SUBJECT: Ph.D. Proposal Presentation
   
BY: Sebastian Herzig
   
TIME: Thursday, April 24, 2014, 8:00 a.m.
   
PLACE: Love Building, 109
   
TITLE: A Bayesian Learning Approach for Inconsistency Management in Model-Based Systems Engineering
   
COMMITTEE: Dr. Christiaan Paredis, Chair (ME)
Dr. Leon McGinnis (ME / ISyE)
Dr. Jonathan Rogers (ME)
Dr. Tommer Ender (GTRI)
Dr. Rahul Basole (CS)
 

SUMMARY

Developing complex engineering systems requires the collaborative effort of a variety of stakeholders. These stakeholders form different views on a system and use models to address their particular concerns of interest. Because these concerns cannot be completely separated, dependencies between the different views exist. Such dependencies can lead to inconsistencies. Identifying and resolving that is, managing inconsistencies is an open challenge. The primary question addressed in this thesis is to what degree the process of managing inconsistencies can be automated in Model-Based Systems Engineering. One essential part of answering this question is to develop a computational method for identifying semantic relationships, overlaps and dependencies in a set of models. In this thesis, the use of stochastic reasoning for the purpose of identifying semantic relationships and inconsistencies in formal models is investigated. The primary focus is on the development of a mathematically sound framework as a basis for a formal computational method. A stochastic reasoning approach is chosen due to the inherently incomplete, heterogeneous and structurally dissimilar nature of the models used in the process of developing engineering systems. More specifically, a particular stochastic machine learning algorithm Bayesian learning is investigated as a practical approach towards assisting humans in identifying inconsistencies. Bayesian approaches have the advantage that newly provided information can be used as evidence to update a prior belief. Features of models are used as evidence and are extracted by means of pattern matching. A prediction is then made to update the belief of a particular inconsistency existing. To the best of knowledge of the author of this thesis, the use of Bayesian learning has not yet been investigated within the context of inconsistency management. In order for both the effectiveness and the possible gain in efficiency resulting from the use of this approach to be measurable, a prototype implementation and a case study are developed. The case study is formulated in close collaboration with Boeing Research & Technology and focuses on managing inconsistencies in the information and knowledge contained in a holistic model of a landing gear system that is composed of both design- and manufacturing-related models.