During my MS in Engineering Design, I primarily worked "Towards Computationally Efficient Reinforcement Learning Frameworks for the Design of Cyber-physical Systems". Beyond my thesis, I contributed to a wide range of individual and group projects at the intersection of machine learning, design, and manufacturing, including natural language processing for experiential design, data-driven topology optimization and deep learning for mass customization.

My experience with human-centered design has enabled me to adopt a multidisciplinary approach that prioritizes user needs, ensuring functional, impactful and accessible outcomes. It has also enhanced my ability to communicate with both specialists and non-specialists, fostering collaboration within interdisciplinary teams.

Research projects

A multi-agent reinforcement learning framework for intelligent manufacturing with autonomous mobile robots

Intelligent manufacturing (IM) leverages Industry 4.0 principles to enhance autonomy and efficiency, but ad hoc system development limits standardization. This work presents a standardized multi-agent framework for integrated job scheduling and navigation control in an autonomous mobile robot-driven shop floor, an increasingly common IM paradigm. We specifically propose a multi-agent framework involving mobile robots, machines, humans. Further, we demonstrate the application of reinforcement learning within this framework, enabling robot collaboration to maximize throughput and safety. Moreover, we observe nuanced behavior in the demonstrations, including the ability to autonomously compensate for processing delays, and machine and robot failures in real time.

Deep learning for rapid topology optimization

Topology optimization aims to generate optimum structures by maximizing the stiffness of the structure, given a set of geometric, loading and boundary conditions. However, these approaches are computationally expensive as it requires many iterations to converge to an optimum solution. This work explores the effectiveness of deep generative models to rapidly generate solutions to a diverse range of topology optimization problems with varying design constraints, loading and boundary conditions. Specifically, four distinctive models — TOP-CNN, U-Net, GAN with U-Net as generator (TOP-GAN), and GAN with CNN regressor as generator (DCGAN), were successfully developed, trained, and evaluated in comparison to a conventional approach.

Deep learning for mass customization

Additive manufacturing offers the ability to create customized structures for body fitted products like prosthetics, helmets and shoes where customer anthropometry is significant. As the product is redesigned for a specific customer, functions of safety and comfort need to be met based on their anthropometry. This often involves the analysis of custom configured structures to evaluate a specific function. However, this analysis can significantly contribute to the total build-to-order time. To this end, we propose a deep learning method that enables the rapid analysis of custom structures. We consider helmets as our use-case product that needs to be analyzed to meet the function of safety in a crash. Specifically, ECE 22 crash simulations were performed on head forms generated using ANSUR II database. Further, a CNN was trained to predict the head form acceleration based on its orthographic views.

Topic modeling and sentiment analysis of airline reviews

Identifying customer pain points is crucial for effective design, and online user-generated data offers rich but unstructured insights. This study enhances NLP techniques by integrating sentiment analysis to extract meaningful information, focusing on aviophobia in the airline industry — a fear of flying that reduces passenger numbers and impacts the market. We propose a novel methodology using Reddit comments, employing topic modeling to identify common themes and sentiment analysis to interpret key insights. Our approach addresses the complex factors contributing to aviophobia and provides recommendations for the airline industry, along with suggestions for future research.

Research primer on 3D anthropometry of heads

This project primer delves into the fundamentals of head shape analysis using advanced 3D anthropometry techniques. It highlights key insights, such as how smaller heads tend to be rounder and larger heads more elongated, and underlines the importance of preserving full geometric details rather than relying on simplified 1D measurements. By examining the latest scanning technologies, discussing typical sources of error, and showcasing the use of statistical shape models and principal component analysis to capture shape variations, this project lays the groundwork for developing more accurate, customized head-related products and presents exciting prospects for mass customization.

Human-centered design projects

Your data for Potsdam

This project focused on enhancing the mobility experience in Potsdam, Germany by promoting a shift from individual to public transportation, aiming to sustain comfort and reduce the environmental footprint. Using the PESTEL Framework and field interviews, we analyzed trends like autonomous driving, carbon tax impacts, and climate change. Key insights were synthesized from user interviews to identify patterns and clusters. Ideas were generated and evaluated using the Ciao-How-Now-Wow scheme, leading to a proposed data-driven transport planning system. The final prototype symbolized this system with a map embedded in a cloud, emphasizing sustainable and efficient public transportation solutions.

Helmet redesign to improve user accommodation and comfort

This project focuses on redesigning unisex bike helmets to improve fit and comfort for Chinese adults (ages 18–66). Head injuries account for 48% of cycling-related injuries, and poor helmet fit contributes to reduced safety and comfort. Using anthropometric data from Chinese populations and correlations from the ANSUR II dataset, head length and breadth are estimated from head circumference. The proposed design features two shell sizes and independent padding adjustments for length and breadth, ensuring even pressure distribution and accommodating a broader range of users. This design improves helmet fit, encourages usage, and enhances cyclist safety by reducing head injuries.

Page updated

Google Sites

Report abuse