Whole-Body Postures during Standing Hand-Force Exertions: Development of a 3D Biomechanical Posture Prediction Model.

Abstract

Hand force and posture are key determinants of body loads during standing hand-force exertions. Current digital human modeling tools lack validated posture prediction algorithms and assume the force requested is representative of the actual hand force. Furthermore, a change in hand force is not reflected in the simulated task posture. To address this need, a three-dimensional whole-body posture-prediction model was developed based on the following general biomechanical hypothesis: People choose postures that maintain shoulder moments below a threshold, and minimize lower-back torsion, while maintaining their torso orientation near vertical. To tune and validate the model, posture data were gathered from nineteen men and women in a laboratory study of force exertions with one and two hands. Participants exerted maximum forces and 25%, 50%, and 75% of maximum at three handle heights. The exertions included pushes, pulls, and vertical up/down exertions against a stationary handle. Posture and force data were analyzed for use in a hybrid biomechanical/empirical model formulation. The off-axis forces that were observed were consistent with participants reducing shoulder moment and increasing ground reaction force to improve foot traction. Shoulder flexion/extension moments were less than or equal to 37 Nm in 90% of trials, regardless of hand force magnitude and direction. Pelvis location and orientation during one-hand exertions support the hypothesis that postures are selected to reduce rotational moments about the lumbar spine. A tradeoff between torso inclination and change in shoulder position with increasing hand force was observed for exertions performed at a mid-thigh handle-height. Across both one- and two-handed exertions, shoulder flexion/extension moment was found to be well predicted with an adjusted R2 value of 0.82 and 0.92, and root mean square error (RMSE) of 10.7 Nm and 5.48 Nm for one- and two-handed trials, respectively. The resulting posture-prediction model showed good performance for push/pull exertions. For up/down exertions, the wide range of tactics demonstrated by the study participants limited the model performance. Agreement between observed and predicted postural metrics was good when validated against 20% of trials withheld from the original dataset, with correlation coefficients ranging from 0.847 to 0.531 for all exertions.