Rocket engine health management system and method for full-life data value mining

Abstract

The invention discloses a rocket engine health management system and method for full-life data value mining, and the system comprises a big data storage and calculation center which receives and stores the information from all modules of the system, and achieves the data storage; the test run data management module is used for creating model, station number, measuring point and working condition labels, and importing, labeling and classifying the test run data; the fault diagnosis strategy module provides functions of strategy construction, training, testing, evaluation and publishing; a health assessment and prediction strategy is formed, and the health assessment and prediction strategy is sent to the intelligent analysis module; the intelligent analysis module is used for developing, managing and evaluating a fault diagnosis strategy and integrating a diagnosis result into the fault knowledge base module; and the fault knowledge base module is used for establishing a fault tree and associated fault modes and case libraries, realizing accumulation and application of fault knowledge and storing related information in a big data storage computing center.

Description

Technical Field

[0001] The present invention relates to a rocket engine health management system and method for mining the value of full-life data, and particularly to a liquid rocket engine health management system and method for mining the value of full-life data, belonging to the technical field of fault diagnosis. Background Art

[0002] Full-life data refers to the data recorded for each individual or unit within the entire time range from entering the study to termination in the research, that is, all data of the liquid rocket engine from design, simulation, test run to flight.

[0003] The cost of hot-fire testing of liquid rocket engines is extremely high, and the data generated during the testing process is very valuable and contains a lot of useful information. At present in China, for the management, analysis and application of test run data, a software platform for massive data analysis, unified evaluation and fault traceability has not been established. A large amount of accumulated test run data has not been effectively explored. In particular, the analysis of fault data has not been deeply studied. The effective information obtained by designers of different types of engines from the tests is very limited. The data can neither support the effective evaluation of different data analysis methods nor support the fault traceability and design improvement of subsequent designers. As a result, a large amount of test data can only become a huge storage burden without generating greater value, and may lead to repeated hot-fire tests of some similar engine systems for different needs, failing to achieve the purpose of reducing development costs and improving product performance.

[0004] Regarding the new situation of the business background requirements of new technologies, high quality and high efficiency for the health assessment task of space rocket engines, how to standardize the working methods, optimize the current working mode and improve the working efficiency is an urgent task at present. By sorting out and analyzing the business status at home and abroad, the pain points are summarized as follows:

[0005] 1) The strategy development and evaluation have not achieved standardization

[0006] For the existing fault analysis algorithms and strategies, the implementation mechanism is not clear, and unified and standardized verification and evaluation have not been carried out;

[0007] The algorithm models studied by multiple parties cannot be jointly integrated and applied.

[0008] 2) The data processing efficiency needs to be improved urgently

[0009] The data processing work intensity is large, and it requires manual sorting, processing and interpretation, which is time-consuming and laborious;

[0010] Data traceability is difficult, it is difficult to obtain data under different working conditions, and historical fault data cannot be conveniently referred to and applied;

[0011] It is difficult to judge data association. It is impossible to efficiently carry out correlation analysis on the slow-changing, fast-changing, and simulation data of multiple test runs, and it is difficult to trace the process data in the interpretation and analysis.

[0012] 3) The data value needs to be further explored

[0013] There is a large amount of historical test run and simulation data, which is scattered and managed unstructured. The valuable value of the data has not been mined by forming fault knowledge.

[0014] Most of the existing strategies are based on mechanisms, with limited application scope, high false alarm and missed alarm rates, and the intelligent diagnosis strategy based on data-driven has not been realized.

[0015] In view of the three problems that urgently need to be solved in the current aerospace engine industry in China, namely, the lack of product full life cycle monitoring and diagnosis technology, the low level of intelligent maintenance, and the lack of fault traceability and optimization ability, it is urgent to rely on health management and prediction technology, aiming at the full life cycle state monitoring, operation fault diagnosis traceability, and optimization of rocket engines, and combining with the urgent need to improve the capabilities of rocket engine test platforms, to establish a health management system for liquid rocket engines that can mine the value of full life data. Summary of the Invention

[0016] The technical problem to be solved by the present invention is: to overcome the deficiencies of the prior art, to provide a rocket engine health management system and method for mining the value of full life data, and to improve the data analysis ability of liquid rocket engines.

[0017] The technical solution of the present invention is:

[0018] The present invention discloses a health management system for liquid rocket engines that can mine the value of full life data, including:

[0019] A big data storage and computing center, adopting a hybrid architecture combining a relational database and big data storage, receiving and storing information from all modules of the system, and realizing data storage;

[0020] A test run data management module, used to create model, test run, measurement point, and working condition labels, and after importing, tagging, and classifying the test run data, transfer the classified data to the intelligent analysis module;

[0021] A fault diagnosis strategy module, receiving the classified data from the test run data management module, providing functions of strategy construction, training, testing, evaluation, and release; forming a health assessment and prediction strategy, and sending the health assessment and prediction strategy to the intelligent analysis module;

[0022] An intelligent analysis module, according to the health assessment and prediction strategy transmitted by the fault diagnosis strategy module, developing, managing, and evaluating the fault diagnosis strategy, and integrating the diagnosis results into the fault knowledge base module;

[0023] The fault knowledge base module receives the diagnostic results from the fault diagnosis strategy module, establishes fault trees and associated fault modes and case bases, realizes the accumulation and application of fault knowledge, and stores relevant information in the big data storage and computing center.

