Supervised ANN-assisted modeling of seated body apparent mass under vertical whole body vibration

The modeling of human body response to whole body vibration has been a challenging task owing to the complex dependence on various factors related to anthropometry, and sitting and vibration conditions. This paper addresses the functionality of Artificial Neural Network (ANN) for prediction of seated body apparent mass under different levels of vibration excitations in the 0.5-20 Hz frequency range, while assuming two different sitting postures (with and without a back support). A multilayer feed-forward neural network with back propagation (BP) algorithm was developed with various structures to identify an optimal configuration. From preliminary simulations, a neural network structure with 20 neurons in the first and 20 neurons in the second hidden layers was selected, which resulted in least mean square error, MSE, and highest coefficient of determination R2, when compared to those of the other model structures considered. Portions of the measured data acquired with 51 adult male and female subjects were used in the training and testing phases, which revealed MSE magnitudes of 4.83 and 5.97 kg2, respectively, with R2 values in excess of 0.96. Subsequently, the predicting ability of the model was assessed using the datasets for 14 unforeseen subjects. It was inferred that a well-trained ANN has the capacity to predict biodynamic responses of seated subjects as functions of the body mass, vibration magnitude and support condition. The model could predict the primary resonance frequency and the corresponding magnitude while the validity in predicting the responses were obtained at MSE of 2.13 and 1.83 kg and with R2 values in excess of 0.98 for the male and female subjects, respectively.