Major project report

on

AN INVERSE DESIGN METHOD FOR BIO-MIMETIC PROPULSION

SYSTEM FOR AUTONOMOUS UNDERWATER VEHICLES USING

ARTIFICIAL NEURAL NETWORKS

Submitted by

MID KSHITIJ MANTRAVADI (3784/101)

MID KDIL SENEVIRATHNE (3739/101)

MID HIMANSHU MALIYAL (3766/101)

MID SARANSH NEGI (3783/101)

in partial fulfillment of the requirements for the award of the degree of

Bachelor of Technology

In

Mechanical Engineering

Faculty of Mechanical Engineering

INDIAN NAVAL ACADEMY
AUTUMN TERM – 21

1
 DECLARATION

We hereby declare that the project titled “AN INVERSE DESIGN METHOD FOR
BIO-MIMETIC PROPULSION SYSTEM FOR AUTONOMOUS UNDERWATER
VEHICLES USING ARTIFICIAL NEURAL NETWORKS” submitted for the Bachelors of
Technology in Mechanical Engineering degree, represents our ideas in our own words and
where others' ideas or words have been included, we have adequately cited and
referenced the original sources. We also declare that we have adhered to all principles of
academic honesty and integrity and have not misrepresented or fabricated or falsified any
idea/data/fact/source in my submission. We understand that any violation of the above will
be cause for penal action from the sources which have thus not been properly cited or
from whom proper permission has not been taken when needed.

MID KSHITIJ MANTRAVADI

MID KDIL SENEVIRATHNE

MID HIMANSHU MALIYAL

MID SARANSH NEGI

2
 CERTIFICATE

This is to certify that the major project report on “ AN INVERSE DESIGN METHOD
FOR BIO-MIMETIC PROPULSION SYSTEM FOR (AUTONOMOUS UNDERWATER
VEHICLE) AUVS USING ARTIFICIAL NEURAL NETWORKS” submitted by MID
KSHITIJ M (3784/101), MID KDIL SENEVIRATHNE (3739/101), MID HIMANSHU
MALIYAL (3766/101) and MID SARANSH NEGI (3783/101) is in partial fulfillment of the
requirement for the degree of Bachelors of Technology in Mechanical Engineering
conducted at Indian Naval Academy, Ezhimala during the academic year 2021 This is a
bonafide record of work carried out under my guidance and supervision.

Asst. Prof Vidhu Manohar

Instructor (ME)

Azmatulla Khan
Captain
HoF (ME)

Place : INA, EZHIMALA

Date :

3
 ACKNOWLEDGMENT

We would like to express our deepest appreciation to all those who provided us with
the opportunity to make this research project a success. We would like to express our
deepest gratitude to our project guide, Asst. Prof Vidhu Manohar, whose contribution in
stimulating creativity, suggesting improvements and encouraging us to explore the
uncharted depths of our research has helped us to coordinate our project and make it a
success.

Furthermore we would also like to acknowledge with much appreciation, the crucial
role of the staff of project bay (ME), who gave the permission to use all required equipment
and the necessary materials to complete design and analysis.

We have to appreciate the guidance given by other supervisors as well as the

panels especially in our project presentation that has improved our presentation skills
thanks to their comments and advice.

MID KSHITIJ M

MID KDIL SENEVIRATHNE

MID HIMANSHU MALIYAL

MID SARANSH NEGI

4
 ABSTRACT

The study for better performing Autonomous Underwater Vehicles (AUVs) have
increased dramatically in size and scope over the past two decades. The applications of
AUVs are numerous and varied; from deep sea exploration, to pipeline surveillance to
mine clearing to patrolling EEZ. However, one limiting factor with the current technology is
the duration of missions due to the limited power that can be carried onboard the AUV.
One of the major contributing factors to this is the efficiency of the propulsion system,
which is usually the marine propellers.

