Robust concept development utilising artificial intelligence and machine learning
Typ
Examensarbete för masterexamen
Master's Thesis
Master's Thesis
Program
Product development (MPPDE), MSc
Publicerad
Författare
Karlsson, Kevin
Alfgården, Hugo
Modellbyggare
Tidskriftstitel
ISSN
Volymtitel
Utgivare
Sammanfattning
The 80-20 rule suggests that design decisions significantly impact downstream effects,
such as product cost, with many of these decisions made during concept generation.
This early commitment limits the ability to make changes later in development.
Early-stage design requires a variety and quantity of concepts, but designers
often fixate on existing designs, limiting innovation. In the aerospace industry,
the complexity of concept development and evaluation is particularly challenging.
Therefore, this study seeks to explore how AI/ML methods can aid designers in the
concept development process.
This thesis was initiated as a results of a 2023 internship at GKN Aerospace, which
involved generating a concept for the sectioning and manufacturing of an existing
part. Recognizing the intricacies of these phases, the authors explored the potential
of AI/ML methods to enhance robustness in concept generation and evaluation.
The aim is to evaluate how GKN Aerospace can effectively integrate AI and ML into
their product development workflows. This involves understanding current methodologies
and identifying gaps to address before implementation. The focus is on
leveraging AI and ML to streamline complex decision-making processes, ultimately
providing actionable insights for robust, efficient concept design aligned with the
Zero Defect paradigm in aerospace.
Additionally, the thesis identifies gaps in the organization that needs to be address
before a possible integration, such as data quality and data secrecy.
The result, building on extensive interview studies and literature studies is that there
is potential in incorporating AI/ML in concept development processes. Although,
AI methods such as LLMs, still have limitations, including confidently producing
incorrect results, a phenomenon known as hallucinations.
The conclusion is that Generative AI, design tools with integrated AI/ML methods,
and LLMs still offers opportunities to simplify concept generation. LLMs can
assist with ideation, creative reasoning, and cognitive task offloading. Fine-tuned
LLMs, trained on internal documentation, provide instant feedback on less complex
tasks, helping designers explore a broader design space, mitigate bias, and enhance
knowledge, facilitating the development of robust design solutions
Beskrivning
Ämne/nyckelord
Product Development , Concept Generation , Concept Evaluation , Set- Based Design , Artificial Intelligence , Machine Learning