Está en la página 1de 9

Application of Artificial Intelligence in Pile foundation


In most of the engineering streams, empirical relationships are often utilized to estimate design
parameters and engineering properties. Generally, this process involves a number of interacting
factors in which relationship between these factors is not precisely known. The procedure utilized to
calculate the output value follows a tedious process and demands human attentions along with the
intuitions. These all errors can be eliminated by utilizing artificial intelligence techniques.

There are many techniques utilized in artificial intelligence as Artificial Neural Networks (ANNs),
Genetic Programming (GP), Evolutionary Polynomial Regression (EPR), Support Vector Machines
(SVM), M5 model trees, and k-nearest neighbors. Here, the path followed to arrive at the result is
back-propagation neural networks. This is the product of artificial intelligence. In back-propagation
neural networks, the mathematical relationships between various parameters are not specified.
Instead, they learn from the data fed to them. In addition, they can generalize correct responses that
only broadly resemble the data in the learning phase.

Training of the neural network is essentially carried out through the presentation of a series of
example patterns of associated input and target (expected) output values. Each hidden and output
neuron processes its inputs by multiplying each input by its weight, summing the product and then
passing the sum through a nonlinear transfer function to produce a result. The S-shaped sigmoid curve
is commonly used as the transfer function. The neural network learns by modifying the weights of the
neurons in response to the errors between the actual output values and the target output values. This
is carried out through the gradient descent on the sum of squares of the errors for all the training
patterns. [1] Training is carried out by repeatedly presenting the entire set of training patterns (with
the weights updated at the end of each cycle) until the average sum squared error over all the training
patterns are minimized and within the tolerance.

At the end of the training phase, the neural network should correctly reproduce the target output
values for the training data provided the errors are minimal, i.e. convergence occurs. The associated
trained weights of the neurons are then stored in the neural network memory. In the next phase, the
trained neural network is fed a separate set of data. In this testing phase, the neural network
predictions (using the trained weights) are compared with the target output values. This assesses the
reliability of the neural network to generalize correct responses for the testing patterns that only
broadly resemble the data in the training set. No additional learning or weight adjustments occur
during this phase. Once the training and testing phases are found to be successful, the neural network
can then be put to use in practical applications.

Application of Artificial Intelligence in Pile foundation

There are many paths such as C program, python, Java, R-language and MATLAB through which back-
propagation neural network can be triggered. For the simplicity and less errors in MATLAB which
utilizes nntool to estimate the output; ultimate bearing capacity of the pile foundation is calculated
using MATLAB.

Pile driving formulae are commonly used to estimate the load capacity of driven piles. These
formulae are essentially derived from impulse-momentum principles. The formulae assume
that there is a correlation between the driving resistance and the ultimate load capacity of
the pile Qu. The important factors influencing the load capacity include the hammer
characteristics, the properties of the pile and soil.

Several pile driving formulae are widely used in actual practice. These include the Engineering
News (EN) [2] formula, the Hiley formula [3] and the Janbu formula [4]. These are represented
in Table 1

Table 1 Pile driving formulae

Formula Equation for 𝐐𝐮 Remarks

Engineering News WH c= 25mm (gravity hammer)
c= 2.5mm (steam hammer)
c= 2.5× (steam hammer

on very heavy piles)

Hiley ef WH W + n2 Wp ef , c1 , c2 , c3 and n are
s + 0.5(c1 + c2 + c3 ) W + Wp tabulated by Chellis [5]
Janbu WH λ 0.5
k u = Cd {1 + (1 + Ce ) }
kus d

λe =
Cd = 0.75 + 0.15

In table 1 W is the hammer weight, H is the hammer drop, L is the pile length, Wp is the pile
weight, A is the pile cross-sectional area and E is the pile modulus of elasticity.
In back-propagation neural networks these formulae are neglected only the data plays a
critical role. Thus, the data has been extracted from different experiments and case studies

Application of Artificial Intelligence in Pile foundation

to estimate the ultimate bearing capacity of the pile foundation. A brief summary of the
source from which the data has been extracted is presented in Table 2.
Table 2 List of data

