction network;

[0010] Among them, the fusion network includes a plurality of fusion modules connected in sequence, and the input of the fusion module is the output of the previous-level fusion module and the true data set.

[0011] In some embodiments, the virtual data is obtained through the following steps:

[0012] Determine the state parameters of the ship engine under the target working condition according to the ship engine operation control model;

[0013] Perform thermodynamic simulation according to the thermodynamic calculation model of the ship engine and the state parameters to obtain in-cylinder simulation data;

[0014] Perform dynamic simulation according to the dynamic model of the ship engine, the state parameters and the in-cylinder simulation data to obtain cylinder wall vibration simulation data;

[0015] Form virtual data according to the state parameters, the in-cylinder simulation data and the cylinder wall vibration simulation data.

[0016] In some embodiments, the step of inputting the true data set and the virtual data set into the fusion network for data fusion to obtain network fusion data includes the following steps:

[0017] Input the true data set and the virtual data set into the first-level fusion module for data fusion to obtain the output of the first-level fusion module;

[0018] Input the output of the (n - 1)-th level fusion module and the true data set into the n-th level fusion module for data fusion to obtain the output of the n-th level fusion module, where n ranges from 2 to N, and N is the number of fusion modules;

[0019] Determine the output of the N-th level fusion module as the network fusion data.

[0020] In some embodiments, the fusion module includes two modal-level decision modules and a fusion unit. The two modal-level decision modules are respectively used to process the two inputs of the fusion module, and the fusion unit is used to process the outputs of the two modal-level decision modules;

[0021] The modal-level decision module is used to perform modal judgment and selection on the input data, and perform modal data fusion operation using the expert network of the corresponding modal according to the selected modal to obtain the output of the modal-level decision module;

[0022] The fusion unit is used to make a fusion operation decision based on the outputs of the two modality-level decision modules, and fuse the outputs of the two modality-level decision modules based on the determined fusion operation to obtain the output of the fusion module.

[0023] In some embodiments, the fusion network is updated through the following steps:

[0024] Determine the classification task loss according to the fault prediction result and the true fault result;

[0025] Determine a first loss function according to the classification task loss and the fusion operation calculation cost in the modality-level decision module, and update the parameters of the modality-level decision module with the goal of minimizing the first loss function;

[0026] Determine a second loss function according to the classification task loss and the fusion operation calculation cost in the fusion module, and update the parameters in the fusion module with the goal of minimizing the second loss function.

[0027] In some embodiments, the step of inputting the network fusion data into the fault prediction network for fault prediction to obtain a fault prediction result includes the following steps:

[0028] Divide the network fusion data according to the data modality to obtain sample data of different modality combinations;

[0029] Extract features from the sample data to obtain a feature fusion covariance matrix;

[0030] Perform feature prediction on the feature fusion covariance matrix to obtain a fault prediction result.

[0031] In some embodiments, the step of performing feature prediction on the feature fusion covariance matrix to obtain a fault prediction result includes the following steps:

[0032] Input the feature fusion covariance matrix into a multi-kernel ridge regression model for feature prediction to obtain a fault prediction result;

[0033] The fault prediction result is used to determine whether the ship engine is faulty and the type of fault.

[0034] To achieve the above object, another aspect of the embodiments of the present application proposes a ship engine fault diagnosis method, including the following steps:

[0035] Obtain the operation data of the ship engine;

[0036] Input the operation data of the ship engine into the ship engine fault diagnosis model for fault prediction to obtain a fault prediction result;

[0037] Among them, the ship engine fault diagnosis model is obtained by training through the ship engine fault diagnosis model training method described in the previous embodiments.

[0038] To achieve the above object, on the other hand, an embodiment of the present application provides an electronic device, which includes a memory, a processor, a program stored on the memory and executable on the processor, and a data bus for realizing connection communication between the processor and the memory. When the program is executed by the processor, the method described in the above embodiments is realized.

[0039] To achieve the above object, on the other hand, an embodiment of the present application provides a storage medium, which is a computer-readable storage medium for computer-readable storage. The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to realize the method described in the above embodiments.

[0040] The ship engine fault diagnosis method, model training method, device and storage medium proposed by the present application obtain a real data set and a virtual data set by corresponding the real data and virtual data of the ship engine under different working conditions, input the real data set and the virtual data set into a fusion network for data fusion to obtain network fusion data, input the network fusion data into a fault prediction network for fault prediction to obtain a fault prediction result, and then update the parameters of the fusion network and the fault prediction network according to the fault prediction result and the real fault result, and then determine the ship engine fault diagnosis model according to the fault prediction network after training. The fusion network of the present application includes a plurality of fusion modules connected in sequence, and the input of the fusion module is the output of the previous-level fusion module and the real data set. Therefore, in the process of supplementing the virtual data set to the real data set, the real data set can be used to adjust the data fusion result each time, so that the network fusion data is closer to the real data, indirectly improving the reliability of the virtual data, and thus improving the accuracy of the ship engine fault diagnosis model. Description of the Drawings

[0041] Figure 1 is a flowchart of the ship engine fault diagnosis model training method provided by an embodiment of the present application;

[0042] Figure 2 is a schematic diagram of the overall design concept of ship diesel engine fault diagnosis based on dynamic virtual-real fusion provided by an embodiment of the present application;

