A Survey of Advanced Information Fusion System: from Model-Driven to Knowledge-Enabled

Advanced knowledge engineering (KE), represented by knowledge graph (KG), drives the development of various fields and engineering technologies and provides various knowledge fusion and knowledge empowerment interfaces. At the same time, advanced system engineering (SE) takes model-based system engineering (MBSE) as the core to realize formal modeling and process analysis of the whole system. The two complement each other and are the key technologies for the transition from 2.0 to 3.0 in the era of artificial intelligence and the transition from perceptual intelligence to cognitive intelligence. This survey summarizes an advanced information fusion system, from model-driven to knowledge-enabled. Firstly, the concept, representative methods, key technologies and application fields of model-driven system engineering are introduced. Then, it introduces the concept of knowledge-driven knowledge engineering, summarizes the architecture and construction methods of advanced knowledge engineering and summarizes the application fields. Finally, the combination of advanced information fusion systems, development opportunities and challenges are discussed.


Introduction
System engineering (SE), also known as optimization technology and simulation technology, builds a cross-domainoriented full life cycle development framework based on the model. Model-based system engineering (MBSE) [1] was first proposed by INCOSE in 2007. It solves the overall development problem through formal modeling throughout the life cycle. On the other hand, knowledge engineering (KE) takes knowledge as the core, expresses knowledge and automatically solves problems by computer with the help of artificial intelligence. The knowledge graph (KG) was first proposed by Google in 2012 [2], trying to solve the problems of semantic expression, knowledge storage and knowledge reasoning through the graph structure.
This survey expounds the researches of SE and KE according to their research timelines and methodologies and summarizes the combination direction of the two from the perspective of double driving. The contributions of this survey are as follows: • This survey introduces the concepts, development, methodological techniques and applications of SE and KE, respectively. • This survey analyzes the typical representative technologies of advanced SE and advanced KE (MBSE &KG). To the best of our knowledge, MBSE can solve cross-domain data islands and achieve physical and digital mapping. At the same time, KG extracts and uses knowledge by drawing structure, which is the key enabling means of cognitive intelligence. Therefore, this survey summarizes the methods and advantages of MBSE and KG to inspire the researchers. • Finally, this survey summarizes the technical advantages of SE and KE and analyzes the research direction, opportunities and challenges of the future development of advanced information fusion system from the perspective of dual-driver combination.
The rest of the survey is organized as follows: in Sect. 1, we systematically elaborate on the concept and development of model-driven SE, analyze the various model approaches of traditional and advanced SE deeply and summarize the key techniques and application areas of MBSE. In Sect. 2, we analyze the concept and development of knowledge-driven KE in the same form, summarize the difference of methods between traditional KE and advanced KE in various stages and summarize the knowledge graph system architecture, construction methods, application directions and application fields. In Sect. 3, We systematically summarized the development trend of MBSE and KG two-way drive of advanced information fusion system and presented the opportunities and challenges in the future research field.

Model-Driven System Engineering
This chapter introduces the concepts and development of system engineering and then introduces the methods, the key technologies and the application fields of SE.

System Engineering Development
System engineering can be traced back as early as the 1940s in the field of military engineering technology. The specific development timeline is shown in Fig. 1.
(1) Application Exploration Period In 1940, Bell Labs divided system development into five stages: planning, research, development, application and general engineering, which is the embryo of SE [3]. In 1942, the Manhattan Project used SE to coordinate the specification of atomic bomb development [4]. In 1970, NASA uses SE to send humans to the moon in the Apollo 13 project. Since then, a new era of SE development has been opened.

Classical Methods of System Engineering
There are many methods in SE development, including traditional SE methods and advanced SE methods.

