Engineering Transactions, 45, 2, pp. 251-263, 1997

Prediction of Concrete Properities Using Neural Networks

J. Kasperkiewicz
Institute of Fundamental Technological Research, Warszawa

To design or optimise concrete mix composition, a mapping is needed from one multidimensional space of the input data into another multidimensional space of the output data. The analytical form of such transformation of the data preserved in various experimental databases is not known. Artificial Neural Network (ANN) is an alternative to using analytical description. Fuzzy ARTMAP - one particular type of ANN was adapted and trained using three different databases to predict concrete properties of strength, workability or cost. The results are encouraging. If properly trained, that is if learning is conducted at appropriate setting of the system parameters, Fuzzy ARTMAP will be able to provide correct predictions. Unlike the case of some other networks and statistical "models", which even in critical cases try to give complete predictions, often of poor accuracy or even totally incorrect, the Fuzzy ARTMAP gives no prediction at all for very non-typical records, which to certain extent protects the user from misinterpretation of the data.

Full Text: PDF
Copyright © Polish Academy of Sciences & Institute of Fundamental Technological Research (IPPT PAN).



G.A. CARPENTER and M.N. GJAJA, Fuzzy ART choice functions, Proc. of the World Congress on Neural Networks (WCNN-94), Lawrence Erlbaum Associates, San Diego, Hillsdale, NJ, I, 713-722, 1994.

J. RACZ, Elements of artificial intelligence in navigation systems of autonomic mobile robots, Doctor Thesis, IPPT PAN, pp. 201, Warszawa 1995.

J. KASPERKIEWICZ, J. RACZ and A. DUBRAWSKI, HPC strength prediction using artifi­cial neural network, ASCE J. Computing in Civil Engng., 9, 4, pp. 279-284, Oct. 1995.

J. KASPERKIEWICZ, Artificial neural networks in concrete design and optimisation (in Polish], Proc. of XLII Conference of KILW PAN and KN PZITB: "Krynica 1996", V. 4, pp. 53-60, 1996.

J. N ORBERG, The rheology of fresh concrete - theory and measuring methods (in Swedish], Cement och Betong Institutet - CBI rapport 2, 94, pp. 72, Stockholm 1994.

A. JOHANSSON and J. NORBERG, The rheology of fresh concrete - tests on ready mixed concrete (in Swedish], Cement och Betong Institutet - CBI rapport 5, 94, pp. 64, Stock­holm 1994.

Ö. PETERSSON, P. BILLBERG, J. NORBERG and A. LARSSON, Effects of the second generation of superplasticizers on concrete properties, Cement och Betong Institutet - CBI rapport 2, 95, pp. 80, Stockholm 1995.

J. KASPERKIEWICZ and Z. PIASTA, Designing and optimisation using large sets of ex­perimental data (in Polish), (in:) Metody optymalizacji materiałów kompozytowych o matrycach cementowych, A.M. BRANDT (Ed.), KILW PAN - Studia z Zakresu Inż., Nr 38, pp. 240-265, Warszawa 1994.