[0043] Figure 3 Schematic diagram of the virtual simulation process provided by an embodiment of the present application;

[0044] Figure 4 Schematic diagram of the control process of the iterative optimization control model provided by an embodiment of the present application;

[0045] Figure 5 Schematic diagram of the fusion network structure provided by the embodiment of the present application;

[0046] Figure 6 It is a flowchart of the ship engine fault diagnosis method provided by the embodiment of the present application;

[0047] Figure 7 It is a schematic diagram of the hardware structure of the electronic device provided by the embodiment of the present application. Detailed implementation manners

[0048] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0049] It should be noted that although the functional modules are divided in the system and the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order from the module division in the system or the order in the flowchart. The terms "first", "second", etc. in the description and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence.

[0050] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application, and are not intended to limit the present application.

[0051] First, several nouns involved in the present application are analyzed:

[0052] The Feature Fusion Covariance Matrix (FFCM) is a statistical tool used to integrate information from different feature sources in multi-modal or multi-variate data fusion. Its core idea is to capture the correlation between different features through the covariance matrix and fuse this information into a unified representation. The FFCM encodes the statistical characteristics (such as correlation and trend of change) of different feature sources into a unified matrix through the covariance matrix, which is convenient for subsequent analysis (such as classification and clustering). The covariance matrix describes the linear relationship between multiple features (variables), the diagonal elements are the variances of each feature, and the non-diagonal elements are the covariances between features. Feature fusion is to combine features from different sensors, modalities or extraction methods to improve the robustness and expressive ability of the model.

[0053] Multiple Kernel Ridge Regression (MRKRR) is a machine learning method that combines Multiple Kernel Learning (MKL) and Ridge Regression, and is used to handle non-linear regression problems. Its core idea is to capture different features or modalities of data by fusing multiple kernel functions, while using the regularization feature of Ridge Regression to prevent overfitting. Ridge Regression is used to solve the overfitting problem in linear regression. Multiple Kernel Learning (MKL) is used to combine multiple kernel functions (such as kernels of different types or different data modalities) to enhance the model's expressive power.

[0054] The embodiments of the present application provide a method for diagnosing faults in a marine engine, a method for training a model, a device, and a storage medium, aiming to improve the accuracy of diagnosing faults in a marine engine.

[0055] The method for diagnosing faults in a marine engine, the method for training a model, the device, and the storage medium provided by the embodiments of the present application are specifically described through the following embodiments. First, the method for training a model for diagnosing faults in a marine engine and the method for diagnosing faults in a marine engine in the embodiments of the present application are described.

[0056] The method for training a model for diagnosing faults in a marine engine and the method for diagnosing faults in a marine engine provided by the embodiments of the present application relate to the technical field of equipment fault detection. The method for training a model for diagnosing faults in a marine engine or the method for diagnosing faults in a marine engine provided by the embodiments of the present application can be applied to a terminal, or can be applied to a server side, or can also be software running on a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or can be configured as a server cluster or a distributed system composed of multiple physical servers, or can also be configured as a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application for implementing the method for training a model for diagnosing faults in a marine engine or the method for diagnosing faults in a marine engine, etc., but is not limited to the above forms.

[0057] This application can be used in numerous general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multi-processor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and so on. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This application can also be practiced in a distributed computing environment where tasks are executed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0058] Figure 1 FIG. is an alternative flowchart of a method for training a ship engine fault diagnosis model provided by an embodiment of this application. Figure 1 The method in may include but is not limited to steps S101 to S104.

[0059] Step S101, obtain the real data and virtual data of the ship engine under different working conditions, and correspondingly obtain a real data set and a virtual data set;

[0060] Step S102, input the real data set and the virtual data set into a fusion network for data fusion to obtain network fusion data;

[0061] Step S103, input the network fusion data into a fault prediction network for fault prediction to obtain a fault prediction result;

[0062] Step S104, update the parameters of the fusion network and the fault prediction network according to the fault prediction result and the real fault result, and determine the ship engine fault diagnosis model according to the fault prediction network after training is completed;

[0063] Among them, the fusion network includes a plurality of sequentially connected fusion modules, and the input of the fusion module is the output of the previous-level fusion module and the real data set.

[0064] Steps S101 to S104 shown in the embodiments of the present application obtain a real dataset and a virtual dataset by corresponding the real data and virtual data of the ship engine under different working conditions, input the real dataset and the virtual dataset into a fusion network for data fusion to obtain network fusion data, input the network fusion data into a fault prediction network for fault prediction to obtain a fault prediction result, and then update the parameters of the fusion network and the fault prediction network according to the fault prediction result and the real fault result. Then, a ship engine fault diagnosis model is determined according to the fault prediction network after training is completed. The fusion network in this embodiment includes a plurality of sequentially connected fusion modules. The input of the fusion module is the output of the previous-level fusion module and the real dataset. Thus, in the process of supplementing the virtual dataset to the real dataset, the real dataset can be used to adjust the data fusion result each time, making the network fusion data closer to the real data, indirectly improving the reliability of the virtual data, and thus improving the accuracy of the ship engine fault diagnosis model.

