基于装配任务与EEG功率信息特征的操作员认知负荷研究

Study on cognitive load of operators based on assembly task and EEG power information features

操作员认知负荷水平的准确分类,对优化企业生产可靠性指标,提高产品质量与效率具有极其重要的意义。现有认知负荷研究大多集中在航空或交通领域,而在机械装配领域研究较少,一般采用基于生理信号与设备信息构建分类模型的方法进行研究,但未针对认知负荷数据特征对分类模型进行改进,分类精度不高。为解决上述问题,设计基于多复杂度减速器装配场景的N-Back次任务脑电试验,收集了主观评价数据、装配数据和脑电信号,分析了数据与认知负荷之间的内在联系,利用机器学习算法,构建了基于互信息量的随机森林模型,实现了复杂装配系统操作员认知负荷的分类评估。结果表明:各特征因素对操作员认知负荷均有显著影响,且随认知负荷增加呈现规律性变化,认知负荷试验设计合理;基于互信息量的随机森林模型分类效果显著优于其他分类模型,多维特征可有效降低算法欠拟合能力,提高分类精度。

Through classification of the operators’ cognitive load, enterprises can optimize safety production, as well as improve production efficiency and product quality. Accurate classification of cognitive load is of great significance to enterprises. At present, the study on cognitive load is mostly carried out in the field of aviation or transportation. The physiological signals and equipment information are used to set up the classification model. However, there is little study in the field of mechanical assembly. Besides, the classification model is not optimized according to the characteristics of the data on cognitive load, thus resulting in poor accuracy in classification. In this article, an N-Back task EEG experiment based on the multi-complexity reducer assembly scene is developed. Through this experiment, efforts are made to collect the subjective evaluation data, assembly data and EEG signals, and analyze the internal relationship between the data and the cognitive load. In combination with the machine-learning algorithm, a random forest model based on mutual information is constructed, in order to realize classification and evaluation of the operators’ cognitive load in the complex assembly system. The results show that each characteristic has a significant effect on the cognitive load of operators, and it changes regularly with the ever-increasing cognitive load. The experimental design of cognitive load is reasonable. The classification effect of the random forest model based on mutual information is significantly better than that of the other classification models. The multi-dimensional features can effectively reduce the under-fitting ability of the algorithm and improve the classification effect.