Diesel engine rail pressure abnormity monitoring method and system based on multi-subspace slow feature analysis

Abstract

The invention relates to a diesel engine rail pressure abnormity monitoring method and system based on multi-subspace slow feature analysis, starting from variable selection, firstly, a subspace on-line division method based on grey correlation degree and dynamic non-similarity is developed, and refined on-line monitoring is carried out on rail pressure signal control quality; then, a distributed modeling strategy is adopted, a weighted fusion method is developed, and static and dynamic characteristics of the signals are synthesized, so that the problems that the rail pressure monitoring is poor in dynamic performance and real-time performance, and the rail pressure signal quality is difficult to measure are solved, and the diesel engine rail pressure signal monitoring efficiency and accuracy are improved.

Description

Technical Field

[0001] This invention belongs to the fields of data processing and analysis, artificial intelligence and machine learning, and intelligent manufacturing, and specifically relates to a method and system for monitoring diesel engine rail pressure anomalies based on multi-subspace slow feature analysis. Background Technology

[0002] Modern diesel engines commonly employ high-pressure common rail fuel injection systems, renowned for their precise injection timing, flexible injection patterns, and extremely high injection pressure. Rail pressure, as one of the key control variables of this system, is the core of the high-pressure common rail system and even the entire diesel engine control system. The dynamic response characteristics and steady-state accuracy of the rail pressure directly determine the quality and consistency of fuel injection. Monitoring the rail pressure signal and promptly, or even in advance, detecting anomalies is crucial for the precise and stable control of the diesel engine.

[0003] In recent years, with the development of technologies such as microcontrollers, sensors, data acquisition, artificial intelligence, and 4G / 5G, signal acquisition and data processing have become easier. Anomaly monitoring methods based on multivariate statistics, as an important branch of data-driven methods, do not require precise system models or in-depth structural principles, and their algorithms are simple and easy to implement. Therefore, they have been applied in engineering practice, gradually realizing remote monitoring and diagnosis of diesel engine operating status. However, most methods only perform simple monitoring and classification of rail pressure operating status, without considering the quality variables in the rail pressure monitoring process. Furthermore, they only perform static analysis of faults, ignoring the dynamic changes of variables and the differences between different samples. This significantly reduces the comprehensiveness and accuracy of the detection results, leading to low reliability of fault diagnosis conclusions. Summary of the Invention

[0004] This invention provides a method and system for monitoring diesel engine rail pressure anomalies based on multi-subspace slow feature analysis, which solves the problems of poor dynamics and real-time performance of rail pressure monitoring and difficulty in measuring the quality of rail pressure signals, thereby improving the efficiency and accuracy of diesel engine rail pressure signal monitoring.

[0005] To address the above technical problems, this invention provides a method for monitoring diesel engine rail pressure anomalies based on multi-subspace slow feature analysis, characterized by the following steps: 1) Obtain the real-time operating parameter information of the diesel engine and perform data resampling and standardization to obtain the recombined dataset i; 2) Based on the recombined dataset i, evaluate whether dataset i can meet the training requirements of the monitoring algorithm. If it does, proceed to the next step. 3) Based on mechanistic knowledge, variables are selected from dataset i to form the process variable set X and the quality variable set Y; 4) Calculate the grey relational degree between each process variable and each quality variable, and construct the grey relational degree matrix GR; 5) Divide the quality-related subspace according to the GR threshold of the grey relational degree matrix. Mass-independent subspace ; 6) In the mass-related subspace Within, the dynamic dissimilarity index (DDI) for each variable is calculated based on the mean and skewness; 7) Based on the DDI value The variables within are sorted, and a high-contribution subspace is divided according to their contribution rate. and low contribution subspace ; 8) In subspace , , Slow feature analysis is performed separately in each subspace to calculate the monitoring statistics and corresponding control limits for each subspace; 9) Based on conditional probability, perform weighted fusion of multidimensional quality variables, calculate fusion statistics and control limits, and determine whether each subspace has a fault; 10) Output the monitoring conclusion of abnormal rail pressure of diesel engine.

[0006] In step 2), if the evaluation dataset i cannot meet the training requirements of the monitoring algorithm, then return to step 1.

[0007] The formula for calculating the grey relational degree matrix GR is: in For the process variable set X ( The i-th variable and the k-th sample, For the set of quality variables Y ( The j-th variable and the k-th sample, resolution coefficient .

[0008] The Dynamic Dissimilarity Index (DDI) includes at least a difference term based on mean change and a difference term based on skewness change.