Document-Based System Engineering
(1) Hall Model An SE three-dimensional analysis model is based on time dimension, logical dimension and knowledge dimension. The main methods are intuitive empirical methods, such as brainstorming (BS) [10] and synectics method [11], etc.; forecasting, such as Delphi method [12], scenario analysis [13], time series forecasting method [14], etc.; structural model method, such as interpretative structural modeling method (ISM) [15], graph theory [16], etc. The Hall model establishes the theoretical framework of DBSE. However, there are some problems such as node knowledge limitation and low portability.
(2) V-Model and V &V Model An SE rapid development model that emphasizes model testing. Based on the waterfall model, the system design and integration process are related through model testing and developed into a V &V model [17], focusing on concurrency and risk management. Common methods include Delphi method and analytical hierarchy process (AHP). For example, Wang [18] proposed the 1 + X education resource evaluation framework based on Delphi method. V model and V &V model can realize visual management in complex systems. However, there are some defects such as difficult backtracking, large rework amount and low flexibility. (

3) Spiral Model
A risk-driven SE method, which carries out risk identification, risk analysis and risk control for activities in each stage in advance. Applicable to interdisciplinary complex SE, such as B The OSCEs (objective structured clinical examinations) proposed by kumaravel [19] et al., is used to evaluate the medical ability of students. Spiral model focuses on the development of prototype system, which has the advantages of flexible design and controllable interaction, but it relies heavily on risk identification. In addition, DBSE includes feedback models, elliptical models and circular models. DBSE has been successfully applied in the space industry in the early years, but gradually unable to cope with the system complexity brought by the development of information and network. Therefore, MBSE came into being.

Model-Based System Engineering
(1) MBSE A formal application of a modeling approach that supports activities such as demand, design, analysis, validation and validation through the life cycle. Compared with DBSE, it has the following advantages: 1) integrated design, covering the full life cycle, guaranteeing consistency management, traceability management and complexity management; 2) multidisciplinary collaborative validation, through simulation technology to achieve system pre-validation and interdisciplinary optimization; and 3) features are unambiguous, graphical models are used to accurately and uniformly describe system functions, standardize system design and transfer information without ambiguity. Currently, MBSE is the mainstream application and research method in large and complex systems in many fields.
(2) MBRSE Model-based reliability system engineering (MBRSE) [20] was proposed by the School of Reliability and System Engineering, Beihang University in 2021. Based on product, fault and environment models, it cognizes and applies failure rules according to model evolution and achieves closed-loop optimization and control of failure. The deep combination of MBSE and reliability system engineering (RSE) is an important development direction of advanced system engineering in the future.

(3) MBSE-Based Digital Twin
A data analysis framework [21] that combines MBSE with cyber-physical systems (CPSs). Based on high-precision digital system model (DSM), this framework constructs a mirror of physical entities in digital space, maps the entire life cycle process of corresponding entities and provides management, integration and analysis capabilities. It has the technical features of all element modeling definition, all data collection and all decision simulation. Essentially, it is to establish a digital mirror (DM) [22] of physical entities in SE.
In summary, SE can be divided into DBSE and MBSE as shown in Table 1.

MBSE Modeling Language
Including DSL, UML, SysML, etc., [23] DSL refers to the setting language for a specific field, which has strong description ability for a specific range and solves the communication and consistency problems of MBSE. It can be divided into internal DSLs such as Kotlin DSL and groovy DSL, and external DSLs such as regular expression, XML and Markdown. SysML is a general graphical modeling language. It is reused and extended based on Unified Modeling Language (UML), which formally represents the system requirements, functions and processes in SE, and effectively solve the consistency problem of different model languages.

MBSE Formal Methods
A method of normalizing formal structure in SE through logic science, combining mathematical analysis and computer verification to improve reliability and robustness, is applied to highly integrated SE that pays attention to safety, including formal specification (FS) and formal verification (FV). FS uses DSL to induce the logical relationship of the system, including process algebra language CSP, CCS, etc., and sequential logic language PLTL, CTL, etc. FV verifies whether the formalized system meets the design requirements can be verified by automatic theorem proving tools such as CoQ [24] and lean [25], or distributed verification tools based on time sequence such as UPPAAL [26], TLA [27], Spin [28,29].

Advanced System Engineering Applications
System engineering has full life cycle data processing capabilities. International enterprises such as SIEMENS, DASSAULT, PTC, Ansys, AVEVA, Microsoft and Unity have continuously developed system engineering applications. Based on industry requirements, model-driven system engineering application systems, including GE Predix, SIEMENS COMOS, PTC ThingWorx and Jupiter Digital Twin Platform have been built [30][31][32][33]. Advanced SE has been studied and practiced in many fields and can be summarized as urban governance, industrial manufacturing and health care.
(1) Urban governance with the increasing scale of the city, MBSE builds a city image in the virtual space using urban facilities and human behavior learning to predict the changes of the city state [34]. Optimize the comprehensive management of traffic information [35]. In addition, by building an environmental assessment model based on MBSE to detect air pollution and greenhouse gas emissions [36], we can also reduce pollution emissions and protect urban ecology.
(2) In industrial manufacturing in the modular production process, MBSE manages each life cycle stage of manufacturing in series [37] and realizes the visual display of product manufacturing [38]. At the same time, MBSE supports cross-domain designer information interaction, including intelligent order scheduling, system decision support and other operations, integrates, analyzes and processes information flow and realizes data exchange more efficiently.

