Seminar Report

On

“APPLICATIONS OF ARTIFICIAL NEURAL

NETWORKS IN CIVIL ENGINEERING”

Submitted on partial fulfilment of requirement for degree of

BACHELOR OF CIVIL ENGINEERING

2012-2013

Presented By:-

Zode Pramey Moreshwar

T80430056

T.E. (Civil Engineering)

Under the Guidance of

Prof. R.R.Sorate.

Civil Engineering Department

Sinhgad Academy of Engineering, PUNE

Sinhgad Technical Education Society‟s

Sinhgad Academy of Engineering.

CERTIFICATE

This to certify that the seminar report on the topic “Applications of Artificial

Neural Networks in Civil Engineering” submitted by Zode Pramey Moreshwar. Roll

No.T80430056 is record of bonafide work carried out by him under my supervision and

guidance satisfactorily in the Department of Civil Engineering as prescribed by university of

Pune, during the academic year 2012-2013.

Prof. R.R.Sorate External Examiner

Seminar Guide

Dr. S. R. Parekar (HOD) Prof. Sorate (Seminar Incharge)

Department of Civil Engineering Department of Civil Engineering

Dr. A. G. Kharat(Principal) SAE, Pune

Acknowledgment

I hereby take this opportunity to express my profound thanks and deep sense of gratitude

towards my guide Prof. R. R. Sorate, Professor, Department of Civil Engineering, SAE. He

gave me a precious time from his busy schedule & his valuable guidance has been a constant

encouragement. I would also like to thank my friend Ankush Kawalkar for his continuous

help.

Chapter 1

INTRODUCTION

1.1 ARTIFICIAL NEURAL NETWORKS

Many tasks involving intelligence or pattern recognition are performed very easily by animals

but are extremely difficult to automate. For instance, animals recognize various objects and

make sense out of the large amount of visual information in their surroundings, requiring

very little effort. This task is performed by animal's neural network. The neural network of an

animal is part of its nervous system, containing a large number of interconnected neurons

(nerve cells). And artificial neural networks refer to computing systems whose central theme

is borrowed from the analogy of biological neural networks.

An artificial brain-like network based on certain mathematical algorithms developed using a

numerical computing environment like MATLAB is called as an „Artificial Neural Network

(ANN)‟. ANN system is modelled on human brain. It tries to obtain a performance similar to

that of human performance while solving problems. It is made up of a large number of simple

and highly interconnected processing elements which process information by its dynamic

state response to external inputs. Computational elements in ANN are non-linear and so the

results coming out through non-linearity can be more accurate than other methods. ANN is

configured for specific applications (such as data classification or pattern recognition)

through a learning process. Learning involves adjustment of synaptic connections that exist

between neurons. ANN can be simulated within specialized hardware or sophisticated

software. ANNs are implemented as software packages in computer or being used to

incorporate Artificial Intelligence in control systems.

Chapter 2

LITERATURE REVIEW

2.1 INTRODUCTION

Most of the water resources problems are of the nature of prediction and estimation of

rainfall, runoff, contaminant concentration, water stages etc. Solving these problems with

conventional techniques is computationally expensive and far from real life situation. An

important feature of ANN is their ability to extract the relation between inputs and outputs of

a process, without the physics being directly provided. They are able to provide a mapping

from one multivariate space to another; given a set of data representing the mapping, even

when the data is noisy. This shows that ANN may be well suited for the problems of

estimation and prediction. A review of the success of ANN in Surface Water Hydrology is

taken in this chapter.

2.2 SURFACE WATER RESOURCES

2.2.1 Rainfall Forecasting

Conversion of the remote sensed signal into rainfall rates, and hence into runoff for a given

river basin, is done by Smith and Eli (1995). This was achieved by using multi layered feed

forward network with Back Propagation training algorithm.