[0024] Furthermore, in the above system, the test run data management system accesses and archives and stores the test run simulation data under the maintained model; tags the test run data according to the operating conditions and fault occurrence situations, and associates the data tags with the fault test points.

[0025] Furthermore, in the above system, the method for forming the health assessment and prediction strategy is as follows:

[0026] Obtain the propagation characteristics of faults according to the complex network theory and the network structure of the engine system;

[0027] Analyze and identify the structural weak points in the network structure according to the propagation characteristics of faults;

[0028] For each weak point, formulate fault prevention and maintenance methods;

[0029] According to the classified data of the test run data management module, mine the inherent correlation performance laws of faults in the engine system and establish a fault performance model;

[0030] Extract the causal relationships of faults according to the fault performance model;

[0031] According to the causal relationships of faults, fault prevention and maintenance methods, comprehensively analyze the safety state and operation development trend of the engine system, and combine prior probability knowledge to quantitatively analyze the possibility of accidents occurring to form a health assessment and prediction strategy.

[0032] Furthermore, in the above system, the fault knowledge base module is used to update the fault mode information, fault handling information and fault case information, and store them in the corresponding knowledge bases; establish the association relationships between the fault trees and the fault mode information and fault case information to form expert experience knowledge.

[0033] Furthermore, in the above system, the intelligent analysis module uses the established algorithm model and fault strategy to form a health assessment and prediction strategy, realize engine fault diagnosis, and visually display the results.

[0034] The present invention discloses a method for health management of liquid rocket engines for full-life data value mining, including:

[0035] Create model, unit, measurement point and test run condition tags for the engine to be tested;

[0036] Conduct a test run on the engine to obtain test run data;

[0037] Tag the test run data to indicate the model, unit number, measurement point, and test run condition to which the test data belongs;

[0038] Classify the tagged test run data into data for training evaluation and data for diagnosis;

[0039] Import the data for training evaluation into the intelligent analysis module;

[0040] Create a diagnosis strategy based on the data for diagnosis;

[0041] Use the data for training evaluation to train the diagnosis strategy to obtain a trained diagnosis strategy;

[0042] Select the trained diagnosis strategy to conduct fault diagnosis and obtain a diagnosis result;

[0043] Organize the diagnosis result to generate a fault knowledge tree;

[0044] Store the diagnosis data, diagnosis strategy, diagnosis result, and fault knowledge tree in the big data storage and computing center.

[0045] The beneficial effects of the present invention compared with the prior art are as follows:

[0046] (1) Aiming at the deficiencies in data management, data utilization, and fault diagnosis capabilities in the existing liquid rocket engine industry, the present invention provides a liquid rocket engine health management system based on data analysis, utilization, and deep mining architecture, improving the data analysis ability of liquid rocket engines.

[0047] (2) Starting from the aspects of full-life cycle state monitoring and operation fault diagnosis traceability and optimization of liquid rocket engines, relying on mainstream big data management and analysis technologies, the present invention constructs a big data platform for rocket engine tests, and constructs an integrated platform for abnormal monitoring and early warning of rocket engines and development, evaluation, and application of fault diagnosis strategies, enhancing the abilities of abnormal monitoring, fault early warning, and fault traceability of rocket engines, and laying a solid foundation for subsequent online diagnosis applications of strategies.

[0048] (3) The set of liquid rocket engine health management systems of the present invention can effectively improve the problem of the small number of effective strategies in the current engineering application process. By constructing a platform framework integrating development, verification, and application, there is a relatively complete process mechanism guarantee and strategy evaluation process data traceability from research to mature availability for diagnosis strategies. At the same time, it improves the poor generality of verified diagnosis strategies among models, playing a greater promoting role in enhancing product performance and efficiency.

[0049] (4) The engine health management system designed by the present invention focuses on managing and analyzing the full-life data of the engine, realizing the value mining of the full-life data of the engine.

[0050] (5) The present invention establishes a framework for the liquid rocket engine health management system with data set management, strategy development, strategy evaluation, and strategy application as the core, realizing the storage management of the full-life data of the engine. At the same time, the full-life data is applied to the development and evaluation of fault diagnosis algorithms and the monitoring of the engine state, and finally a fault knowledge base is formed. The value of the full-life data of the engine is fully mined.

[0051] (6) The present invention realizes the storage management of the test run data and simulation data of each model of the engine through the big data storage and computing center and the test run data management module.

[0052] (7) Engineers can select appropriate test run data through the data set management function of the present invention in combination with the characteristics of the strategies they construct, meeting requirements such as measurement points and working conditions, thus providing a good data basis for the full verification of the strategies. They can flexibly construct exclusive data sets according to the needs of the strategies, or construct data sets with a certain generality that meet the verification of common conditions, so as to unify the basis for strategy verification and better evaluate the strategies.

[0053] (8) The present invention facilitates the development and evaluation of diagnostic strategies through the fault diagnosis strategy module and the intelligent analysis module.

