Big data-driven construction waste composition estimation : an integrated approach of bulk density and computer vision

Abstract

Globally, the construction industry’s approximate 8% contribution to Gross Domestic Product comes at the cost of generating around 40% of the total solid waste stream. Construction waste, i.e., solid waste originating from the new construction, renovation, and demolition activities of buildings and infrastructures, has emerged as a major cause of rampant environmental and social issues. Construction waste management (CWM) has thus become a prominent environmental protection initiative for most governments.

Waste composition information plays a fundamental role in onsite CWM, as waste disposal destination, procedure, levy rate, and the like vary depending on compositions. Researchers have proposed various waste composition estimation methodologies since the 1990s. However, due to concentrating only on characterizing the typical waste composition in specific regions and identifying pre-defined composition types, existing methodologies are ill-suited for real-time construction waste composition estimation (RCWCE) at given job sites like landfills, recycling facilities, and construction sites. RCWCE requires swiftly and accurately measuring the quantity of different components in mixed construction waste, to support subsequent CWM activities. It is pivotal for the smooth and efficient operation of CWM systems in contemporary cities. Yet, little research has engaged in this problem.

The primary aim of this study is to develop a RCWCE methodology by integrating the strengths of bulk density-based estimation and computer vision-based recognition. The research involves: (1) reviewing composition estimation studies across the solid waste research field and funnelling to examine existing construction waste composition estimation approaches; (2) developing the methodological framework of the integrated approach of bulk density and computer vision; (3) investigating and comparing the performance of the approach based only on bulk density or computer vision and the integrated approach; and (4) analysing the applicability and generalizability of the proposed methodology. Interdisciplinary research methods, e.g., literature research, comparative experimentation, big data analytics, and computer vision, were coupled to deliver the research objectives. All datasets required for this research were collected from the off-site construction waste sorting facilities in Hong Kong.

The theoretical contribution of the research is five-fold. Firstly, it sheds light on a widespread but underexplored research problem (i.e., RCWCE). Secondly, this research developed a generalizable RCWCE approach by combining bulk density and computer vision. Thirdly, this thesis, for the first time, uncovered the convergent bulk density probability distribution of inert and non-inert construction waste. Fourthly, this research provided empirical evidence for the correlation between bulk density and construction waste contents, and the feasibility of predicting the real quantity of different components in mixed waste by computer vision. Lastly, the research offered an important policy implication for improving construction waste composition measurement practice.

This research is also of practical significance. On the one hand, it contributes a pragmatic solution to supporting the smooth operation of composition-dependent CWM systems. On the other hand, by real-time guiding waste producers dumping destination selection so that different waste can be sent to distinct disposal facilities, this research has the potential to foster construction waste material circularity, echoing the practice of circular economy principles in construction.