The discrete counterpart of a class of Hopfield neural networks with periodic impulses and finite distributed delays is introduced. A sufficient condition for the existence and global exponential stability of a unique periodic solution of the discrete system considered is obtained.

The paper discusses methods of diagnosing the technical condition of reinforced concrete beams, based on the change in dynamic characteristics. The objects of research were 12 reinforced concrete (RC) beams. Testing of RC beams included both static and dynamic tests. A series of step loaded static tests was aimed to produce successive damage to the beams. After each load step (at the moment of displacement and strain stabilization), dynamic testing followed. To carry out the concept of concrete beams diagnosis, on the basis of frequency changes, Artificial Neural Networks (ANNs) were applied.
Keywords: estimation, reinforced concrete beam, dynamics, artificial neural networks

This paper uses an interval and fuzzy finite element approach for the eigenfrequency analysis of a mechanical structure with uncertain parameters. The component mode synthesis method is applied for the numerical reduction of the structure, in order to reduce the calculation time of the interval and fuzzy analyses. Special attention is paid to the effect of uncertainties on the description of the substructuring technique and the consequences on the calculation time. All concepts are illustrated through a benchmark structure example.

This paper deals with the second-order CH of a heterogeneous material undergoing small displacements. Typically, in this approach an RVE of a heterogeneous material is investigated. A given discretized microstructure is determined a priori, without focusing on details of specific discretization techniques. Application of BNN as a tool for identification of characteristic length of a microstructure is discussed. An indentation test was analyzed under plane strain constraints for generating pseudo-experimental patterns by means of FEM. A single input of BNN was formulated due to the application of PCA. The BNN of structure 1-16-1 with sigmoid hidden neurons was designed. The Bayesian inference approach was applied to obtain pdf of the characteristic length. Numerical efficiency of the proposed approach is demonstrated in the paper.
Keywords: micro- and macrolevels, second order continuum, computational homogenization (CH), representative volume element (RVE), finite element method (FEM), Bayesian neural network (BNN), probability density function (pdf), principal component analysis (PCA), indentation test.

A concept is presented of a system for automatic processing of the civil engineering data. It may concern designing, optimisation or diagnostics of constructional materials. The main point of interest was concrete and various concrete like composite materials. The applied methods are a combination of various soft computing techniques, like artificial neural networks, machine learning and certain techniques originating in statistics. The system is aimed at taking advantage of varied information available in publications, reports, monographs and direct experimental results, perhaps including even the grey information resources. After preparation of a database collected from laboratory or in-situ observations concerning the behaviour of various concrete materials, and gathered during the two last decades, a number of experiments were performed on the system dedicated mainly to prediction of compressive strength and frost resistance of concrete. The proposed approach allows more efficient control of information in problems of concrete technology.

This paper describes an application of feedforward neural network to analyse the SASW (Spectral Analysis of Surface Waves) measurements of the soil. The free field dynamic experiment was performed to determine the soil dynamic properties. An inversion process is based on the comparison of experimental and theoretical phase velocity curves. The results of the experiment are pre-processed by a neural network. The dynamic soil profile is compared with the real soil profile based on the geotechnical site prospect.

A new procedure based on layered feed-forward neural networks for the microplane material model parameters identification is proposed in the present paper. Novelties are usage of the Latin Hypercube Sampling method for the generation of training sets, a systematic employment of stochastic sensitivity analysis and a genetic algorithm-based training of a neural network by an evolutionary algorithm. Advantages and disadvantages of this approach together with possible extensions are thoroughly discussed and analyzed.

The paper deals with an application of neural networks for computation of fundamental natural periods of buildings with load-bearing walls. The identification problem is formulated as a relation between structural and soil parameters and the fundamental period of building. The patterns are based on long-term tests performed on actual structures. Various splitting up of the set of patterns into training and testing sets are considered in the analysis. The carried out analysis leads to conclusion that, even in "the worst'' case of randomly selected testing patterns, the natural periods of vibrations of buildings are obtained with accuracy quite satisfactory for engineering practice.

This article presents recent developments in the field of stochastic finite element analysis of structures and earthquake engineering aided by neural computations. The incorporation of Neural Networks (NN) in this type of problems is crucial since it leads to substantial reduction of the excessive computational cost. In particular, a hybrid method is presented for the simulation of homogeneous non-Gaussian stochastic fields with prescribed target marginal distribution and spectral density function. The presented method constitutes an efficient blending of the Deodatis-Micaletti method with a NN based function approximation. Earthquake-resistant design of structures using Probabilistic Safety Analysis (PSA) is an emerging field in structural engineering. It is investigated the efficiency of soft computing methods when incorporated into the solution of computationally intensive earthquake engineering problems.