Numerous studies have been conducted into the bio-mimetic fish-like propulsion
systems and robotic fish over the past decade with many claims being made as to the
benefits of a fish like propulsion system over conventional marine propulsion system.
These claims include increased efficiency and greater maneuverability. However, very few
direct experimental comparisons have been made between bio-mimetic and conventional
propulsion systems. Also there is very less experimental data to learn about the deep sea
conditions in which the fin will be more effective over a marine propeller. CAE modeling is
an effective solution to this problem where we can simulate the deep sea conditions and
work upon the effectiveness of fins.

This work attempts firstly to create a database of thrust, moment and drag in
underwater conditions where fin designs modeled in ANSYS space claim is
simulated to replicate fish like movements. This database can be then fed as input
to a neural network built using MATLAB that after training will be able to predict
dimensions for a fin shape that might be suitable to work most efficiently under the
simulated conditions.

5
 TABLE OF CONTENTS

List of Abbreviations 08
List of Figures 09
List of Tables 10
1. Introduction 11
2. Review of Literature 12
3. Propulsion systems in AUVs 14
3.1 Conventional propulsion systems 14
4. Propulsion system of fishes 16
4.1 Locomotion of Fishes 16
4.2 Fins of Fishes 17
5. Fin Selection 19
5.1 BCF Propulsion 19
6. Methodology for CFD analysis 20
6.1 Selection Of Fin Shape 20
6.2 Motion modelling 20
6.3 Meshing and Enclosure 22
6.4 Strouhal no. Significance 27
6.5 Setup 27
6.6 Calculation and Result 29
6.7 Validation 30
7. Data conversion 31
7.1 Necessity for conversion 31
7.2 Selection of python application 32
7.3 Python code for conversion 33
8. Artificial Neural Networks 34
8.1 Introduction to ANN 34
8.2 Feed forward Back-propagation 35
8.3 Regression line analysis 36
8.4 Training the neural network 37
9. Results 38
9.1 Network 1 38

6
9.2 Network 2 39
9.3 Verdict 41
References 43

7
 LIST OF ABBREVIATIONS

AUV Autonomous Underwater Vehicle

ROV Remotely Operated Vehicles

CFD Computational Fluid Dynamics

DPIV Digital Image Particle Velocimetry

LEV Leading Edge Vortex

FSI Fluid Structure Interaction

FFF Fluid Flow Fluent

UDF User Define Function

ANN Artificial Neural Network

RLI Regression Line Analysis

EEZ Exclusive Economic Zones

ANN Artificial Neural Networks

8
 LIST OF FIGURES

fig.1 Types of Propulsion in AUV 14

fig.2 Fins of fishes 18
fig.3 Basic terminology for fins of fish 20
fig.4 Optimal caudal fin shape 21
fig.5 Selected geometry and dimensions 22
fig.6 User defined function 23
fig.7 Dimensions of enclosure of water 24
fig.8 Meshing details 25
fig.9 Enclosure as viewed from space claim software 26
fig.10 Hybrid mesh viewed post updating 26
fig.11 Enlarged view of fin meshing 27
fig.12 Reference values of water 28
fig.13 Output graph of drag vs flow-time 29
fig.14 Output graph of scaled residual vs iterations 29
fig.15 Lift, drag and moment as output files from ANSYS 30
fig.16 Graph of strouhal no. Vs thrust for 60mm caudal peduncle 31
fig 17 graph of strouhal no. Vs input power 31
fig.18 Illustration of neuron and ANN 34
fig.19 Illustration of back-propagation 36
fig.20 Creation of neural network using NNtool 38
fig.21 Structure of network 1 39
fig.22 Output for network 1 39
fig.23 Structure of network 2 41
fig.24 Regression line analysis for network 2 41

LIST OF TABLES

9
Table.1 Types of fins and uses
Table.2 Velocities for various strouhal no. For 1hz
Table.3 Velocities for various strouhal no. For 2hz
Table.4 Raw and normalized data sets for carious lengths at 1 hz
Table.5(a) Transposed raw data from ANSYS for 75 mm fin
Table.5(b) Transposed normalized data from ANSYS for 75mm fin
Table.6 Comparison of output from both networks

10
 CHAPTER 1

1.1 INTRODUCTION

1.1.1 Remotely operated AUVs have been envisioned to augment and replace
manned platforms and fixed systems in a wide variety of scientific, commercial, and
military applications such as collecting oceanographic data, inspecting undersea
infrastructure, monitoring fish farms, and sweeping mines. In many cases, these
vehicles need to be versatile, capable of performing multiple tasks with different
locomotion gaits. For example, an ocean sampling robot should be able to swim
efficiently to achieve long endurance and hold position under strong ocean waves. A
pipeline inspection vehicle needs to be able to cruise efficiently and navigate in
complex environments. These versatility requirements are often beyond the capability
of conventional propellers, such as marine propellers and underwater gliders.

