Knowledge-driven Safety Analysis: An Approach to Enhance Occupational Health and Safety in Construction

The construction industry suffers from a high amount of accidents and fatalities compared to other industries. The reasons for the horrifying statistics are that the industry is highly dynamic and dangerous. The dynamic nature arises as construction is essentially production in a constantly changing and evolving environment, and the dangers arise from the unfinished building, heavy objects, and dangerous machinery. The current practices of ensuring safe work environments for construction workers are predominantly manual and labor-intensive. Safety planning is challenging and cumbersome to do digitally in 3D and even more in 4D. This is why it is often performed in a paper-based approach on the 2D plan view, which is harder to distribute among all affected parties and stakeholders. Thus, sharing the safety plan requires toolbox meetings, which are rarely assisted by digital tools and demand a lot of manual preparation.

Safety inspection is currently performed in weekly or bi-weekly walk-throughs where regulation compliance issues are identified and resolved. The manual safety efforts can not provide a temporal and spatial resolution corresponding to the construction site’s evolvement, leaving the safety planning, mitigation, and inspection a step behind the construction progress. The manual approach also does not resonate well with the future vision of a Digital Twin-driven construction industry.

This Thesis aims to propose solutions that could assist the safety-responsible person by automating some monotonous design, planning, and inspection tasks to increase the spatiotemporal resolution and enhance occupational health and safety for construction workers while freeing time for critical matters requiring experts’ attention. The solutions have been proposed with transparency and relevance for practitioners in mind, which should facilitate their acceptance and adoption of the proposed automated strategies. Besides automatization, this thesis also improves construction safety insight, exploitation, and cross-domain utilization of safety analysis.

The overall methodology to automate and provide additional construction safety insight is knowledge-based. The knowledge-based approach is built on three main steps: understanding the domain, capturing it in a computer-interpretable format, and allowing the computer to identify and capture the potential violations. The knowledge bases comprise the information from the BIM model, regulation, and domain expert knowledge, which allows for automated digital safety planning. The safety planning analysis is based on spatial artifacts and brings new possibilities to the domain of safety planning, such as safety-based schedule comparison, mitigation equipment demand estimations, and safety-related task planning. The digital safety plan is subsequently used for automated inspections to know where to search for safety protective measures and their kind to lower the identification complexity drastically. Afterward, Domain knowledge is captured in a rule-base to mimic the human compliance assessment approach, which together creates a simplistic but well-performing compliance checking framework. Capturing and sharing the safety insights across the construction site also enables stakeholders outside the safety domain to assess the impacts of safety and how safety is impacted by effects that are not directly connected, such as delays.

The present PhD project and thesis investigated, identified, and proposed novel automated solutions to assist in preventing hazards through safety design and planning, performing compliance checking on safety mitigation measures, and identifying safety-related incidents during construction project execution. The proposed solutions all feature transparency, extensibility, scalability, explainability, and stability under low data availability, which is essential when human lives are at stake. The presented results show significant potential for the solutions and the insights obtained.