EEG-based floor vibration serviceability evaluation using machine learning

The emergence of novel floor systems has made the vibration serviceability evaluation, conducted through testing and finite element simulation, a time-consuming process that struggles to accurately reflect the user experience. Even when occupant feelings are taken into account, the evaluations may be subjective. This study introduced electroencephalogram (EEG) to grade the ranges of peak accelerations (ACCs) and maximum transient vibration values (MTVVs). To achieve this goal, a supervised learning algorithm, the eXtreme Gradient Boosting (XGBoost), were adopted to conduct binary classification to recognize the threshold of peak ACCs and MTVVs. Accelerometers and an EEG acquisition instrument were utilized to simultaneously capture the ACC dataset and the EEG of volunteers. Characteristic EEG time-domain curves were then obtained by averaging the preprocessed curves corresponding to specific ranges of peak ACCs. The vibration-induced event-related potential components, P50 (the positive potential component with 50 ms latency) and N200 (the negative potential component with 200 ms latency), were identified based on the averaged EEG curves. Additionally, the peak amplitude and peak latency were calculated using cognitive neuroscience methods. The study results suggest that the potential for floor vibration-induced components is around 10 μV. Lastly, XGBoost was utilized to identify the thresholds of different ranges of peak ACCs and MTVVs using EEG time-domain and frequency-domain features. The classification accuracy according to MTVV using XGBoost can reach up to 99 %. This study quantified human perception of floor vibrations based on EEG and optimized peak ACC and MTVV threshold for the cold-formed steel floor vibration serviceability evaluation.