1.1.3 Many aquatic animals have great speed, high efficiency and rapid
maneuverability. One of the best known examples is Robotuna, an eight link tendon
and pulley driven, whose external shape has form of bluefin tuna, capable of
emulating the motion of a live tuna. Also, Ghostswimmer, a navy prototype vehicle
that swims by manipulating its fins.

1.1.2 As fishes move very easily in water, the efficiency of the propelling system of
the fishes are far better than that of the conventional underwater vehicles using
propellers. Hence various attempts have been made in order to imitate the
locomotion of fishes into AUV so that they can achieve better efficiency, endurance,
speed and maneuverability.

11
 CHAPTER 2

REVIEW OF LITERATURE

2.1 REVIEW OF LITERATURE

2.1.1 UUVs fall into one of two main categories: remotely operated vehicles (ROVs)
and Autonomous Underwater Vehicles (AUVs). ROVs, as the name suggests, are
AUVs that receive commands from operator and power via a tethered umbilical
cable. This umbilical cable connects the ROV to a surface ship,relay unit or a
platform where a human operator can operate the vehicle remotely. The main
advantage of this type of vehicle is that it does not need to carry its own power
supply onboard (i.e. no onboard batteries) as it is supplied externally from an
independent source. This independence allows the vehicle to operate for a longer
period of time as the power source is provided externally. However, the umbilical
physically restricts the operation range of the vehicle and its maneuverability. This
severely limits the utilization capability of the ROVs.

2.1.2 In contrast, AUVs can operate without commands from a human operator and
therefore do not require a physical umbilical connection. Although this provides
maneuvering freedom and a larger operational range than ROVs, it has to carry its
own power supply, navigation and control systems. Both add onto the payload of the
vehicle and the capacity of the onboard power supply limits the range as well as
endurance of the vehicle. Naturally, extending the operational lifespan of the vehicle
through improvements in power consumption is very desirable. Nevertheless, the
main advantage of the AUV is the flexibility that the autonomy provides The onboard
power source for most AUVs is usually electrical batteries. These batteries have to
power all the onboard systems, that include the navigation systems, cameras,
sensors, control systems, processing systems, and other auxiliary support systems.
Of these the major portion of the power is consumed by the propulsion systems.

2.1.3 This reliance on batteries limits the range and operational time available for
the AUV to carry out its mission, which is usually in the region of 1 around 20 hours
for mid-size AUVs[ ]. There are also a number of mission types that require an AUV

12
to maintain a low speed such as for inspection or intervention work. AUVs are usually
designed for high speed and are usually flight vehicles i.e. they rely on control
surfaces and the lift they provide to maneuver. Using this type of design can lead to
the performance of the AUV at low speed to be inefficient, which is a waste of
valuable battery life and can therefore further reduce the mission duration[ ].

2.1.4 A wide variety of caudal fin shapes are seen in BCF swimming modes
(especially, carangiform and thunniform) and choice of the shape for a particular
application in robotics is often challenging [ ]. Numerical studies are often performed
with Computational Fluid Dynamics (CFD) simulations, and experiments using Digital
Image Particle Velocimetry (DPIV) [ ]. It is attempted to compare different shapes by
analyzing various shapes of fish caudal fins. The scaling studies by Geder et al. state
that a three-fold increase in the surface area increased thrust production by nine
times [10]. Hence shapes with same surface area are being considered to avoid any
such kind of interference.

2.1.3 A robotic fish should be more maneuverable than a marine vehicles with
conventional propulsion systems. Flapping foil propulsion is more efficient (about
80%) than conventional screw propulsion systems( around 60%) [11] .

