高处坠落风险智能监控方法及关键技术研究

The construction industry has a higher injury and fatality rate than most of the other industries. Among the root causes of accidents in construction, falling from height accounts for nearly half of the construction accidents. Existing safety practice on fall prevention heavily measures after falling happens; actual prevention measures are mostly manual: on-site safety operation inspection and worker’s awareness on unsafe situation based on her training. However, the worker and safety inspection cannot always be aware of their safety situation when working at height. With the development of the autonomous fall prevention system, workers can now focus more on their works and let the fall prevention system monitor their safety situations. Existing fall prevention monitoring systems mostly rely on location information and compare it to dangerous areas to give warnings to the workers when applicable. It did not capture the comprehensive safety knowledge of working at height (especially in narrow spaces). Therefore, the research aims to provide a more practical and accurate intelligent fall prevention monitoring system by integrating data from location-aware sensors, accelerometers, compass, and others based on the safety knowledge extracted from a more comprehensive study on falling-related accidents. The research advanced the knowledge in three areas: First, I will develop the specification for fall accident prediction in a sensed environment based on accident investigation, expert interview and literature review; second, I will translate the specification to computer readable decision making process and deploy them in a distributed intelligent fall prevention monitoring system considering data fusion. Third, I will evaluate the system in both mocked lab environment and real jobsite to evaluate the system performance and explore approaches to improve the systems performance. To the community of practice, the research provides a more reliable extra layer to monitor the safety situations of working at height and gives warnings to them when they are under the risks of falling from height.

建筑业是伤亡率最高的行业之一，我国年均建筑安全事故总损失达1800~2500亿元，已经超过建筑企业的平均利润。统计数据表明，高处坠落占建筑安全事故的近50%。目前高处坠落安全管理主要集中在坠落后的安全防护、日常现场安全巡查等人工手段。智能预警系统应用了信息技术，提高了安全管理的实时性和全面性，但大多简单依赖距离信息来判断工人的安全状况，准确度有待提高。本研究旨旨在提取高空坠落防护的安全知识，整合多种传感数据，设计更科学可靠的高处坠落风险评估算法，建立更为可靠的坠落预警方法。预期研究成果将有助于研究人员更好的了解高空坠落发生机制及预测方法，对于建筑从业人员提供实用的高处坠落预警系统及配套硬件，具有理论创新性及高度的实用性。研究成果提供了积极有效的高处坠落防范措施，将有效减少安全事故对人、企业和社会造成的损害，实现保护人类财富，减少无益损失，保障劳动条件和维护社会经济增长。

建筑业是伤亡率最高的行业之一，我国年均建筑安全事故总损失达1800~2500亿元，已经超过建筑企业的平均利润。统计数据表明，高处坠落占建筑安全事故的近50%。目前高处坠落安全管理主要集中在坠落后的安全防护、日常现场安全巡查等人工手段。智能预警系统应用了信息技术，提高了安全管理的实时性和全面性，但大多简单依赖距离信息来判断工人的安全状况，准确度有待提高。.本研究从"高坠风险判断模型"、"工人室内定位系统"、"安全围栏检测模型"、"高坠可能性分析模型"、"高坠后果分析"等五个方面展开研究。形成的研究成果主要包括：.1) 简化的高空坠落风险判断模型、高坠风险的影响因素及潜在数据源；2）建立了无需环境学习可快速低成本部署的工人室内定位模型；3）使用深度学习算法建立了准确度达97%的安全围栏检测算法；4）使用穿戴式传感器获取姿态数据预测高坠概率；5) 对美国过去10年的高坠事件进行分析确定影响高坠后果的关键因素。研究成果提供了积极有效的高处坠落防范措施，将有效减少安全事故对人、企业和社会造成的损害，实现保护人类财富，减少无益损失，保障劳动条件和维护社会经济增长。