Tokar and Johnson (1999) applied ANN technique to rainfall-runoff modelling. Daily runoff

was forecasted by giving input of daily precipitation, temperature and snowmelt multi layered

feed forward network with Back propagation algorithm was applied.

Burian.et.al. (2000) tried to disaggregate hourly rainfall values into sub hourly time

increments with the help of ANN. Feed forward back propagation steepest descent learning

algorithm was used with unipolar activation function.

Jain and Indurthy applied ANN technique for rainfall runoff modeling. Two networks were

developed. First one consisted of single hidden layer and second one consisted of multiple

hidden layers. A back propagation algorithm was applied as a training algorithm.it was

concluded that ANN out performs all its conventional counterparts.

2.2.2 Runoff Prediction

Elshorebagy.et.al. (2000) compared technique of ANN with linear and non-linear regression

techniques for spring runoff prediction. In most cases ANN models showed superiority.

However, in some situations the performance of other techniques was better.

Thirumaliah and Deo (2000) applied ANN technique to real time forecasting of hourly flood

runoff and daily river stage and to the prediction of rainfall sufficiency for India. ANN was

found to be superior to the linear multiple regression model. Runoff Forecasting was tried

with back propagation, conjugate gradient and cascade correlation training algorithm.

ANN model was developed to predict both runoff and sediment yield on a daily as well as

weekly basis from simple information of rainfall and temperature. (Raghuwanshi et. al.

2006). It was shown that ANN model performs better than the linear regression based

models. Also ANN model with double hidden layer were observed to be better than single

hidden layer. Also performance of model increased with no. of hidden neurons and input

variables. Prediction was found to be accurate.

2.2.3 Reservoir operation

An auto regressive integrated moving average time series model and an ANN based model

were fitted to the monthly inflow data series and their performances were compared (Jain

et.al1999). ANN was found better prediction high flows whereas for low flows other model

was found suitable. Also ANN was found powerful tool for input output mapping and can be

used for reservoir inflow forecasting and operation.

Neelakanthan and Pundarikanthan (2000) found ANN based simulation optimization

performs satisfactorily as compared to the conventional stimulation model for reservoir

operation.

For multivariate reservoir forecasting, Coulibaly and Anctil (2001) developed and compare

the three different types of neural network architectures, an input delayed neural network

(IDNN) and a recurrent neural network (RNN) and multi layered perceptron (MLP). It was

found RNN was the best suitable architecture.

Chandramauli and Raman (2001) developed a dynamic programming based neural network

model for optimal reservoir operation. It was compared with the regression based approach

and single reservoir-dynamic programming neural network model.

2.2.4 Streams flow prediction

Flows in streams are main input for design of any hydraulic structure or environmental

impact assessment.

Karunanithi.et.al. (1994) used cascade correlation algorithm instead of designing and trying

different types of architectures and choosing the best performing architecture for predicting

daily stream flow data. This type of algorithm uses incremental architecture in which training

starts with minimal network and goes on increasing size as proceeds.

Thirumalaiah and Deo (1998) used ANN approach to forecast level of water in river. They

compared different types of algorithms like hack propagation, conjugate gradient and cascade

correlation. Cascade correlation algorithm was found to be the fastest for the training of the

network.

Jain and Chalisgaonkar (2000) used ANN to establish rating curves. Three layered back

propagation training algorithm was used. ANN results were found significantly better than

conventional curve fitting techniques. Liong et.al. (2000) also used ANN for forecasting river

stage; in addition sensitivity analysis was also done to investigate importance of each of the

neurons. Some neurons were found less effective in accuracy of prediction and removal of

these did not affect the output much.

2.2.5 Estimation of evapotranspiration

Kumar et.al. (2002) investigated utility of ANN for estimation of daily grass reference crop

evapotranspiration and compared the performance with conventional method. Multi layered

feed forward network with back propagation was used. Results showed that single hidden

layer was sufficient to account for non-linear relationship between climatic variables and

corresponding evapotranspiration.

