Predicting the performance of mechanical systems

using machine learning

K kiran kumar B.E.,M.E., (PHD)

Assistant Professor at IIIT, Telangana
Department of Mechanical Engineering
E-mail: [email protected]
Linkedin: https://in.linkedin.com/in/kiranmek3
Web Portal: https://eit-ae.academia.edu/kirankumar

1
DATA SCIENCE

2
What is Data Science ?

 Also known as “Data Driven Science” that makes use of different scientific methods, processes,
algorithms and systems to extract knowledge or insights with its goal to discover hidden patterns from
the raw data.
 In other words, it is about finding and exploring data in the real world and then using that knowledge
to solve business problems.
 Data Science is applied by the Data Scientist whose primary role is to design and create processes for
the complex as well as large scale datasets.
 He is involved in processing, cleaning, verifying all the integrities of the data for the analysis. He also
build predictive models using Machine Learning Algorithms.

3
Programming Languages for Data Science

4
5
MACHINE LEARNING

6
What is Machine Learning ?

 It is a type of Artificial Intelligence that allows software applications to learn from the data and
become more accurate in predicting outcomes without human intervention.

 In other words, it allows software apps to learn of its own by following a set of instructions.

 The basic Idea behind Machine Learning is to mimic the way as our brain works.

7
Machine Learning - Flow

8
Machine Learning - Types

SUPERVISED LEARNING
1
2 UNSUPERVISED LEARNING

REINFORCEMENT LEARNING
3
9
SUPERVISED LEARNING

It is where we have a input variable(X) and output variable(Y) and we use an algorithm to learn the
mapping function from the input to the output.

Y = f(X)

LINEAR REGRESSION LOGISTIC REGRESSION DESCISION TREE

NAIVE BAYES
RANDOM FOREST KNN
CLASSIFIER

10
 SUPERVISED LEARNING - ALGORITHMS

LINEAR REGRESSION

It is used to estimate real values like cost of house, salary prediction, number of sales in a company etc.

LOGISTIC REGRESSION
Used for estimating discrete values i.e. binary values 0 or 1, Yes/No, True/False like spam detection, whether a person
can buy certain product or not etc.

11
 SUPERVISED LEARNING - ALGORITHMS

DECISION TREE
Basically used for classification problems, when we want to classify data into different categories. Example - Review
greater than 4 is excellent, less than 2 is worst.

RANDOM FOREST
Quite similar to Decision tree but its accuracy is high

12
 SUPERVISED LEARNING - ALGORITHMS

NAIVE BAYES CLASSIFIER

Based on Bayes Theorem where we have no relation between b/w different predictors.
Example - Probability of playing golf at different weather conditions.

KNN (K-NEAREST NEIGHBORS)

An object is classified by a majority vote of its neighbours, or case based on a similarity measure.
Example – Search Results

13
UNSUPERVISED LEARNING

 It is the training of a model using information that is neither classified not labelled in which
there is no explanation of the data.

 It is also called "Clustering Analysis"

 Example - Unlabelled picture or audio downloaded from internet

K-MEANS CLUSTERING HIERARCHICAL CLUSTERING

14
REINFORCEMENT LEARNING

 It is an area of machine learning where an RL agent learns from the consequences of its
actions, rather than being taught explicitly.

 It selects the actions on the basis of its past experiences(exploitation) and also by new choices
(exploration).

 Here machine takes the decision on its own.

Q - LEARNING

15
16
Neural Networks

17
Neurons / Perceptron
Very important part of brain through which the signals are transmitted.
Artificial Neural Network is based on Natural Biological Systems.
ANN is a software based approach to replicate the biological neurons.

Let’s discuss about Biological Neurons !!

18
Biological Neurons

19
Artificial Neurons
Bias

20
Perceptron Mathematical Model

21
 Multiple Perceptron Networks

INPUT LAYER OUTPUT LAYER

HIDDEN LAYER

22
Activation Functions
Main role is to calculate and decides the output.
Types –
. Step Function
. Sigmoid Function
. Hyperbolic Tangent
.Rectified Linear Unit (ReLU)
Softmax Function

Let’s discuss about these one by one !!