The paper deals with application of AI tools in experimental modal analysis. The example of Stabilization Diagram processing, that is an intermediate stage of modal parameter estimation procedure, was selected. In order to automate decision-making carried out during Stabilization Diagram processing a set of tools employing: fuzzy reasoning and artificial neural nets was applied. The application of these tools enabled to ease and shorten execution time of Stabilization Diagram processing. Additionally, the result of processing has become operator-independent.

The paper deals with the application of soft computing used in uncertainty analysis in the field of structural dynamics. Employing Genetic Algorithms, fuzzy sets theory as well as interval algebra authors show quite useful extension of well known approaches of solving eigenproblems considering assumed model uncertainties. During performed calculation, ranges of the first natural frequency of a simple FE model are found and then compared to those ones obtained with Monte Carlo simulation. As input uncertain parameters some of material properties are taken into account. The main objective of the work is to highlight possible advantages of the application in terms of reducing computation time meant for uncertainty analyses.

Gaussian mixture models (GMM) and support vector machines (SVM) are introduced to classify faults in a population of cylindrical shells. The proposed procedures are tested on a population of 20 cylindrical shells and their performance is compared to the procedure, which uses multi-layer perceptrons (MLP). The modal properties extracted from vibration data are reduced into low dimension using the principal component analysis and are then used to train the GMM, SVM and MLP. It is observed that the GMM gives 98% classification accuracy, SVM gives 94% classification accuracy while the MLP gives 88% classification accuracy. Furthermore, GMM is found to be more computationally efficient than MLP which is in turn more computationally efficient than SVM.

This paper is devoted to the application of the evolutionary algorithms and artificial neural networks to uncertain optimization problems in which some parameters are described by fuzzy numbers. The special method of global optimization: Two-Stages Fuzzy Strategy (TSFS) for structures in uncertain conditions is proposed. As the first stage of the TSFS the fuzzy evolutionary algorithm is used. As the second stage the local optimization method with neuro-computing is proposed. The presented approach is applied in the identification problems of mechanical structures, in which material parameters and loadings are uncertain. To solve the direct problem the fuzzy boundary element method (FBEM) is used. Several numerical tests and examples are presented.

This paper proposes a neural network model using genetic algorithm for a model for the prediction of the damage condition of existing light structures founded in expansive soils in Victoria, Australia. It also accounts for both individual effects and interactive effects of the damage factors influencing the deterioration of light structures. A Neural Network Model was chosen because it can deal with `noisy' data while a Genetic Algorithm was chosen because it does not get `trapped' in local optimum like other gradient descent methods. The results obtained were promising and indicate that a Neural Network Model trained using a Genetic Algorithm has the ability to develop an interactive relationship and a Predicted Damage Conditions Model.

The objective of this paper is to investigate the efficiency of nonlinear Bayesian regression for modelling and predicting strength properties of high--performance concrete (HPC). A multilayer perceptron neural network (MLP) model is used. Two statistical approaches to learning and prediction for MLP based on the likelihood function maximization and Bayesian inference are applied and compared. Results of experimental data sets show that Bayesian approach for MLP offers some advantages over classical one.
Keywords: Bayesian inference, regression, high-performance concrete, neural network.

The identification of the industrial processes is a complex problem, especially in the case of signals denoising. The holistic approaches used for signal denoising processes are recently considered in various types of applications in the domain of experimental simulations, feature extraction and identification. A new signal filtering method based on the dynamic particles (DP) approach is presented. It employs physics principles for the signal smoothing. The presented method was validated in the identification of two kinds of input data sets: artificially generated data according to a given function y=f(x) and the data obtained in laboratory mechanical tests of metals. The algorithm of the DP method and the results of calculations are presented. The obtained results were compared with commonly used denoising techniques including weighted average, neural networks and wavelet analysis. Moreover the assessment of the results' quality is introduced.

Mathematical-model-based structural identification algorithms for the damage detection and performance evaluation of civil engineering structures have been widely proposed and their performance for small and simple structural models has been studied in the past two decades. Actual civil engineering structures, however, usually have a great number of degrees of freedom (DOFs). It is unpractical to directly apply these conventional methods for the identification of large-scale structures, because excessive computation time and computer memory are necessary for the search of optimal solutions in inverse analysis, which is often computationally inefficient and even numerically unstable. Moreover, for the identification of large-scale structures, it is difficult to obtain unique estimates of all structural parameters by the optimization search processes involved in the conventional identification algorithms requiring the use of secant, tangent, or higher-order derivatives of the objective function. The ability of artificial neural networks to approximate arbitrary continuous function provides an efficient soft computing strategy for structural parametric identification. Based on the concept of localized and decentralized information architecture, novel decentralized and localized identification strategies for large-scale structure system by the direct use of structural vibration response measurements with neural networks are proposed in this paper. These methodologies does not require the extraction of structural frequencies and mode shapes from the measurements and have the potential of being a practical tool for on-line near-real time and damage detection and performance evaluation of large-scale engineering structures.