2.1.4 For practical usability it is often tedious work for designers to continuously
evaluate the requisite efficiency provided for the thrust force, hence for the purpose
of convenience and time saving it is easier to develop a program that is able to
understand and learn the design. This study aims to obtain an inverse design of the
bio-mimetic fin using the iterations provided from the ANSYS Fluent simulations as
the learning variables to provide an optimal design corresponding to highest
efficiency.

13
 CHAPTER 3

PROPULSION SYSTEMS IN AUVs

3.1 PROPULSION SYSTEMS IN AUVs

3.1.1 Over the past few decades, ocean research and exploration have gained a lot
of momentum driven by the advances in research, development and successful
implementation of AUV technologies. AUVs provide a new marine platform that has
proven to be an asset in many areas of oceanographic research and exploration.
Consequently, there has been a growing interest in AUV research among several
communities. These include:
(a) Host of scientific and engineering communities for data collection
(b) Oceanographic measurements in adverse weather.
(c) To map bathysphere around oil wells.
(d) Surveillance or mine information gathering by the navies

3.1.2 The wide range of applications has resulted in development of AUVs with a
variety of shapes, sizes, working depth limits, sources of energy and means of
propulsion.

By propellers

Types of Populsion
Jet pumps

Bouyacy driven

fig.1 Types of Propulsion in AUVs

14
3.3 THE DIFFERENCE BETWEEN BIO-INSPIRED AND EXISTING SYSTEMS

3.3.1 Current underwater vehicle technology can appear slower and cumbersome
compared to the characteristics of some fishes. The benefits of bio-mimetic
propulsion techniques over conventional methods may include improved propulsion
efficiency, increased maneuverability and less disturbance of the surrounding water.

3.3.2 The efficiency and low noise nature would allow for longer and more detailed
studies of aquatic life while minimizing disturbance to the natural environment.
Increased maneuverability in confined spaces and efficiency of bio-mimetic vehicles
allow for increased mission duration and ability to visit places which are unsuited for
conventional vehicles. The ability to cause less disturbance than a conventional
vehicle and a fish like appearance will give a stealth like element which could be used
for military purposes.

3.3.3 The conventional propellers are not deformed by the surrounding fluid, but in
the flapping foils undergo various deformation based on the characteristics of the
surrounding fluid. Thus there is need to study about this deformation. FSI (Fluid
Structure Interaction) is the interaction of some movable or deformable structure with
an internal or surrounding fluid flow. FSI can be stable or oscillatory. In oscillatory
interactions, the strain induced in the solid structure causes it to move such that the
source of strains reduced, and the structure returns to its former state only for the
process to repeat.

15
 CHAPTER 4

PROPULSION SYSTEM OF FISHES

4.1 FISH LOCOMOTION

4.1.1 In an fluid environment propulsion system of fishes are second to none. They
are more efficient and produce less noise compared to the conventional propulsion
system used in vehicles. Fishes swim using a combination of three different
propulsive techniques:

Undulation
It is the process of fish propulsion through wavelike motions of a portion of a fish’s
body. These wavelike movements push backward on the water thus propelling the
fish forward. To optimize the forward motion thrust must be maximized while drag
and any lateral forces minimized. Obtained by motion of caudal tail.

Rowing
It is obtained by the motions of pectoral and pelvic fins which also known as the
paired fins. The method is two stroke. The power stroke is the first half of the motion
where the paired fins move backward with respect to the body. When the fins flip
backward they push on the water behind them. The Recovery stroke creates drag in
opposite direction of that of power stroke.

Lift-based
In this method, animals do not push the water backwards. Animals swim using their
fins as wings. Lift is created when there is a pressure difference between top and
bottom of the wing.

16
4.2 FINS OF FISHES

4.2.1 The propulsion of fishes are mainly done by using their fins. There are
different types of fins for different types of fishes. They are as follows:

SR. FIN USE

NO.
1. Dorsal fin They are found on the back of the fish. There may be up to
three of them. The fish uses this fins for balance in the
water and sudden movements.

