Construction site safety and quality analysis utilizing KHNN Model

Abstract

Improving construction site safety through effective hazard identification and mitigation is critical. This study aims to predict rework, defects, and associated costs using artificial neural networks and optimization algorithms. Traditional safety planning approaches lack pre-construction hazard analysis. To examine deficiencies, various metrics were analyzed, including rework costs per $1M scope and injury rates. Ineffective safety practices like inadequate training and protection have led to accidents. This work identifies approaches to enhance worker safety performance through hazard identification. Inputs to a neural network model predict rework workers, defects, and costs. Safety execution aims to systematically identify hazards before construction. Model performance using actual data was evaluated. Two soft computing methods - artificial neural network and optimization algorithms - were implemented in MATLAB. Krill herd and grey wolf optimization techniques optimized hidden neuron weights in the neural network structure. Predictions from these algorithms outperformed other existing methods like particle swarm and genetic algorithms. This study provides a framework to quantitatively forecast rework, defects, and associated costs through systematic pre-construction hazard analysis and modeling. The proposed optimizationenhanced neural network models can help construction managers implement targeted safety improvements.