Comparing Automated vs. Manual Data Analytic Processing of Long Duration International Space Station Post Mission Crew Feedback

摘要

Qualitative data collected from International Space Station (ISS) postflight crew debriefs was used to evaluate the performance of a convolutional neural network (ConvNet) model. While the ISS postflight debriefs cover a broad range of spaceflight and on-orbit operations related topics, this model was specifically trained and tested to classify debrief comments as safety related or not, based on a previously coded subset of debrief comments that were manually evaluated by human factors engineers to determine if a comment had safety implications. This evaluation revealed that a ConvNet can adequately determine whether textual debrief comments contain safety data. These methods can potentially save large amounts of manual effort on the part of human factors engineers and improve the ability to identify and act on crew knowledge that informs or identifies risk to spaceflight crew.