(3) Health Care
With the help of medical Internet of Things equipment, MBSE has established a new medical simulation method to conduct real-time perception of the patient's body data [39], track health care data such as diet, sleep and exercise, evaluate health status and give diagnosis results and treatment plans. In addition, combined with VR, AR and other technologies, digital twin technology based on MBSE makes remote surgery possible [40].

Knowledge-Enabled Knowledge Engineering
This chapter first introduces the development of knowledge engineering and then introduces the architecture, methods and applications of advanced KE.

Knowledge Engineering Development
Knowledge engineering can be traced back to the expert system in the last century. The knowledge graph development timeline is shown in Fig. 2.
(1  [42], which uses artificial intelligence, takes knowledge as the object of processing and uses computers to solve problems. In 1980, McCarthy proposed ontology [43], which describes the world by building ontologies. In 1989, Tim Berners-Lee invented the World Wide Web (WWW) [44], and in 1998 introduced the semantic web (SW) [45], which made network data machinereadable. After the WWW was combined resource description framework (RDF), knowledge representation and reasoning were acquired. In 2006, Berners Lee proposed linked data [46], a semantic web ecosystem for data disclosure and data linking.
In 2012, Google introduced the concept of knowledge graph to improve the return quality and query efficiency of search engines and provide structured information around topics. Since then, KG has become the key technology of advanced KE. In 2021, the Institute of Artificial Intelligence of Tsinghua University and others released the Research Report of Cognitive Graph, which uses KG, cognitive reasoning and logical expressions to form a cognitive graph, enabling knowledge to be understood and used by machines, realizing intelligence from perception to a critical breakthrough in mental intelligence.
KE has mainly experienced three development stages, and the data comparison of each development stage of KE are shown in Table 2.

Advanced Knowledge Engineering Architecture
The architecture of KE is shown in Fig. 3. It first obtains and processes raw data (unstructured, semi-structured and structured), then extracts information elements through Fig. 2 Knowledge engineering development timeline entity extraction, relation extraction, attribute extraction and other knowledge extraction methods and performs knowledge fusion of entity links, entity resolution, entity alignment, etc., based on knowledge representation methods such as EDF, RDFs, OWL, etc., extracts knowledge facts from the original database and third-party databases, and builds knowledge base to realize knowledge computing of graph mining calculation, entity reference, rule-based reasoning, etc., and finally realizes knowledge application.

Advanced Knowledge Engineering Methods
The main methods of KE can be divided into knowledge extraction, knowledge fusion and knowledge processing.

Knowledge Extraction
Knowledge extraction (KE) focuses on extracting important information elements from heterogeneous data, mainly divided into named entity recognition and relationship extraction.
(1) Name entity recognition Name entity recognition (NER) identifies entity types with specific meanings from text. The main methods include rule-based NER identifies entities by matching rules between the marked data and the natural text according to the expert rule template; Statistical-learning-based NER focuses on the problem of sequence labeling through (RE), after NER, establishes semantic links between entities to form triple knowledge for discretized semantic relationships. The main methods include rule-based RE, which constructs rules manually, has the advantages of high accuracy and strong targeting, but relies heavily on manual operation and high cost; statisticallearning-based RE divided into three categories. Supervised RE acquire features through sequence markers and context relationships. Semi-supervised RE requires only a small amount of manual label data. Unsupervised RE do not require manual labeling and have a strong migration ability; deep learning-based RE automatically extracts features and mines hidden information through a large number of training, is the main RE method at present. The common knowledge extraction methods is shown in Table 3.