2. Caudal fins They are also called tail fins. This is what we refer to as the
fish’s tail.

3. Anal fins They are on the ventral surface of the fish, behind anus.
Fishes use these for stability while swimming.

4. Pectoral fins They are located on each side of the fish, around where
the head meets the body.

5. Pelvic fins They are located on each side of the fish, near its bottom
middle. These help fish go up and down.

6. Adipose fins They are soft fins found between dorsal and caudal fins.

Table.1 types of fins and uses

17
fig.2 Fins of fishes

18
 CHAPTER 5

FIN SELECTION

5.1 BODY CAUDAL FIN PROPULSION

5.1.1 Most fish swim by generating undulatory waves that propagate down the body
through the caudal fin. This form of undulatory locomotion is termed Body Caudal
Fin (BCF) swimming on the basis of the body structures used; it includes anguilliform,
sub-carangiform, carangiform, and thunniform locomotory modes, as well as the
oscillatory ostraciiform mode [10]. Because BCF swimming relies on more caudal
body structures that can direct powerful thrust only rearwards, this form of locomotion
is particularly effective for accelerating quickly and cruising continuously. BCF
swimming is, therefore, inherently stable and is often seen in fish with large migration
patterns that must maximize efficiency over long periods [9].

The fin which has more involvement is propulsion of the fish is Caudal fin or
Tail fin [9]. So we are selecting Caudal fin for design and analysis.

5.1.3 The caudal peduncle length is the distance from the insertion of the anal fin to
the end of the hypural plate. The caudal peduncle depth is the least depth of this
structure from the mid-line of the ventral surface. The caudal fin length is the length
of the caudal fin measured diagonally.

19
Fig.3 Basic terminology for fins of fish

20
 CHAPTER 6

METHODOLOGY FOR ANALYSIS

6.1 SELECTION OF FIN SHAPE

6.1.1 Having selected the caudal fin for our analysis, the right geometry selection
for representation of the fin is important. For this a previously presented paper is
used to select the most optimal fin geometry [ ]. Based on the most optimal geometry,
10 fin designs of varying length of caudal peduncle were modeled in ANSYS space
claim. The length of the caudal peduncle varied from 0.25mm to 0.75mm at interval
of 5mm each.

Fig.4 Optimal caudal fin shape

21
 fig.5 Selected geometry and dimensions

6.2 MOTION MODELLING

6.2.1 The actual motion of the fin is a combination of heaving and pitching motions,
a rigid fin was considered in our simulations for simplicity. The aim is to identify the
optimal shape. The swimming speed of the foil was simulated with a flow at velocity
U (m/s) through the inlet boundary condition.

The transverse motion of the foil is given by,

h(t)=Asin(2πft+ψ)
where,
A = tail flapping amplitude
f = flapping frequency
ψ = phase difference between heaving and pitching

22
Angular motion is given by,
θ(t)=θmaxsin(2πft)
where,
θmax is the pitching amplitude.

The axis of pitching is placed at 50% of the chord length[ ].

This motion of the fin is simulated in ANSYS using the UDF, as shown below [ ].

fig.6 User defined function

23
6.3 MESHING AND ENCLOSURE

6.3.1 The dimensions of the water enclosure and parameters of mesh used is
shown here.

Fig.7 Dimensions of enclosure of water

24
fig.8 Details of meshing

25
fig.9 enclosure as viewed from space claim software

fig.10 hybrid meshing post updating

26
 fig.11 Enlarged view of fin meshing

6.4 STROUHAL NO. AND SIGNIFICANCE

6.4.1 In dimensional analysis, the Strouhal number is a dimensional number that

describes the oscillating flow mechanisms. In animal flight or swimming, propulsive
efficiency is high over a narrow range of Strouhal constants, generally peaking in the
0.2 < St < 0.4 range[13]. However, in ANSYS there is no direct method to vary the
strouhal number. So, it is indirectly varied by changing the inlet velocity (u) according
to the formula,

