A method for positive diagnosis and reverse positioning feedback control of internal combustion engine based on multi-source data

Abstract

The present application aims at providing a kind of internal combustion engine fault positive diagnosis and reverse positioning feedback control method based on multi-source data, belongs to internal combustion engine fault diagnosis field. Including the following steps: data integration, feature analysis, positive diagnosis reverse positioning and feedback control, the present application applies multiple technologies to internal combustion engine fault diagnosis, involves data processing, feature analysis, simulation, artificial intelligence algorithm and multiple technologies, constructs fault diagnosis closed loop feedback control, establishes the parameter relationship of internal combustion engine positive diagnosis and reverse positioning by multi-source data driving, combined with machine learning algorithm, the present application promotes the intelligent diagnosis control of internal combustion engine fault diagnosis, realizes the monitoring of internal combustion engine fault state.

Description

Technical Field

[0001] The present invention relates to a method for diagnosing internal combustion engine faults. Background Technology

[0002] Internal combustion engine fault diagnosis serves as a bridge connecting the physical engine with the digital world. It safeguards safety, conserves resources, protects the environment, and continuously drives the entire power industry towards a smarter, more reliable, and more sustainable future. From a safety perspective, it protects life and property; from an economic perspective, it reduces costs and increases efficiency, creating direct value; from an environmental and regulatory perspective, it protects the blue sky and fulfills its responsibilities; and from a technological and social perspective, it promotes progress and optimizes the user experience.

[0003] Multi-source fault data for internal combustion engines is the "cornerstone" and "fuel" for achieving advanced fault diagnosis and health management. It is not merely a simple accumulation of data, but a crucial leap that elevates diagnostic capabilities from singular, vague, and passive to comprehensive, precise, and proactive. Single data sources (such as relying solely on a vibration signal or a fault code) are susceptible to interference, offer limited information, and are prone to false alarms or missed alarms. Acquiring multi-source fault data for internal combustion engines is fundamental for intelligent diagnosis, health management, and predictive maintenance. Its sources and methods can be broadly categorized into direct measurement and indirect acquisition. Acquiring multi-source fault data is a systematic engineering challenge, integrating testing technology, experimental design, and data science. Multi-source fault data transforms the internal combustion engine from a potentially faulty "black box" into a transparent, assessable, predictable, and optimizable intelligent asset, serving as the core data infrastructure for achieving intelligent, highly reliable, long-life, and efficient operation and maintenance of power systems. In the context of Industry 4.0 and intelligent manufacturing, its strategic value is increasingly prominent.

[0004] With the development of science and technology, internal combustion engine fault diagnosis methods can be divided into three stages: traditional fault diagnosis methods, digital fault diagnosis methods, and intelligent fault diagnosis methods. Traditional fault diagnosis methods mainly rely on experience, manual inspection, and basic tools, combined with logical analysis of fault phenomena. Digital fault diagnosis methods utilize sensors, data analysis, and intelligent algorithms, significantly improving the accuracy and efficiency of internal combustion engine fault diagnosis. Intelligent fault diagnosis methods represent a further upgrade of digital technology, deeply integrating advanced technologies such as artificial intelligence (AI), the Internet of Things (IoT), edge computing, and cloud computing to achieve autonomous perception, intelligent decision-making, and predictive maintenance of internal combustion engine faults. Summary of the Invention

[0005] The purpose of this invention is to provide a method for forward diagnosis and reverse positioning feedback control of internal combustion engine faults based on multi-source data, which can realize intelligent diagnostic control for internal combustion engine fault diagnosis.

