Multi-objective optimization-based method for kinematic posture prediction: development and validation

摘要

Posture prediction plays an important role in product design and manufacturing. There is a need to develop a more efficient method for predicting realistic human posture. This paper presents a method based on multi-objective optimization (MOO) for kinematic posture prediction and experimental validation. The predicted posture is formulated as a multi-objective optimization problem. The hypothesis is that human performance measures (cost functions) govern how humans move. Twelve subjects, divided into four groups according to different percentiles, participated in the experiment. Four realistic in-vehicle tasks requiring both simple and complex functionality of the human simulations were chosen. The subjects were asked to reach the four target points, and the joint centers for the wrist, elbow, and shoulder and the joint angle of the elbow were recorded using a motion capture system. We used these data to validate our model. The validation criteria comprise R-square and confidence intervals. Various physics factors were included in human performance measures. The weighted sum of different human performance measures was used as the objective function for posture prediction. A two-domain approach was also investigated to validate the simulated postures. The coefficients of determinant for both within-percentiles and cross-percentiles are larger than 0.70. The MOO-based approach can predict realistic upper body postures in real time and can easily incorporate different scenarios in the formulation. This validated method can be deployed in the digital human package as a design tool.