Implementation of Artificial Neural Networks to Predict Void Content
Void content is one of the governing factors to determine the properties of concrete. Basically determination of void content is a tough job. This paper deals with the usage of neural network to predict void content. It also eradicates the hectic work and consumes less amount of cement, which also saves time. The aim is to identify the suitable number of hidden neurons that minimizes the error with minimum void content. Void content of the aggregate mixture is determined by using four networks with different number of hidden neurons. These are optimized with levenberg-marquadt algorithm and then compared with each other. Different volume proportions of river sand, intermediate aggregate (maximum size of 10mm (I.A)) and coarse aggregate (maximum size of 20mm (C.A)) are used as inputs. Experimental void content is used as target. The network which has N3 number of neuron in hidden layer shows good results when compared with other networks.
Key words: Void content, neural networks, neurons, levenberg-marquadt alogorithm and hidden layers.