[0065] In step S101 of some embodiments, the different working conditions of the ship engine refer to the operating conditions of the ship engine under different working states, which are usually determined by the combination of key parameters such as load, speed, temperature, and fuel supply. The changes in these parameters directly affect the power output, efficiency, emissions, and reliability of the ship engine. The real data comes from various operating state data of the ship engine in a real environment during normal or faulty conditions, and each collected real data forms a real dataset. The virtual data comes from various operating data and state data of the ship engine in a virtual simulation environment during normal or faulty conditions, and each collected virtual data forms a virtual dataset. The operating state data refers to data such as temperature, pressure, torque, and output power of the ship engine during operation. The ship engine in the embodiments of the present application is a device that converts a certain type of energy (such as chemical energy of fuel, electrical energy, etc.) into mechanical energy and is used to drive a ship. Exemplarily, the ship engine can be a ship diesel engine. When collecting its virtual data, thermodynamic modeling and dynamic modeling are performed through the actual parameters of the ship diesel engine, and the two are jointly simulated to obtain a multi-domain simulation model of the ship diesel engine. Then, a control algorithm is added to realize the automatic adjustment of the multi-domain simulation model, and simulation signals of different fault forms of the ship diesel engine under different working conditions are obtained to establish a virtual dataset. Through a real ship diesel engine fault diagnosis test bench, real signals of different fault forms of the ship diesel engine under different working conditions are obtained, thereby establishing a real dataset.

[0066] In step S102 of some embodiments, virtual data is fused into real data through a fusion network to form network-fused data, thereby supplementing the real data. Since there are certain differences between virtual data and real data, the fusion network adopted in the embodiments of the present application includes multiple sequentially connected fusion modules. The input of the fusion module is the output of the upper-level fusion module and the real data set, and the fusion module is used to perform feature fusion on the input data. The input of the fusion module is the output of the upper-level fusion module and the real data set. It adjusts the fused features output by the upper-level fusion module using the real data set, so that the network-fused data can be closer to the data from the real environment, thereby improving the accuracy of subsequent model training.

[0067] The method of this embodiment is a dynamic fusion method based on simulated virtual signals and real signals, which adaptively fuses virtual data and real data and generates a data-dependent forward path during the inference process. The fusion network in the embodiments of the present application can be a global gating network, that is, the fusion modules in the fusion network combine gating functions to provide data fusion functions at the modal level and fusion functions between real data and simulation data.

[0068] In step S103 of some embodiments, after the real data set and the virtual data set are input into the fusion network for data fusion, sample data can be constructed based on the network-fused data, and then the sample data is input into the fault prediction network for fault prediction to obtain a fault prediction result. The fault prediction result is used to determine whether the input ship engine sample data is faulty. If the sample label includes fault classification, the fault prediction result can further be used to determine the fault type. The fault prediction network includes a feature extraction part and a feature classification part. The feature extraction part can extract the feature fusion covariance matrix (FFCM) of the sample data, and the feature classification part can use a multi-kernel ridge regression (MRKRR) model for data classification. Fault diagnosis is performed on the fused signal based on the feature fusion covariance matrix (FFCM) and multi-kernel ridge regression (MRKRR). The FFCM is a simple and effective feature-level fusion descriptor that can maintain the interaction relationship between multi-sensor features, thereby obtaining a more comprehensive and accurate feature representation, being able to more fully express fault information, and realizing more accurate and reliable ship diesel engine fault diagnosis. It can be understood that other feature extraction algorithms and feature classification algorithms can also be used for feature extraction and feature classification, and the embodiments of the present application do not make specific limitations.

[0069] In step S104 of some embodiments, the classification task loss is calculated based on the fault prediction result and the true fault result, and then the parameters of the fusion network and the prediction network are updated backward in combination with the classification task loss, so that the fusion network can dynamically fuse data considering real data and virtual data in each batch training, enabling the fault prediction network to learn the mapping relationship between input features and fault labels and achieve accurate prediction. After training is completed, a ship engine fault diagnosis model is determined according to the fault prediction network therein. It can be understood that in the backward update process, the fusion network and the fault prediction network can set different training objectives in combination with the classification task loss respectively to optimize the fusion network and the fault prediction network.

[0070] In an exemplary embodiment, multi-modal signals of a marine diesel engine are obtained, including vibration, temperature, pressure, power, etc. The simulation signal and the actual signal are fused by a dynamic fusion method. Finally, the fault diagnosis of the fused signal of virtual and real data is carried out by using the feature fusion covariance matrix and the kernel ridge regression model. Specifically, according to the actual parameters of the marine diesel engine, thermodynamic modeling and dynamic modeling are carried out, and the two are jointly simulated to obtain a multi-domain simulation model of the marine diesel engine. A control algorithm is added to realize the automatic adjustment of the multi-domain simulation model, and simulation virtual signals of different fault forms of the marine diesel engine under different working conditions are obtained to establish a virtual data set. Then, through a real marine diesel engine fault diagnosis test bench, real signals of different fault forms of the marine diesel engine under different working conditions are obtained to establish a real data set. Secondly, based on the dynamic fusion method of simulation virtual signals and real signals, virtual data and real data are adaptively fused, and a data-related forward path is generated during the inference process. Finally, fault diagnosis of the fused signal is carried out based on the feature fusion covariance matrix (FFCM) and the multi-kernel ridge regression (MRKRR). FFCM is a simple and effective feature-level fusion descriptor, which can maintain the interaction relationship between multi-sensor features, thereby obtaining a more comprehensive and accurate feature representation, being able to more fully express fault information, and realizing more accurate and reliable fault diagnosis of marine diesel engines.