Knowledge Fusion
Knowledge fusion (KF) transforms the obtained facts into a structured knowledge system. The main methods are as follows: (1) Entity Disambiguation Entity disambiguation (ED) solves the problem of polysemy, is divided into cluster disambiguation and link disambiguation. Clustering disambiguation aggregates entities with high similarity by comparing similarity. Link disambiguation is to disambiguate the corresponding target by referring to the linked entity. (2) Entity Alignment Entity alignment (EA) uses the attribute information of entities to determine whether different entities can be aligned and calculates the entity similarity method by mapping entities and relationships to low-dimensional vectors. The common knowledge fusion methods are shown in Table 4.

Knowledge Processing
Knowledge processing (KP) transforms the obtained facts into a structured knowledge system. The main methods are as follows: (1) Ontology Construction Ontology construction obtains the ontological database through the similarity calculation of entity relations, the extraction of superior and subordinate relations and the generation of ontologies, and provides the definition, principle and theory of specific fields.
(2) Knowledge Reasoning Knowledge reasoning infers entity relationships and constructs new relationships based on entity relationship data in the knowledge base, thus expanding and enriching the knowledge base network.

(3) Quality Evaluation
Quality evaluation screens knowledge, ABC quantifies the trustworthiness of knowledge and discards low-quality knowledge data to ensure the quality of knowledge base.
The common methods of knowledge reasoning are shown in Table 5:

Knowledge Engineering Application
Structured KE has the ability of efficient data processing and knowledge reasoning, and has been applied in various fields.
(1) Smart finance KE stores massive data such as financial information and investment data in the form of triples, mines relationships, carries out financial recommendation, investment inquiry, customer mining and so on. At the same time, KG can realize intelligent risk identification and independent early warning. (2) Smart City KE extracts citizen attributes, excavates service site loads and proposes optimal policies to provide effective decisionmaking and reference for city management. Implement traffic analysis and modeling of urban services, urban traffic management and public data mining. Moreover, the City-KG can generate dynamic management information to dynamically warn the city security and avoid potential risks.

Advanced Information Fusion System
This section firstly introduces the information fusion system of KE and SE and summarizes the development ideas and emerging technologies of KG for MBSE and MBSE for KG.

Information Fusion System of KE and SE
In the era of AI 3.0, perceptual intelligence is turning to cognitive intelligence. The fusion of KE and SE can effectively help the realization of knowledge empowerment. Recently, due to the increasing requirements of information systems on data, models and knowledge, the development of advanced KE and SE has made the information fusion systems important. On the basis of retaining the original system characteristics of advanced KE and SE methods, the advanced information fusion system has been upgraded from model-driven to knowledge-enabled through a variety of fusion forms of KG and MBSE. Advanced information fusion system was mainly applied in the aerospace field in the early stage. For example, NASA integrated digital thread with knowledge architecture and realized the domain integration of knowledge management, information architecture and data science through digital twins [76,77]. At present, informatization and digitalization have promoted information fusion systems to many fields. Based on the original MBSE information system, SIEMENS, DASSAULT, Microsoft and other enterprises integrate KG technology, expand the range of associated data, map multisource production models and provide visualization of complex networks [33,78].
The combination of SE and KE brings a new vision for future development. KG enables MBSE, which intelligently implements structural design, reliability design, knowledge penetration and so on. Meanwhile, MBSE offers KG formal modeling, data chain governance, digital twinning, and more from a new perspective. This chapter summarizes the idea and technology of combining KE and SE in advanced information fusion system.

KG for MBSE
MBSE data processing is based on empirical methods and domain rules. The introduction of KG provides specific solutions to the structural design, reliability and security problems of system engineering.

Design Method of Span Structure
MBSE relies on traditional data storage methods and has optimization space for storage, query and reasoning. The introduction of KG brings the following advantages: (1) Data Integration The existing MBSEs a have weak ability to integrate the same model, different sources and different forms. KG can use GOPPRRE, OWL and other methods to regularize the pre-processing of MBSE model, build a structured knowledge map for the whole life cycle and ensure the consistency and integrity of module data. For example, Zuheros [79,80] et al. proposed to use LSTM to encode the context information of the target entity without using external resources such as knowledge base.
(2) Graph Fusion MBSE has problems such as data maintenance, redundant documents and understanding ambiguity. KG can provide RDF standard data model [81]. It can use visualization technology to describe the knowledge and relationship of MBSE, such as DB2RDF [82] and SQLGraph [83]. It takes into account the physical structure of relational tables and the logical structure of graph models [84]. (3) Multimodal Embedding MBSE relies on expert experience to artificially represent data and query models, and traditional data query methods cannot meet large-scale dynamic query requirements. Deep learning-based KG can automatically learn hidden features and identify synonym models from MBSE when domain knowledge is scarce. For example, large-scale concept ontology for multimedia (LSCOM) [85][86][87][88], core ontology for multimedia (COMM) [89], etc.