[0054] (9) The functional framework provided by the present invention, which integrates strategy development, evaluation, and application, can conveniently and quickly reuse existing strategies and develop new strategies, greatly shortening the overall development cycle of the strategies. At the same time, it provides good functional support for strategy evaluation and optimization, enabling the developed strategies to be applied to model analysis and fault knowledge accumulation faster.

[0055] (10) The present invention constructs a fault knowledge base, which is convenient for subsequent rapid call and auxiliary analysis reference, etc.

[0056] (11) The present invention provides functions such as fault tree design, fault mode library, fault case library, and support for strategy development environment. The formed standardized knowledge base such as fault modes / cases is convenient for the accumulation of knowledge or experience such as faults, fault cases, and strategies, realizing the efficient sharing and reuse of the existing knowledge of the engine, laying a foundation for the online application of models and strategies, and comprehensively strengthening the professional collaborative analysis and result sharing capabilities of business analysts. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 is the functional architecture diagram of the present invention;

[0058] Figure 2 is the operation flow chart of the system architecture of the present invention;

[0059] Figure 3 is the process diagram of data storage and processing of the present invention;

[0060] Figure 4 is the logic flow diagram of the test run data management function of the present invention;

[0061] Figure 5 is the logic flow diagram of the fault diagnosis strategy module of the present invention;

[0062] Figure 6 is the logic flow diagram of the personal strategy management function of the present invention;

[0063] Figure 7 is the logic flow diagram of the model strategy evaluation function of the present invention;

[0064] Figure 8 is the logic flow diagram of the intelligent analysis module of the present invention;

[0065] Figure 9 is the logic flow diagram of the fault knowledge base function of the present invention. Detailed implementation manners

[0066] The following further describes the present invention patent in detail with reference to the accompanying drawings and specific implementation manners.

[0067] The present invention discloses a health management system for liquid rocket engines for full-life data value mining, including:

[0068] A big data storage and computing center, adopting a hybrid architecture combining a relational database and Hadoop big data storage, receiving and storing information from all modules of the system, and realizing data storage; all modules of the system include a test run data management module, a fault diagnosis strategy module, an intelligent analysis module, and a fault knowledge base module.

[0069] A test run data management module, used to create model, batch, measurement point, and working condition tags, and after importing, tagging, and classifying the test run data, transfer the classified data to the intelligent analysis module;

[0070] A fault diagnosis strategy module, receiving the classified data from the test run data management module, providing functions of strategy construction, training, testing, evaluation, and release; forming a health assessment and prediction strategy, and sending the health assessment and prediction strategy to the intelligent analysis module;

[0071] An intelligent analysis module, according to the health assessment and prediction strategy transmitted by the fault diagnosis strategy module, developing, managing, and evaluating the fault diagnosis strategy, and integrating the diagnosis result into the fault knowledge base module;

[0072] The fault knowledge base module receives the diagnostic results from the fault diagnosis strategy module, establishes fault trees and associated fault modes and case bases, realizes the accumulation and application of fault knowledge, and stores relevant information in the big data storage and computing center.

[0073] Preferably, the test run data management system accesses and archives and stores the test run simulation data under the maintained model; tags the test run data according to the operating conditions and fault occurrence situations, and associates the data tags with the fault test points.

[0074] Preferably, a health assessment and prediction strategy is formed. The specific method is as follows:

[0075] According to the complex network theory and the network structure of the engine system, obtain the propagation characteristics of faults;

[0076] According to the propagation characteristics of faults, analyze and identify the structural weak points in the network structure;

[0077] For each weak point, formulate fault prevention and maintenance methods;

[0078] According to the classified data of the test run data management module, mine the inherent correlation performance laws of faults in the engine system, and establish a fault performance model;

[0079] According to the fault performance model, extract the causal relationships of faults;

[0080] According to the causal relationships of faults, fault prevention and maintenance methods, comprehensively analyze the safety state and operation development trend of the engine system, and combined with prior probability knowledge, quantitatively analyze the possibility of accidents occurring to form a health assessment and prediction strategy.

[0081] Preferably, the fault knowledge base module is used to update the fault mode information, fault handling information and fault case information, and store them in the corresponding knowledge bases; establish the association relationships between the fault trees and the fault mode information and fault case information to form expert experience knowledge.

[0082] Preferably, the intelligent analysis module uses the established algorithm model and fault strategy to form a health assessment and prediction strategy, realize engine fault diagnosis, and perform visual display of the results.

[0083] The present invention discloses a liquid rocket engine health management method for mining the value of full-life data, including:

[0084] Create model, batch, measurement point and test run condition tags for the engine to be tested;

[0085] Conduct a test run on the engine to obtain test run data;

[0086] Tag the commissioning data to indicate the model, unit number, measurement point, and commissioning condition of the test data;

[0087] Classify the tagged commissioning data into data for training and evaluation and data for diagnosis;

[0088] Import the data for training and evaluation into the intelligent analysis module;

[0089] Create a diagnostic strategy based on the data for diagnosis;

[0090] Use the data for training and evaluation to train the diagnostic strategy to obtain the trained diagnostic strategy;

[0091] Select the trained diagnostic strategy for fault diagnosis to obtain the diagnostic results;

[0092] Organize the diagnostic results to generate a fault knowledge tree;

[0093] Store the diagnostic data, diagnostic strategy, diagnostic results, and fault knowledge tree in the big data storage and computing center.

