You are here

A method to predict how different soil types respond to earthquakes

Earthquakes can cause severe devastation, leading to loss of human life and infrastructural damage. Unfortunately, it is almost impossible to predict exactly when and where an earthquake will occur. A researcher from IISc has developed an improved technique which can predict whether a given type of soil will give in to tremors during an earthquake or not. This could in turn help in determining whether it is safe to construct buildings, roads, dams, bridges on such soil or not.

Many times, it is buildings that kill people, rather than the earthquake itself.

A lot of the damage to lives and property can be prevented if care is taken before the construction of the buildings. Surveying properties of the soil on which buildings are going to be constructed is a crucial first step. One of the methods used for such an investigation is the cone penetration test (CPT), in which a long cylindrical tool with a conical cap is pushed inside the ground at slow speed just like the drilling of a submersible pump. The parameters thus obtained are used to study the soil behaviour under different conditions like vibration stress, which occurs during earthquake.

What really happens to the soil during earthquake is -- the soil liquefies. This may sound unbelievable, but because of the shaking, the soil substantially loses its strength and stiffness, and starts to behave like a liquid -- a process known as soil liquefaction. Incidentally, the same phenomenon is responsible for quicksand. Any structure built over it starts to float as if on a water body; because of lack of a solid support, the construction breaks apart.

Pijush Samui from the Department of Civil Engineering, IISc, has developed a method called Relevance Vector Machine (RVM) which can predict whether a given soil will liquefy during earthquake or not based on the CPT data. This method is better than other methods of predictions as it is more accurate, takes lesser computational time and needs lesser of components for predictive analysis. Moreover, it can always be updated to yield even better results as new data becomes available.

Researchers have used various mathematical methods known as algorithms or models which can be first made to “learn” and then latter make predictions about certain things, like artificial neural networks (ANN). ANNs are based on the concept of interconnection of neurons found in our brain, and are used for problem solving and prediction. However, such methods require heavy computational power and time, and cannot be generalized easily. In contrast, Samui’s method—Relevance Vector Machine (RVM) overcomes these problems.

The RVM method has been earlier utilised in computer based tasks such as converting 2-D image into 3-D, optical detection of cancer etc. Samui checked if he could apply the same method for predicting soil liquefaction. In the present work, he used a database which contains the data for different soils, categorised under whether they underwent liquefaction during earthquakes or not. Associated with each soil type, there are a total of seven parameters such as the grain size of soil, magnitude of earthquake etc. Using this database as the backbone for his study Samui developed the RVM model which used 68% of the data for “learning” and the rest for prediction. He found that a particular model shows good performance achieving as high as 100% accuracy with as small as 4 number of components. When he compared his new method with that of the existing method he found that his RVM outperforms the ANN method which uses many more parameters in data analysis as compared to only one in RVM.

Samui says, “Using this method we can find whether a given soil will liquefy or not during the earthquake. If it does then we can amend the soil properties as well as use other methods so that any construction over such a soil remains stable during earthquake”. This can greatly benefit the geotechnical engineers who are involved in looking at the suitability of sites for construction of buildings, roads, dams and bridges.

About the author:

Prof. Pijush Samui did his M.Sc (Engg.) and PhD from the Department of Civil Engineering, Indian Institute of Science, Bangalore. He is currently a Professor and the Director (In-charge) at Centre for Disaster Mitigation, VIT University, Vellore, India.


The paper has appeared in the journal Earthquake Engineering and Engineering Vibration, and is available on the following link: