Estimation of EMG Activity of Trunk Muscles in Manual Lifting Tasks Based on Trunk Dynamics Using the Fuzzy Relational Rule Network

The main objective of this study was to demonstrate feasibility of predicting the electromyographic (EMG) activities of trunk muscles in manual lifting tasks. The applied data-driven fuzzy model with relational rule network utilized trunk dynamics as input variables, including sagittal and lateral trunk moments, pelvic tilt and rotation angles, and sagittal, lateral, and twist trunk angles. The data for model training and testing were randomly selected from a data set of EMG time domain values collected for 20 male college students. The utilized EMG data represented a total of 24 combinations of weight (15, 30, 50) lifted, asymmetry (0–60 degrees), and the origin and destination of lift (floor-waist, floor-102 cm, knee-waist, knee-102 cm), with two replications of each condition. The model was trained using the data for ten subjects and 18 randomly selected trials, and was then tested based on the EMG data for another ten subjects using randomly selected 6 trials. The model allowed for estimating EMG responses (in time domain) for the ten trunk muscles with the mean absolute error ranging from 0.91% to 11.8%. The study showed that it is feasible to estimate the time domain EMG responses of human trunk muscles due to manual lifting tasks with the acceptable level of accuracy.