[0094] Embodiment

[0095] As Figure 1 shown, this embodiment provides a health management system for a liquid rocket engine for full-life data value mining. The system consists of a big data storage and computing center, a commissioning data management module, a fault diagnosis strategy module, an intelligent analysis module, and a fault knowledge base module. The system first creates model, unit number, measurement point, and condition tags in the commissioning data management module for the engine commissioning experiment. Subsequently, the engine is commissioned, and the commissioning data is imported into the commissioning data management module. The commissioning data is tagged to indicate which model and unit number the data belongs to, which measurement points the data contains, and the commissioning condition for convenient management. After completing the tagging operation, the commissioning data is classified, with a part used for training and evaluation and a part used for diagnosis. After completing the data organization, the data for training and evaluation is imported into the fault diagnosis strategy module, where a diagnostic strategy is created and the strategy is trained, tested, evaluated, and released. These strategies are selected in the intelligent analysis module for fault diagnosis, and the module combines and applies the strategies in the intelligent analysis module through visual dragging to form diagnostic methods to support various diagnostic analyses. Subsequently, a large number of diagnostic results are organized to generate a fault knowledge tree. Finally, all the information is stored in the big data storage and computing center. The overall process is as Figure 2 shown.

[0096] Each module is specifically as follows:

[0097] Big Data Storage and Computing Center: The platform storage architecture plans to adopt a hybrid architecture that combines a relational database and Hadoop big data storage to achieve secure, efficient storage and usage support for various types of data in the system. The Big Data Storage and Computing Center adopts a reliable distributed file system (HDFS), big data redundancy elimination, and high-efficiency and low-cost big data storage technology to complete the storage of massive manufacturing data, establish corresponding association indexes, and use ZooKeeper application coordination services to provide consistency services for the cluster, including configuration maintenance, distributed synchronization, group member management, etc., for unified platform management and application invocation. The stored data mainly comes from structured and unstructured data at each existing business stage. The relational database and distributed storage synchronization technology are used to achieve two-way data synchronization under different storage mechanisms. According to the actual situation of this project, the Sqoop component in the Hadoop ecosystem can be adopted, and customized development can be carried out based on open-source products according to different data collection characteristics, supporting concurrent mutual exchange between Mysql, HDFS, and HBASE. Sqoop is called in the command-line mode, with a mature version and wide application. For data such as commissioning data, simulation data, and image information in the commissioning data, due to the unclear file data structure and frequent changes, the distributed storage of the big data platform can adopt low-cost storage, or the data can be parsed into structured or semi-structured data for storage. Based on the Flume component in the Hadoop ecosystem, Flume can collect and load text data distributed on different nodes and machines into HDFS, Hive, or HBase in real time, and has the ability of online data conversion and direct messaging. Based on the big data hybrid architecture, massive data storage and query are realized. In the data storage link, when the data files are stored in a distributed data packet format, a flexible multi-level index mechanism is established. The relational database serves as an index to improve the structured query ability of HBase. The data itself is stored in the distributed database, and the data that needs to establish an index is stored in the relational database. Each record in the database contains a "pointer" to the distributed database. The relational database returns the query result set. The query planner queries the distributed database according to the "pointer", and the query result set and the distributed database query result are merged and returned to the user.

[0098] After the original file is uploaded, it is stored in separate tables according to slow-changing, fast-changing, static simulation, and dynamic simulation. The preliminary design of Hbase storage is as follows: It is divided into two types of column families according to whether the attributes are commonly used or not (the names of column families and columns should not be too long). For commonly used ones, the original column content can be merged and stored according to the relationship between measurement points, and the same principle applies to those that are not commonly used. As for whether the attributes that are not commonly used occupy a total of 1 column or are divided into multiple columns, it needs to be evaluated based on the comprehensive performance of actual data reading and writing (the product of sub-items). The evaluation indicators can be TPS (data volume per second) and processing time, where the former is the larger the better, and the latter is the shorter the better.

[0099] The entire data storage and processing process is divided into: original file -> web system -> hbase storage -> data preprocessing -> feature quantity calculation and conversion into WAV files ==> After the feature value calculation is completed, it will be written into the attribute column of hbase. Due to the limitation of the front-end data processing performance of the web system, the size of a single csv and txt data file should not be too large, especially for rapidly changing data files, which can be stored in segments in full volume. The storage and processing process is as Figure 3 shown:

[0100] Test run data management module: It mainly realizes functions such as list recording of test run simulation data of model batches, display of model trees, import of batch data, synchronization of data in the test data management system, parsing and warehousing, data management, data tagging, deletion, data preview, download, query, and viewing of diagnostic records. The test run data management module is used to integrate test run data, simulation data, etc. of the engine, store and manage them uniformly; integrate and manage tests to form an engine database and a fault case database; provide no less than 10 general data preprocessing tools such as sparse noise reduction and sparse morphological component analysis; support abnormal monitoring of sensor data, data preprocessing, etc., and provide an open interface for algorithm or strategy applications. The functional logic flow is as Figure 4 shown.

