Goran Turk

Prediction of concrete strength using ultrasonic pulse velocity and artificial neural networks.

Ultrasonic pulse velocity technique is one of the most popular non-destructive techniques used in the assessment of concrete properties. However, it is very difficult to accurately evaluate the concrete compressive strength with this method since the ultrasonic pulse velocity values are affected by a number of factors, which do not necessarily influence the concrete compressive strength in the same way or to the same extent. This paper deals with the analysis of such factors on the velocity-strength relationship. The relationship between ultrasonic pulse velocity, static and dynamic Youngs modulus and shear modulus was also analyzed. The influence of aggregate, initial concrete temperature, type of cement, environmental temperature, and w/c ratio was determined by our own experiments. Based on the experimental results, a numerical model was established within the Matlab programming environment. The multi-layer feed-forward neural network was used for this purpose. The paper demonstrates that artificial neural networks can be successfully used in modelling the velocity-strength relationship. This model enables us to easily and reliably estimate the compressive strength of concrete by using only the ultrasonic pulse velocity value and some mix parameters of concrete.

Prediction of subsidence due to underground mining by artificial neural networks

Alternatively to empirical prediction methods, methods based on influential functions and on mechanical model, artificial neural networks (ANNs) can be used for the surface subsidence prediction. In our case, the multi-layer feed-forward neural network was used. The training and testing of neural network is based on the available data. Input variables represent extraction parameters and coordinates of the points of interest, while the output variable represents surface subsidence data. After the neural network has been successfully trained, its performance is tested on a separate testing set. Finally, the surface subsidence trough above the projected excavation is predicted by the trained neural network. The applicability of ANN for the prediction of surface subsidence was verified in different subsidence models and proved on actual excavated levels and in levelled data on surface profile points in the Velenje Coal Mine.

A kinematically exact finite element formulation of elastic-plastic curved beams

Abstract A finite element, large displacement formulation of static elastic–plastic analysis of slender arbitrarily curved planar beams is presented. Non-conservative and dynamic loads are at present not included. The Bernoulli hypothesis of plane cross-sections is assumed and the effect of shear strains is neglected. Exact non-linear kinematic equations of curved beams, derived by Reissner are incorporated into a generalized principle of virtual work through Lagrangian multipliers. The only function that has to be interpolated in the finite element implementation is the rotation of the centroid axis of a beam. This is an important advantage over other classical displacement approaches since the field consistency problem and related locking phenomena do not arise. Numerical examples, comprising elastic and elastic–plastic, curved and straight beams, at large displacements and rotations, show very nice computational and accuracy characteristics of the present family of finite elements. The comparisons with other published results very clearly show the superior performance of the present elements.

GPS-derived Geoid using Artificial Neural Network and Least Squares Collocation

Abstract The geoidal undulations are needed for determining the orthometric heights from the Global Positioning System GPS-derived ellipsoidal heights. There are several methods for geoidal undulation determination. The paper presents a method employing the Artificial Neural Network (ANN) approximation together with the Least Squares Collocation (LSC). The surface obtained by the ANN approximation is used as a trend surface in the least squares collocation. In numerical examples four surfaces were compared: the global geopotential model (EGM96), the European gravimetric quasigeoid 1997 (EGG97), the surface approximated with minimum curvature splines in tension algorithm and the ANN surface approximation. The effectiveness of the ANN surface approximation depends on the number of control points. If the number of well-distributed control points is sufficiently large, the results are better than those obtained by the minimum curvature algorithm and comparable to those obtained by the EGG97 model.

MODELLING OF RADIONUCLIDE MIGRATION THROUGH THE GEOSPHERE WITH RADIAL BASIS FUNCTION METHOD AND GEOSTATISTICS

Abstract The modelling of radionuclide transport through the geosphere is necessary in the safety assessment of repositories for radioactive waste. A number of key geosphere processes need to be considered when predicting the movement of radionuclides through the geosphere. The most important input data are obtained from field measurements, which are not available for all regions of interest. For example, the hydraulic conductivity, as input parameter, varies from place to place. In such cases geostatistical science offers a variety of spatial estimation procedures. To assess the long term safety of a radioactive waste disposal system, mathematical models are used to describe the complicated groundwater flow, chemistry and potential radionuclide migration through geological formations. The numerical solution of partial differential equations (PDEs) has usually been obtained by finite difference methods (FDM), finite element methods (FEM), or finite volume methods (FVM). Kansa introduced the concept of solving PDEs using radial basis functions (RBFs) for hyperbolic, parabolic and elliptic PDEs. The aim of this study was to present a relatively new approach to the modelling of radionuclide migration through the geosphere using radial basis functions methods and to determine the average and sample variance of radionuclide concentration with regard to spatial variability of hydraulic conductivity modelled by a geostatistical approach. We will also explore residual errors and their influence on optimal shape parameters.