where,
f is the flapping frequency
L is the characteristic length of fin

27
 U is the inlet velocity.

For frequency of 1 hz

STROUHAL NO. VELOCITY

0.15 0.667
0.20 0.5
0.25 0.4
0.30 0.333
0.35 0.2857
0.40 0.25

For frequency of 2 hz

STROUHAL NO. VELOCITY

0.15 1.333
0.20 1.00
0.25 0.8
0.30 0.667
0.35 0.5714
0.40 0.5

6.5 SETUP

fig.12 Setting up strouhal no. using inlet velocity

28
6.6 CALCULATIONS AND RESULTS
The graph of drag vs flow time is plotted of given time for 200 time
steps. Further a graph of scaled residual vs iterations is also plotted as output in
order to monitor convergence. These graphs are as follows.

fig.13 Output graph of drag vs flow time

fig.14 Output graph scaled residual vs iterations

29
 fig.15 Lift, moment and drag output files from ANSYS

6.7. VALIDATION

5.5.1 With well-refined grids, Fluent is known to perform well for various other
studies, For validation, results from oscillating foil experiments conducted by
Anderson et al. (1998) were used. A NACA0012 profile with chord length C = 0.1m,
Amplitude A=0.075m, velocity U=0.4m/s, Reynolds number Re = 40,000, phase
angle between heaving and pitching ψ=90, St= 0.2, 0.25 and 0.3 is used. For each
St, the frequency of oscillation f=1Hz respectively. Other parameters were similar to

30
those used in their experiments. A coefficient of thrust as defined by Anderson et al.
is given in above where Fx is the average thrust force over a cycle. A sufficiently
refined mesh obtained from the grid independence study was used for validation.

Fig.16 strouhal no. Vs thrust for 60mm caudal peduncle

St vs. Thrust Force

fig.17 strouhal no. Vs input power

St No vs. Power Input

31
 CHAPTER 7

CONVERSION OF DATA

7.1 NECESSITY OF CONVERSION

7.1.7 The output of the CFD analysis is in terms of lift, drag and moment files for
every time step. These values cannot be used as input to the neural network. So they
are converted to avg thrust, maximum thrust, maximum power input and efficiency.
This is done by use of a python code. The basic formulae used for conversion are as
follows.

The average output power over one cycle can be calculated as:
Po=(1/T) 0∫T Fx(t).Udt
where Fx(t) is the thrust force generated at any time t.

The average input power delivered by the motors over a cycle is calculated as:
Pi=(1/T) 0∫T Fy(t) (dh/dt) dt + (1/T) 0∫T Mz(t) (dθ/dt) dt

where FY is the force required for heaving motions and M z is the moment required for
pitching.

Efficiency is defined as:

η =P0/Pi

7.2 SELECTION OF PYTHON APPLICATION

7.2.1 Python was selected as the programming language due to it’s flexibility and
availability. We build our program in pandas dataframe. Pandas is a fast, powerful,
flexible and easy to use open source data analysis and manipulation tool, built on top
of the Python programming language. To enable our program to execute numerical
computations we used Numpy library. NumPy is an open source project aiming to

32
 enable numerical computing with Python. It was built on the early work of the
Numeric and Numarray libraries.

7.3 PROGRAM TO CONVERT THE CFD DATA FILES TO READABLE ANN DATA.
#import libraries
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
#import data file in .csv format
df = pd.read_csv('all.csv')
column_name = "drag"
#aqcire input variables from user
u=float(input("velocity of flow:"))
s=float(input("strahall number:"))
n=float(input("time steps:"))
f=float(input("frequency:"))
#calculations and updating the table
T=1/f
df['Dht']=.05*2*3.14*f*np.cos(2*3.14*f*df['time_step']*n)
df['Dtt']=25*2*3.14*f*np.cos(2*3.14*f*df['time_step']*n)
column_sum = df[column_name].sum()
p_out=f*n*u*column_sum
df['p_in']=f*(df['lift']*df['Dht']+df['moment']*df['Dtt'])
pin_sum = df['p_in'].sum()
#printing the output
print("__________RESULTS_____________")
print( "total time for one oscillation is",T ,"sec" )
print("total power in put is ", round(pin_sum,2) ,"N")
print("total power output is ", round(-p_out,2) ,"N")
print("efficiency is ", round((-p_out/pin_sum*100),2) ,"%" )

33
 CHAPTER 8

ARTIFICIAL NEURAL NETWORKS

Fig.18 Illustration of structure of neuron and ANN (www.researchgate.com)

8.1 INTRODUCTION TO ANN

8.1.1 An ANN is based on a collection of connected units or nodes called artificial

neurons, which loosely model the neurons in a biological brain. Each connection, like
the synapses in a biological brain, can transmit a signal to other neurons. An artificial
neuron that receives a signal then processes it and can signal neurons connected to
it.