[0101] This module supports docking with the test data management interface method to obtain slow-changing and rapidly changing data under a specified model; offline import of simulation data from the simulation platform, so as to ensure the accuracy, smoothness and traceability of the subsequent utilization of test run data, and provide a data basis for subsequent analysis. The test run data management module is used to maintain the test run simulation data under the model, access and archive and store and manage the data, and at the same time, it can process the test run data by tagging data according to the operating conditions and fault occurrence situations, and mapping the associated parameters of the measuring points, etc. The test run simulation data is the source of the training and verification set data and an important input data for training strategies and diagnostic analysis. This functional module can effectively solve the problems of data management and utilization.

[0102] Fault Diagnosis Strategy Module: It mainly creates fault diagnosis or condition monitoring strategies for the entire engine or its various components. By selecting the objects to be analyzed and their data, and developing corresponding analysis strategies, it lays a foundation for carrying out analyses such as diagnosis, monitoring, and prediction. The fault diagnosis strategy module takes the core subsystems of the engine body as basic units, based on the monitoring data of the engine test run system. According to the normal operation model and fault knowledge model of the whole machine, it applies data mining technology and modern statistical learning theory to establish a comprehensive state evaluation model and prediction model for the engine system, comprehensively analyzes its safety state and operation development trend, and combines prior probability knowledge to quantitatively analyze the possibility of accidents. From a system perspective, the whole machine system is regarded as a network system composed of different nodes (components) that are interconnected and restricted. Based on complex network theory, it studies the inherent structural characteristics of the system itself and the propagation characteristics of faults; studies the robustness and vulnerability of the body system, analyzes and identifies the safety-sensitive links and structural weak points in the system, formulates system-level fault prevention and maintenance mechanisms, and improves the safe operation performance of the system. Based on the monitoring data of the test run monitoring process unit, it mines the inherent correlation manifestation rules of faults in the subsystems, establishes corresponding fault manifestation models, extracts the causal relationships of faults, and solves the problem of quickly identifying the source fault equipment of complex faults in the engine system.

[0103] This module provides an integrated environment for strategy construction, training, testing, evaluation, and release. It combines and applies the algorithms in the algorithm library in a visual drag-and-drop manner to form strategies that support various diagnostic analyses; provides general graphic components, supports the graphical output of result data during the strategy construction process, and also supports custom graphical output through programming; supports insight analysis during the strategy training process to assist in optimizing the strategy model; establishes strategy evaluation indicators to support the evaluation of strategies; supports strategy result management. After the strategy training is completed, it can be published as a diagnostic analysis service, and the user visible range of the strategy results can be authorized and managed; supports the call of algorithms in multiple programming languages, such as Java, Python, R, and Matlab (the server operating environment needs to be provided by Party A), and supports the upload, update, enabling, and disabling of algorithms, forming the accumulation, precipitation, and reuse of algorithms.

[0104] This module is targeted at fault diagnosis strategy researchers, and mainly solves problems such as the complex process of strategy development, the difficulty in generalizing the developed strategies, and the inconsistent strategy evaluation criteria, supporting the quick formation of strategies in a visual drag-and-drop algorithm combination manner and the evaluation of strategies based on standard evaluation methods. The specific process is as Figure 5 shown.

[0105] Provides a strategy development and evaluation environment, enabling users to develop, manage, and evaluate algorithm models and strategies through this module. It also supports strategy sharing, publishing, copying, and generating evaluation reports from strategy evaluation records, and provides strategy evaluation report downloading functions, thereby ensuring strategy version control, strategy reuse between models, and offline use of evaluation results, and providing available strategies for intelligent analysis. Figure 6 shown.

[0106] This module contains two sections: personal strategy and model strategy.

[0107] The personal strategy section is used to maintain and manage all personal strategies and support the design and evaluation of strategies. At the same time, the personal strategy function module can train, evaluate, optimize, share and publish the user-designed strategy process through AI's own general algorithms and custom algorithms, provide a platform for strategy development and management for system users, and provide strategy information for the strategy application module. This module mainly implements the functions of strategy list recording, strategy tree display, strategy addition, basic information editing, strategy editing, associated data set, evaluation, evaluation record, attachment, maturity evaluation, publishing, sharing, unsharing, copying, querying, and deleting. The functional logic flow is as follows: Figure 6 shown.

[0108] Model strategy is used to maintain all published strategies under the model, support data set association and evaluation for each model strategy, provide a model strategy management platform for system users, and provide strategy information for strategy application modules. This module mainly implements strategy list records, strategy tree display associated data sets, evaluation, evaluation records, attachment query, view details, enable, disable and other operations. Functional logic flow is as follows Figure 7 shown.

[0109] Intelligent analysis module: It realizes fault diagnosis or status monitoring of the entire engine or its components, and conducts diagnosis by selecting the analyzed object and its data and the corresponding analysis strategy.