[0071] According to some embodiments of the present application, please refer to Figure 2 , the overall design concept of the fault diagnosis of marine diesel engines based on dynamic virtual-real fusion in this embodiment is as follows:

[0072] A multi-domain simulation model module constructed based on the principles of dynamics and thermodynamics, combined with the monitoring data of the marine diesel engine in physical experiments, constructs a dynamic model and a thermodynamic model of the marine diesel engine, and realizes joint simulation through pressure data to simulate the operation of the marine diesel engine.

[0073] Based on the data acquisition and preprocessing module for the operation process of the marine diesel engine multi-domain simulation model, the data collected during the operation of the marine diesel engine multi-domain simulation model is preprocessed and data partitioned to obtain multiple simulation training signals to form a virtual data set.

[0074] Based on the marine diesel engine fault diagnosis physical test bench, a variety of real signals of various fault types of the marine diesel engine under various working conditions are collected, and then through preprocessing and data partition, multiple real training signals are obtained to form a real data set.

[0075] Based on the virtual data set obtained from the marine diesel engine multi-domain simulation model and the real data set obtained from the physical test bench, the dynamic fusion network is used to dynamically fuse the two, adaptively fuse the virtual data and the real data, so as to reduce data noise and improve the model performance.

[0076] Construct a fault prediction network (i.e., a fault diagnosis model) based on the Feature Fusion Covariance Matrix (FFCM) and the Multi-Kernel Ridge Regression (MRKRR), and train the model by combining the data output by the above fusion network. FFCM fuses the information of different feature sources into a covariance matrix, so as to extract more comprehensive and accurate feature information. MRKRR can better fit the non-linear relationship by combining multiple kernel functions, improving the accuracy and generalization ability of regression. Apply the trained fault diagnosis model to the actual working conditions for marine engine fault prediction to obtain the fault diagnosis result.

[0077] According to some embodiments of the present application, the virtual data of the embodiments of the present application can be obtained through the following steps, but not limited to:

[0078] Step S201, determine the state parameters of the marine engine under the target working condition according to the marine engine operation control model;

[0079] Step S202, perform thermodynamic simulation according to the thermodynamic calculation model and state parameters of the marine engine to obtain in-cylinder simulation data;

[0080] Step S203, perform dynamic simulation according to the dynamic model, state parameters and in-cylinder simulation data of the marine engine to obtain cylinder wall vibration simulation data;

[0081] Step S204, form virtual data according to the state parameters, in-cylinder simulation data and cylinder wall vibration simulation data.

[0082] In this embodiment, an iterative optimization control model (i.e., the marine engine operation control model) constructed based on the difference between the instantaneous speed and the speed set value of the marine diesel engine is used to simulate various control strategies and regulators of the diesel engine to realize the monitoring, control and optimization of the operation state of the marine diesel engine. Please refer to Figure 3, the iterative optimization control module is integrated into the virtual simulation model to achieve automatic control of the ship engine during the virtual simulation process (including the thermodynamic simulation process and the kinetic simulation), so as to better simulate the control behavior of the real diesel engine under different working conditions and more accurately evaluate the performance and response characteristics of the diesel engine.

[0083] Please refer to Figure 3 , the thermodynamic simulation process using the thermodynamic model and the kinetic simulation process using the kinetic model involved in the virtual simulation process, as well as the iterative optimization control model for ship engine simulation control will be elaborated below. The iterative optimization control model is constructed based on the difference between the instantaneous speed and the set speed of the ship diesel engine.

[0084] The thermodynamic simulation process using the thermodynamic model is as follows:

[0085] First, establish a thermodynamic model. Based on the fundamentals of thermodynamic analysis, the following thermodynamic model of the ship engine can be constructed:

[0086] Instantaneous cylinder volume:

[0087]

[0088] In the formula: D is the cylinder diameter; S is the stroke; λ is the connecting rod crank ratio; is the crankshaft angle.

[0089] Instantaneous heat release rate of in-cylinder fuel combustion:

[0090] In the formula: Q B is the instantaneous heat release; g f is the cyclic fuel injection quantity; H u is the lower calorific value of fuel combustion. Generally, for light fuel oil, it is 43961 kj / kg; x is the ratio of the mass of the burned fuel to g f at a certain crankshaft angle; is the combustion heat release rate.

[0091] Heat dissipation rate of the working medium in the cylinder to the cylinder wall:

[0092] In the formula: α g is the instantaneous average heat transfer coefficient; A is the heat transfer area; T is the instantaneous temperature of the working medium in the cylinder; T wi is the average wall temperature; i = 1, 2, 3, representing the cylinder head, piston, and cylinder liner respectively.

[0093] Compression process:

[0094] Combustion process:

[0095] Expansion process:

[0096] Scavenging process:

[0097] Then, according to the above thermodynamic model, the state parameters are input into the thermodynamic software for simulation operation to generate simulation results such as in-cylinder pressure data and temperature data. Through the simulation results and analog signals displayed by the host computer, the in-cylinder simulation data is obtained.

