| Principal Investigator: | Ing. Bulko Roman, PhD. |
| Project Duration: | 9/2020 - 8/2021 |
| The project will follow on from the already completed UNIZA grant project entitled "Evaluation of Static Penetration Test Using Neural Networks". The principle of the Static Penetration Test (CPT) test is the static pushing of a measuring tip using a system of steel rods into the examined environment at a constant speed. Direct evaluation of the parameters of constitutive models is quite difficult in the case of the CPT test, but shows some potential when using alternative approaches such as constitutive models based on neural networks. In the completed project, I worked on the implementation of an artificial neural network (ANN) model for predicting complex soil profiles using the results of the static penetration test. 5926 CPT readings were collected from 9 conducted field survey sites. The main idea was to identify soil types from the results of the CPT tests, which were based primarily on Robertson graphs. The work presents the development of an ANN model using multilayer perceptrons trained by the "feed-forward" "back-propagation" (FFBP) algorithm. This follow-up project builds on the conclusions of a previous project, where the ability of an artificial neural network (ANN) to learn and solve similar geotechnical problems was demonstrated, while the proposed ANN approach well replaced soil classification according to Robertson graphs. However, for better learning of the ANN, it will be necessary to process a larger amount of data than was collected in the previous project. We could consider at least 5000 data points per soil type or soil parameter as a sufficient amount. | |
| Principal Investigator: | Ing. Gago Filip, PhD. |
| Researchers: | Ing. Bulko Roman, PhD., Ing. Masarovičová Soňa, PhD., doc. Ing. Nguyen Giang, CSc., doc. Mgr. Sitányiová Dana, PhD., Ing. Vlček Jozef, PhD. |
| Project Duration: | 4/2021 - 3/2024 |
| Artificial intelligence and machine learning are dynamically developing fields. Machine learning (ML) allows computers to learn from existing data. Geotechnical survey generates a lot of data, and various ML methods can also be used in geotechnical engineering. The project aims to explore the possibilities of using ML for geotechnical engineering problems, where uncertainty in rock environment parameters is a common part of predictive models for foundation soil behavior. Rock environment parameters can be determined using laboratory and field tests or estimated using empirical or numerical correlations that are derived from regression analysis on a data set. During the project, we plan to use artificial neural networks (ANNs) for multivariate nonlinear modeling of rock environment and prediction of its parameters. ANN models will be developed and trained using local data sets. The performance of the models will be compared with the results of laboratory and field tests. | |