Virtual human posture simulation method based on neural network
In the manufacturing industry,the ergonomic assessment of workers'working postures is an important work for the prevention of Musculoskeletal Disorders(MSD).In this study,a simulation method for virtual human working posture was proposed to address the problems of insufficient authenticity and low efficiency of virtual hu-man working posture adjustment that exist in traditional MSD assessment methods.A virtual human skeleton model based on 6D rotation was established.A virtual human joint rotation generation model based on encoder-decoder ar-chitecture using Recurrent Neural Network(RNN)and Attentional Mechanism(AM)was designed to solve the vir-tual human joint poses with virtual human hands,feet and root node coordinates as inputs.Based on the balance method,the virtual human posture generation was iterated to generate the virtual human carrying posture.The pose was filtered according to the Rapid Upper Limb Assessment(RULA)score to obtain the final pose of the virtual hu-man.The error of the average joint angle between the virtual human working posture generated by the proposed method and the actual operating posture of the operator was 5.79°,and the accuracy rate of the RULA score was 85.4％,which showed good practicability.
NETL
NSTL
万方数据