34
 8.1.2 The "signal" at a connection is a real number, and the output of each neuron is
computed by some non-linear function of the sum of its inputs. The connections are
called edges. Neurons and edges typically have a weight that adjusts as learning
proceeds. The weight increases or decreases the strength of the signal at a
connection. Neurons may have a threshold such that a signal is sent only if the
aggregate signal crosses that threshold. Typically, neurons are aggregated into
layers. Different layers may perform different transformations on their inputs. Signals
travel from the first layer (the input layer), to the last layer (the output layer), possibly
after traversing the layers multiple times.

8.2 FEED FORWARD BACK PROPAGATION

8.2.1 Back-propagation is a training algorithm consisting of two steps Feed

forward the values and calculate the error and propagate it back to the earlier layers.
Thus, forward-propagation is part of the back propagation algorithm but comes
before back-propagating. It is the method of fine-tuning the weights of a neural
network based on the error rate obtained in the previous epoch (i.e., iteration).
Proper tuning of the weights allows you to reduce error rates and make the model
reliable by increasing its generalization.

8.2.2 Back-propagation in neural network is a short form for "backward propagation

of errors." It is a standard method of training artificial neural networks. This method
helps calculate the gradient of a loss function with respect to all the weights in the
network.

8.2.3 The Back propagation algorithm in neural network computes the gradient of
the loss function for a single weight by the chain rule. It efficiently computes one layer
at a time, unlike a native direct computation. It computes the gradient, but it does not
define how the gradient is used. It generalizes the computation in the delta rule.
Consider the following Back propagation neural network example diagram.

8.2.4 Inputs X, arrive through the reconnected path Input is modeled using real
weights W. The weights are usually randomly selected. Then Calculate the output for

35
every neuron from the input layer, to the hidden layers, to the output layer. Calculate
the error in the outputs.

Error = Actual Output – Desired Output

Travel back from the output layer to the hidden layer to adjust the weights such that
the error is decreased. Keep repeating the process until the desired output is
achieved.

8.2.5 Following advantages can be obtained from using feed forward back
propagation method.
Back propagation is fast, simple and easy to program.
(a) It has no parameters to tune apart from the numbers of input.
(b) It is a flexible method as it does not require prior knowledge about the
network.
(c) It is a standard method that generally works well.
(d) It does not need any special mention of the features of the function to be
learned.

Fig.19 Back propagation illustration

36
8.3 REGRESSION LINE ANALYSIS

8.3.1 In simpler words, it means fitting a function from a selected family of

functions to the sampled data under some error function. Regression analysis is one
of the most basic tools in the area of machine learning used for prediction. Using
regression you fit a function on the available data and try to predict the outcome for
the future or hold-out datapoints.

8.3.2 Regression analysis helps in the prediction of a continuous variable. There

are various scenarios in the real world where we need some future predictions such
as weather condition, sales prediction, marketing trends, etc., for such case we need
some technology which can make predictions more accurately.

8.4 TRAINING THE NEURAL NETWORK FOR INVERSE DESIGN

8.4.1 The results of the CFD simulations are compiled in a format that can be
determined and input being fed into the NN Tool Program. The inputs are hence
plotted on an excel sheet in a matrix format for ease of interpretation. To simplify
matters for our ANN, we have normalized the set values I.e range from 0-1 of the
maximum number making it easy for the Neural Network to reduce regression line
error that occurs as a result of Big Data Analytic.

