Ruxandra Duca
(Advisor: Prof. Mavris)

will defend a doctoral thesis entitled,

A Graph-Based Methodology for Model Inconsistency Identification and Robust Architecture Exploration and Analysis

On

Wednesday, April 13 at 2:00 p.m.

In the
Collaborative Visualization Environment (CoVE)
Weber Space Science and Technology Building (SST II)

And

https://bluejeans.com/423711991/7610

Abstract
The ongoing rise in complexity in aircraft design leads to errors discovered late when changes are costly. Leading causes include a distributed design with isolated but interdependent models and the lack of a holistic understanding of the multidisciplinary aircraft. With novel technologies and non-conventional subsystems, reliance on subject matter expertise must be supplemented by more systematic methods. Several gaps are identified before a methodology is formulated to mitigate inconsistencies throughout the process, enabling more robust exploration and analysis of new subsystem architectures.

In Model-Based Systems Engineering (MBSE), a formal, central model can be created to give a holistic view of the aircraft. However, modeling languages are too general and require customization through ontologies, to avoid ad-hoc constructs. These are often done for each application, with no cross-disciplinary standard for complex systems. When new subsystems are introduced, interfaces must be considered to reduce integration errors. Few architecture exploration methods tackle connectivity. They require significant upfront effort to construct a feasible architecture, lacking support for incremental changes, as done in practice. Even if an internally consistent system is conceptualized, analysis tools embed assumptions that are not explicit via inputs or outputs, meaning they can use an inconsistent system definition. Promising methods for inconsistency identification between models rely on manually specifying their semantic overlap. Finally, the multi-disciplinary analysis (MDA) is set up based on expertise and common parameters, making it prone to errors when applied to novel architectures.

A methodology is proposed to (1) leverage MBSE to define a complete and internally feasible candidate architecture, (2) ensure that external analysis models are consistent with it, and (3) systematically extract cross-tool dependencies for MDA setup.

In the first step, an interface-based ontology was created using a standardized physics-based set of all possible interactions within a complex system. Rules about component terminals were added, to allow exploration by replacing functionally equivalent components. Then, an extensive semi-automated query-and-action process was formulated to find the components and connections that must be added or removed after a local change. The process was demonstrated by sequentially electrifying subsystems of a conventional baseline, resulting in numerous changes and restoring the system’s internal feasibility.

In the second step, the search for semantic overlap across models was automated to enable methods for inconsistency identification. Labeled property multidigraphs were used to encode data from models according to the ontology in the first step. An algorithm was extensively modified to find the maximum common subgraph based on this semantic data and demonstrated on graphs representing the electrified architecture from the first step and a conventional aircraft as seen by an analysis tool. After finding the equivalent elements, the rest of the inconsistency detection method was demonstrated.

The last step leveraged the results of the first two: analysis tools linked to a holistic, physics-based descriptive model. Using the central model as an intermediary, its elements were partitioned to extract coupling variables and constraints, even when the relevant parameters were not exposed as inputs or outputs of the analysis tools. This was demonstrated between the analysis tool in the second step and a newly introduced, localized thermal model.

With a holistic view of a candidate architecture and a set of properly configured tools whose interactions were extracted systematically, this methodology enables the exploration of subsystem architectures during preliminary design with less effort than current methods, and with the prospective of fewer errors being discovered in later stages of design.

Committee

• Prof. Dimitri Mavris – School of Aerospace Engineering (advisor)
•  
• Prof. Daniel Schrage – School of Aerospace Engineering
•  
• Prof. Ümit Çatalyürek – School of Computational Science and Engineering
•  
• Dr. Jimmy Tai – School of Aerospace Engineering
•  
• Dr. Jean Charles Domerçant – Georgia Tech Research Institute
•