[0110] The module has the following features:

[0111] 1. Minimalistic modeling process

[0112] The entire test data intelligent analysis process is designed based on drag-and-drop node operation, wired process connection, and guided parameter configuration. The fault analysis process can be quickly constructed by simply dragging and configuring. The module has built-in data processing, data fusion, feature engineering, extended programming and other functions, and can flexibly use a variety of processing methods to pre-process data and improve the quality of modeling data. At the same time, the rich algorithm library also provides users with more choices for modeling. Combined with the "loop line" function, batch modeling is realized to help users efficiently model and quickly tap the hidden value of data.

[0113] 2. One - key Modeling

[0114] The intelligent analysis module supports the one - key modeling function for classification or regression tasks. Users only need to input data, set the roles of the data, then connect to the one - key modeling node, configure the automatic feature generation, selection, model configuration, automatic parameter selection, cross - validation related parameters, and the optimal model evaluation criteria to complete the modeling. After the node runs successfully, the best - performing model and the corresponding model results will be given automatically. This saves the user's modeling time and improves the modeling efficiency.

[0115] 3. Node Data Statistics

[0116] After the process is executed successfully, if the "Insight Light" of the node in the process is set to the on state, the statistical information of the corresponding data set of the node can be viewed in the insight module, and all the statistical content of the node can also be exported as a Word report for offline reporting.

[0117] 4. Perfect Breakpoint Caching Technology

[0118] The intelligent analysis module provides a breakpoint caching mechanism for nodes. Breakpoint function: The process stops after executing from the beginning to the specified node. Caching function: When the process is executed for the first time, a cache is generated at the specified node to save the execution result of the current node. When the process is executed for the second time, it can directly execute from the node with the cache enabled until the end of the process.

[0119] 5. Result Insight

[0120] The intelligent analysis module supports directly exporting the insight content of each node, including the data situation before and after processing and the model content, as a word report. In addition, when an error occurs at a certain node during the process execution, the insight information that was successfully executed before the error - reporting node can also be viewed to assist users in quickly tracing the problem.

[0121] This module can perform functions such as policy tree viewing, policy version selection, selection of diagnostic data, policy mapping, parameter configuration, execution of calculations, diagnostic records, etc. The functional logic flow is as Figure 8 shown.

[0122] Fault Knowledge Base Module: It mainly realizes the construction of the fault tree and the display of related fault modes, cases, and content, facilitating subsequent quick calls and auxiliary analysis and reference, etc. Establish a fault knowledge base module, including a fault mode library containing fault handling information, a fault case library, etc., to form the accumulation, precipitation, and application of engine fault knowledge. Support the maintenance of the model - specific fault tree, provide the maintenance of fault mode information, fault handling information, and fault case information, and store them in the corresponding knowledge base. Establish the association relationship between the fault tree and the fault mode and fault case information, and perform structured storage to form expert experience knowledge.

[0123] The fault knowledge base provides users with the ability to construct and maintain the fault trees of models, as well as the fault modes and fault cases on the fault trees, realizing the structured management of model fault knowledge, facilitating subsequent rapid call and auxiliary analysis reference. The detailed function diagram is as Figure 9 shown.

[0124] In view of the three problems that urgently need to be solved in the liquid rocket engine, namely, the lack of product full-life cycle monitoring and diagnosis technology, the low level of intelligent maintenance, and the lack of fault tracing and optimization ability, the present invention designs a health management system for liquid rocket engines that mines the value of full-life data. This system uses a big data analysis and calculation center and a test run data management module to store and manage the full-life data of the design, simulation, test run, and flight of various models of liquid rocket engines, and designs a fault diagnosis strategy module and an intelligent analysis module to mine the value of the engine's full-life data, develop, manage, and evaluate the engine fault diagnosis algorithm models and strategies, and finally complete the fault diagnosis or status monitoring of the whole engine or its various components. Finally, using the stored and managed full-life data of the engine, a fault knowledge base is established for different models and different faults of the engine.

[0125] First of all, the overall architecture of the liquid rocket engine health management system includes: big data storage and calculation for storing the full-life data of the rocket engine, test run data management for supporting analysis, development and evaluation of fault diagnosis strategies, intelligent analysis of the engine, and a fault knowledge base. The overall function architecture is shown in the following figure.

[0126] Big Data Storage and Computing Center: The platform storage architecture plans to adopt a hybrid architecture that combines a relational database and Hadoop big data storage to achieve secure, efficient storage and usage support for various types of data in the system. The Big Data Storage and Computing Center adopts a reliable distributed file system (HDFS), big data redundancy removal, and high-efficiency and low-cost big data storage technology to complete the storage of massive manufacturing data, establish corresponding association indexes, and use the ZooKeeper application coordination service to provide consistency services for the cluster, including configuration maintenance, distributed synchronization, group member management, etc., for unified platform management and application invocation. The stored data mainly comes from structured and unstructured data in existing business stages. The relational database and distributed storage synchronization technology are used to achieve two-way data synchronization under different storage mechanisms. According to the actual situation of this project, the Sqoop component in the Hadoop ecosystem can be adopted, and customized development based on open-source products can be carried out according to different data collection characteristics, supporting concurrent mutual exchange between Mysql, HDFS, and HBASE. Sqoop is called in the command-line manner, with a mature version and wide application. For data such as test run data, simulation data, and image information in the test run data, due to the unclear file data structure and frequent changes, the distributed storage of the big data platform can adopt low-cost storage, or the data can be parsed into structured or semi-structured data for storage. Based on the Flume component in the Hadoop ecosystem, Flume can collect and load text data distributed on different nodes and machines into HDFS, Hive, or HBase in real time and has the ability of online data conversion and direct messaging. Based on the big data hybrid architecture, massive data storage and query are realized. In the data storage link, when the data files are stored in a distributed data packet manner, a flexible multi-level index mechanism is established. The relational database serves as an index to improve the structured query ability of HBase. The data itself is stored in the distributed database, and the data that needs to establish an index is stored in the relational database. Each record in the database contains a "pointer" to the distributed database. The relational database returns the query result set. The query planner queries the distributed database according to the "pointer", and the query result set is merged with the distributed database query result and returned to the user.

