A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Discussion on questions about using artificial neural network for predicting of concrete property
Several questions about predicting concrete property using BP artificial neural network have been discussed, including the selection of network structure, the determination of sample capacity and grouping method, the protection from over-fitting, and the comparison on precision of prediction. For the network-structure, it has been found that directly apply the consumption of raw-material and other crucial quality indices as the units of input can bring about a satisfactory result of prediction, in which a single hide layer holds 10 units and the workability along with the strength and durability formed two sub-networks simultaneously. For the sample capacity and grouping method, at least 100 sets of samples are necessary to find the intrinsic regularity, among them 1/3-1/4 should be taken as test samples. A new tactics for error-tracking has been proposed which is verified effective to avoid the over-fitting. The comparison of effectiveness and feasibility between BP neural networks and linear regression algorithm showed that BP neural networks have better performance in accuracy of prediction. Finally, an applicable software has been developed and used as examples to predict 163 sets of mixes for a ready-mixed concrete plant, to show its application in detail.
Discussion on questions about using artificial neural network for predicting of concrete property
Several questions about predicting concrete property using BP artificial neural network have been discussed, including the selection of network structure, the determination of sample capacity and grouping method, the protection from over-fitting, and the comparison on precision of prediction. For the network-structure, it has been found that directly apply the consumption of raw-material and other crucial quality indices as the units of input can bring about a satisfactory result of prediction, in which a single hide layer holds 10 units and the workability along with the strength and durability formed two sub-networks simultaneously. For the sample capacity and grouping method, at least 100 sets of samples are necessary to find the intrinsic regularity, among them 1/3-1/4 should be taken as test samples. A new tactics for error-tracking has been proposed which is verified effective to avoid the over-fitting. The comparison of effectiveness and feasibility between BP neural networks and linear regression algorithm showed that BP neural networks have better performance in accuracy of prediction. Finally, an applicable software has been developed and used as examples to predict 163 sets of mixes for a ready-mixed concrete plant, to show its application in detail.
Discussion on questions about using artificial neural network for predicting of concrete property
Chen Bin, (author) / Mao Qian, (author) / Hu Zhaoyuan, (author) / Wu Wei, (author)
2011-04-01
832914 byte
Conference paper
Electronic Resource
English
Artificial neural network for predicting drying shrinkage of concrete
| Online Contents | 2013
Artificial neural network for predicting drying shrinkage of concrete
| British Library Online Contents | 2013
Predicting the Workability of Self-Compacting Concrete Using Artificial Neural Network
| British Library Conference Proceedings | 2011
Discussion: Questions in Reinforced Concrete Design
| ASCE | 2021
Predicting the compressive strength of self compacting concrete using artificial neural network
| British Library Conference Proceedings | 2009