Determination of compressive strength of concrete using Self Organization Feature Map (SOFM)

Abstract

Plurality of parameters that influenced the compressive strength of concrete and the nonlinear relationship of parameters and concrete properties caused researchers to turn to the Self Organization Feature Map systems. In a SOFM network, the processing units are located at the nodes of a one-dimensional, two-dimensional, or more dimensional network. In a competitive learning process, units are regulated based on input patterns. The location of the units in the network is regulated so that a significant coordinates system develops on the network for input features. Competitive learning which is employed in such networks is called unsupervised learning. In the present study, 173 concrete samples with different characteristics have been used. Networks used in the current research are Self Organization Feature Map networks of constant weight, including Kohonen network. Slump parameters, water/cement ratio, maximum size of gravel, sand and cement content were considered as the input values, and concrete compressive strength was calculated using this model. The structures of all SOFM systems were optimized using genetic algorithms. To verify the accuracy of the model, it was compared with statistical models and artificial neural networks. The results showed that the Self Organization Feature Map systems, which were optimized using genetic algorithm, had more accuracy than other models in predicting the compressive strength of concrete.