Combining physics-based and Kriging models to improve the estimation of noise exposure

Abstract

Worker exposure to occupational hazards is traditionally measured by equipping workers with wearable exposure monitors. An emerging alternative measurement first generates time-varying hazard maps from permanent monitors within the facility, then estimating worker exposure by integrating hazard levels traversed in map, following the tracked movement of workers. Complex environments may require many monitors to produce a hazard map with the necessary accuracy, but effective interpolation functions can reduce the required number of monitors needed. This work assesses the effectiveness of three models for accurately interpolating hazard levels among monitors: a traditional Kriging model, a physics-based model, and a hybrid model that combines the Kriging and physics-based models. The effectiveness of each interpolation function was tested with sound levels collected in four environmental settings. These detailed experimental data were used to generate over 10,000 simulation trials, where each trial configured the experimental data into a unique arrangement of simulated monitoring and sampling positions. For each simulation trial, the effectiveness of the three models was assessed with the root mean square error of the sound levels at the simulated sampling positions, using the simulated monitoring positions as input. The interpolated values between the monitored positions were analyzed separately from the extrapolated values beyond the monitored positions. The hybrid model consistently reported among the lowest errors in each trial. The Kriging model performed best for the densest networks (those with the most monitors). Even in these cases, the hybrid model performed within 10% of the Kriging model with less than a third of the monitors. The experiment demonstrates that the hybrid model is highly effective at estimating hazardous sound levels; future work may demonstrate similar advantages for gas and aerosol hazards.