The paper presents a discussion of some applications of Artificial Neural Networks (ANNs) in geoengineering using the analysis of the following six geotechnical problems, related mainly to prediction and classification purposes: 1) prediction of Overconsolidation Ratio (OCR), 2) determination of potential soil liquefaction, 3) prediction of foundation settlement, 4) evaluation of piles bearing capacity, 5) prediction of compaction parameters for cohesive soils, 6) compaction control of embankments built of cohesionless soils. The problems presented are based on the applications of the Multi-Layered Perceptron (MLP) neural networks.


Keywords: geotechnical problems, artificial neural networks (ANNs), Multi-Layer Perceptron (MLP), Overconsolidation Ratio (OCR), bearing capacity of piles, settlement of foundation, soil liquefaction, compaction control.

The paper describes the applications of back propagation neural networks with the ability to process input and output variables expressed as fuzzy numbers. The presentation of an algorithm for finding fuzzy neural network weights is followed by three examples of applications of this technique to the problems of implicit modelling of material and structure behaviour. The following problems are considered: prediction of concrete fatigue failure, high performance concrete strength prediction, and prediction of critical axial load for eccentrically loaded reinforced concrete columns.


Keywords: neural networks, fuzzy weight neural network, strength of high performance concrete, buckling of reinforced columns.

An advisory system for repairs of industrial concrete floors is a supporting tool for making material and technological decisions in the sphere of problems of recurrent character. The presented advisory system has the character of a hybrid system. Various elements of tools from the artificial intelligence group have been used in it. Artificial neural networks are of particular importance for functioning of the system. They act as an inference engine. The article presents, inter alia, an approach in the sphere of teaching artificial neural networks on the basis of an expert's knowledge, as well as utilization of fuzzy sets for data transformation and for increasing the size of the case set. The conclusions indicate the profits resulting from utilization of artificial neural networks like speed of operation or absence of the need to possess complete knowledge.


Keywords: neural networks, advisory system, repairs, industrial floors.

The paper is a continuation of [9], where new experimental data were analysed. The Multi-Layered Perceptron and Semi-Bayesian Neural Networks were used. The Bayesian methods were applied in Semi-Bayesian NNs to the design and learning of the networks. Advantages of the application of the Principal Component Analysis are also discussed. Two compaction characteristics, i.e. Optimum Water Content and Maximum Dry Density of granular soils were identified. Moreover, two different networks with two and single outputs, corresponding to the compaction characteristics, are analysed.


Keywords: granular soils, compaction characteristics, Optimum Water Content (OWC), Maximum Dry Density (MDD), neural networks, Multi-Layered Perceptron (MLP), Semi-Bayesian NN (SBNN), Principal Component Analysis (PCA).

The article is related to the results of research on Node Decoupled Extended Kalman Filtering (NDEKF) as a learning method for the training of Multilayer Perceptron (MPL). Developments of this method made by the author are presented. The application of NDEKF and MPL and other methods (pruning of MLP, Gauss Process model calibrated by Genetic Algorithm and Bayesian learning methods) are discussed on the problem of hysteresis loop simulations for tests of compressed concrete specimens subjected to cyclic loading.


Keywords: Artificial Neural Networks (ANN), Kalman Filter (KF), Node Decoupled Extended Kalman Filtering (NDEKF), Multilayer Perceptron (MPL), Genetic Algorithm (AG), Bayesian methods, concrete specimens, cyclic loading, hysteresis loops.

The application of FEM/NMM/p-EMP computational hybrid system in formulation of the Neural Material Model (NMM) for granular soils is presented. NMM is a Multi Layer Preceptron formulated 'on-line'. The cumulative algorithm of the autoprogressive method was implemented into the FEM program. The patterns for NMM training were generated in the rigid strip footing analysis. Pseudo-empirical equilibrium paths p-EMP for verification of the NMM were computed by a FEM program for the elastic-plastic Drucker-Prage material model. The discussed inverse problem of NMM identification is illustrated by two study cases of footing: 1) rigid strip footing on plane semispace, 2) inclined slope analysis. It was numerically proved that the NMM identified in the first study case can be successfully applied to the analysis of the latter one.


Keywords: Artificial Neural Network (ANN), Neural Material Model (NMM), hybrid computational system, constitutive modelling.

This paper gives a concise overview of concrete properties prediction using advanced nonlinear regression approach and Bayesian inference. Feed-forward layered neural network (FLNN) with Markov chain Monte Carlo stochastic sampling and Gaussian process (GP) with maximum likelihood hyperparameters estimation are introduced and compared. An empirical assessment of these two models using two benchmark problems are presented. Results on these benchmark datasets show that Bayesian neural networks and Gaussian processes have rather similar prediction accuracy and are superior in comparison to linear regression model.


Keywords: nonlinear regression, Bayesian methods, concrete, neural network, Gaussian process.

In this paper, an artificial neural network (ANN) is used to approximate response of deep excavation numerical model on input parameters. The approximated model is then used in minimization procedure of the inverse problem, i.e. minimization of the differences between the response of the model (now, neural network) and the field measurements. ANN based objective function is continuous and differentiable thus gradient based optimization algorithm can be efficiently used in this problem. It is showed that initial approximation of the numerical model by means of ANN increase efficiency of the identification process without loss of accuracy.


Keywords: artificial neural network, parameter identification, deep excavation.