[0006] The objective of this invention is achieved as follows: This invention discloses a method for forward diagnosis and reverse location feedback control of internal combustion engine faults based on multi-source data, characterized by comprising the following steps: (1) Conduct internal combustion engine bench tests and construct internal combustion engine numerical simulation models, modify internal combustion engine parameters, set multiple fault states, collect multi-source fault data, monitor the operating status in real time through sensors, collect real-time data of internal combustion engine, including internal combustion engine operating parameters and time series data, collect relevant parameters that cannot be directly measured by bench tests through simulation, and supplement the parameters of fault data. (2) Integrate the data samples of the internal combustion engine operating status obtained directly and indirectly in step (1), reduce the noise of the vibration signal, manually extract the time domain features and frequency domain features or use CNN to automatically extract the data features, verify the simulation model for the simulation data, determine the accuracy of the simulation data, integrate the fault data obtained directly and indirectly, and construct a multi-source fault dataset for different fault types. (3) Perform feature analysis on the multi-source fault dataset, use correlation analysis and sensitivity analysis methods to analyze the strong correlation parameters of different fault types, perform feature screening and dimensionality reduction, and based on the idea of ​​transfer learning, reconstruct the features of fault data under different working conditions of internal combustion engines according to the feature screening results. (4) Import the multi-source fault dataset into machine learning and deep learning algorithms, construct a fault diagnosis model, predict the fault type of internal combustion engine, perform forward diagnosis, monitor the operating status of internal combustion engine, combine with optimization algorithms to optimize the fault diagnosis model, or expand the data sample to improve the model prediction accuracy when there is less internal combustion engine fault data, and perform reverse positioning of internal combustion engine fault components based on the fault diagnosis model diagnosis results. (5) Use machine learning and deep learning algorithms to construct a regression model of internal combustion engine operating state parameters, determine the relationship between internal combustion engine operating parameters and fault types, determine the internal combustion engine operating state, and use the internal combustion engine operating parameters for feedback control based on the internal combustion engine operating state regression model to maintain the normal operation of the internal combustion engine; (6) Based on the multi-source fault data of the internal combustion engine, combined with feature analysis, forward diagnosis and reverse positioning, the internal combustion engine operating status is controlled in a closed loop, and the internal combustion engine operating status is monitored and adjusted in feedback so that the internal combustion engine is kept in a healthy state.

[0007] The present invention may also include: 1. Direct field measurements include the entire internal combustion engine and sensors, including temperature sensors, pressure sensors, speed sensors, and knock sensors. The temperature, pressure, and air mass flow parameters of the internal combustion engine are measured, and the vibration displacement or acceleration time-domain data of the cylinder wall and cylinder head are acquired to construct a multi-source fault dataset for the internal combustion engine.

[0008] 2. Indirect acquisition includes the construction of internal combustion engine simulation models and the operation of simulation software. The simulation models include zero-dimensional, one-dimensional, and three-dimensional models, which are adjusted according to the internal combustion engine model. The internal combustion engine is simulated in zero-dimensional, one-dimensional, and three-dimensional models using AVL-BOOST, GT-POWER, and ANSYS simulation software to obtain simulation data and expand the parameters of the fault types in the internal combustion engine fault data set.

[0009] 3. Step (3) Perform data analysis on the multi-source fault dataset, including correlation analysis and sensitivity analysis, including information gain ratio, global sensitivity, principal component analysis, noise reduction of vibration signals, and analysis of the importance of internal combustion engine operating status and data characteristics, and screening and dimensionality reduction of data characteristics.

[0010] 4. In step (3), based on the idea of ​​transfer learning, including model-based transfer learning and feature-based transfer learning, the fault data of internal combustion engines under different working conditions are reconstructed to reflect the relationship between fault types and data features under different working conditions.

[0011] 5. In step (4), during the positive diagnosis, machine learning algorithms and deep learning algorithms are used to construct a fault diagnosis model for multi-source fault data. The machine learning algorithms include support vector machine, neural network, and random forest. The model is adjusted according to the fault data. When there is less fault data, the model parameters are optimized by combining particle swarm optimization algorithm, genetic algorithm, and gray wolf optimization algorithm. Alternatively, adversarial neural network is used to expand the fault data sample to improve the prediction accuracy of the fault diagnosis model and locate the specific faulty parts of the internal combustion engine.

[0012] 6. In step (5), during reverse positioning, a regression model of the internal combustion engine operating state parameters is constructed. The regression model algorithm includes neural network, radial basis function, response surface method, and long short-term neural network to determine the operating state. Based on the relationship between the internal combustion engine operating parameters and the fault type in the regression model, the internal combustion engine parameters are controlled to maintain the normal operating state of the internal combustion engine.

[0013] 7. The forward diagnosis and reverse positioning feedback control method for internal combustion engine faults based on multi-source data in step (6) is a closed-loop control. The fault status of the internal combustion engine is monitored through forward diagnosis and reverse positioning to maintain the healthy state of the internal combustion engine. The method and simulation model are updated for different models of internal combustion engines.

