Artificial Intelligence (AI) in Model-based Systems Engineering

Model-based Systems Engineering (MBSE) is a growing research field that helps to cope with the complexity of modern technical systems. A multitude of MBSE tools, methods, and application fields have been established over the last years. At the same time, Artificial Intelligence (AI) is being established as a key technology for innovations in technical systems as well as their development processes. In the following, we provide an overview of existing and possible application areas for AI-based assistants in MBSE to support their adoption in practice.

This list is updated and maintained by the Fraunhofer Institute for Mechatronic Systems Design. If you would like additional MBSE assistants to be added to this list, please contact us.

Application Areas for Artificial Intelligence in MBSE

This matrix provides an overview of the possible possible application areas of AI in MBSE. The rows conain relevante engineering processes, the columns contain concrete MBSE activities. Each field in the matrix is an application area for AI in MBSE. The numbers within the field refer to known AI use cases in MBSE, which are listed below.

You can view the matrix directly on this page or download by clicking on the link below.

List of Identified AI Use Cases in MBSE

This list provides an overview of AI use cases in MBSE. It is continiously extended and should not be considered as complete.

[1]Structure textual requirementsThe assistant takes a list of textual requirements as input and structures them according to a set of pre-defined structuring criteria (e.g., functional dependencies). The output are the structured requirements as well as identified conflicts or inconsistencies. To this end, the system utilizes mainly text mining procedures. First, the requirements are tokenized and analyzed with respect to the structuring criteria. Next, information retrieval appraoches (e.g., clustering) are used to identify the resulting categories.
[2]Knowledge-based validation of system designsTextual test and inspection reports are automatically analyzed and preselected with regard to their relevance for the current system design. The system engineer can display relevant test and inspection reports for each element of the architecture. For this purpose, NLP methods such as similarity learning or named entity recognition can first be used to analyze the test reports and system models/system designs. The comparison and matching of the system elements with the matching test reports can be done with similarity learning.
[3]Ensure requirements qualityA given requirements specification is evaluated for quality based on defined and learned rules. In addition, for requirements that fall below a defined level of quality, the violated rules are indicated. The potential AI methods differ based on the quality criteria, but come from the field of NLP and machine learning. For uniqueness, a search for “weak words” can be performed by natural language patterns. Completeness can be checked with Part of Speech Tagging. The testability and understandability of requirements can be checked using ML techniques such as text classification.
[4]Visualize related elements for requirementsThis assistant supports impact analyses from requirements. It uses the trace links in the model to identify related elements and a knowledge base to select elements for visualization depending on the user preferences. Moreover, linked data visualization is used to visualize the elements in a user-friendly manner and to enable navigation within the visualization.
[5]Intelligent comparison of requirements documentsOnly “relevant” differences of (requirements) documents are displayed, timestamps etc. can be hidden. First, the requirements documents must be broken down into individual parts using the text mining method tokenization. The comparison can then take place by using Information Retrieval or Sentence Analysis, i.e. procedures from text mining or NLP.
[6]Create data model automaticallyThe systems analyzes the avaialable data of a company and generates an abstract data model. This data model is updated as the compnaies data changes. It serves as baseline for Engineering IT management. In the process of creating the data model, data types (e.g., product ID, construction drawing) and the relationships between the data types are determined. In the context of data governance or data management concept, data mining can be used to gather the necessary information for a data overview. Through methods such as dependency parsing and prediction, the relationships between data types can be identified, described and given a probability. Classification or clustering methods are then applied to form the data overview, often in the form of a model, and to divide it into areas.
[7]Prioritize and weight requirementsRequirements are prioritized and weighted based on the priority and weight of prior requirements. The main AI approach is the identification of similarities between new and existing (i.e. prioritizied) requirements based.
[8]Reuse model requirementsAuto-complete assistance during the creation of a requirements model based on inventory data. The system utilizes NLP and clustering approaches to extract similar requirements from inventory data and suggests these requirements while the user is creating a new requirement.
[9]Reuse textual requirementsAuto-complete assistance during the creation of a textual requirements specification based on inventory data. The system utilizes NLP and clustering approaches to extract similar requirements from inventory data and suggests these requirements while the user is creating a new requirement.
[10]Generate test scenariosThe system creates test scenarios based on a given requirements specification. To this end, the system identifies similar requirements in prior system designs, extracts the connected test scenarios and integrates them into the system model.
[11]Create benchmark / decision basis for system designThe system outlines the implications of a change in the system model as a basis for making decisions during system design. To this end, the system searches in prior system models for examples were the planned change was conducted and summarizes the difference to the current system model.
[12]Impact analyses on change requestsWhen a change request is made, the systems engineer must perform a cross-domain impact analysis and create domain-specific sub-change requests for the necessary design changes. Automatically map the likely impact of a change request and propose appropriate design updates and associated sub-change Requests based on historical data.
[13]Identify design updatesWhenever the system design is changed the assistant identifies possible required follow-up changes (similar to [11] and [12]) and notifies the engineers working on the parts of the system that might require changes.
[14]Propose design updatesThe assistant performs an automated failure root cause analysis in the case of failed tests or simulations and and proposes design changes to solve the issue. The analysis is performed by searching historical data for similar situations (similar system type, similar failure etc.)
[15]Identify user groups and preferencesFor the (further) development of customer-specific market services (related to individual users), information about the user preferences and behaviors of the respective user is required. In particular, the personal usage data of the user of a market service (customer) can be evaluated with regard to user behavior, e.g., with regard to the click rate or the duration of use.
