Computational & Technology Resources
an online resource for computational,
engineering & technology publications
PROCEEDINGS OF THE SIXTH INTERNATIONAL CONFERENCE ON ENGINEERING COMPUTATIONAL TECHNOLOGY
Edited by: M. Papadrakakis and B.H.V. Topping
Utilisation of Computational Intelligence Techniques for Stabilised Soil
A.H. Alavi1, A.A. Heshmati1, A.H. Gandomi2, A. Askarinejad3 and M. Mirjalili4
1College of Civil Engineering, Iran University of Science and Technology, Tehran, Iran
A.H. Alavi, A.A. Heshmati, A.H. Gandomi, A. Askarinejad, M. Mirjalili, "Utilisation of Computational Intelligence Techniques for Stabilised Soil", in M. Papadrakakis, B.H.V. Topping, (Editors), "Proceedings of the Sixth International Conference on Engineering Computational Technology", Civil-Comp Press, Stirlingshire, UK, Paper 175, 2008. doi:10.4203/ccp.89.175
Keywords: stabilised soil, multilayer perceptron, linear genetic programming, textural properties of soil, cement, lime, asphalt, unconfined compressive strength.
In the present study, two branches of computational intelligence techniques namely, the multilayer perceptron (MLP) and linear genetic programming (LGP), are employed to simulate the complex behavior of the strength improvement in a chemical stabilisation process. Due to a need to avoid extensive and cumbersome experimental stabilisation tests on soils on every new occasion, it was decided to develop mathematical models to be able to estimate the unconfined compressive strength (UCS) as a quality of the stabilised soil after both compaction and curing by using particle size distribution, liquid limit, plasticity index, linear shrinkage as the properties of natural soil before compaction and stabilisation and the quantities and types of stabiliser. A comprehensive and reliable set of data including 219 previously published UCS test results were utilised to develop the prediction models.
Based on the values of performance measures for the models, it was observed that all models are able to predict the UCS value to an acceptable degree of accuracy. The results demonstrated that the optimum MLP model with one hidden layer and thirty six neurons outperforms both the best single and the best team program that have been created by LGP. It can also be concluded that the best team program evolved by LGP has a better performance than the best single evolved program. This investigation revealed that, on average, LGP is able to reach a prediction performance similar to the MLP model. Moreover, LGP as a white-box model provides the programs of an imperative language or machine language that can be inspected and evaluated to provide a better understanding of the underlying relationship between the different interrelated input and output data.
purchase the full-text of this paper (price £20)