[0098] The dynamic simulation process using the dynamic model is as follows:

[0099] First, establish a three-dimensional assembly model and export the model through the interface program.

[0100] Then, establish a dynamic calculation model. According to the basis of dynamic analysis, the following dynamic models can be constructed:

[0101] Displacement of the piston:

[0102] Acceleration of the piston:

[0103] Angular displacement of the connecting rod swing: β = arcsin(λsinα);

[0104] Angular velocity of the connecting rod swing:

[0105] Angular acceleration of the connecting rod swing:

[0106] In the formula: α is the angle turned by the crank from the top dead center; R is the crank length; L is the connecting rod length; λ is the crank connecting rod ratio; ω is the crank angular velocity.

[0107] Then, based on the thermodynamic simulation process, extract and generate an in-cylinder pressure data text document, import the in-cylinder pressure data and assign it to the unidirectional force, and then use the dynamic software for simulation operation to generate in-cylinder wall vibration simulation data.

[0108] Finally, integrate the state parameters, in-cylinder simulation data, and in-cylinder wall vibration simulation data involved in the virtual simulation process to form virtual data.

[0109] Among them, please refer to Figure 4 , and the control process of the iterative optimization control model for determining the state parameters is described as follows:

[0110] First, set the rotational speed of the marine diesel engine. Then, through a series of limiting links, such as load limiting, torque limiting, etc., for signal regulation, finally obtain the output signal of the governor.

[0111] Then, the actuator controls the position of the throttle lever according to the command signal output by the governor, thereby driving the fuel pump rack of the diesel engine to adjust the fuel supply amount, and thus controlling the output torque of the diesel engine.

[0112] Finally, the output amount is determined according to the difference between the instantaneous speed of the diesel engine and the set value. Through the calculations of the proportional, derivative, and integral parts, the output rack displacement is sent to the fuel system to adjust the fuel supply amount, and thus finally control the output torque of the diesel engine. The control law and transfer function are as follows:

[0113]

[0114] In the formula, k p is the proportional coefficient, T i is the integral time constant, T D is the derivative time constant, and e(t) is the difference between the set speed and the measured speed.

[0115] According to some embodiments of the present application, step S102 may include but is not limited to the following steps:

[0116] Step S301, input the real data set and the virtual data set into the first-level fusion module for data fusion to obtain the output of the first-level fusion module;

[0117] Step S302, input the output of the (n-1)th-level fusion module and the real data set into the nth-level fusion module for data fusion to obtain the output of the nth-level fusion module, where n ranges from 2 to N, and N is the number of fusion modules;

[0118] Step S303, determine the output of the Nth-level fusion module as the network fusion data.

[0119] In this embodiment, as Figure 5 shown, the fusion network of the embodiments of the present application may be a global gating network. The fusion network includes N-level fusion modules. Figure 5 As shown, the fusion network includes four-level fusion modules. Two modules (i.e., the subsequent modal-level decision module) 1 and the fusion unit 1 form the first-level fusion module. Two modules 2 and the fusion unit 2 form the second-level fusion module, and so on, which will not be elaborated here.

[0120] The two inputs of the first-level fusion module are the real data set x1 and the virtual data set x2 respectively. The two inputs of the second-level and subsequent fusion modules are the output of the previous-level fusion module and the real data set respectively.

[0121] According to some embodiments of the present application, the fusion module includes two modality-level decision modules and a fusion unit. The two modality-level decision modules are respectively used to process the two inputs of the fusion module, and the fusion unit is used to process the outputs of the two modality-level decision modules.

[0122] The modality-level decision module is used to perform modality judgment and selection on the input data, and perform modality data fusion operations using the expert network of the corresponding modality according to the selected modality, to obtain the output of the modality-level decision module.

[0123] The fusion unit is used to make a fusion operation decision according to the outputs of the two modality-level decision modules, and fuse the outputs of the two modality-level decision modules based on the determined fusion operation, to obtain the output of the fusion module.

[0124] Specifically, in the modality-level decision module, first define that the input data has M modalities, such as modality data of voltage, pressure, temperature, etc., represented by the vector x = (x1, x2, x3,..., x M ). Then design a set of expert networks, and each expert specializes in a subset of all M modalities. For example, if M = 3, there are seven expert networks: E1(x1), E2(x2), E3(x3), E4(x1, x2), E5(x1, x3), E6(x2, x3), E7(x1, x2, x3). The expert network is essentially an event-driven neural network, and the linear and non-linear processing of the expert network is more complex because it is based on a knowledge base and an inference engine, constructs an inference network according to the knowledge base, and defines the linear and non-linear processing functions of network nodes using the inference rules of the expert system for judging modality inputs. B is the number of expert networks, and G(x) is a gating network that determines which expert network should be activated. The final output is expressed as G(x), where x i represents the subset of modalities that are input to the 8th expert.