[0014] The advantages of this invention are as follows: This invention establishes the parameter relationship between forward diagnosis and reverse localization of internal combustion engines by using multi-source data and combining machine learning algorithms, thereby promoting intelligent diagnostic control of internal combustion engine fault diagnosis and realizing the monitoring of internal combustion engine fault status. Attached Figure Description

[0015] Figure 1This is a framework diagram of the present invention; Figure 2 This is a flowchart of the present invention. Detailed Implementation

[0016] The invention will now be described in more detail with reference to the accompanying drawings: Combination Figure 1-2 This invention addresses internal combustion engine fault diagnosis by combining data and model fusion, including data integration, feature analysis, forward diagnosis, reverse localization, and feedback control. Data integration involves acquiring various types of data samples of the internal combustion engine's operating status through direct and indirect measurements, real-time measurement data from multiple sensors based on internal combustion engine bench tests, simulation data based on internal combustion engine simulation models, supplementing fault data, and integrating the electrical and vibration signals of sensors and the results of numerical simulation processes according to different operating states of the internal combustion engine. Feature analysis analyzes the characteristics of integrated data on the operating status of internal combustion engines. Sensitivity analysis and correlation analysis are used to analyze the importance of the operating status and data characteristics of internal combustion engines. Data features are screened and dimensionality reduced. Transfer learning is used to reconstruct the data features of multiple operating conditions, reducing the time and economic consumption of multi-condition fault tests and simulations, and making them adaptable to the feature analysis of the operating status of internal combustion engines under various operating conditions. Forward diagnosis and reverse localization are based on machine learning, deep learning and other technologies. They study the relationship between the operating status of internal combustion engines and data characteristics, build fault diagnosis models, accurately predict the operating status of internal combustion engines, monitor the health status of internal combustion engines, improve the accuracy of fault diagnosis when the number of data samples is small, and reverse localize the specific mechanical parts of the internal combustion engine fault status. Feedback control constructs a regression model of internal combustion engine operating state parameters based on data samples of internal combustion engine operating state, determines normal state parameters based on fault diagnosis results, and adjusts the fault-related parameters of internal combustion engine through the internal combustion engine operating state regression model to maintain the internal combustion engine in a normal operating state.

[0017] like Figure 2 As shown in the diagram, steps 1 and 2 involve data integration, step 3 involves feature analysis, step 4 involves forward diagnosis and reverse control, step 5 involves feedback control, and step 6 involves closed-loop integration. The modules of this invention are closely interconnected, with the most crucial aspect being the collaborative work between multi-source data from internal combustion engines and machine learning algorithms. The specific steps are as follows: Step 1: Conduct bench tests on the internal combustion engine and build a numerical simulation model of the internal combustion engine. Modify the parameters of the internal combustion engine, set multiple fault states, collect multi-source fault data, monitor the operating status in real time through multiple sensors, and collect real-time data of the internal combustion engine, such as internal combustion engine operating parameters and timing data. Through simulation, collect relevant parameters that cannot be directly measured by bench tests and supplement the parameters of the fault data. Step 2: Integrate the various types of data samples of the internal combustion engine operating status obtained directly and indirectly in Step 1, reduce noise for vibration signals, manually extract time-domain and frequency-domain features, or use CNN to automatically extract data features. For simulation data, the simulation model needs to be verified to determine the accuracy of the simulation data. Integrate the fault data obtained directly and indirectly to construct a multi-source fault dataset for different fault types. Step 3: Perform feature analysis on the multi-source fault dataset. Utilize correlation analysis and sensitivity analysis methods to analyze the strong correlation parameters of different fault types, perform feature selection and dimensionality reduction, and reconstruct the features of fault data under different operating conditions of internal combustion engines based on the feature selection results, using the idea of ​​transfer learning. Step 4: Import the multi-source fault dataset into machine learning and deep learning algorithms to build a fault diagnosis model, predict the fault type of internal combustion engine, perform forward diagnosis, monitor the operating status of internal combustion engine, optimize the fault diagnosis model by combining optimization algorithms, or expand the data sample to improve the model prediction accuracy when there is less internal combustion engine fault data, and perform reverse localization of faulty internal combustion engine components based on the diagnosis results of the fault diagnosis model. Step 5: Construct a regression model of internal combustion engine operating state parameters using machine learning and deep learning algorithms, determine the relationship between internal combustion engine operating parameters and fault types, identify the internal combustion engine operating state, and implement feedback control of internal combustion engine operating parameters based on the internal combustion engine operating state regression model to maintain the normal operation of the internal combustion engine. Step 6: Based on the multi-source fault data of the internal combustion engine, combined with feature analysis, forward diagnosis and reverse positioning, the internal combustion engine operating status is controlled in a closed loop. The internal combustion engine operating status is monitored and feedback is adjusted to keep the internal combustion engine in a healthy state.