2.2.6 Draught Analysis

A draught is generally defined as an extreme deficiency of water available in the hydrologic

cycle over an extended period of time. Draught forecasting plays a vital role in the control

and management of water resources systems.

ANNs are used to forecast draught (Kim, 2003). The results indicate that the conjunctive

models significantly improve the ability of ANN to forecast the indexed regional draught.

Three layered feed forward network with back propagation training algorithm is used.

Accurately predicted draughts allow water resources decision makers to prepare efficient

management plans and proactive migration programs that can reduce draught related social,

environmental and economic impact significantly

2.2.7 Soil water storage

Jain et.al(2004) applied knowledge of ANN to analyse the soil water retention data. A three

layered feed forward network was used in input layer, one neuron represented the matric

potential values and the only output neuron represented corresponding moisture content.

Hidden layer had 5 neurons and sigmoid transfer function was used. Back propagation

training algorithm was used.

2.2.8 Flood Routing

Abede and Price (2004) tried application of information theory and neural networks for

managing uncertainty in flood routing. The approach is based on the application of parallel

ANN model that uses state variables, input and output data and previous model errors at

specific time steps to predict the errors of a physically based model. It was concluded that

ANN models not only remove the errors of physical based models hut also reduces the

prediction uncertainty.

2.2.9 Model Drainage Pattern

Kao (1996) used ANN to automatically determine the drainage pattern from digital elevation

model (DEM). Three-layered network with back propagation algorithm is used for training

2.2.10 Classification of River Basins

Thandaveswara and Sajikumar used pattern clustering and pattern mapping capabilities of

ANN for classifying river basins. An unsupervised ANN architecture viz. Adaptive

Resonance Theory (ART) is used for pattern clustering that is grouping of basins of

hydrological homogeneity. Multi layered perceptron is used for pattern mapping.

Chapter 3

STRUCTURE OF ANN

3.1 BIOLOGICAL NEURON

The most basic element of the human brain is a specific type of cell, which provides us with

the abilities to remember, think, and apply previous experiences to our every action. These

cells are known as neurons, each of these neurons can connect with up to 200000 other

neurons.

All natural neurons have four basic components, which are dendrites, soma, axon and

synapses. Basically, a biological neuron receives inputs from other sources, combines them in

some way, performs a generally non-linear operation on the result, and then output the final

result. The fig. 3.1 below shows a simplified biological neuron and the relationship of its four

components.

Fig. 3.1 Structure of Biological Neuron

3.2 ARTIFICIAL NEURON

The basic unit of neural networks, the artificial neurons, simulates all the four basic

components of natural neurons. Artificial neurons are much simpler than the biological

neurons. The figure below shows the basic structure of an artificial neuron.

Fig. 3.2 Structure of Artificial Neuron

Various inputs to the network are represented by the mathematical symbol, xn. Each of these

inputs are multiplied by its weight, these weights are represented by wn. In the simplest case,

these products are simply summed, fed through a transfer function to generate a result, and

then output.

3.3 NEURAL NETWORKS

Artificial neural networks emerged from the studies of how brain performs. The human brain

consists of many millions of individual processing elements called neurons that are highly

interconnected.

ANNs are made up of simplified individual models of the biological neurons that are

connected together to form a network. Information is stored in the network in the form of

weights or different connection strengths associated with the synapses in the artificial neuron

models.

Many different types of neural networks are available and multi-layered neural network are

the most popular which are extremely successful in pattern reorganization problems. Each

neuron input is weighted by wi. Changing the weights of an element will alter the behaviour

of the whole network. The output y is obtained summing the weighted inputs and passing the

result through a non-linear activation function. Fig 3.3 shows a typical artificial neural

network.

Fig.3.3 An artificial neural network

Artificial neural networks are also referred to as "neural nets," "artificial neural systems,"

"parallel distributed processing systems," and "connectionist systems."

