A Statistical and Machine Learning Analysis of Pullout Resistance of Power actuated fasteners (PAF)

Abstract

Power Actuated Fasteners (PAFs) are widely used in construction for non-structural, light-duty applications due to their efficient and cost-effective installation. These fasteners are often installed in sets to improve reliability. This study focuses on predicting the pull-out capacity of individual PAFs based on experimental measurements using a machine learning approach. A Random Forest model is developed and trained on an extensive dataset of test results conducted across various concrete configurations, including both traditional concrete and fiber-reinforced concrete, using steel and synthetic fibers. Key experimental parameters such as fiber type and dosage, nail curvature, embedment depth, and surface damage characteristics are incorporated into the model. The model is thoroughly tested, and its predictive performance evaluated using standard metrics such as MAE, MSE, RMSE, and R². The results demonstrate the model's ability to capture complex relationships between the input parameters and the pull-out capacity, offering an interpretable and data-driven tool for estimating fastener performance. This approach enhances the reliability of fastening systems by enabling performance assessment based on measurable input parameters—without the need for additional destructive testing. The methodology can be extended to other fastening technologies and construction scenarios, contributing to safer and more reliable structural design.