Data-Driven Research On Engineering Design Thinking And Behaviors In Computer-Aided Systems Design: Analysis, Modeling, And Prediction
Date of Graduation
Doctor of Philosophy in Engineering (PhD)
Second Committee Member
Third Committee Member
Fourth Committee Member
Deep learning, Engineering design thinking, Machine learning, Sequential decision making
Research on design thinking and design decision-making is vital for discovering and utilizing beneficial design patterns, strategies, and heuristics of human designers in solving engineering design problems. It is also essential for the development of new algorithms embedded with human intelligence and can facilitate human-computer interactions. However, modeling design thinking is challenging because it takes place in the designer’s mind, which is intricate, implicit, and tacit. For an in-depth understanding of design thinking, fine-grained design behavioral data are important because they are the critical link in studying the relationship between design thinking, design decisions, design actions, and design performance. Therefore, the research in my dissertation aims to develop a new research platform and new research approaches to enable fine-grained data-driven methodology that helps foundation ally understand the designers’ thinking and decision-making strategies in engineering design. To achieve this goal, my research has focused on modeling, analysis, and prediction of design thinking and designers’ sequential decision-making behaviors.
In the modeling work, different design behaviors, including design action preferences, one step sequential decision behavior, contextual behavior, long short-term memory behavior, and reflective thinking behavior, are characterized and computationally modeled using statis tical and machine learning techniques. For example, to model designers’ sequential decision making, a novel approach is developed by integrating the Function-Behavior-Structure (FBS) design process model into deep learning methods, e.g., the long short-term memory (LSTM) model and the gated recurrent unit (GRU) model.
In the work on analysis, this dissertation focuses primarily on different clustering analysis techniques. Based on the behaviors modeled, designers showing similar behavioral patterns can be clustered, from which the common design patterns can be identified. Another analysis performed in this dissertation is on the comparative study of different sequential learning techniques, e.g., deep learning models versus Markov chain models, in modeling sequential decision-making behaviors of human designers. This study compares the prediction accuracy of different models and helps us obtain a better understanding of the performance of deep-learning models in modeling sequential design decisions.
Finally, in the work related to prediction, this dissertation aims to predict sequential design decisions and actions. We first test the model that integrates the FBS model with various deep-learning models for the prediction and evaluate the performance of the model. Then, to improve the accuracy of the prediction, we develop two approaches that directly and indirectly combine designer-related attributes (static data) and designers’ action sequences (dynamic data) within the deep learning-based framework. The results show that with ap propriate configurations, the deep-learning model with both static data and dynamic data outperforms the models that only rely on the design action sequence. Finally, I developed an artificial design agent using reinforcement learning with a data-driven reward mechanism based on the Markov chain model to mimic human design behavior. The model also helps validate the hypothesis that the design knowledge learned by the agent from one design problem is transferable to new design problems.
To support fine-grained design behavioral data collection and validate the proposed approaches, we develop a computer-aided design (CAD)-based research platform in the application context of renewable engineering systems design. Data are collected through three design case studies, i.e., a solarized home design problem, a solarized parking lot design problem, and a design challenge on solarizing the University of Arkansas (UARK) campus. The contribution of this dissertation can be summarized in the following aspects. First, a novel research platform is developed that can collect fine-grained design behavior data in support of design thinking research. Second, new research approaches are developed to characterize design behaviors from multiple dimensions in a latent space of design thinking. We refer to such a latent representation of design thinking as design embedding. Furthermore, using deep learning techniques, several different predictive models are developed that can successfully predict human sequential design decisions with prediction accuracy higher than traditional sequential learning models. Third, by analyzing designers’ one-step sequential design behaviors, common and beneficial design patterns are identified. These patterns are found to exist in many high-performing designers in the three respective design problems studied. Fourth, new knowledge has been obtained on the ability of deep learning-based models versus traditional sequential learning models to predict sequential design decisions of human designers. Finally, a novel research approach is developed that helps test the hypothesis of transferability of design knowledge. In general, this dissertation creates a new avenue for investigating designers’ thinking and decision-making behaviors in systems design context based on the data collected from a CAD environment and tested the capability of various deep-learning algorithms in predicting human sequential design decisions.
Rahman, M. (2022). Data-Driven Research On Engineering Design Thinking And Behaviors In Computer-Aided Systems Design: Analysis, Modeling, And Prediction. Graduate Theses and Dissertations Retrieved from https://scholarworks.uark.edu/etd/4650
Artificial Intelligence and Robotics Commons, Computer-Aided Engineering and Design Commons