Chapter 4

DESIGNING OF ANN

4.1 PROCEDURE FOR ANN SYSTEM DESIGN

In realistic application the design of ANNs is complex, usually an iterative and interactive

task. The developer must go through a period of trial and error in the design decisions before

coming up with a satisfactory design. The design issues in neural network are complex and

are the major concerns of system developers.

Designing of a neural network consists of:

Arranging neurons in various layers.

Deciding the type of connection among neurons of different layers, as well as among

the neurons within a layer.

Deciding the way neurons receive input and produces output.

Determining the strength of connection that exists within the network by allowing the

neurons learn the appropriate values of connection weights by using a training data

set.

The process of designing a neural network is an iterative process.

The fig. 4.1 describes its basic steps.

Fig. 4.1 Steps for ANN System Design

As the figure above shows, the neurons are grouped into layers. The input layer consists of

neurons that receive input from external environment. The output layer consists of neurons

that communicate the output of the system to the user or external environment. There are

usually a number of hidden layers between these two layers. The figure above shows a simple

structure with only one hidden layer.

When the input layer receives the input, its neurons produce output, which become input to

the other layers of the system. The process continues until certain condition is satisfied or

until the output layer is invoked and fire their output to the external environment.

Chapter 5

FEATURES OF ANN

ANNs have several attractive features:

Their ability to represent non-linear relations makes them well suited for non-linear

modelling in control systems.

Adaptation and learning in uncertain system through off line and on line weight

adaptation.

Parallel processing architecture allows fast processing for large-scale dynamic system.

Neural network can handle large number of inputs and can have many outputs.

ANNs can store knowledge in a distributed fashion and consequently have a high

fault tolerance.

Chapter 6

LEARNING TECHNIQUES

An ANN can been seen as a union of simple processing units, based on neurons that are

linked to each other through connections similar to synapses. These connections contain the

“knowledge” of the network and the pattern of connectivity express the objects represented in

the network. The knowledge of the network is acquired through a learning process where the

connections between processing elements is varied through weight changes.

Learning rules are algorithms for slowly alerting the connection weights to achieve a desired

goal such as minimization of an error function. Learning algorithms used to train ANNs can

be supervised or unsupervised. In supervised learning algorithms, input/output pairs are

furnished and the connection weights are adjusted with respect to the error between the

desired and obtained output. In unsupervised learning algorithms, the ANN will map an input

set in a state space by automatically changing its weight connections. Supervised learning

algorithms are commonly used in engineering processes because they can guarantee the

output.

In this power system restoration scheme, a multi-layered perceptron (MLP) was used and

trained with a supervised learning algorithm called back-propagation. A MLP consists of

several layers of processing units that compute a nonlinear function of the internal product of

the weighted input patterns. These types of network can deal with nonlinear relations between

the variables; however, the existence of more than one layer makes the weight adjustment

process for problem solution difficult.

Learning rules are algorithm for slowly alerting the connections weighs to achieve a desirable

goal such a minimization of an error function. The generalized steps for any neural network

learning algorithm are as follows. These are the commonly used learning algorithm for neural

networks.

Multi-layer neural net (MLNN)

Error back propagation (EBB)

Radial basis functions (RBF)

Reinforcement learning

Temporal deference learning

Adaptive resonance theory (ART)

Genetic algorithm

6.1 MLNN IN SYSTEM IDENTIFICATION

There has been an “explosion” of neural network application in the areas of process control

engineering in the last few years. Since it become very difficult to obtain the model of

complex non-linear system due its unknown dynamics and a noise environment.it necessitates

the requirement for a non-classic technique which has the ability to model the physical

process accurately. Since nonlinear governing relationships can be handled very contendly by

neutral network, these networks offer a cost effective solution to modelling of time varying

chemical process.

Using ANN carries out the modelling of the process by using ANN by any one of the

following two ways:

Forward modelling

Direct inverse modelling

6.1.1 FORWARD MODELING

The basic configuration used for non-linear system modelling and identification using neural

network. The number of input nodes specifies the dimensions of the network input. In system

identification context, the assignment of network input and output to network input vector.

