Lecture 5 25 Final

You might also like

Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 37

Simulation Models and Optimization in

Biotechnology

Dr. Shirsendu
Dr.Mitra
Garima Mishra
Assistant Professor
Chemical Engineering
School of Technology
Pandit Deendayal Energy University Gandhinagar

Dr. Garima
Dr. Shirsendu Mitra LECTURE 25 1
OVERVIEW

 Quick Recap
 Introduction
 Significance
 Optimisation techniques
 Artificial Neural Networks (ANN)
 Case Studies

Dr. UNIT 4 2
QUICK RECAP: Video 1

Dr. Shirsendu Mitra 3


INTRODUCTION

• Simulation and modelling is the creation and analysis of digital


prototype of a real model to predict the performance of the
model. It is used to help us understand the conditions of
performance and failure.

• Optimization is the improving of any existing situation. The


main goal is to maximise economic potential of the process by
taking safety, ethics and environment into consideration.

Dr. Shirsendu Mitra 4


Importance of Simulation and Modelling
• Solves real world problems safely and efficiently and enable
experimentation on real world system in virtual environment.
• Provides safe way to test and explore what-if scenarios before
experimenting physically.
• Cost effective and consumes less time.
• Models can be animated in both 2-D and 3-D, allowing
concepts to be easily verified and understood.
• Allows the observation of behaviour of systems over time.
• Can capture more details than analytical model, with
increased accuracy and more precise forecasting.
• Uncertainty in operation and outcome can be represented
using simulation models, allowing risk quantification, and for
finding solutions .
Process Optimization
• Objective is to produce greatest quantity of product at least
cost to maximise profit.
• To effect this, process model is required spanning upstream
and downstream processes, which can be used for defining
strategies for optimizing sub-processes.
• Since global process modelling is difficult, process modelling is
applied for major sub-processes which have the greatest
impact on quality, productivity and cost.
• Since for commodity products, the relative cost of raw
materials is higher than that of processing costs, yield is
preferred for optimization to reduce operating cost.
• Optimized models must be continually updated or superseded
to maintain competitive position in markets.
Basic Terminologies
• Local optimal point: A point or solution “x” is said to be a local optimal point, if there exists no
point in the neighbourhood of “x” which is better than “x”. For minimization problems, a point
“x” is a locally minimal point if no point in the neighbourhood has a function value smaller
than f(x).
• Global optimal point: A point or solution “x” is said to be a global optimal point, if there exists
no point in the entire search space which is better than “x”. For minimization problems, a
point “x” is a globally minimal point if no point in the entire search space has a function value
smaller than f(x).

Dr. Shirsendu Mitra


Basic Terminologies (Contd.)

• Inflection point: A point “x” is said to be an inflection point if the function value
increases locally as “x” increases and decreases locally as “x” reduces or if the
function value decreases locally as “x” increases and increases locally as “x”
deceases.

Dr. Shirsendu Mitra


Optimization Basics (Calculus)

For a function f(x), in the neighbourhood of a point x = x a:


1. There exists a maxima at xa when f’(xa) = 0 and f’’(xa)<0.
2. There exists a minima at xa when f’(xa) = 0 and f’’(xa)>0.
3. There exists an inflection point at xa when f’(xa) = 0 and f’’(xa)=0.

Exercise:
Find the maximum and minimum values of 3x4 - 2x3 – 6x2 + 6x + 1 in
the interval (0,2).

Dr. Shirsendu Mitra


Single Variable Optimization

• Optimization of functions involving only one real


variable.
• Examples:
1. Minimize f(x) = x2
2. Maximize f(x) = 3x + 5x2 – 4/x + log(x)

Popular Algorithms for single variable optimization:


• Dichotomous Search Method
• Fibonacci Search Method
• Golden Section Search Method

Dr. Shirsendu Mitra


Dichotomous Search Method

• The dichotomous search method computes the midpoint (a+b)/2


and then moves slightly to the either side of the midpoint to
compute two points (a+b)/2 ± ε where ε is a very small real
number.
• Steps to Minimize f(x) within the interval (a,b):
1. Define an objective function
2. Define n=number of iterations
3. Compute the midpoint (a+b)/2
4. Compute x1 = (a+b)/2 - ε and x2 = (a+b)/2 + ε
5. If f(x1) ≤ f(x2) then redefine a=a, b= x2, k=k+1 and go to step 2
6. If f(x1) > f(x2) then redefine a= x1, b= b, k=k+1 and go to step 2
Repeat the process until f(x1) ≈ f(x2) or k = n
Dr. Shirsendu Mitra
Dr. Garima
Numerical Understanding