Modelling soil behaviour in uniaxial strain conditions by neural networks

Abstract The feed-forward neural network was used to simulate the behaviour of soil samples in uniaxial strain conditions, i.e. to predict the oedometer test results only on the basis of the basic soil properties. Artificial neural network was trained using the database of 217 samples of different cohesive soils from various locations in Slovenia. Good agreement between neural network predictions and laboratory test results was observed for the test samples. This study confirms the link between basic soil properties and stress–strain soil behaviour and demonstrates that artificial neural network successfully predicts soil stiffness in uniaxial strain conditions. The comparison between the neural network prediction and empirical formulae shows that the neural network gives more accurate as well as more general solution of the problem.

Mechanical analysis of glulam beams exposed to changing humidity

This study deals with the mechanical analysis of glulam beams during changing relative humidity of the surrounding air. The computational part of the article includes two separate numerical procedures. First, the diffusion equation is solved to determine the temporal and spatial distribution of water content in the cross-section of the beam. The results of the first computational stage are used as the input data for the numerical analysis of mechanical response of the beam. The displacements and stress distribution at some characteristic cross-sections are presented. In the article some experimentally determined values of vertical displacements in the middle of span are shown and compared to the results of numerical analysis.

Institutional repository as an important part of scholarly communication

Institutional repositories have been established as a good practice for quite some time. The European Commission requires archiving of research articles in institutional repositories in order to grant international project funding. Therefore, the interest in institutional repositories should increase in Slovenia as well. In 2011 the institutional repository DRUGG was built at the Faculty of Civil and Geodetic Engineering of the University of Ljubljana. By the end of 2012 more than 1400 scholarly publications (B.Sc., M.Sc., and Ph.D. theses) and nearly 150 research articles were archived in it. The repository DRUGG provides open access to scholarly publications and increases the visibility of the Facultys scientific publications. Building a repository is a complex project, in which the whole institution has to be involved. Library offers all technical support to the authors by archiving publications to the repository. Thus, the importance and the reputation of the library have increased, since it brings a significant added value to the quality of all activities at the faculty. The statistics of repository visits and downloads confirms its importance in Slovenia and abroad.

Statistical Properties of Strain and Rotation Tensors in Geodetic Network

This article deals with the characteristics of deformation of a body or a figure represented by discrete points of geodetic network. In each point of geodetic network kinematic quantities are considered normal strain, shear strain, and rotation. They are computed from strain and rotation tensors represented by displacement gradient matrix on the basis of known point displacement vector. Deformation analysis requires the appropriate treatment of kinematic quantities. Thus statistical properties of each quantity in a single point of geodetic network have to be known. Empirical results have shown that statistical properties are strongly related to the orientation in single point and local geometry of the geodetic network. Based on the known probability distribution of kinematic quantities the confidence areas for each quantity in a certain point can be defined. Based on this we can carry out appropriate statistical testing and decide whether the deformation of network in each point is statistically significant or not. On the other hand, we are able to ascertain the quality of the geometry of the geodetic network. The known characteristics of the probability distributions of two strain parameters and rotation in each point can serve as useful tools in the procedures of optimizing the geometry of the geodetic networks.

A comparison of the effectiveness of using the meshless method and the finite difference method in geostatistical analysis of transport modeling

Disposal of radioactive waste in geological formations is a great concern with regards to nuclear safety. The general reliability and accuracy of transport modeling depends predominantly on input data such as hydraulic conductivity, water velocity, radioactive inventory, and hydrodynamic dispersion. The most important input data are obtained from field measurements, but they are not always available. One way to study the spatial variability of hydraulic conductivity is geostatistics. The numerical solution of partial differential equations (PDEs) has usually been obtained by finite difference methods (FDM), finite element methods (FEM), or finite volume methods (FVM). These methods require a mesh to support the localized approximations. The multiquadric (MQ) radial basis function method is a recent meshless collocation method with global basis functions. Solving PDEs using radial basis function (RBF) collocations is an attractive alternative to these traditional methods because no tedious mesh generation is required. We compare the meshless method, which uses radial basis functions, with the traditional finite difference scheme. In our case we determine the average and standard deviation of radionuclide concentration with regard to spatial variability of hydraulic conductivity that was modeled by a geostatistical approach.