6.1.2 DIRECT INVERSE MODELING:

This approach employs a generalized model suggested by Psaltis et al to learn the inverse

dynamic model of the plant as a feed forward controller. Here, during the training stage, the

control inputs are chosen randomly within there working range. And the corresponding plant

output values are stored, as a training of the controller cannot guarantee the inclusion of all

possible situations that may occur in future. Thus, the developed model has taken of

robustness.

The design of the identification experiment used to guarantee data for training the neural

network models is crucial, particularly, in-linear problem. The training data must contain

process input-output information over the entire operating range. In such experiment, the

types of manipulated variables used are very important.

The traditional pseudo binary sequence (PRBS) is inadequate because the training data set

contains most of its steady state information at only two levels, allowing only to fit linear

model in over to overcome the problem with binary signal and to provide data points

throughout the range of manipulated variables. The PRBS must be a multilevel sequence.

This kind of modelling of the process play a vital role in ANN based direct inverse control

configuration.

6.2 ERROR BACK PROPOGATION ALGORITHM

This method has proven highly successful in training of multi-layered neural networks. The

network is not just given reinforcement for how it is doing on a task. Information about errors

is also filtered back through the system and is used to adjust the connections between the

layers, thus improving performance of the network results. Back-propagation algorithm is a

form of supervised learning algorithm.

Chapter 7

CASE STUDY

7.1 TIDAL LEVEL FORCASTING

Tidal level record is an important factor in determining constructions or activity in maritime

areas. Tsai and Lee applied the back propagation neural network to forecast the tidal level

using the historical observations of water levels. However, their model is used only for the

instant forecasting of tidal levels, not a long-term prediction. To demonstrate the ANN

model, D.S. Jeng, D. H. Cha and M. Blumenstein (2003) used different data based in the

training procedure to predict the one-year tidal level in Taichung Harbour. Based on the 15-

day collected data (1-15 Jan 2000), the one-year prediction of tidal level (Jan 2000- Dec,

2000) against the observation is illustrated in Fig. 7.1. In the figure, solid lines denote the

observation data, and dashed lines are the predicted values. The prediction of the present

model overall agree with the observation. The correlation coefficient over one year is 0.9182,

which is reasonable good.

Fig. 7.1 Comparison of observed tide levels with those predicted over one year for Taichung

Harbor (4/1996, 10/1996, 2/1997) (Jeng, Cha & Blumenstein, 2003)

7.2 EARTH RETAINING STRUCTURES

Goh et al. (1995) developed a neural network model to provide initial estimates of maximum

wall deflections for braced excavations in soft clay. The input parameters used in the model

were the excavation width, soil thickness/excavation width ratio, wall stiffness, height of

excavation, soil undrained shear strength, undrained soil modulus/shear strength ratio and soil

unit weight. The maximum wall deflection was the only output. The results produced high

coefficients of correlation for the training and testing data of 0.984 and 0.967, respectively.

Some additional testing data from actual case records were also used to confirm the

performance of the trained neural network model. The agreement of the neural network

predicted and measured wall deflection was encouraging, as shown in Table 4. The study

intended to use the neural network model as a time-saving and user friendly alternative to the

finite element method.

Table 7.1 Comparison of neural network predictions and field measurements (Goh, 1995)

7.3 PILE CAPACITY

Design of axial loaded pile can be done be solving equations of static equilibrium whereas

design of lateral loaded piles requires solution of nonlinear differential equations (Poulos &

Davis, 1980). Other semi-empirical methods used for lateral load capacity of piles are due to

Hansen (1961), Broms (1964) and Meyerhof (1976). Predicting pile capacity is a difficult

task because there are a large number of parameters affecting the capacity which have

complex relationships with each other. It is extremely difficult to develop appropriate

relationships between various essential parameters. Baik (2002) illustrated that these factors

include the soil condition (type of soil, density, shear strength, etc.), information related to

the piles‟ shape (diameter, penetration depth, whether the tip of pile is open-ended or closed-

ended, etc.), and other information (driving method, driving energy, set-up effect, etc.).

