Thousands of residential buildings have been built of prefabricated elements in Poland. Some of the buildings were repaired and modernized in insufficient degree, often occasionally only. Now this is an important issue in Poland. After years of usage, one question is asked frequently: what is the technical deterioration of these buildings? This is a serious technical as well as a social problem. Assessment of the technical deterioration of a building can be performed by many methods. These methods consider different number of analyzed details, they require expertise or are based on expert’s reports thus making their execution time consuming. These methods show different deficiencies: labor-consumption, inaccuracy, lack of objectiveness and others. Even the most exact, but labor-consuming, “visual methods” are defective – two assessments for the same building, done by different specialists, can disagree. This is the result of differences in condition assessment for a building element, particularly if it is hidden under paints, parquets, and other finishing layers. These elements can be evaluated indirectly only, through vision inspection of apparent flaws, cracks, delaminations etc. There is a need for new solutions which would help analyze current values of the technical deterioration for many buildings and would allow for prediction of the future changes. The method presented below is based on the extracted data processing by means of artificial neural networks. The aim is to learn the artificial neural network configurations for a set of data containing values of the technical deterioration and information about building repairs in last years (or other information and building parameters) and next to analyze new buildings by the instructed neural network. The profit from using ANN is the reduction of the number of parameters. Instead of more then forty parameters describing a building, about ten are usually sufficient for satisfactory accuracy. Three types of ANN has been used: MLP - Multilayer Perceptron, RBF - Radial Basis Function, SVM - Support Vector Machine. The algorithm for obtaining results from artificial neural networks (ANN) consists of: 1. preparing full database, 2. selecting data for the next step of analysis, 3. normalizing the data, 4. selecting components from the records, 5. selecting records for groups of data: for training, for cross validation, for testing, 6. choosing the type of artificial neural net, selecting topology, 7. teaching the net, verifying results, 8. repeating steps (5,6,7) for optimization of the network architecture, 9. analyzing new data with the optimal net. The net of Multilayer Perceptron type reached the best result (with architecture 10-4-1). Estimating the technical deterioration of building using ANN: has lower accuracy, can work on incomplete data, is more resistant to errors in data, needs small range of data as an input, it can be applied as one of many methods for estimating technical deterioration.
Abstract
Authors
Resource added by
Keywords
Fields of science
Civil Engineering