8.4.2 The program is set using MATLAB Simulink. For the same we consider
different sets of data inputs. In accordance with trial and error methodology we used
different combinations of Neural Networks as follows.

37
table.5(b) Transposed normalized data form ANSYS

fig.20 creation of network using NNtool

38
 CHAPTER 9
RESULTS AND ANALYSIS

9.1 NEURAL NETWORK 1

fig.21 Structure of network 1

Network 01 was found to be unsuccessful in getting the required results on the

basis of trial and error methodology. As in the figure, the network is not complex
enough (Lack of layers and neurons), to attain a successful fit.

fig.22 Output of network 1

The regression value of the fit was found to be 0.87 will led to the value
of the fit being much lesser than the required value of the output. This led to the

39
 inaccurate error of the value. The final values of network 01 are shown in the table as
follows.

Table 6. Results of prediction 1

For Prediction 01 the value should be 0.7333

Value for N1 = 0.93

For the value of the second prediction,

The required accurate value of the normalized caudal length should have been
0.866667
Actual Value = 0.8757
Hence the network is unsuccessful in prediction training.

9.2 NEURAL NETWORK 2

Network 2 is of more complex nature (greater number of neurons and layer)
as compared to network 1.
The performance for both the predictions is as follows;

For Predict 1
Expected Value = 0.7333
Actual Value = 0.9985
The value is beyond experimental range.
The network has failed

40
For Predict 2
Expected Value = 0.86667
Actual Value = 0.866667
The Network is successful in training and getting the answer.

Fig.23 Structure of network 2

fig.24 regression line analysis of network 2

41
FINAL TABLE RESULTS

PREDICT EXPECTED OBTAINED OBTAINED REMARKS

VALUE VALUE N1 VALUE N2
01. 0.76666 0.93 0.9965 FAILED FOR BOTH

02. 0.86667 0.8757 0.86667 FAILED FOR N1, BUT

PASSED FOR N2

Table.6 comparison of both networks.

CONCLUSION
The predictions for both networks for the value of “predict_1” is inaccurate by far, the
reason for this is determined as the lack of data that is present in the value of
strouhal number less that 0.2. This reduces the graphical density of data in that
region which cases limited data points during the fitting of the curve in regression
analysis. Hence the incorrect values.

42
 REFERENCES

[1] Richard Bainbridge, “Caudal fin and body movement in the propulsion of some
fish”,Zoological Laboratory, Cambridge,pp 24-25.
[2] Watts, Christopher Mark, “A comparison study of biologically inspired propulsion
systems for an autonomous underwater vehicle”,University of Glasgow, pp 24-50.
[3] Naga Praveen, Babu Mannam and Parameswaran Krishnankutty, “Hydrodynamic
study of flapping foil propulsion system fitted to surface and underwater
vehicles” ,Taylor and Francis group, Ships and offshore structures 2018, vol. 13, no.
6.
[4] Howard Chung , Shunxiang Cao , Michael Philen, Philip S. Beran , Kevin G.
Wang ,”CFD-CSD coupled analysis of underwater propulsion using a biomimetic
fin-and-joint system”, Department of Aerospace and Ocean Engineering, Virginia
Polytechnic Institute and State University, Blacksburg.
[5] Arun Krishnadas, Santhosh Ravichandran , Prabhu Rajagopal, “Analysis of
biomimetic caudal fin shapes for optimal propulsive efficiency”,Ocean Engineering,
journal homepage: www.elsevier.com/locate/oceaneng.
[6] Anderson, J.M., Streitlien, K., Barrett, D.S., Triantafyllou, M.S., 1998. Oscillating
foils of high propulsive efficiency. J. Fluid Mech. 360, 41–72.
[7] Lauder, G.V., 2015. Fish locomotion: recent advances and new directions. Annu.
Rev. Mater. Sci. 7, 521–545.
[8] Shu-yanWang, Jun Zhu, Xin-guo Wang, Qin-feng Li, and Hui-yun Zhu, School of
Mechanical Engineering, Jiangsu University of Science and Technology, Kinematics
Modeling and Simulation of a Bionic Fish Tail System Based on Linear Hypocycloid.
[9] Researchgate.com

43