Although many methods in this regard have been presented, they did not appropriately

consider the various parameters that affect pile resistance. The main criticism of these

methods is that they oversimplify the complicated mechanism of pile resistance, and the soil

characteristics, type of pile, and information on driving conditions are not properly taken into

account. Hence, ANN models could be an alternate approach for the above case.

Park and Cho (2010) applied an artificial neural network (ANN) to predict the resistance of

driven piles in dynamic load tests. They collected 165 data sets for driven piles at various

construction sites in Korea. Predictions on the tip, shaft, and total pile resistance were made

for piles with available corresponding measurements of such values. The results indicate that

the ANN model serves as a reliable and simple predictive tool to appropriately consider

various essential parameters for predicting the resistance of driven piles. The proposed neural

network model has seven nodes in the input layer, eight nodes in the hidden layer, and three

nodes in the output layer (Fig.7.3). In order to find an appropriate combination of transfer

functions providing good correlation in training and testing stage, various combinations using

log-sigmoid, tan-sigmoid and linear was applied to hidden layer and output layer. The

combination of transfer functions applied to the hidden layer and output layer neurons are

tan-sigmoid (2 / (1 + e-2n

)-1) and linear, respectively.

Fig. 7.3 Architecture of the artificial neural network model (Park & Cho, 2010)

Fig. 7.4 Comparison of predicted and measured pile resistance (Park and Cho, 2010)

Chapter 8

CONCLUSIONS

It is evident from this review and case study that ANNs have been applied successfully to

many civil engineering areas like hydrology, pile capacity prediction, tide level prediction

and deflection of retaining walls, etc. Based on the results of case studies, it is evident that

ANNs perform better than, or as well as, the conventional methods. In many situations in

civil engineering, many problems are encountered that are very complex and not well

understood. Most of the mathematical models fail to simulate the complex behaviour of these

problems. In contrast, ANNs are based on the input-output data alone in which the model can

be trained. Moreover, ANNs can always be updated to obtain better results by presenting

new training examples as new data become available. Thus ANN have a number of

significant benefits that make them a powerful and practical tool for solving many problems

in the field of civil engineering and are expected to be applicable in future.

Chapter 9

FUTURE SCOPE OF STUDY

The energy from subsurface of earth at specific location that has ability to change the normal

functioning of human system is called as Geopathic Stress. Geopathic stresses affect human

body in a significant way. But many people being unaware of this fact fall prey to its adverse

effects. Geopathic stress could be detected using various techniques like changes in human

blood pressure, heart rate, body voltage and reaction time, light interference technique. This

field data being readily available through various experiments done by researchers, in future a

generalised ANN model could be created which would give the user an almost clear idea

whether the Geopathic Stress is present or not in his/her area of interest so that Geopathic

stress identification would become easy way for even a non-expert person.

REFERENCES

D.S. Jeng, D. H. Cha and M. Blumenstein (2003), "Application of Neural Network in

Civil Engineering Problems"

Mohamed A. Shahin, Mark B. Jaksa and Holger R. Maier (2001), "ARTIFICIAL

NEURAL NETWORK APPLICATIONS IN GEOTECHNICAL ENGINEERING,"

Australian Geomechanics – March 2001

Hyun Il Park, "Study for Application of Artificial Neural Networks in Geotechnical

Problems"

Mohamed A. Shahin, Mark B. Jaksa, Holger R. Maier, "State of the Art of Artificial

Neural Networks in Geotechnical Engineering," EJGE

R.R.Sorate, "Project report on Applications of Artificial Neural Networks in Civil

Engineering"