Integrating computer vision into quality assurance of structural concrete in Sri Lankan construction industry

Abstract

The Sri Lankan construction industry struggles with persistent quality issues in structural concrete works, including cracks, honeycombing, spalling, and misaligned reinforcement, resulted in cost overruns, delays, and safety risks due to inefficient manual inspections. This study investigates the integration of computer vision (CV) technologies leveraging convolutional neural networks (CNNs) and deep learning algorithms into quality management processes to enable proactive, automated defect detection during construction. A literature review identifies CV's efficacy for real-time monitoring and its superiority over traditional methods, while highlighting adoption barriers including data scarcity, high initial costs, skill shortages, and infrastructure limitations specific to Sri Lanka. Semi-structured interviews with ten industry experts reveal strong consensus on CV's potential for accuracy, standardisation, and productivity gains, but they also confirm the sector's incompetence, particularly among small contractors facing a shortage of human and financial resources. Key recommendations include government-incentivised pilot projects, localised datasets, workforce training, awareness programs, and hybrid LLM-mobile applications as scalable entry points. Findings highlight the need for a tailored framework to transition from reactive to predictive quality assurance. By addressing contextual challenges through phased implementation and policy support, CV integration promises enhanced structural integrity, reduced rework, and sustainable development in Sri Lanka's construction landscape. This research provides a roadmap for digital transformation, with implications for similar developing economies.