Estudo da resistência mecânica e detecção de defeitos em estruturas de concreto a partir de funções de resposta em frequência e técnicas de aprendizado de máquina

Abstract

The mechanical performance and durability of concrete are essential for ensuring the safety and longevity of structures used in civil and agricultural engineering. In response to the increasing incidence of pathological manifestations and the need for more precise monitoring methods, this study investigates the integrated application of experimental modal analysis and machine learning techniques to evaluate mechanical strength, identify defects, and compare the dynamic behavior of conventional concrete and bioconcrete incorporating macaúba endocarp (Acrocomia aculeata). Impact excitation tests were performed on specimens of different strength classes, recording frequency response functions (FRF) at two standard instrumentation positions to extract natural frequencies and damping ratios. The dynamic data were used in supervised machine learning models, particularly Artificial Neural Networks and Random Forest, to predict compressive strength and classify defect severity using an information- leakage-free pipeline. The results showed that dynamic properties—especially natural frequencies and damping—are sensitive to microstructural variations caused by both integrity loss and compositional changes in the concrete. The machine learning models exhibited robust performance in predicting strength and assessing integrity, highlighting the effectiveness of FRF as non-destructive predictors. The comparison between conventional concrete and bioconcrete revealed systematic dynamic differences, indicating that the partial replacement of coarse aggregate with macaúba endocarp influences dynamic stiffness without compromising structural behavior within the evaluated limits. This study therefore advances non-destructive evaluation techniques, demonstrates the feasibility of combining computational models with modal analysis, and proposes a methodological foundation for future applications in structural health monitoring and in the development of environmentally sustainable alternative materials.