Advanced Modeling of Problem-solving Behavior in Virtual Reality Manufacturing Environments

In the United States, new jobs have been consistently emerging due to the swift changes in the current manufacturing landscape. However, there have been increasing concerns in recent years about the problem-solving skills of engineering students, who will soon become the next generation of the U.S. workforce. Such circumstances call for a systematic evaluation and better understanding of the gap in engineering students' problem-solving skills. This paper developed a virtual reality (VR) manufacturing environment and adopted a deep learning-based analytical workflow to evaluate engineering students' nonlinear problem-solving behavior. Students performed assembly tasks to produce car toys that satisfy pre-defined requirements within the VR manufacturing system. Spatiotemporal eye movement data was collected and incorporated into the problem-solving performance evaluation via an LSTM-based auto-encoder. Experimental analysis showed that the results obtained from the encoded vectors outperform those solely based upon the performance metrics. The present investigation improves the evaluation consistency of performance metrics-based evaluation with the nonlinear dynamics summarized from the in-situ production process. The proposed evaluation workflow has a broader impact on facilitating the problem-solving assessment across a wide range of manufacturing systems.