Implementation of Highly Reliable Model
MBRSE is an important research direction in SE. The introduction of KG helps the combination of MBSE and RSE in the following ways: (1) Sensitive Data Supervision MBSE's model data are stored and transmitted on a variety of device ports. Sensitive data leaking will lead to information loss and system security problems. Security knowledge map can protect MBSE's network data, establish ontology-based security knowledge ontology framework and construct structured knowledge base of smart security domain by threat modeling, such as STUCCO [90], UCO, MALOnt [91], etc. MBSE's single-step attack warning cannot meet the requirements of network early warning. Security reasoning KG can determine attack source, attack medium and attack path. It is an important technology for MBSE from passive to active defense, such as CVE [92], CCE [93], CVSS [94] and CAPEC [95]. At the same time, associating security knowledge map with log semantics can analyze attack path by graph, such as extracting the semantics of attack behavior proposed by Zeng J [96], inferring node semantics, enumerating all behaviors.

MBSE for KG
KG is conducting further research in several fields, and the intelligent integration with MBSE provides specific ideas for the development of cognitive intelligence.

Data Chain Governance
Accurate extraction and rapid aggregation of KG knowledge relies on high-quality data, and MBSE offers the following data chain governance advantages: (1) Discovery of All Cycle Data KG automatically finds datasets from a data warehouse (DWH). Combined with MBSE, it provides data integration and description for each stage. At the same time, ML-based data discovery can further enhance the ability of data query [97][98][99]

Digital Twin Management
KG lacks complete cross-domain engineering management capabilities. MBSE-based digital twins (DT) provides a grip for accelerating KG applications.
(1) DT-based System Analysis Systems analysis of KG applications is incomplete, and the introduction of DT can improve interpretability. In the digital simulation system, cluster method, factor analysis, co-word analysis is processed, and the feature vector of the knowledge map is centrally sampled to make the model explanatory. Finally, the model is further understood in the form of visualization. (2) DT-based Security Perception KG relies on traditional security detection methods and has insufficient processing power to determine hidden threats, dynamic threat assessment, threat event mining and so on. MBSE-based DT can effectively analyze data and monitoring systems and build a knowledge base of KG's heterogeneous cyber security information (HCSI) intelligent security domain to achieve visual security awareness and detection.

Conclusion
Advanced SE takes MBSE as its core and can process large data throughout its life cycle. Meanwhile, advanced KE takes KG as its representative and is the key enabling tool for cognitive intelligence. Both SE and KE are looking for new interdisciplinary possibilities. This survey reviews the research and application of SE and KE and introduces their concepts, development, methods and applications, respectively. In addition, we analyze and discuss the development form of advanced information fusion system from model-driven to knowledge-enabled. Hoping this review can provide theoretical references and innovative ideas for the research and development of advanced information fusion system.
Acknowledgements Throughout the writing of this dissertation I have received a great deal of support and assistance. I would first like to thank my supervisor Bohan Li, whose expertise was invaluable in formulating the research questions and methodology. Your insightful feedback pushed me to sharpen my thinking and brought my work to a higher level. I would particularly like to acknowledge my teammate, for their wonderful collaboration and patient support. I would also like to thank my tutors Bohan Li, for their valuable guidance throughout my studies. You provided me with the tools that I needed to choose the right direction and successfully complete my dissertation. In addition, I would like to thank my parents for their wise counsel and sympathetic ear. You are always there for me. Finally, I could not have completed this dissertation without the support of my friends, who provided stimulating discussions as well as happy distractions to rest my mind outside of my research Author Contributions DZ gave his contribution to the collation and analysis of references and the compilation of manuscripts; BL provided the idea and gave analysis through constructive discussions; HY, YX, JW, BZ updated the references in the digital thread section; YC and ZY updates the references in the map of knowledge section; HW and ZY summarized the information fusion system section.

Availability of Data and Materials
All data generated or analyzed during this survey are included in this article.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.