[0125] In the fusion unit, assume that its input data has a total of M modalities, that is: x = (x1, x2,..., x M ), define the set O i of corresponding fusion operations, and the unit output is The fusion unit performs fusion operations on the outputs of the respective expert networks selected in the modality-level decision module to obtain the final output result. As Figure 5 shown, each fusion module has a set of candidate fusion operations {O i} and a gating network G(x), and h represents the output of the fusion unit. Figure 5 The network framework of represents a dynamic multi-modal architecture with stacked fusion cells. The gating networks G(x) in the four fusion modules are integrated into a global gating network G(x), and the global gating network G(x) outputs the decisions of the four units at one time.

[0126] According to some embodiments of the present application, the fusion network of the embodiments of the present application can be updated through the following steps:

[0127] Step S401: Determine the classification task loss according to the fault prediction result and the true fault result.

[0128] Step S402: Determine the first loss function according to the classification task loss and the operation cost of the fusion operation in the modality-level decision module, and update the parameters of the modality-level decision module with the goal of minimizing the first loss function.

[0129] Step S403: Determine the second loss function according to the classification task loss and the operation cost of the fusion operation in the fusion module, and update the parameters in the fusion module with the goal of minimizing the second loss function.

[0130] In this embodiment, for the design of the modality-level decision module and the fusion unit, the calculation of each expert network E i (or fusion operation O i ) is different. Usually, a computationally intensive expert network (or fusion operation) has strong representation ability. If the network is directly trained by minimizing the loss specific to the classification task, the gating network may learn a trivial solution that always selects the computationally expensive branch. To achieve efficient inference, this embodiment introduces a resource-aware loss function into the training objective. Let C(E i ) represent the computational cost of executing an expert network E i , and C(O i,j ) represent the computational cost of the i-th fusion operation in the j-th unit. The first loss function at the modality level and the second loss function at the fusion level are constructed for the training of the fusion network as follows:

[0131] The first loss function:

[0132] The second loss function:

[0133] In the formula, L task represents the classification task loss; λ is a hyperparameter that controls the relative importance of the two loss terms; g i represents the i-th decision vector given within the fusion unit; represents the i-th decision vector given by the j-th fusion unit; B is the number of expert networks; F is the number of fusion units.

[0134] During the optimization process of the fusion network or the fault prediction network parameters, reparameterization techniques can be adopted. Reparameterization techniques are used to change the representation of model parameters to improve the training efficiency, stability, or performance of the model. In many cases, reparameterization techniques can help the optimizer better search the parameter space, speed up the training, reduce overfitting, or improve the numerical stability of the model. The Umbel-softmax is an improved softmax activation function, which is usually used to handle multi-label classification tasks. In the traditional softmax function, each category has a corresponding probability, while in multi-label classification tasks, a sample may belong to multiple categories simultaneously. The Umbel-softmax takes into account the correlation between categories and improves the parameterization form of the softmax distribution to better adapt to the multi-label classification scenario. During the training process, reparameterization techniques and Umbel-softmax are introduced, and a real-valued soft vector is adopted: In the formula, represents the real-valued soft vector. First, each expert network is pre-trained to ensure that each path of the dynamic network is uniformly optimized, and then reparameterization techniques are introduced to jointly optimize the dynamic network and the gating network in an end-to-end manner.

[0135] In this embodiment, during the adaptive dynamic fusion process of virtual data and real data, the fusion network adopts a gating function to provide multi-modal functions and resource-aware loss functions for decisions at the modality level or the fusion level, thereby improving the computing efficiency.

[0136] According to some embodiments of the present application, step S103 may include but is not limited to the following steps:

[0137] Step S501, dividing the network fusion data according to the data modality to obtain sample data of different modality combinations;

[0138] Step S502, extracting features from the sample data to obtain a feature fusion covariance matrix;

[0139] Step S503, performing feature prediction on the feature fusion covariance matrix to obtain a fault prediction result.

[0140] According to some embodiments of the present application, step S503 may include but is not limited to the following steps:

[0141] Step S601, inputting the feature fusion covariance matrix into a multi-kernel ridge regression model for feature prediction to obtain a fault prediction result;

[0142] Among them, the fault prediction result is used to judge whether the ship engine is faulty and the type of fault.

[0143] In this embodiment, the fault prediction network includes a feature extraction part and a feature classification part. The feature extraction part can extract the Feature Fusion Covariance Matrix (FFCM) of the sample data, and the feature classification part can use the Multi-kernel Ridge Regression (MRKRR) model for data classification. After the fault prediction network is trained, it can be used as a ship engine fault diagnosis model for actual ship engine fault diagnosis. The training process of the fault prediction network is as follows:

[0144] S11. Divide the dynamically fused data (i.e., network fusion data) into several sample data, and design a feature matrix according to the statistical parameters of different modal data after fusion, as follows:

[0145]

[0146] where: M is the number of modes of the fused signal; d represents the number of extracted statistical parameters. Randomly select the obtained FFCM to form a training data set and a test data set;

[0147] S12. To achieve the purpose of multi-modal data fusion, extract the feature fusion covariance matrix of the sample data through the Feature Fusion Covariance Matrix model (FFCM). The feature fusion covariance matrix model is expressed as follows:

[0148]

[0149] where: P represents the characteristic parameters of the signal in the time domain and frequency domain; d is the number of all parameters; μ is the average value of all characteristic parameters; G is the characteristic parameter of the feature fusion covariance matrix model. It can be seen that the calculation of FFCM is very simple and does not require any artificial adjustment of free parameters. Randomly select the obtained FFCM to form a training data set and a test data set.