23
Step Function

24
Sigmoid Function

25
Hyperbolic Tangent Function

26
Rectified Linear Unit (ReLU) Function

F(x) =max (0,x)

27
Softmax Function

28
Cost Function
It calculates difference between the actual and predicted values, i.e. it calculates the error.

It has one drawback as it slows down the learning of neural network, So in this case we use
“Cross Entropy”

29
Cross Entropy
It reduces the cost function and increases the learning of neural network.

y = actual values a = predicted values

30
Gradient Descent
It is an optimization
algorithm for finding the
minimum value of a function to
reduce cost.

31
Back-Propagation
It is simply a revision of weights.
In this using Feed Forward Network, weights are changed till we get the lease cost function
(error)

32
 CASE STUDY-1
PREDICTION OF COMPLIANCE INDUCED ERROR IN RRP SERIAL
MANIPULATOR

Reach 600 mm
Pay load- rated 2 kg
Pay load- maximum 5 kg

Joint range (maximum)

Joint 1 +/-1050
Joint 2 +/-1500
Joint 3 210 mm

Joint speed(maximum)
Joint 1 6.28 (rad/sec)
Joint 2 11.73 (rad/sec)
Joint 3 1100 (mm/sec)

33
Experimental testing

34
Result obtained after applying neural networks

35
 CASE STUDY-2
Design of human arm prosthesis

36
 CASE STUDY-3
SYNTHESIS, KINEMATIC, DYNAMIC ANALYSIS AND OPTIMAL
TRAJECTORY DESIGN OF PLANAR MECHANISM
Trajectory of the parallel mechanism
20
Coupler point trace of parallel mechanism
10

y axis co-ordinates
-10

-20

-30

-40

-50

-60
-40 -30 -20 -10 0 10 20 30 40 50
x axis co-ordinates

37
 Error traced by coupler curve upon retraining the network
TRAIL-1 TRAIL-2 TRAIL-3
COUPLER CURVE TRAJECTORY
20 COUPLER CURVE TRAJECTORY COUPLER CURVE TRAJECTORY
20 20
desired
desired desired
10 Actual Actual Actual
10 10

0 0 0
y axis co-ordinates

y axis co-ordinates

y axis co-ordinates
-10 -10 -10

-20 -20 -20

-30 -30
-30

-40 -40
-40

-50 -50
-50

-60 -60
-60 -40 -30 -20 -10 0 10 20 30 40 50 -40 -30 -20 -10 0 10 20 30 40 50
-40 -30 -20 -10 0 10 20 30 40 50 x axis co-ordinates x axis co-ordinates
x axis co-ordinates

TRAIL-4 20
COUPLER CURVE TRAJECTORY

desired
TRAIL-5
10 Actual COUPLER CURVE TRAJECTORY
20
desired
0 10 Actual
y axis co-ordinates

-10 0

y axis co-ordinates
-20 -10

-30 -20

-40 -30

-50 -40

-60 -50
-40 -30 -20 -10 0 10 20 30 40 50
x axis co-ordinates
-60
-40 -30 -20 -10 0 10 20 30 40 50
x axis co-ordinates
CASE STUDY-4

Determining moisture content

39
CASE STUDY-5

Determining Thermal parameters

40
CASE STUDY-6

Modeling of CO Emissions

41
CASE STUDY-8

Image processing in quality control

42
Demo on training MACHINE LEARNING using
MATLAB

43
TOOL BOX IN MATLAB

44
NEURAL NETWORK TOOL BOX

45
NEURAL NETWORK FITTING APP

46
CLUSTERING APPLICATION

47
TIME SERIES

48
RESULT VALIDATION BY NEURAL NETWORK
OPTIMIZATION TECHNIQUE
How to start training network

50
Neural Network creation requirements

51
TRAINING BY BACK PROPAGATION TECHNIQUE IN
NEURAL NETWORKS
TRAINING NETWORK IN NN
54