A Machine Learning Approach to Automatically Capture Designers’ Affective States during Prototyping Activities

Open Access
- Author:
- Bezawada, Shruthi Reddy
- Graduate Program:
- Industrial Engineering
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- July 18, 2016
- Committee Members:
- Conrad S Tucker, Thesis Advisor/Co-Advisor
Conrad S Tucker, Committee Member
Conrad S Tucker, Committee Member - Keywords:
- affective computing
engineering design
interest
comfort
engineering machines
intelligent monitoring systems - Abstract:
- Prototyping helps design teams explore multiple approaches and ideas, hereby reducing risk and ensuring that all the design requirements are met. The iterative prototyping process can be time-consuming as well as physically and mentally demanding. It has been shown that positive affective states such as happiness, excitement or interest influence productivity. Most studies show a positive relationship between a designer’s engagement or “interest” and productivity during the prototyping process. Interest in a particular prototyping task can sometimes be the deciding factor in the success of the final product. In order to increase growth and sustainability of design teams by promoting productivity, it is important to measure affective states such as interest of designers. Though interest during the prototyping process is important, the use of power- operated machines and other such equipment require designers to be comfortable with using this kind of equipment. Comfort and interest can be measured in various ways such as filling out surveys or questionnaires at the end of a particular task. However, these methods are disadvantageous as they do not provide real-time feedback and have cost as well as scalability concerns. Addressing these issues, semi-automated/automated technologies have been developed to capture designers’ internal representations using text, speech or body language. However, analyzing designers’ internal representations using these modalities may be impractical due to interference with the task at hand. To mitigate this challenge, this thesis proposes a machine learning approach to model designers’ affective states such as interest and comfort by capturing their facial expressions. Automatic prediction of these affective states will provide real-time feedback and thus help in building intelligent systems which have the potential to improve efficiency and productivity during the prototyping process. Moreover, they will be extremely useful in the preparation of workforce training protocols used to train the new engineering design workforce. A machine learning approach is proposed to detect these affective states using trained Support Vector Machines (SVMs). An SVM classification model is used to predict interest and the accuracy is found to be 72%. An SVM regression model is used to predict comfort and it yielded an R^2 value of 0.68. The two case studies illustrate that the prediction of affective states such as interest and comfort is possible with a reasonably good accuracy. This thesis has the potential to transform the manner in which design teams utilize engineering equipment, towards more efficient concept generation and prototype creation processes.