Preface and Editorial

This state-of-the-art-paper is a resume of research activity of a non-formal Research Group on Artificial Neural Networks (RGANN) applications in Civil Engineering (CE). RGANN has been working at the Cracow University of Technology, Poland, since 1996 under the supervision of the author of this paper. Ten years 1996-2005 of the research and teaching activity of RGANN was reported in paper [61]. The present paper briefly reports on the activities originated in the ten year period and their continuation after 2005. The main attention is focused on new research carried out in the five year period 2006-2011. The paper discusses some selected problems which are included in fourteen supplementary papers, marked in references of these papers as published in this CAMES Special Issue. The attention is focused on: Hybrid Computational Systems, development of modifications of ANNs and methods of their learning, Bayesian neural networks and Bayesian inference methods, damage identification in CE structures, structure health monitoring, applications of ANNs in mechanics of structures and materials, joining of ANNs with measurements on laboratory models and real structures, development of new non-destructive measurement methods, applications of ANNs in health structure monitoring and repair, applications of ANNs in geotechnics and geodesy. The paper is based on the supplementary papers which were presented at the Special Session on Applications of ANNs at the 57th Polish Civil Engineering Conference in Krynica, 2011, see [74].


Keywords: Civil Engineering Artificial (CE), Neural Network (ANN), Standard NN (SNN), Multi-Layred Perceptron (MLP), Fuzzy Weight NN (FWNN), Kalman Filtering (KF), Bayesian NN (BNN), True-BNN (TBNN), Semi-Bayesian NN (SBNN), Gaussian Process (GP), Recurrent Cascade NN (RCNN), Principle Component Analysis (PCA), Hybrid Computational System (HCS), Finite Element Method (MES), Em- pirical Data (EMP), Hybrid Monte Carlo Method (HMCM), Hybrid Updated Algorithm (HUA), Neural Material Model (NMM), Structural Health Monitoring (SHM).

A study of the capabilities of artificial neural networks in respect of selected problems of the analysis of mine-induced building vibrations is presented. Neural network technique was used for the prediction of building fundamental natural period, mapping of mining tremors parameters into response spectra from ground vibrations, soil-structure interaction analysis, simulation of building response to seismic- type excitation. On the basis of the experimental data obtained from the measurements of kinematic excitations and dynamic responses of actual structures, training and testing patterns of neural networks were formulated. The obtained results lead to a conclusion that the neural technique gives possibility of efficient, accurate enough for engineering, analysis of structural dynamics problems related to mine-induced excitations.


Keywords: neural network, neural simulation, data compression, data pre-processing, mining tremors, experimental data.

The article presents possibilities of using different artificial neural networks and neuro-fuzzy systems to solve certain engineering geodesy tasks. Special attention is paid to tasks connected with the construction of a numerical terrain model, transformation of coordinates from the "1965" system into the "2000" system, and prediction of a time series on the basis of results of GPS measurements. The paper also includes a short description of those neural networks and neuro-fuzzy systems that provided good quality solutions of the tasks undertaken. The goal of the article is to review the papers published in the years 2005-2010.


Keywords: neural networks, neuro-fuzzy systems, geodesy.

The paper presents a structure test system developed for monitoring structural health, and discusses the results of laboratory experiments conducted on notched strip specimens made of various materials (aluminium, steel, Plexiglas). The system takes advantage of elastic wave signals actuated and sensed by a surface-mounted piezoelectric transducers. The structure responses recorded are then subjected to a procedure of signal processing and feature's extraction, which includes digital filters, wavelets decomposition, Principal Components Analysis (PCA), Fast Fourier Transformation (FFT), etc. A pattern database defined was used to train artificial neural networks and to establish a structure diagnosis system. As a consequence, two levels of damage identification problem were performed: novelty detection and damage evaluation. The system"s accuracy and reliability were verified on the basis of experimental data. The results obtained have proved that the system can be used for the analysis of simple as well as complex signals of elastic waves and it can operate as an automatic Structure Health Monitoring system.


Keywords: artificial neural networks, damage detection, structural health monitoring, elastic waves.

The paper presents the application of Artificial Neural Networks (ANNs) for finite element (FE) models updating. The investigated structures are beams and frames, their models are updated by ANNs with input vectors composed of dynamic characteristics of structures measured on laboratory models. The ANNs (multi layer feed-forward networks and Bayesian neural networks) are trained on numerical data disturbed by an artificial noise. The responses of the structures are measured on laboratory models. The updating procedure is also applied in identification of defects or additional masses attached to the structure.


Keywords: artificial neural networks, updating, dynamics, vibrations, identification.

The paper develops the idea of [8], i.e., the application of Artificial Neural Networks (ANNs) in probabilistic reliability analysis of structures achieved by means of Monte Carlo (MC) simulation. In this method, a feed-forward neural network is used for generating samples in the MC simulation. The patterns for network training and testing are computed by a Finite Element Method (FEM) program. A high numerical efficiency of this Hybrid Monte Carlo Method (HMC) is illustrated by two examples of the reliability analysis that refer to a steel girder [4] and a cylindrical steel shell [2].


Keywords: reliability, Artificial Neural Networks (ANNs), Finite Element Method (FEM), Hybrid Monte Carlo Method (HMC), steel girder, cylindrical steel shell.

The paper is a development of research originated in [8]. The identification problem deals with searching the location of a small mass attached to a steel plate. The corresponding inverse problem is based on measurement of dynamic plate responses on a laboratory model of the plate, taking into account only the bending plate eigenfrequencies. In the inverse analysis the Recurrent Cascade Neural Network was applied, developed in [3]. Much attention is paid to recognition of identification possibilities of RCNN. The testing process is in fact an unsupervised learning, which can lead to unstable and inaccurate recurrence procedure. That is why the verification testing process was carried out adopting the barrier bound approach. These problems are discussed in the present paper.


Keywords: laboratory model of plate, plate eigenfrequencies, Recurrent Cascade Neural Network (RCNN), supervised and unsupervised learning, verification testing, barrier bound.