Abstract
Appropriate use of helmets as industrial personal protective gear is a long-standing challenge. The dilemma for any user wearing a helmet is thermal discomfort versus the chances of head injuries while not wearing it. Applying helmet microclimate psychrometry, we propose a logistic regression- (LR) based machine learning (ML) algorithm coupled with low-cost and readily available MEMS sensors to determine if a helmet was worn (W) or not worn (NW) by a human user. Experiment runs involving human subject (S) and mannequin experiment control (C) groups were conducted across no mask (NM) and mask (M) conditions. Only ambient-microclimate humidity difference (AMHD) was a feasible parameter for helmet wearing determination with 71 to 85% goodness of fit, 72 to 76% efficacy, and distinction from control group. Ambient-microclimate humidity difference's rate of change (AMHDROC) had high correlation to helmet wearing and removal initiations and was quantitatively better in all measures. However, its feasibility was doubtful for continuous use beyond 1 min due to plateauing AMHD response. Experiments with control groups and temperature measurement showed invariant response to helmet worn or not worn with goodness of fit and efficacy consolidation to 50%. Results showed the algorithm can make helmet-wearing determinations with combination of analysis and use of data that was individually authentic and non-identifiable. This is an improvement as compared to state of the art skin-contact mechanisms and image analytics methods in enabling safety enhancements through data-driven worker safety ownership.