[0150] S13. Construct a multi-kernel ridge regression model for the training data set The objective function of the multi-kernel ridge regression model is expressed as:

[0151]

[0152] where ρ is the kernel coefficient of the kernel function; represents the kernel function, whose function is to map non-linear data to a higher Hilbert feature space; W represents the feature matrix; ξ represents the training loss matrix; θ represents the regularization parameter that controls the model complexity; ‖.‖ F represents the Frobenius norm; l i represents the one-hot label, L = [l1,…,l N ; k represents the Riemannian manifold kernel function; β represents the Lagrange multiplier.

[0153] S14, according to the formula β t ←(Γ(·, ·, |ρ|) + θ -1 I) -1 L T and the formula That is, β and ρ are updated through multiple iterations of the above formula, where t represents the number of updates, until the formula max{|ρ t+1 - ρ t | ≤ 10e - 5} is satisfied. Based on the optimal β and ρ, the discriminant function for unknown samples is designed as:

[0154]

[0155] Based on the patterns learned from the training data, the above formula is used to determine which class or state a new, unseen data sample belongs to.

[0156] S15, verify the performance of MRKRR using the test data set. After passing the verification, the fault prediction network can be used as a ship engine fault diagnosis model for actual ship engine fault diagnosis.

[0157] This embodiment performs fault diagnosis on the fused signal based on the Feature Fusion Covariance Matrix (FFCM) and the Multi - Kernel Ridge Regression (MRKRR). The FFCM is a simple and effective feature - level fusion descriptor that can maintain the interaction relationship between multi - sensor features, thereby obtaining a more comprehensive and accurate feature representation, being able to more fully express fault information, and achieving more accurate and reliable ship diesel engine fault diagnosis.

[0158] Please refer to Figure 6 , this application embodiment also provides a ship engine fault diagnosis method, which may include but is not limited to the following steps:

[0159] Step S701, obtain the ship engine operation data;

[0160] Step S702, input the ship engine operation data into the ship engine fault diagnosis model for fault prediction to obtain a fault prediction result;

[0161] In this embodiment, the ship engine fault diagnosis model is trained through the ship engine fault diagnosis model training method as in the previous embodiment.

[0162] An embodiment of the present application further provides an electronic device, which includes: a memory, a processor, a program stored on the memory and executable on the processor, and a data bus for implementing connection and communication between the processor and the memory. When the program is executed by the processor, it implements the above-mentioned ship engine fault diagnosis model training method or ship engine fault diagnosis method. The electronic device can be any intelligent terminal including a tablet computer, an in-vehicle computer, etc.

[0163] Please refer to Figure 7 , Figure 7 , which schematically shows the hardware structure of an electronic device according to another embodiment. The electronic device includes:

[0164] A processor 901, which can be implemented in ways such as a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided by the embodiments of the present application;

[0165] A memory 902, which can be implemented in forms such as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store an operating system and other application programs. When implementing the technical solutions provided by the embodiments of this specification through software or firmware, the relevant program codes are stored in the memory 902 and are called by the processor 901 to execute the ship engine fault diagnosis model training method or ship engine fault diagnosis method of the embodiments of the present application;

[0166] An input / output interface 903, which is used to implement information input and output;

[0167] A communication interface 904, which is used to implement communication and interaction between this device and other devices, and can achieve communication through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.);

[0168] A bus 905, which transmits information between various components of the device (such as the processor 901, the memory 902, the input / output interface 903, and the communication interface 904);

[0169] Among them, the processor 901, the memory 902, the input / output interface 903, and the communication interface 904 achieve communication connections with each other inside the device through the bus 905.

[0170] The embodiments of the present application also provide a storage medium, which is a computer-readable storage medium for computer-readable storage. The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the above-mentioned ship engine fault diagnosis model training method or ship engine fault diagnosis method.

[0171] As a non-transitory computer-readable storage medium, the memory can be used to store non-transitory software programs and non-transitory computer-executable programs. In addition, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories can be connected to the processor through a network. Examples of the above network include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0172] The embodiments described in the embodiments of the present application are for more clearly illustrating the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided by the embodiments of the present application. Those skilled in the art will know that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

[0173] Those skilled in the art can understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than those shown in the figures, or combine certain steps, or different steps.

[0174] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0175] Those of ordinary skill in the art can understand that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices can be implemented as software, firmware, hardware, and appropriate combinations thereof.

[0176] In the description of this application and the above-mentioned drawings, terms such as "first", "second", "third", "fourth", etc. (if any) are used to distinguish similar objects and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances so that the embodiments of this application described here can be implemented in an order different from those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that comprises a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0177] It should be understood that in this application, "at least one (item)" means one or more, and "a plurality" means two or more. "And / or" is used to describe the association relationship of associated objects and indicates that three relationships can exist. For example, "A and / or B" can mean: only A exists, only B exists, and both A and B exist simultaneously. Here, A and B can be singular or plural. The character " / " generally indicates that the associated objects before and after are in an "or" relationship. "At least one (one) of the following" or a similar expression means any combination of these items, including any combination of single items (ones) or plural items (ones). For example, at least one (one) of a, b, or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0178] In several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the above-mentioned division of units is only a logical function division, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of systems or units can be in electrical, mechanical, or other forms.