[0009] The Dynamic Dissimilarity Index (DDI) is calculated as follows: Mean difference term: Skewness difference item: in, Variables representing the current fault state. Variables representing the normal state, It is a random variable The average value, for standard deviation For the sample size, For quality-related sub-blocks The number of variables contained within; The mean, This refers to the skewness.

[0010] Fusion statistics and control limits The calculation method is as follows: Calculate the fault weights for each subspace ; fusion statistics are calculated based on Bayesian inference. ; Set control limits If the value exceeds the control limit, a fault is considered to have occurred; otherwise, it is considered to be in a normal state.

[0011] A diesel engine rail pressure anomaly monitoring system based on multi-subspace slow feature analysis, characterized in that it includes: Data acquisition module: used to acquire data from the ECU and various sensors in real time; Data processing module: used to preprocess and extract features from the collected raw data, converting the raw signals into information that can characterize the health status or fault characteristics of the diesel engine, so that the processed data can be directly stored and called by the rail pressure monitoring module; Rail pressure monitoring module: Utilizing the feature parameters extracted by the data processing module, the monitoring method described above is executed to diagnose abnormal rail pressure conditions in the diesel engine; Data storage module: Used to store raw data, processed data, and fault diagnosis results; Network service module: used to provide data access and communication services; User interaction module: used to display test results and provide an operation interface.

[0012] The data acquisition module interacts with CAN, USB, RS232, and RS485 interfaces by calling the interfaces provided by the hardware abstraction layer.

[0013] The user interaction module adopts a graphical design, displaying the diesel engine's rail pressure signal data and anomaly monitoring results in the form of charts, curves, and lists.

[0014] Beneficial effects: This invention can monitor the anomalies of diesel engine rail pressure signals based on multidimensional mass variables, thereby improving the dynamics and accuracy of anomaly monitoring of diesel engine rail pressure signals.

[0015] Starting from the logic and completeness of rail pressure anomaly monitoring, this method begins with variable selection. First, it develops an online subspace partitioning method based on grey relational analysis and dynamic dissimilarity, and uses a slow feature algorithm to perform refined online monitoring of rail pressure signal control quality. Then, employing a distributed modeling strategy, it develops a weighted fusion method that integrates the static and dynamic features of the signal and fuses the quality monitoring results of multiple subspaces. This solves the problems of poor dynamics and real-time performance in rail pressure monitoring, making it difficult to measure rail pressure signal quality, and improves the efficiency and accuracy of diesel engine rail pressure signal monitoring. Attached Figure Description

[0016] Figure 1 Flowchart of a diesel engine rail pressure anomaly monitoring method based on multi-subspace slow feature analysis Figure 2 A diesel engine rail pressure anomaly monitoring system based on multi-subspace slow feature analysis. Detailed Implementation

[0017] To make the objectives, contents, and advantages of the present invention clearer, the specific embodiments of the present invention will be described in further detail below.

[0018] This invention proposes a diesel engine rail pressure anomaly monitoring method based on multi-subspace slow feature analysis. Starting with variable selection, firstly, an online subspace partitioning method based on grey relational analysis and dynamic dissimilarity is developed. By improving the basic slow feature algorithm, refined online monitoring of rail pressure signal control quality is achieved. Then, a weighted fusion method is developed using a distributed modeling strategy to integrate the static and dynamic features of the signal and fuse multiple quality monitoring results from three subspaces. This solves the problems of poor dynamics and real-time performance in rail pressure monitoring, and difficulty in measuring rail pressure signal quality, thereby improving the efficiency and accuracy of diesel engine rail pressure signal monitoring.

[0019] Specifically, the following steps are included: 1) Obtain the real-time operating parameter information of the diesel engine and perform data resampling and standardization to obtain the recombined dataset i; 2) Based on the recombined dataset i, evaluate whether dataset i can meet the training requirements of the monitoring algorithm in terms of variable dimensions, sample size, missing values, and data units. If yes, proceed to step 3); otherwise, return to step 1). The monitoring algorithm can be a slow feature algorithm or an algorithm that can achieve refined online monitoring of the quality of rail pressure signal control; the monitoring algorithm belongs to a conventional method.

[0020] 3) Based on mechanistic knowledge, variables are selected from dataset i to form the process variable set X and the quality variable set Y; 4) Calculate the grey relational degree between each process variable in the process variable set X and each quality variable in the quality variable set Y, and construct the grey relational degree matrix GR. The method for calculating the grey relational degree matrix GR is as follows: in, For the process variable set X ( The i-th variable and the k-th sample, For the set of quality variables Y ( The j-th variable and the k-th sample, The resolution coefficient is given according to the actual situation. . For the k-th sample, and Grey relational degree, for and The grey relational vector, For the process variable set X and The grey relational matrix.

