Computational & Technology Resources
an online resource for computational,
engineering & technology publications
Civil-Comp Proceedings
ISSN 1759-3433
CCP: 74
PROCEEDINGS OF THE SIXTH INTERNATIONAL CONFERENCE ON THE APPLICATION OF ARTIFICIAL INTELLIGENCE TO CIVIL AND STRUCTURAL ENGINEERING
Edited by: B.H.V. Topping and B. Kumar
Paper 27

A Neural Network to Estimate Concrete Strength Based on Non-Destructive Test Results

S.A. Mourad+, A.W. Sadek* and A.A. El-Fayoumy$

+Structural Engineering Department, Faculty of Engineering, Cairo University, Egypt
*Kuwait Institute For Scientific Research, Kuwait
$Channel Maintenance Research Institute, National Water Research Center, Egypt

Full Bibliographic Reference for this paper
S.A. Mourad, A.W. Sadek, A.A. El-Fayoumy, "A Neural Network to Estimate Concrete Strength Based on Non-Destructive Test Results", in B.H.V. Topping, B. Kumar, (Editors), "Proceedings of the Sixth International Conference on the Application of Artificial Intelligence to Civil and Structural Engineering", Civil-Comp Press, Stirlingshire, UK, Paper 27, 2001. doi:10.4203/ccp.74.27
Keywords: neural network, concrete strength, non-destructive testing, rebound number, pulse velocity.

Summary
Neural networks have demonstrated versatility as a problem-solving tool with various applications in civil engineering problems[1]. In this research, a neural network model was established to estimate the compressive strength of concrete by using the combination of two of the most widely used Non-Destructive Tests (NDT); rebound number and pulse velocity tests. This technique is utilized in an attempt to increase the reliability of the non-destructive tests in detecting the strength of concrete[2].

The study was conducted by constructing a back propagation neural network model. Different network topologies were tried by changing the number of hidden layers and nodes in each layer in order to select the best network configuration. The first attempt was using a neural network with one hidden layer, and the number of neurons was changed from one to ten neurons. However, all the networks failed to yield reasonable results. The second trial was performed using two hidden layers instead of only one hidden layer. Eighty networks with different configurations were tested to include all possible combinations of eight neurons in the first hidden layer and ten neurons in the second hidden layer at a high level of precision (target error set to 0.01). The same number of networks was tested at a lower level of precision (target error set to 0.05). The most suitable network configuration was judged based on the root mean square error for both training and testing.

Three networks were established. For network A, the input parameters for each network were the rebound number and the pulse velocity whereas the output parameter specified the corresponding compressive strength. The network was trained using a set of readings for pulse velocity, rebound hammer readings and cube strength[3]. The input and output parameters for training were taken from different mixes and at different ages to make sure that the applicability of the chosen neural network is not limited to a specific mix or age of concrete.

The best neural network configuration was determined for both target error values of 0.01 and 0.05. It was judged that the two hidden layers network with three neurons in the first hidden layer and one neuron in the second hidden layer yielded the best results. The same neural network with the same configuration was trained and tested using the rebound number only (network B) and the pulse velocity only (network C) as a single input parameter for the network and the strength as the output parameter to check the network validity in the case of using the non- destructive test individually. However, networks B and C were less successful in predicting the concrete strength[4].

The networks were used to determine concrete strength for two case studies in order to validate the results. The first field study was performed on a set of concrete slab specimens, and the second on a set of concrete columns. Applying neural network A on the data from the first study resulted in differences between actual cube strength and network results from 2% to 22%, which is considered acceptable. In general, network A provided output that was in good agreement with cube strength values recorded earlier.

References
1
Flood, I., and Kartam, N., "Neural networks in Civil Engineering I: Principles and Understanding", Journal of Computing in Civil Engineering, Vol. 8, No. 2, pp. 131-148, April, 1994. doi:10.1061/(ASCE)0887-3801(1994)8:2(131)
2
Tanigawa, Y., Baba, K. and Mori, H., "Estimation of concrete strength by combined non-destructive testing method", ACI special publication 82, 1984.
3
Higazi B., "Estimation of quality and integrity of concrete members by non- destructive testing", A Thesis submitted in Partial Fulfilment of the Requirements of M. Sc. Degree, Cairo University, 1989.
4
El-Fayoumy, A.A., "A neural network to detect concrete strength based on combination of rebound hammer and ultrasonic pulse velocity tests", A Thesis submitted in Partial Fulfilment of the Requirements of M. Sc. Degree, Cairo University, 2000.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description