Aplicação de inteligência artificial em problemas geotécnicos de otimização experimental e ensaios de campo
Saved in:
| Main Author: | |
|---|---|
| Publication Date: | 2024 |
| Format: | Doctoral thesis |
| Language: | por |
| Source: | Repositório Institucional da UPF |
| Download full: | https://repositorio.upf.br/handle/123456789/9604 |
Summary: | The technological perspective with the advent and development of Artificial Intelligence/AI has enabled the implementation of machine learning techniques in various branches of science. From the geotechnical engineering standpoint, Artificial Neural Networks/ANNs, Random Forest, and similar algorithms are academically applied for various purposes. Additionally, the main applications in this field include the use of models in predicting geotechnical parameters such as soil resistance, soil bearing capacity, soil-structure interaction, and soil classification through images. Therefore, the primary objective of this doctoral thesis was to conduct a bibliometric and systematic research to identify gaps in the application of machine learning in geotechnical engineering, thus identifying applicability’s for the developed work. Complementary to this, a continuous literature review was conducted, systematically listing new techniques/algorithms and their potential applications in geotechnical engineering. As a second objective, Random Forest and ANN algorithms were applied to optimize experimental design in microbially induced calcite precipitation (MICP) processes, yielding superior results compared to traditional methods such as response surface methodology (RSM), with R2 between 0.93 and 0.94. Furthermore, machine learning and deep learning algorithms were employed as a third objective in predicting geotechnical parameters such as tip resistance (qc) and lateral resistance (fs) for cone penetration test (CPT) experiments, based on soil classification parameters. This achieved satisfactory coefficients of determination (R2) of 0.92 and 0.82 for qc and fs, respectively. Thus, the objectives demonstrate the effectiveness and generalization capability, especially when applying the Random Forest and ANN models to both problem statements studied. Moreover, the last experimental chapter of the thesis aimed to continue the research by applying decision trees like Random Forest and XGBoost regarding the predictive capacity of qc and fs, expanding the research to include CPT tests from different locations: Germany/Austria and Brazil. This involved creating a mixed database and analyzing the predictive and generalization capabilities in these cases. Therefore, the goal of the thesis was to analyze, in the chapters, the various possibilities of applying machine learning in the geotechnical environment, yielding promising results in terms of experimental design optimization and field tests like CPT. Ultimately, this allows for a data-driven site characterization (DDSC) approach with the potential implementation of these techniques in locations with data availability. |
Similar Items: Aplicação de inteligência artificial em problemas geotécnicos de otimização experimental e ensaios de campo
- Buscando a Nanoformulação Perfeita (BANP) : otimização laboratorial utilizando design experimental, inteligência artificial e aprendizado de máquina
- Estudo numérico-experimental de otimização topológica utilizando elementos finitos
- Autoeficácia no uso do computador em situações de aprendizagem : uma análise da literatura internacional
- Uma arquitetura de few-shot learning para classificação de insetos na agricultura usando poucas amostras
- Avaliação do comportamento geotécnico de um aterro experimental de rejeito filtrado de flotação de minério de ferro
- Classificação de imagens médicas em modalidades usando visão computacional