Source Location NO. OF DATA SOIL TYPE

Muhs et al. Iraq 08 Cohesionless Soil
WeiB Iraq 18 Cohesionless Soil
Muhs andWeiB Iraq 23 Cohesionless Soil
Muhs andWeiB Iraq 15 Cohesionless Soil
Briaud and Gibbens Iraq 12 Cohesionless Soil
Gandhi Iraq 07 Cohesionless Soil
- Iraq 45 Cohesionless Soil
Total 128
Haraz-Iran 16 Cohesive Soil
Mahshahr-Iran 12 Cohesive Soil
Gulf of Mexico (Stockard) 09 Cohesive Soil
1. Goh AT. Empirical India (Stockard,1986) 15 Cohesive Soil
design in
geotechnics using
neural networks,
Geotech 1995;
Alton-Illinois (Larry, 1988) 18 Cohesive Soil
Bandar Imam-Iran 12 Cohesive Soil
2. Goh ATC. Pile Shiraz-Iran 16 Cohesive Soil
driving records
reanalyzed using
neural networks, J
Geotech Eng 1996;
122(6): 492-5.2
Ontario (Fellenius and
Altaee, 2002) Bandar 18 Cohesive Soil
Bandar Imam 10 Cohesive Soil
Mahshahr-Iran 06 Cohesive Soil
Isfahan-Iran 06 Cohesive Soil
Bandar Abbas 08 Cohesive Soil
Isfahan-Iran 18 Cohesive Soil
Total 164
(yet references have to be quoted)
Utilizing the data sets which are mentioned in table 2; training and testing has been done
using nntool in MATLAB for both the cohesive and cohesionless soil samples.
Application of Artificial Intelligence in Pile foundation

Figure 1 Neural network tool in MATLAB

Experiments were carried out using several combinations of input parameters to determine
the most reliable neural network model. Generally, the reliability of the model improved as
the number of input parameters increased. The neural network model with seven input
neurons representing E, A, Wp , H, W, s, and the hammer type (H type) and two hidden
neurons was found to be most reliable. The neural network model is represented in figure 2.

Figure 2 Neural network model

After creating the neural network in the training phase, the parameters have to be changed
to obtain the optimal results. At 1000 epochs and 1000 validation check the neural network
presented the optimal results.

Application of Artificial Intelligence in Pile foundation

Figure 3 Training phase

From complete set of data 60% of the data is utilized for training and 40% is utilized for training. The
regression obtained from the data for cohesive soil is shown in figure 4 and for cohesionless soil is
shown in figure 5.

Application of Artificial Intelligence in Pile foundation

Figure 4 Regression for cohesive soil

Application of Artificial Intelligence in Pile foundation

Figure 5 Regression for cohesionless soil

From above regression the convergence has been achieved. This can infer from figure 4 and 5. The
validation of the results also been checked by comparing the results obtained from the neural network
and experimental results. This is depicted in figure 6 for cohesive soils and figure 7 for cohesionless

Figure 6 Check for cohesive soil sample

Application of Artificial Intelligence in Pile foundation

Figure 7 Check for cohesionless soil

The results for both cohesion less and cohesive soil indicates that convergence was achieved
for the training phase. The scatter of the predicted Q, values versus the measured Q, values
were assessed using regression analysis. The results from the testing phase suggest that
although the model was not explicitly trained for these data, the neural network was capable
of generalization and generally gave reasonable predictions. The results indicate that the
neural network was successful in modelling the nonlinear relationship between Qu and the
other parameters.


Statistical methods are utilized in the development of standard relationships between various
factors. This is often complex and circuitous, particularly for nonlinear relationships. Also, to
formulate the statistical model, the important parameters must be known. By comparison,
the modelling process in back-propagation neural networks is more direct, as there is no
necessity to specify a mathematical relationship between the input and output variables.
Neural networks can be effective for analysing a system containing several variables, to
establish patterns and characteristics not previously known. In addition, it can generalize
correct responses that only broadly resemble the data in the training set. During training,
irrelevant input variables are assigned low connection weights. These variables can then be
omitted from the model. In neural networks, quantitative as well as qualitative information
can be considered. As new data become available, the neural network model can be readily
updated by retraining with patterns which include these new data.
Application of Artificial Intelligence in Pile foundation


1. Rumelhart, D. E., Hinton, G. E. & Williams, R. J. Learning internal representation by

error propagation. In Parallel Distributed Processing, ed. R. H. Rumelhart & J. L.
McClelland. MIT Press, Cambridge, MA, 1986.

2. Wellington, A. M. The iron wharf at Fort Monroe, VA. Trans. ASCE, 27 (1892) 129-37.

3. Hiley, A. The emciency of the hammer blow, and its effects with reference to piling.
Engineering, 2 June (1922) 673

4. Janbu, N. Une analyse energetique du battage des pieux a l'aide de parametres sans
dimension. Norwegian Geotechnical Institute, Oslo, 1953, pp. 63—4 (in Norwegian).

5. Chellis, R. D. Pile Foundations, 2nd edn. McGraw-Hill, New York, 1961