[0127] Test Run Data Management Module: It supports connecting to the test data management interface to obtain slow-changing and fast-changing data under a specified model; offline import simulation data of the simulation platform to ensure the accurate, smooth and traceable subsequent utilization of test run data, and provide a data basis for subsequent analysis. The test run data management module is used to maintain the test run simulation data under the model, access and archive the data for storage management, and at the same time, it can process the test run data by tagging data according to operating conditions and fault occurrences, as well as mapping related parameters of measurement points. The test run simulation data is the source of training and validation set data and also an important input data for training strategies and diagnostic analysis. This functional module can effectively solve the problems of data management and utilization.

[0128] This module mainly realizes functions such as list recording of test run simulation data for model batches, display of model trees, import of batch data, synchronization of data from the test data management system, parsing and warehousing, data management, data tagging, deletion, data preview, download, query, and viewing diagnostic records. The test run data management module is used to integrate test run data, simulation data, etc. of the engine, store and manage them uniformly; integrate and manage the test to form an engine database and a fault case database; provide no less than 10 general data preprocessing tools such as sparse noise reduction and sparse morphological component analysis; support abnormal monitoring and data preprocessing of sensor data, and provide an open interface for algorithm or strategy applications.

[0129] Fault Diagnosis Strategy Module: It provides an integrated environment for strategy construction, training, testing, evaluation, and release. Through visual dragging, algorithms in the algorithm library are combined and applied to form strategies that support various diagnostic analyses; it provides general graphic components, supports graphic output of result data during strategy construction, and at the same time supports custom graphic output through programming; it supports insight analysis during the strategy training process to assist in optimizing the strategy model; it establishes strategy evaluation indicators to support the evaluation of strategies; it supports strategy result management. After the strategy training is completed, it can be published as a diagnostic analysis service, and the user visible range of the strategy results can be authorized and managed; it supports the call of algorithms in multiple programming languages such as Java, Python, R, and Matlab (the server operating environment needs to be provided by Party A), and supports the upload, update, enablement, and disablement of algorithms to form the accumulation, precipitation, and reuse of algorithms.

[0130] Intelligent analysis module: It can realize fault diagnosis or status monitoring of the whole engine or its components, and carry out diagnosis by selecting the analyzed object, its data and the corresponding analysis strategy. The intelligent analysis module applies the liquid rocket engine health management platform, which is a new generation of intelligent analysis model construction and application scenario design tool independently developed. Adhering to the design concept of "intelligent modeling, easy application", the liquid rocket engine health management platform has opened up the whole process of "from data to model, from model to scenario application"; relying on the powerful algorithm engine and big data processing capabilities, it creates an intelligent and easy-to-use artificial intelligence analysis and application construction platform for "national data scientists" to help digital operations in the AI ​​era. The liquid rocket engine health management platform meets the functional nodes of the user's full process modeling needs, graphical, drag-and-drop modeling experience, full process model insights, multiple automatic learning functions, intelligent algorithm selection, rapid completion of analysis model construction, support for one-click service release of analysis results, and can be seamlessly integrated with the company's existing business systems.

[0131] The beneficial effects brought by the technical solution of the present invention are:

[0132] This project customized the development of a liquid rocket engine health management system, which can effectively improve the problem of the small number of effective strategies in the current engineering application process. By building a platform framework that integrates development, verification and application, the diagnostic strategy has a relatively complete process mechanism guarantee from research to maturity and the strategy evaluation process data is traceable. At the same time, it improves the problem of poor universality between verified diagnostic strategy models, which has greatly promoted the improvement of product performance and efficiency.

[0133] The engine health management system designed by the present invention focuses on managing and analyzing the engine's full life data, and realizes value mining of the engine's full life data.

[0134] The present invention establishes a liquid rocket engine health management system framework with data set management, strategy development, strategy evaluation and strategy application as the core, realizes the storage management of engine life cycle data, and uses the life cycle data to develop and evaluate fault diagnosis algorithms and monitor engine status, and finally forms a fault knowledge base, fully tapping the value of engine life cycle data.

[0135] This engine realizes the storage and management of test data and simulation data of various engine models through the big data storage computing center and the test data management module.

