סמינר מחלקה של שי אלכסנדרוני -חיפוש מושגים מספקים תחת מטרות מרובות
School of Mechanical Engineering Seminar
Wednesday, January 11, 2023 at 14.00
Wolfson Building of Mechanical Engineering, Room 206
Searching for Satisficing Concepts under Multiple Objectives
Shai Alexandroni
M.Sc. student of Dr. Amiram Moshaiov
The Computational Intelligence Research Group
Concept selection is one of the critical decisions of the engineering design process. However, decision-making during the conceptual design stage is challenged by the large variety of possible designs, high uncertainties, and the biases of the Decision Makers (DMs). Thus, computer-supported conceptual design methods have been suggested in recent years. One such method is the Window-of-Interest based Satisficing Concept Search (WoISCS). This method involves the representation of a conceptual solution as a set of particular solutions and on predefined performance requirements. These requirements are presented as a set of acceptable performance vectors which has been termed Window-of-Interest (WoI). A concept is considered as satisficing if it is associated with at least one particular solution that has a performance vector within the WoI. Prior to this study, WoISCS has been researched to a limited extent, and only one dedicated search algorithm, which is termed WoISR, has been suggested in the literature.
The current research goal is to develop and compare new search algorithms for WoISCS. First, dedicated methods and tools are developed for evaluating WoISCS algorithms. Then, a first attempt on algorithm development is carried out by repurposing of existing Multi-Objective Evolutionary Algorithms (MOEAs), which were originally designed for Pareto optimization. A total of eight MOEAs were tested and compared to WoISR. The MOEAs performed surprisingly well, and some were comparable to the dedicated WoISR algorithm. Subsequently, a novel algorithm for WoISCS was developed based on the lessons of the MOEAs evaluation. The new algorithm was tested and compared to the other algorithms. Based on the numerical experiments it has been statistically inferred that the proposed algorithm outperforms both WoISR and the MOEAs in nearly all the test problems.
Join Zoom Meeting
https://tau-ac-il.zoom.us/j/4962025174?pwd=bVJUeElXRUUya3BERisyNllLOE9EZz09