Deep-learning Networks for Construction Quality Control

用于施工质量控制的深度学习网络

• 批准号: 557090-2020
• 负责人: Gulliver, Aaron
• 金额: $ 1.46万
• 依托单位: University of Victoria
• 依托单位国家: 加拿大
• 项目类别: Alliance Grants
• 财政年份: 2020
• 资助国家: 加拿大
• 起止时间: 2020-01-01 至 2021-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=706286
• 关键词: Deep; learning; Networks; Construction; Quality

The focus of this project is to design and implement an effective deep learning network to identify and predict quality problems in construction projects. This will involve deep learning-based analysis of text to automatically identify quality defects from data generated by the building information management (BIM) technology during the construction process. This analysis will be done by extracting keywords and the context of surrounding words to obtain an accurate text classification model. The partner company has developed an information-based management model to reduce costs and maximize profits over the construction process. It uses traditional data analysis methods built on project intelligence to facilitate cost optimization. It processes internal historical data (e.g. planned start and end dates) as well as external factors (e.g. project size and contract) to provide cost estimates. Although this model can be used to reduce project costs, performing an analysis over all stages of a construction project is unmanageable due to the complexity of tasks that must be performed and the massive amounts of data collected by the BIM technology. This reduces the platform utility and impacts construction productivity. The BIM technology collects data from all project phases, so this data can be used as a basis to answer numerous task-specific questions. The first goal of this project will be to enable the platform to process project data from the entire construction project life cycle to answer task-specific questions through data-driven insights. A predictive model will be developed to examine current and past data to detect quality defects. This model will be applicable to a wide range of construction projects to predict quality issues in areas such as framing, drywall, electric, plumbing, and flooring. The predictive model will be based on deep learning algorithms to ensure minimal defects as well as project execution according to required standards. It will also be used to classify quality defects into categories such as minor, major and critical. This hierarchical classification can have a significant impact on improving areas such as schedule management, subcontractor management, construction site environment monitoring, and safety.

该项目的重点是设计和实现一个有效的深度学习网络，以识别和预测建筑项目中的质量问题。这将涉及基于深度学习的文本分析，以自动识别建筑信息管理（BIM）技术在施工过程中生成的数据中的质量缺陷。这种分析将通过提取关键词和周围词的上下文来完成，以获得准确的文本分类模型。该合作伙伴公司开发了一种基于信息的管理模式，以降低成本，并在施工过程中实现利润最大化。它使用基于项目智能的传统数据分析方法来促进成本优化。它处理内部历史数据（如计划的开始和结束日期）以及外部因素（如项目规模和合同），以提供成本估计。虽然该模型可以用于降低项目成本，但由于必须执行的任务的复杂性以及BIM技术收集的大量数据，因此对建筑项目的所有阶段进行分析是无法管理的。这降低了平台的效用并影响施工生产率。BIM技术从所有项目阶段收集数据，因此这些数据可以作为回答许多特定任务问题的基础。该项目的第一个目标是使该平台能够处理整个建设项目生命周期的项目数据，通过数据驱动的见解回答特定于任务的问题。将开发一个预测模型来检查当前和过去的数据，以检测质量缺陷。该模型将适用于广泛的建筑项目，以预测框架、干墙、电气、管道和地板等领域的质量问题。预测模型将基于深度学习算法，以确保最小的缺陷以及根据所需标准执行项目。它还将用于将质量缺陷分类为轻微、重大和严重等类别。这种层次分类可以对改进诸如进度管理、分包商管理、施工现场环境监控和安全等领域产生重大影响。