[0018] In step 1, data integration involves direct measurement of the entire internal combustion engine and various sensors, such as temperature sensors, pressure sensors, speed sensors, and knock sensors. Parameters such as temperature, pressure, and air mass flow rate of the internal combustion engine are measured, and time-domain data such as vibration displacement or acceleration at locations such as the cylinder wall and cylinder head are acquired to construct a multi-source fault dataset for the internal combustion engine.

[0019] In step 2, data integration indirectly involves the construction of internal combustion engine simulation models and the operation of simulation software. The simulation models are zero-dimensional, one-dimensional, and three-dimensional, and are adjusted according to the internal combustion engine model. Simulation software such as AVL-BOOST, GT-POWER, and ANSYS are used to perform zero-dimensional, one-dimensional, and three-dimensional simulations of the internal combustion engine to obtain simulation data and expand the parameters of the fault types in the internal combustion engine fault data set.

[0020] Step 3: Feature analysis. Data analysis is performed on the multi-source fault dataset, including correlation analysis and sensitivity analysis, which includes methods such as information gain ratio, global sensitivity, and principal component analysis. Noise reduction is performed on the vibration signal, and the importance of the internal combustion engine operating status and data features is analyzed. Data features are then filtered and dimensionality reduced.

[0021] In step 3, feature analysis, based on the idea of ​​transfer learning, including model-based transfer learning and feature-based transfer learning, reconstructs the features of fault data under different operating conditions of internal combustion engines. The reconstructed features can reflect the relationship between fault types and data features under different operating conditions.

[0022] In step 4, the forward diagnosis utilizes machine learning and deep learning algorithms to construct a fault diagnosis model from multi-source fault data. There are many types of machine learning algorithms, such as support vector machines, neural networks, and random forests. These need to be adjusted according to the fault data. When there is a small amount of fault data, it is necessary to combine particle swarm optimization, genetic algorithms, and gray wolf optimization algorithms to optimize the model parameters, or use adversarial neural networks to expand the fault data samples, thereby improving the prediction accuracy of the fault diagnosis model and locating the specific faulty parts of the internal combustion engine.

[0023] In step 5, reverse localization is performed to construct a regression model of the internal combustion engine's operating state parameters. Regression model algorithms, such as neural networks, radial basis functions, response surface methodology, and long short-term neural networks, are used to determine the operating state. Furthermore, based on the relationship between the internal combustion engine's operating parameters and fault types in the regression model, feedback control is applied to the internal combustion engine parameters to maintain the normal operating state of the internal combustion engine.

[0024] In step 6, the internal combustion engine fault forward diagnosis and reverse positioning feedback control method based on multi-source data is a closed-loop control. It monitors the fault state of the internal combustion engine through forward diagnosis and reverse positioning to maintain the healthy state of the internal combustion engine. The method and simulation model are updated for different models of internal combustion engines.

Claims

1. A method for forward fault diagnosis and reverse location feedback control of internal combustion engines based on multi-source data, characterized by: Includes the following steps: (1) Conduct internal combustion engine bench tests and construct internal combustion engine numerical simulation models, modify internal combustion engine parameters, set multiple fault states, collect multi-source fault data, monitor the operating status in real time through sensors, collect real-time data of internal combustion engine, including internal combustion engine operating parameters and time series data, collect relevant parameters that cannot be directly measured by bench tests through simulation, and supplement the parameters of fault data. (2) Integrate the data samples of the internal combustion engine operating status obtained directly and indirectly in step (1), reduce the noise of the vibration signal, manually extract the time domain features and frequency domain features or use CNN to automatically extract the data features, verify the simulation model for the simulation data, determine the accuracy of the simulation data, integrate the fault data obtained directly and indirectly, and construct a multi-source fault dataset for different fault types. (3) Perform feature analysis on the multi-source fault dataset, use correlation analysis and sensitivity analysis methods to analyze the strong correlation parameters of different fault types, perform feature screening and dimensionality reduction, and based on the idea of ​​transfer learning, reconstruct the features of fault data under different working conditions of internal combustion engines according to the feature screening results. (4) Import the multi-source fault dataset into machine learning and deep learning algorithms, construct a fault diagnosis model, predict the fault type of internal combustion engine, perform forward diagnosis, monitor the operating status of internal combustion engine, combine with optimization algorithms to optimize the fault diagnosis model, or expand the data sample to improve the model prediction accuracy when there is less internal combustion engine fault data, and perform reverse positioning of internal combustion engine fault components based on the fault diagnosis model diagnosis results. (5) Use machine learning and deep learning algorithms to construct a regression model of internal combustion engine operating state parameters, determine the relationship between internal combustion engine operating parameters and fault types, determine the internal combustion engine operating state, and use the internal combustion engine operating parameters for feedback control based on the internal combustion engine operating state regression model to maintain the normal operation of the internal combustion engine; (6) Based on the multi-source fault data of the internal combustion engine, combined with feature analysis, forward diagnosis and reverse positioning, the internal combustion engine operating status is controlled in a closed loop, and the internal combustion engine operating status is monitored and adjusted in feedback so that the internal combustion engine is kept in a healthy state.

