Join our welcoming International Design Community.
Complete this application form to join now.
Author: D’Amelio, Valentina
Supervisor: Tomiyama, T.
Institution: Technische Universiteit Delft
In the last decades, the complexity of products has increased exponentially. This com- plexity is due to the ever-growing market demand for new and innovative products with multiple functionalities, to the product size and component density that grow to a level that only a handful of experts can deal with, and to the multi-disciplinarity or inter-disciplinarity nature of products such as mechatronics products. As a conse- quence of this product complexity, the product development process is becoming complex as well due to the multiple stakeholders involved in the process and their rapidly changing roles.
A well-known procedure for product development is the V-model. In the V-model, a product design consists firstly of requirements analysis, which considers the possible conflicting needs of various stakeholders, secondly of the system design that derives system specifications from requirements and chooses main concepts of the product. Thirdly, the overall system concept is quickly decomposed in subsystems that provide the design with more details and eventually components are designed. Then, the design verification starts by gradually integrating and testing components and subsystems.
Subsystems are clearly defined and understood because they belong to one small mod- ule of a product. Although each subsystem has clear definition and it is supposed to behave as specified, designers can still be surprised by unpredicted problems that dete- riorate the performance of the product. Unpredicted problems could be hard to solve or even hard to detect. As a consequence, reaching a final satisfactory product can be time-consuming and cost-inefficient because of additional iterations between design and design verification in the development process. In order to reduce these iterations, we need to detect such unpredicted problems as early as possible in the design process.
The aim of the research presented in this thesis is to diminish unpredictable problems of complex products such as multi-disciplinary products by taking measures in the conceptual design phase of those products.
To do so, a Design Interferences Detector (DID) has been developed. DID employs the Function-Behavior-State (FBS) model as a representational method and Qualitative Physics (QP) as a reasoning engine. Basically, first, the designer builds the FBS model of the design to analyze, and then a qualitative reasoning system infers physical phe- nomena that were not predicted by the designer. The FBS model incorporates func- tional, behavioral and architectural information of the product. QP infers qualitative information by referring to qualitative knowledge of physical systems.
QP is considered potentially useful in building a computational support for the con- ceptual phase of a design because precise and complete information is not required and concept solutions do not have to be evaluated on a detailed level before proceed- ing to further developments. Another advantage is that it allows building a model of a fairly complex system in a short time that can be handy for big systems.
However, qualitative reasoning has disadvantages as well, which makes difficult its use in practice. For instance, qualitative reasoning generates too many spurious and neg- ligible solutions, which need to be removed from the list of inferred results. This is a soundness problem. In certain situations, the system could not predict all possible in- terferences, which is a completeness problem. Furthermore, qualitative reasoning has intrinsic ambiguities due to the lack of quantitative information of the design model.
To solve these problems, we developed various methods that became part of DID. A first method, the prioritization method, alleviates the completeness problem by act- ing on implicit relations of components. Two filtering methods, the contrast method and the interaction finding method, alleviate the soundness problem by using heu- ristics. These methods are useful to classify physical phenomena into spurious and negligible, predicted and unpredicted. A third method, the ambiguity solver method, has been developed to reduce ambiguities from the qualitative results based on intel- ligent user interventions.
DID has been tested and applied on various case studies in order to show its efficiency. The methods are explained one by one in detail so that they can easily be reproduced for further developments of the tool. Finally, these methods are integrated in a single framework so that, in total, they can increase a product’s predictability.