摘要
施工人员的活动监控对施工安全管理及预防职业疾病至关重要。为提升施工安全的智能化管理水平,对楼板钢筋工程施工中的8个主要活动进行识别。为弥补单一传感技术采集维度不足的缺陷和集成传感技术对系统灵活性的限制,采用智能手机内置加速度传感器和陀螺仪采集试验人员模拟施工人员活动时的加速度和倾角数据,并提取平均值、标准差、协方差、四分位距(IQR)为活动的特征矢量。通过决策树中的CART算法建立分类训练模型,采用“交叉验证法”对模型进行评估和验证。测试结果表明:对于样本个体的平均分类准确率为95.28%,预测准确率为92.86%;样本总体的分类准确率为89.67%,预测准确率为94.82%。研究表明,基于智能手机采集数据的决策树模型可以用于施工人员的活动识别。
In order to raise the construction safety intelligent management level,this paper intends to search for ways to properly collect 8 kinds of intelligent activities and effective identification in the engineering practice of the slab reinforcing bar. As a matter of fact,it is crucial to monitor the construction workers’ activities to heighten the construction safety management and prevention from the occupational diseases. For the said purpose,before collecting the data needed,two smartphones had been used to fix the subject’s right wrist and upper right leg on with the armbands to collect the ways of their acceleration and angle data for imitating their activities,with the Orion-CC appplication being used to extract the data acquired and then stored in the smartphones. Besides,since two smartphones were used to acquire the data and information needed from the construction workers,we have to use two devices to stress the accuracy of the activity identification and the acquisition time difference. In such cases,the paper has managed to use the preprocessing technique to control the time difference and then extracted the domain features,including the mean variation and the standard variation,the covariance and the IQR. Furthermore,we have managed to adopt the CART algorithm of a decision tree to build up a classification training model based on the Gini index. Before inputting the data into the system,each activity had been labeled with a unique or particular label. The effectiveness of the model has then been evaluated and verified through the cross-validation. The final results indicate that the average classification accuracy of the individual samples can reach 95. 28% with their prediction accuracy going up to 92. 86%. And,with all the samples taken,the classification can be said as accurate as up to 89. 67% with the accuracy of prediction being up to 94. 82%. The feasibility has been verified of using smartphones as data-acquisition tools also in the construction management. Moreover,it proves that the combination of a decision-tree algorithm with the help of smartphones as a supplementary device is in the position to achieve complex activity classification and identification. In addition,the use of smartphones can help to enhance the efficiency and quality of the data acquisition. Hence,the activity identification model can be adopted to implement safety monitoring evaluation for individual and groups of workers,which is of high practical value.
作者
张明媛
陈硕
赵雪峰
ZHANG Ming-yuan;CHEN Shuo;ZHAO Xue-feng(Faculty of Infrastructure Engineering,Dalian University of Technology,Dalian 116024,Liaoning,China;School of Civil Engineering,Dalian University of Technology,Dalian 116024,Liaoning,China)
出处
《安全与环境学报》
CAS
CSCD
北大核心
2019年第3期861-866,共6页
Journal of Safety and Environment
基金
中央高校基本科研业务费项目(DUT18JC44)
大连市青年科技之星项目支持计划项目(2016RQ002)
关键词
安全工程
传感器
智能手机
机器学习
活动识别
safety engineering
sensor
smartphone
machine learning
activity identification