PI: AI-enabled Construction Quality Management System
Eyes and intelligence for construction site management
Construction is one of the pillars of Hong Kong’s economy, but the city is notorious for its high building cost – highest in Asia and third highest in the world[1]. When it comes to multi-billion-dollar projects, every penny counts. Managing a construction site is complicated as many activities of various natures happen simultaneously, entailing exceptional experience, shrewd judgment, and meticulous resource and progress management. Similarly, much manpower is invested in assuring quality and safety in construction sites.
In light of this, Prof. Heng Li, Director of Smart Construction Lab, Department of Building and Real Estate, led a research team to develop PI, a vision-based AI-enabled construction site management system that uses computer vision and deep learning to perform laborious management tasks. The smart system can be adopted in almost all aspects of site management including resource allocation, quality control, safety and progress management, greatly relieving the managers and inspectors of their heavy burden.
Issues with manual inspections and management
Quality inspections on a construction site are time-consuming and labour-intensive; the records are sometimes poorly documented and other times, not accurate. “In extreme cases, reports can be written without an actual inspection being conducted. The multi-tier subcontracting practice further complicates the situation as subcontractors may not uphold the same high standards as required by the prime contractor,” explained Prof. Li. Sub-standard works in a project are always costly to fix and are likely to cause delays, and may even jeopardise safety of workers and users.
Similarly, tracking progress, analysing productivity and resource levelling on a construction site are equally painstaking. So far, most construction managers in Hong Kong are taking a manual approach to all such tasks, which require paramount commitment and experience. “In big projects, it takes time to just travel from one end to the other end of a site. That means it is almost impossible for the manager to grasp the real status of the whole project at any one point of time precisely.”
PI: Intelligence and eyes for project management
The project that Prof. Li and his team developed was christened "PI", which stands for both "Project Intelligence" and "Project Eyes". It was designed to help managers and inspectors with most of the tedious tasks associated with onsite monitoring, data collection and report writing, so that they can spend time on more important matters. The system basically uses images simultaneously acquired from surveillance cameras within the construction site for management purposes. The images are analysed by an AI algorithm that identifies 22 classes of objects commonly found on construction sites, including workers, materials, products, equipment and general vehicles. By recognising the activities that are being conducted using deep learning technology, the system sheds light on the status, level of completion and even the quality of the work performed.
“PI uses images from surveillance cameras that are already being used on construction sites, so that workers are not required to wear any sensors. It is non-intrusive and requires little additional cost. Besides, the AI algorithm can be trained to identify different activities with deep learning, meaning PI can be adopted for almost any on-site management task, such as resource and manpower management, quality control and even safety monitoring,” said Prof. Li. The system performs well with over 80% and 95% accuracy in objection detection and action recognition respectively. This is quite impressive in the domain of construction informatics. A local public utility company will adopt the system in one of its construction sites later this year.
In April 2019, PI system won a gold medal at the 47th International Exhibition of Inventions of Geneva, Switzerland.