[0136] Engineers can select appropriate test drive data through the dataset management function in combination with the characteristics of their own constructed strategies to meet requirements such as measuring points and operating conditions, thereby providing a good data foundation for the full verification of the strategies. They can flexibly construct exclusive datasets according to the needs of the strategies or construct datasets with certain generality that meet the verification of common conditions, so as to unify the basis for strategy verification and better evaluate the strategies.

[0137] This engine facilitates the development and evaluation of diagnostic strategies through a fault diagnosis strategy module and an intelligent analysis module.

[0138] The functional framework provided by the present invention, which integrates strategy development, evaluation, and application, can conveniently and quickly reuse existing strategies and develop new strategies, greatly shortening the overall development cycle of the strategies. At the same time, it provides good functional support for strategy evaluation and optimization, enabling the developed strategies to be applied to model analysis and fault knowledge accumulation faster.

[0139] The present invention constructs a fault knowledge base to facilitate subsequent rapid call and auxiliary analysis reference, etc.

[0140] The present invention provides functions such as fault tree design, fault mode library, fault case library, and support for the strategy development environment. The standardized knowledge base formed by fault modes / cases, etc., greatly facilitates the accumulation of knowledge or experience such as faults, fault cases, and strategies, realizes the efficient sharing and reuse of existing engine knowledge, lays a foundation for the online application of models and strategies, and comprehensively enhances the professional collaborative analysis and result sharing capabilities of business analysts.

[0141] Although the content of the present invention has been described in detail through the above preferred embodiments, it should be recognized that the above description should not be considered as a limitation of the present invention. After those skilled in the art have read the above content, various modifications and substitutions to the present invention will be obvious. Therefore, the protection scope of the present invention should be defined by the appended claims.

[0142] The content not described in detail in the specification of the present invention belongs to the well-known technology of those skilled in the art.

Claims

1. A liquid rocket engine health management system based on full life cycle data value mining, characterized in that: include: The big data storage computing center uses a hybrid architecture that combines relational databases and big data storage to receive and store information from all modules of the system to achieve data storage; The test data management module is used to create labels for models, units, measuring points and working conditions, and import, label and classify the test data, and then pass the classified data to the intelligent analysis module; The fault diagnosis strategy module receives the classified data from the test data management module, provides strategy construction, training, testing, evaluation and release functions; forms health assessment and prediction strategies, and sends the health assessment and prediction strategies to the intelligent analysis module; The intelligent analysis module develops, manages and evaluates the fault diagnosis strategy based on the health assessment and prediction strategy delivered by the fault diagnosis strategy module, and integrates the diagnosis results into the fault knowledge base module; The fault knowledge base module receives the diagnosis results from the fault diagnosis strategy module, establishes the fault tree and the associated fault modes and case libraries, realizes the accumulation and application of fault knowledge, and stores the relevant information in the big data storage and computing center.

2. According to claim 1, a liquid rocket engine health management system for full life cycle data value mining is characterized in that: The test data management system accesses and archives the test simulation data under the maintenance model; labels the test data according to the operating conditions and fault conditions, and associates and maps the data labels with the fault test points.

3. According to claim 1, a liquid rocket engine health management system for full life cycle data value mining is characterized in that: The health assessment and prediction strategy is formed by: According to the complex network theory and the network structure of the engine system, the propagation characteristics of the fault are obtained; Analyze and identify structural vulnerabilities in the network structure based on the propagation characteristics of faults; Develop failure prevention and maintenance methods for each vulnerable point; Based on the classified data from the test data management module, the inherent correlation performance rules of faults in the engine system are mined and a fault performance model is established; Extract the cause and effect relationship of the fault according to the fault manifestation model; Based on the cause-effect relationship, fault prevention and maintenance methods of the fault, a comprehensive analysis is conducted on the safety status and operation development trend of the engine system. Combined with prior probability knowledge, a quantitative analysis is conducted on the possibility of accidents to form a health assessment and prediction strategy.

4. According to claim 1, a liquid rocket engine health management system for full life cycle data value mining is characterized in that: The fault knowledge base module is used to update fault mode information, fault handling information and fault case information, and store them in the corresponding knowledge base; Establish the correlation between the fault tree and the fault mode information and fault case information to form expert experience knowledge.

5. The liquid rocket engine health management system for full life cycle data value mining according to claim 1 is characterized in that: The intelligent analysis module uses the established algorithm model and fault strategy to form a health assessment and prediction strategy, realize engine fault diagnosis, and visualize the results.

6. A liquid rocket engine health management method based on full life cycle data value mining, characterized in that: include: Create labels for the model, unit number, measurement point and test condition for the engine to be tested; Test the engine and obtain test data; Label the test data to indicate the model and unit number, measurement point and test condition to which the test data belongs; The labeled test data is classified into data for training and evaluation and data for diagnosis; Import the data used for training evaluation into the intelligent analysis module; Create diagnostic strategies based on the data used for diagnosis; Using the data used for training and evaluation, the diagnostic strategy is trained to obtain a trained diagnostic strategy; Select the trained diagnostic strategy to perform fault diagnosis and obtain the diagnostic result; Arrange the diagnosis results and generate a fault knowledge tree; The diagnostic data, diagnostic strategies, diagnostic results and fault knowledge trees are stored in the big data storage computing center.

Images