[0021] 5) Determine the GR threshold and divide the quality-related subspace according to whether it exceeds the threshold. Mass-independent subspace ; 6) Obtain fault data and standardize it in the quality-related subspace. Within the scope, the dynamic dissimilarity index (DDI) for each variable is calculated based on the mean and skewness. The DDI calculation method is as follows: in, The mean, For skewness, Variables representing the current fault state. This represents a variable under normal conditions. It is a random variable The average value, for standard deviation For the sample size, For quality-related sub-blocks The number of variables contained within.

[0022] 7) Based on the DDI value The variables within are sorted, and the top 85% are selected based on their contribution rate. The last 25% ; 8) In three subspaces , , The monitoring algorithm is run in the middle, the slow feature analysis algorithm is selected first, and each monitoring statistic and its corresponding control limit are calculated. The calculation method is as follows: Control limits are set to determine whether a fault has occurred in each subspace, and the quality variable set Y is tested separately. Steps 4)-8) are executed for the p variables in the p-dimensional mass variables to complete the monitoring of rail pressure anomalies. 9) Based on the monitoring results corresponding to the p-dimensional quality variables obtained in step 8), calculate the fusion statistic. and fusion control limits Based on conditional probability, a weighted average is applied to determine whether each subspace in the fused system has experienced a fault. The calculation method is as follows: in, , For the b-th quality variable sequence, Let be the mean of the b-th quality variable. Standard deviation , The number of fault samples. This represents the number of sample points whose statistic exceeds the control limit. The control limit is the quality statistic under normal conditions. If the proportion of the b-th quality variable that exceeds the normal state is high, it indicates that the variable has a significant impact on the fault.

[0023] Fusion statistics The calculation method is as follows: Among them, samples The failure condition probability is: ; In the formula, , For confidence level, , , When the statistic Corresponding to different subspaces ( , , When ), the merged statistics There are different calculation forms ( , , ), fusion Statistic Exceeding fusion control limits If a malfunction is detected, it is considered to be in a normal state; otherwise, it is considered to be in a normal state.

[0024] 10) Output the diesel engine rail pressure anomaly monitoring conclusion based on the algorithm running results.

[0025] This invention provides a diesel engine rail pressure anomaly monitoring system based on multi-subspace slow feature analysis, comprising the following: Data Acquisition Module: Deployed at the diesel engine, this module is responsible for real-time data acquisition from the ECU and various sensors such as vibration, cylinder pressure, temperature, pressure, flow rate, and acoustic signature sensors. By calling interfaces provided by the hardware abstraction layer, this module interacts with data acquisition interfaces such as CAN, USB, RS232, and RS485 to acquire data from different types of sensors and transmit the data to the data processing module.

[0026] Data processing module: Deployed in the cloud, the data processing module preprocesses and extracts features from the collected raw data, transforming the raw signals into information that can characterize the health status or fault characteristics of the diesel engine, so that the processed data can be directly stored and called by the rail pressure monitoring module.

[0027] Rail pressure monitoring module: Deployed in the cloud, the rail pressure monitoring module is one of the core modules of the system. It uses the feature parameters extracted by the data processing module to run the diesel engine rail pressure anomaly monitoring method based on multi-subspace slow feature analysis to diagnose the abnormal rail pressure state of the diesel engine.

[0028] Data storage module: Deployed in the cloud, this module is responsible for storing the collected raw data, processed data, and fault diagnosis results on the terminal's hard drive. To improve data storage efficiency and reliability, the data storage module employs a reasonable data storage structure and file management strategy.

[0029] Network Service Module: Deployed in the cloud, this module enables the terminal to act as a server, connecting with other devices via Ethernet to access and display data. Based on the TCP / IP protocol, it utilizes network communication protocols such as HTTP and MQTT to provide data interfaces and services.

[0030] User Interaction Module: Deployed in the cloud, the user interaction module uses an HDMI interface to implement graphical development and debugging functions, providing users with an intuitive and user-friendly interface. The module employs a graphical design, displaying diesel engine rail pressure signal data, anomaly monitoring results, and other information in the form of charts, curves, and lists.

[0031] Innovation of this invention: 1) By combining rail pressure monitoring with quality variable analysis, a subspace online partitioning method based on grey relational degree and dynamic dissimilarity was developed to monitor signal control quality online; 2) For multidimensional quality variables, a weighted fusion method was developed using a distributed modeling strategy.

