Abstract: Traditional methods have low sensitivity to changes in input variables for the risk assessment of work-related musculoskeletal diseases,resulting in insufficient accuracy and reliability of the risk assessment outputs.To conduct human factors engineering risk assessment in a more accurate way,a Rapid Upper Limb Assessment(RULA) method was proposed based on Adaptive Neuro Fuzzy Inference System (ANFIS).Based on the convolutional neural network,the key points of the human working posture were detected and recognized in the video,with the joint angles calculated.Because of the ANFIS,the method was improved in the rapid upper limb assessment,and a risk assessment framework was constructed for work-related musculoskeletal diseases to solve the problem of obtaining the same score when different postures were evaluated.The joint angle data of different working postures were randomly selected to train and test the network,and it was adjusted the optimal parameters of the work-related musculoskeletal disease risk prediction model according to the ANFIS as well as the rapid upper limb assessment method.The first 15 working postures in the joint angle dataset were selected for correlation verification,whose results were compared with those of the original rapid upper limb assessment method.Also,the operation process of the branch pruning tool was applied to analyze the case to achieve real-time dynamic assessment of risk scores.The results showed that the optimized rapid upper limb assessment method was more sensitive than the original one,which verified that the adaptive neuro fuzzy inference system could effectively improve the rapid upper limb assessment method and predict risk scores in real time.

Key words: rapid upper limb assessment, adaptive neuro fuzzy inference system, fuzzy control, key point detection, human factors engineering risk

摘要： 传统方法对工作相关肌肉骨骼疾病风险评估的输入变量变化敏感性较低,导致风险评估输出结果的精确性和可靠性不足。为更加准确地进行人因工程风险评估,提出了基于自适应神经模糊推理系统的快速上肢评估方法(RULA)。首先,基于卷积神经网络对视频中人体工作姿势的关键点进行检测及识别,并计算关节角度;其次,基于自适应神经模糊推理系统对快速上肢评估方法进行改进,搭建工作相关肌肉骨骼疾病风险评估架构以解决评估不同姿势时获得相同评分的问题;再次,随机选取不同工作姿势的关节角度数据对网络进行训练和检测,调整基于自适应神经模糊推理系统和快速上肢评估方法的工作相关肌肉骨骼疾病风险预测模型的最佳参数;最后,选取关节角度数据集里的前15个工作姿势进行相关性验证,将结果与原始快速上肢评估方法的结果进行比较,应用树枝修剪工具的操作过程进行案例分析以实现风险得分的实时动态评估。结果表明,优化后的快速上肢评估方法比原始方法更敏感,验证了利用自适应神经模糊推理系统能够有效改进快速上肢评估方法并实时预测风险得分。

关键词: 快速上肢评估法, 自适应神经模糊推理系统, 模糊控制, 关键点检测, 人因工程风险