[0179] The units described above 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 can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0180] In addition, in each embodiment of the present application, each functional unit can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0181] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes multiple instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in the various embodiments of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store programs.

[0182] The preferred embodiments of the embodiments of the present application have been described above with reference to the accompanying drawings, which does not limit the scope of rights of the embodiments of the present application. Any modifications, equivalent replacements, and improvements made by those skilled in the art without departing from the scope and essence of the embodiments of the present application shall fall within the scope of rights of the embodiments of the present application.

Claims

1. A method for training a ship engine fault diagnosis model, characterized in that Including the following steps: Obtain the real data and virtual data of the ship engine under different working conditions, and correspondingly obtain the real data set and the virtual data set; Input the real data set and the virtual data set into a fusion network for data fusion to obtain network fusion data; Input the network fusion data into a fault prediction network for fault prediction to obtain a fault prediction result; According to the fault prediction result and the real fault result, update the parameters of the fusion network and the fault prediction network, and determine a ship engine fault diagnosis model according to the fault prediction network after training is completed; Wherein, the fusion network includes a plurality of fusion modules connected in sequence, and the input of the fusion module is the output of the previous-level fusion module and the real data set.

2. The method according to claim 1, characterized in that, The virtual data is obtained through the following steps: Determine the state parameters of the ship engine under the target working condition according to the ship engine operation control model; Perform thermodynamic simulation according to the thermodynamic calculation model of the ship engine and the state parameters to obtain in-cylinder simulation data; Perform dynamic simulation according to the dynamic model of the ship engine, the state parameters and the in-cylinder simulation data to obtain cylinder wall vibration simulation data; Form virtual data according to the state parameters, the in-cylinder simulation data and the cylinder wall vibration simulation data.

3. The method for training a ship engine fault diagnosis model according to claim 1, wherein The step of inputting the real data set and the virtual data set into a fusion network for data fusion to obtain network fusion data includes the following steps: Input the real data set and the virtual data set into the first-level fusion module for data fusion to obtain the output of the first-level fusion module; Input the output of the (n - 1)-th level fusion module and the real data set into the n-th level fusion module for data fusion to obtain the output of the n-th level fusion module, where n ranges from 2 to N, and N is the number of fusion modules; Determine the output of the N-th level fusion module as the network fusion data.

4. The method for training a ship engine fault diagnosis model according to claim 3, characterized in that The fusion module includes two modal-level decision modules and a fusion unit. The two modal-level decision modules are respectively used to process the two inputs of the fusion module, and the fusion unit is used to process the outputs of the two modal-level decision modules; The modal-level decision module is used to perform modal judgment and selection on the input data, and perform modal data fusion operation using the expert network of the corresponding modal according to the selected modal to obtain the output of the modal-level decision module; The fusion unit is used to make a fusion operation decision according to the outputs of the two modal-level decision modules, and fuse the outputs of the two modal-level decision modules based on the decision-making fusion operation to obtain the output of the fusion module.

5. The method for training a ship engine fault diagnosis model according to claim 4, wherein The fusion network is updated through the following steps: Determine the classification task loss according to the fault prediction result and the real fault result; Determine a first loss function according to the classification task loss and the fusion operation calculation cost in the modal-level decision module, and update the parameters of the modal-level decision module with the goal of minimizing the first loss function; Determine a second loss function according to the classification task loss and the fusion operation calculation cost in the fusion module, and update the parameters in the fusion module with the goal of minimizing the second loss function.

6. The method for training a ship engine fault diagnosis model according to claim 1, wherein Inputting the network fusion data into a fault prediction network for fault prediction to obtain a fault prediction result, which includes the following steps: Dividing the network fusion data according to data modalities to obtain sample data with different modality combinations; Performing feature extraction on the sample data to obtain a feature fusion covariance matrix; Performing feature prediction on the feature fusion covariance matrix to obtain a fault prediction result.

7. The method for training a ship engine fault diagnosis model according to claim 6, wherein The performing feature prediction on the feature fusion covariance matrix to obtain a fault prediction result includes the following steps: Inputting the feature fusion covariance matrix into a multi-kernel ridge regression model for feature prediction to obtain a fault prediction result; Wherein, the fault prediction result is used to determine whether the marine engine is faulty and the type of the fault.

8. A method for diagnosing faults in a marine engine, characterized in that, Including the following steps: Obtaining the operating data of the marine engine; Inputting the operating data of the marine engine into a marine engine fault diagnosis model for fault prediction to obtain a fault prediction result; Wherein, the marine engine fault diagnosis model is trained by the marine engine fault diagnosis model training method according to any one of claims 1 to 7.

9. An electronic device, characterized in that, The electronic device includes a memory, a processor, a program stored on the memory and executable on the processor, and a data bus for realizing the connection and communication between the processor and the memory. When the program is executed by the processor, the steps of the method according to any one of claims 1 to 8 are realized.

10. A storage medium, the storage medium being a computer-readable storage medium for computer-readable storage, characterized in that, The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to realize the steps of the method according to any one of claims 1 to 8.

Images