[0032] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for monitoring diesel engine rail pressure anomalies based on multi-subspace slow feature analysis, characterized in that, Includes the following steps: 1) Obtain the real-time operating parameter information of the diesel engine and perform data resampling and standardization to obtain the recombined dataset i; 2) Based on the recombined dataset i, evaluate whether dataset i can meet the training requirements of the monitoring algorithm. If it does, proceed to the next step. 3) Based on mechanistic knowledge, variables are selected from dataset i to form the process variable set X and the quality variable set Y; 4) Calculate the grey relational degree between each process variable and each quality variable, and construct the grey relational degree matrix GR; 5) Divide the quality-related subspace according to the GR threshold of the grey relational degree matrix. Mass-independent subspace ; 6) In the mass-related subspace Within, the dynamic dissimilarity index (DDI) for each variable is calculated based on the mean and skewness; 7) Based on the DDI value The variables within are sorted, and a high-contribution subspace is divided according to their contribution rate. and low contribution subspace ; 8) In subspace , , Slow feature analysis is performed separately in each subspace to calculate the monitoring statistics and corresponding control limits for each subspace; 9) Based on conditional probability, perform weighted fusion of multidimensional quality variables, calculate fusion statistics and control limits, and determine whether each subspace has a fault; 10) Output the monitoring conclusion of abnormal rail pressure of diesel engine.

2. The method for monitoring diesel engine rail pressure anomalies based on multi-subspace slow feature analysis according to claim 1, characterized in that, In step 2), if the evaluation dataset i cannot meet the training requirements of the monitoring model, then return to step 1).

3. The method for monitoring diesel engine rail pressure anomalies based on multi-subspace slow feature analysis according to claim 1, characterized in that, The formula for calculating the grey relational degree matrix GR is: in For the process variable set X ( The i-th variable and the k-th sample, For the set of quality variables Y ( The j-th variable and the k-th sample, resolution coefficient .

4. The method for monitoring diesel engine rail pressure anomalies based on multi-subspace slow feature analysis according to claim 1, characterized in that, The Dynamic Dissimilarity Index (DDI) includes at least a difference term based on mean change and a difference term based on skewness change.

5. The method for monitoring diesel engine rail pressure anomalies based on multi-subspace slow feature analysis according to claim 4, characterized in that, The Dynamic Dissimilarity Index (DDI) is calculated as follows: Mean difference term: Skewness difference item: in, Variables representing the current fault state. Variables representing the normal state, It is a random variable The average value, for standard deviation For the sample size, For quality-related sub-blocks The number of variables contained within; The mean, This refers to the skewness.

6. The method for monitoring diesel engine rail pressure anomalies based on multi-subspace slow feature analysis according to claim 1, characterized in that, Fusion statistics and control limits The calculation method is as follows: Calculate the fault weights for each subspace ; fusion statistics are calculated based on Bayesian inference. ; Set control limits If the value exceeds the control limit, a fault is considered to have occurred; otherwise, it is considered to be in a normal state.

7. A method for monitoring diesel engine rail pressure anomalies based on multi-subspace slow feature analysis according to claim 6, characterized in that, Fusion statistics The calculation method is as follows: ； Among them, samples The failure condition probability is: ； In the formula, , For confidence level, , , When the statistic Corresponding to different subspaces ( , , When ), the merged statistics There are different calculation methods: , , , Statistic Exceeding control limits If a malfunction is detected, it is considered to be in a normal state; otherwise, it is considered to be in a normal state.

8. A diesel engine rail pressure anomaly monitoring system based on multi-subspace slow feature analysis, characterized in that, include: Data acquisition module: used to acquire data from the ECU and various sensors in real time; Data processing module: used to preprocess and extract features from the collected raw data, converting the raw signals into information that can characterize the health status or fault characteristics of the diesel engine, so that the processed data can be directly stored and called by the rail pressure monitoring module; Rail pressure monitoring module: Utilizing the feature parameters extracted by the data processing module, it executes the monitoring method described in any one of claims 1-7 to diagnose abnormal rail pressure conditions of the diesel engine; Data storage module: Used to store raw data, processed data, and fault diagnosis results; Network service module: used to provide data access and communication services; User interaction module: used to display test results and provide an operation interface.

9. The monitoring system according to claim 8, characterized in that, The data acquisition module interacts with CAN, USB, RS232, and RS485 interfaces by calling the interfaces provided by the hardware abstraction layer.

10. The monitoring system according to claim 8, characterized in that, The user interaction module adopts a graphical design, displaying the diesel engine's rail pressure signal data and anomaly monitoring results in the form of charts, curves, and lists.

Images