Quantitative assessment for postural stresses of automobile assembly jobs using a neural network method

摘要

In this study, a methodology for quantitatively evaluating postural stresses of automobile assembly jobs is proposed using a new postural coding system and neural network approach. Initially, psychophysical discomfort values were determined for varying postures at five body joints: wrist, shoulder, neck, back and leg using free modulus magnitude estimation. In this study, subjective discomfort ratings were obtained from 19 student subjects for 42 representative assembly tasks including diverse ranges of postural workloads from a lab simulation. Each subject simulated a total of 42 different work postures for one mm each, and gave their whole-body discomfort ratings through the magnitude estimation technique with free modulus. The subjective discomfort rating data were then normalized by mm-max standardization method to reduce the inter-individual variations. The postures were ranked on their ratings and the relationships between the whole body discomfort ratings and the individual joint discomfort ratings were analyzed. In the neural network model, we used the postural classification and the postural stress indices of individual joints as input data. Output data, target values, included subjective ratings of whole-body discomfort for 42 postures. When the predicted postural stress values obtained by the model were compared with the observed values, they showed a good linear relationship with r{sup}2=0.591. The postural stress prediction model proposed in this study, which are specific to automobile assembly jobs, can be applied to another types of tasks in different industrial settings with a minor modification.