2. The method for forward diagnosis and reverse location feedback control of internal combustion engine faults based on multi-source data as described in claim 1, characterized in that: Direct measurements include the entire internal combustion engine and sensors, including temperature sensors, pressure sensors, speed sensors, and knock sensors. The temperature, pressure, and air mass flow parameters of the internal combustion engine are measured, and the vibration displacement or acceleration time-domain data of the cylinder wall and cylinder head are acquired to construct a multi-source fault dataset for the internal combustion engine.

3. The method for forward diagnosis and reverse location feedback control of internal combustion engine faults based on multi-source data as described in claim 1, characterized in that: indirect... The process involves constructing internal combustion engine simulation models and running simulation software. The simulation models include zero-dimensional, one-dimensional, and three-dimensional models, which are adjusted according to the internal combustion engine model. The AVL-BOOST, GT-POWER, and ANSYS simulation software are used to perform zero-dimensional, one-dimensional, and three-dimensional simulations of the internal combustion engine to obtain simulation data. The parameters of the central fault types in the internal combustion engine fault data set are then expanded.

4. The method for forward diagnosis and reverse location feedback control of internal combustion engine faults based on multi-source data according to claim 1, characterized in that: Step (3) Perform data analysis on the multi-source fault dataset, including correlation analysis and sensitivity analysis, including information gain ratio, global sensitivity, principal component analysis, noise reduction of vibration signals, analysis of the importance of internal combustion engine operating status and data characteristics, and screening and dimensionality reduction of data characteristics.

5. The method for forward diagnosis and reverse location feedback control of internal combustion engine faults based on multi-source data according to claim 1, characterized in that: In step (3), based on the idea of ​​transfer learning, including model-based transfer learning and feature-based transfer learning, the fault data of internal combustion engines under different operating conditions are reconstructed to reflect the relationship between fault types and data features under different operating conditions.

6. The method for forward diagnosis and reverse location feedback control of internal combustion engine faults based on multi-source data according to claim 1, characterized in that: In step (4), during the positive diagnosis, machine learning algorithms and deep learning algorithms are used to construct a fault diagnosis model for multi-source fault data. The machine learning algorithms include support vector machines, neural networks, and random forests. The model is adjusted according to the fault data. When there is little fault data, the model parameters are optimized by combining particle swarm optimization algorithm, genetic algorithm, and gray wolf optimization algorithm. Alternatively, adversarial neural networks are used to expand the fault data samples to improve the prediction accuracy of the fault diagnosis model and locate the specific faulty parts of the internal combustion engine.

7. The method for forward diagnosis and reverse location feedback control of internal combustion engine faults based on multi-source data according to claim 1, characterized in that: In step (5), a regression model of the internal combustion engine operating state parameters is constructed. The regression model algorithm includes neural network, radial basis function, response surface method, and long short-term neural network to determine the operating state. Based on the relationship between the internal combustion engine operating parameters and fault types in the regression model, the internal combustion engine parameters are controlled to maintain the normal operating state of the internal combustion engine.

8. The method for forward diagnosis and reverse location feedback control of internal combustion engine faults based on multi-source data according to claim 1, characterized in that: In step (6), the internal combustion engine fault forward diagnosis and reverse positioning feedback control method based on multi-source data is a closed-loop control. The fault status of the internal combustion engine is monitored through forward diagnosis and reverse positioning to maintain the healthy state of the internal combustion engine. The method and simulation model are updated for different models of internal combustion engines.

Images