Keywords: neural network, genetic algorithm, clamping, pruning, expansive soils, light structures.

Expansive soils have been identified as the most common attribute to ground movement around the world including Australia. Numerous light structures founded on expansive soils in Victoria, Australia suffer from ground movement due to edge heave or under flooring drying settlement in the clay beneath the structures. This is caused by swelling and shrinking of the expansive soils surrounding the property. The structure that is most susceptible to swelling and shrinkage is the foundation. The presence of expansive soils alone doesn't necessarily is the main cause of the problem. The presence of other factors such as vegetation, pipe leakages, climate factors, types of construction materials, geology type and workmanship are also the contributing factors of the damage. Three quarters of damage to light structures on expansive soils in Victoria resulted from local drying settlement caused by trees and shrubs planted too close to the structures.

Although many assumptions and guessing work have been made in regards to damage to light structures on expansive soils in Victoria, little has been done to actually distinguish the important factors that cause the damage. It is from here, that the aim of this work is to investigate where the problems of light structures on expansive soils originate and the correlation of the important factors if any that may have a vast impact on the damage. This is done by developing a model of predictive damage condition of light structure on expansive soils in Victoria Australia. It is hoped that the model can assist the government or the building trade authorities to finally recognize and identify the parameters that are most affecting damage to light structures on expansive soils. In addition to that, the model could help them predict which class the damage condition of the light structure is with only a click of a button. That way any serious and urgent repairs can be identified and action can be taken without delays.

The paper covers the development and analysis of the proposed model using different techniques. The aim of this work is to rank the importance of individual factors for damage to light structures. Artificial Intelligence techniques were used for the development of the proposed model. The neural network and genetic algorithm toolboxes from MATLABRversion 7.1 were used for the Artificial Intelligence method. Clamping and Pruning techniques were adopted to analyse the proposed model. This paper analyses and discusses the outcomes of the methods.

The main challenge for any inspector is to investigate technically which one of these is predominant in any particular case. Hence in this work, the challenge is answered by developing a predictive damage condition model. This work has helped identify the parameters that influence damage to light structures on expansive soils using the proposed model. The relative importance of inputs was determined using clamping and pruning methods. The average results from the weights of both the methods were used as the final result. The results for the individual parameters showed that ChgTMI was the most important parameter that influences the damage to light structures on expansive soils. This was followed by three other important parameters which were construction footing, construction wall and geology respectively. These parameters were said to be susceptible to damage to light structure and hence it is not surprising to see that they are the four top ranked of the input parameters which are the most influential to the damage of light structure.

It was also found that artificial intelligence techniques such as a neural network trained with a genetic algorithm which was used in the development of the proposed model performed better than the more conventional method such as statistical methods.

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
purchase this book (price £62 +P&P)