Surveys of individual users or individual customer tests are also possible. With the help of the evaluated customer data, valid statements can be made about user preferences and behavior. Based on this, specific updates can be offered to existing customers or highly individualized new market services can be derived for individual users.
[16]Check consistency of MBSE system architecturesCompanies are increasingly applying model-based specification techniques in systems engineering to address growing system complexity. This involves describing the system using (semi-)formal modeling languages. The models produced in this process can be used to identify inconsistencies in the system specification at an early stage and to initiate corrective measures. However, the manual analysis of these specifications represents a time-consuming and repetitive process. For this reason, automating consistency checking offers great potential for artificial intelligence in PE. The system engineer establishes a set of design guidelines. Using these design guidelines, inconsistencies in existing models are automatically identified and displayed along with opportunities for change.
[17]Structure model requirementsDuring the specification of systems, requirements for the system occur at various points. These are taken, for example, from the project description or derived from the system use cases. Often, these are modeled directly in tools, such as in requirements diagrams in SysML. Accordingly, the requirements are described with model elements, which in turn contain texts and properties. Requirements can be structured with respect to various criteria. For example, with regard to the system level, assemblies or functional or with regard to their origin. Due to the multiplicity of requirements with the development of complex systems it is on the one hand time-consuming to structure the requirements, on the other hand the structuring is accomplished by several persons, whereby it can come to inconsistencies in the structuring. The available requirements are structured according to the indication of structuring criteria in the model. Hierarchy levels are created automatically. Conflicts are marked in the process. 
[18]Trace unrelated requirements through the development processTracing requirements to any artifacts in the development process is a relevant part in the validation of systems. Especially for safety-critical systems, this is required by various standards. In order to ensure this traceability, it is necessary to link requirements to other artifacts. Due to the large number of requirements and the constant refinement of requirements, this linkage is often incomplete or non-existent. To set these linkages afterwards is not only complex, but requires also a large system understanding, which cannot be carried with today’s complex systems by individual persons. The linking of requirements is supported by suggesting missing links from requirements to stakeholders, use cases, functions, logical elements and real elements.
[19]Exploration of the design spaceExploration of the design space follows directly from CAD. Since a design has to fulfill a multitude of functions, the final design is created in iterative steps until design requirements are sufficiently mapped. Today’s product complexity makes a manual approach to the final design very time-consuming. However, the authoring tools available on the market do not yet have any features to support users in generating alternative designs.  
[20]Evaluation of requirementsUsing AI, engineering teams can more quickly identify poorly written, incomplete and ambiguous requirements. NLP technologies help to score requirements on the basis of quality indicators and make recommendations for more clear, consistent and complete requirements.
[21]AI assistant for engineersEmpowering engineers to accelerate their product development workflow by reducing the need for prototyping or iterative testing. By harnessing data from across the entire product engineering process and using AI technologies, it helps to understand the effects of design changes.
[22]Technology summarizerAutomatically and systematically extracting information from patent documents, textbooks, online sources and generating summaries with the help of NLP and machine learning techniques.
[23]Automated fault analysisThe use of NLP and inference rules on the quality data, inspection areas and error reports and feed the result back into the  specification model in order to recognize weak points of product design and implement counter-measures.
[24]Patent analysis of the specification modelThe examination of specification model against patent data for detecting potential patent infringements in order to identify the weak points of the specification models.
[25]Design-for-X evaluationDesign rules are formalized in a knowledge base and with the help of the pattern recognition and supervised learning the design is checked against design rules for example assessment of being suitable for production or assembly. 
[26]Proposal for test proceduresThe system reacts to changing test procedures in selected test institutions (e.g. Stiftung Warentest). These new or modified test procedures are recorded by the system and submitted to the test departments of the company. Finally, it makes conformity assessment of the test specifications to the standard and provide suggestions for improvement.
[27]Digitization assistanceUse of NLP and computer vision technologies in order to creaete accurate and re-usable documents of performed activities in the design workshops.
[28]Defect detectionComputer-aided evaluation of digital camera images, with the aim of automated optical inspection. It allows to differentiating parts, anomalies, and characters, by imitating human visual inspection.
[29]AI-supported product innovationCollecting ideas from external sources into the organization, and bringing those adapted, transformed and enriched ideas from social crowd to product innovation.
[30]Consistent syntaxAutomatic layout of the program, know-how in the companies, Maintaining consistent syntax in system architecture
[33]Optimization of system architecturesCollecting product data in the field and using this data for further development of the product segment. It helps identifying valid product modules and transferring them to the next product generation. 
[34]Test design based on data from predecessorsUse of existing software architecture knowledge to design, reuse and evaluation of software development and automate the search for an optimal architecture design with respect to a set of quality attributes and constraints. Additionally, simulation of the SW architectures enables to find weak points and eliminate them.
[35]Requirements generationDeriving test specifications for new products by analyzing usage data of predecessors or similar products 
[36]Smart system designCreating formalized requirements from poorly formulated product „shall“ descriptions using Natural language understanding and generation techniques in order to support requirements analysis process.  
[37]Traceability analysis based on documentsDesign proposal of different frameworks, methods and tools by considering project goals and  evaluating trends of different techniques.