Experimental study and deep learning model for evaluating the impact strength of a concrete U-shaped specimen retrofitted with a UHPFRC layer

<h2>Abstract</h2> Normal concrete (NC) structures are susceptible to impact-induced brittle failure, necessitating effective retrofitting solutions to enhance durability and safety. This study examines the application of ultra-high-performance fiber-reinforced concrete (UHPFRC) layers as a retrofit strategy to enhance the impact resistance of NC specimens. U-shaped NC beams were overlaid with UHPFRC layers of 10 mm and 15 mm thickness and subjected to repeated drop-weight impact tests. Acoustic emission signals during testing were analyzed using Fast Fourier Transform (FFT) and deep learning models. The results demonstrate that the 15 mm UHPFRC layer significantly enhances impact resistance, requiring 376% and 394% more blows to initiate the initial crack (N₁) and ultimate failure (N₂), respectively, compared to plain NC. The ductility index increased by 21% with the 15 mm layer, indicating a clear transition from brittle to ductile behavior. FFT analysis revealed balanced energy distribution in UHPFRC-retrofitted specimens, confirming improved energy absorption. A convolutional neural network (CNN) model achieved 93% accuracy in classifying crack stages using acoustic data, offering an efficient alternative to traditional testing. Regression models were developed to correlate layer thickness with energy absorption and ductility, providing practical tools for engineering design. This research validates UHPFRC retrofitting as a viable method for enhancing impact resistance and introduces a data-driven framework for optimizing retrofit parameters in critical infrastructure.