ALGORITHMIC DESIGN OF COMPLEX SYSTEMS

Abstract

The creative process of design influences nearly every aspect of our environment. Although humans surpass current computational devices in creativity, relying solely on human designers presents major challenges. The quality and quantity of design options depend on the knowledge, experience, and resources available. Both overstaffed and understaffed design teams impede optimal design and are prone to human biases such as anchoring, confirmation bias, and cognitive dissonance. There is an inability to fully explore the solution space, systematically capture or utilize historical learnings, and provide scientific explainability. Moreover, the design representation is based on graphical elements and focuses on the "what" and "how" but fails to consider the “why” and optimality aspects. Inevitably, the design process produces sub-optimal outcomes, resulting in significant tangible and intangible losses. While simple systems can be designed using automated techniques, the design of complex systems has traditionally remained the domain and expertise of humans, as it requires predicting and managing emergent variables. The increase in stakeholder requirements, agile approach, Industry 5.0, and other aspects have made design exceptionally challenging. Tremendous computing power and machine learning capabilities are available today, but the traditional design approaches are incompatible with leveraging them. The inability of algorithms to contribute to the process of design is a major roadblock in advancing the engineering profession. This thesis investigates whether and how algorithms can be used to design complex systems. It develops an algorithmic framework and evaluates its performance on diverse design problems. The framework standardizes the design process by segregating the design problem, process, and outcome. It investigates the full range of possible solutions, manages trade-offs among various objectives, and emulates human engineering behavior while overcoming limitations of the human-reliant approach. It facilitates the algorithmic representation and assessment of design options. The framework is modular, scalable, and enables engineers to concentrate on problem formulation. Additionally, the thesis outlines potential future research to further enhance the framework's capabilities.