The durability of reinforced concrete (RC) structures is strongly influenced by carbonation, a phenomenon governed by material and environmental interactions. This study applied machine learning (ML) techniques—Random Forest (RF), Support Vector Regression (SVR), and Artificial Neural Networks (ANNs)—to predict carbonation depth using a synthetic dataset of 20,000 instances generated from the validated Possan equation. Model performances were evaluated across multiple scenarios, with compressive strength and exposure time identified as the most influential features, while relative humidity and exposure conditions had intermediate effects. SVR consistently captured linear and nonlinear trends, the ANN achieved the highest R² values but showed minor overestimations, and RF exhibited lower adaptability to feature variations. The results highlight the applicability of ML models for durability assessments, particularly under complex conditions where traditional approaches are limited. Moreover, this study reinforces the strategic value of synthetic datasets in developing predictive models when experimental data collection is time-consuming or impractical. The methodologies developed here can be extended beyond carbonation modeling to other deterioration processes, supporting data-driven strategies for maintenance planning and resilience design in RC structures.

Este artigo aborda a criação de um método computacional baseado em dicionário léxico, que seja capaz de analisar textos em idioma português brasileiro e de polarizar as palavras de acordo com os sentimentos por elas transmitidos. O estudo realizado visa detectar os sentimentos expressos nos textos da Web e comparar a abordagem proposta com métodos existentes, utilizando a aplicação iFeel, que retorna os resultados dos principais métodos para a análise de sentimentos textuais. Pode-se concluir que, por meio de experimentos em uma base de dados do Twitter em português brasileiro, o método proposto retorna os sentimentos associados nos textos analisados com uma acurácia melhor do que os métodos descritos em trabalhos relacionados.