Q. Consider a solar thermal application where the hot water produced by a solar collector is kept in a
cylindrical storage tank and its use regulated so that it is also available during night time. The storage tank
has a capacity of 4000 L. Convective losses from the tank have to be minimized. Relative losses can be
neglected. Ambient temperature and convective coefficient are constant. The hot water temperature may be
assumed constant in the analysis. Take initial search interval for cylinder radius as 0.5 ≤ r ≤ 3.5 m. Perform
4 iterations using:
1. Dichotomous Search
2. Fibonacci Search
3. Golden Section Search

Dr. Shirsendu Mitra


Dichotomous search algorithm:
Objective function: f(r) = 2πr2 + 8/r r(0.5, 3.5), ε = 0.01
Number of iterations = 4 (given)

Iteration 1: a = 0.5, b = 3.5, Compute the midpoint (a+b)/2= 2


r1 = (a+b)/2 - ε = 1.99 f(r1=1.99) = 28.9 m2
r2 = (a+b)/2 + ε = 2.01 f(r2=2.01) = 29.36 m2
a r1 r2 b

Iteration 2: a = 0.5, b = 2.01, Midpoint (a+b)/2= 1.255


r1 = (a+b)/2 - ε = 1.245 f(r1=1.245) = 16.16 m2
r2 = (a+b)/2 + ε = 1.265 f(r2=1.265) = 16.37 m2

Iteration 3: a = 0.5, b = 1.265, Midpoint (a+b)/2= 0.882


r1 = (a+b)/2 - ε = 0.872 f(r1=0.872) = 13.95 m2
r2 = (a+b)/2 + ε = 0.892 f(r2=0.892) = 13.96 m2

Iteration 4: a = 0.5, b = 0.892, Midpoint (a+b)/2= 0.696


r1 = (a+b)/2 - ε = 0.686 f(r1=0.686) = 14.61 m2 a r1 r2 b
r2 = (a+b)/2 + ε = 0.706 f(r2=0.706) = 14.46 m2
Dr. Shirsendu Mitra
Fibonacci Search Method

• In this method the search interval is reduced according to Fibonacci numbers.


• Fibonacci series is represented as: Fn = Fn-1 + Fn-2 i.e. the nth term is sum of the
two previous terms.
• Fibonacci series: 0, 1, 1, 2, 3, 5, 8, 13, 21, 34……
• Steps to Minimize f(x) within the interval (a, b):
1. Define n=number of iterations
2. Compute L = b-a
3. Compute L1 = (Fn-1 /Fn)×L
4. Set x1 = a + L1 and x2 = b – L1
5. Compute f(x1) and f(x2)
6. If f(x1) ≤ f(x2) then redefine a=a, b= x2, k=k+1, L= L1 and go to step 3
7. If f(x1) > f(x2) then redefine a= x1, b= b, k=k+1, L=L1 and go to step 3
Repeat the process until f(x1) ≈ f(x2) or k=n
Fibonacci search algorithm:

Fibonacci series: 0, 1, 1, 2, 3, 5, 8, 13, 21, 34……


Objective function: f(r) = 2πr2 + 8/r
Number of iterations = 4 (given); Taking Fn = 34 a r1 r2 b

Iteration a b L = b- L1= (Fn-1 /Fn) × r1= r2 = f(r1) f(r2)


No. a L b-L1 a+L1

1 0.5 3.5 3 (21/34)x 3 = 1.65 2.35 21.95 38.10


1.85
2 0.5 2.35 1.85 (13/21)x 1.85 1.21 1.64 15.81 21.77
= 1.14
3 0.5 1.64 1.14 (8/13)x 1.14 0.94 1.2 14.06 15.71
= 0.70
4 0.5 1.2 0.7 (5/8)x 0.7 = 0.76 0.94 14.155 14.06
0.44

Dr. Shirsendu Mitra


Golden Section Search Method

• In this method the search interval is reduced according to


Golden ratio (=0.618 or 61.8%).
• Steps to Minimize f(x) within the interval (a,b):
1. Define n=number of iterations and k=2 (iteration counter)
2. Compute L = b-a
3. Compute Lk = 0.618×L. Set x1 = a + Lk and x2 = b – Lk
4. Compute f(x1) and f(x2)
5. If f(x1) ≤ f(x2) then redefine a=a, b= x2, k=k+1 and go to step 2
6. If f(x1) > f(x2) then redefine a= x1, b= b, k=k+1 and go to step 2
Repeat the process until f(x1) ≈ f(x2) or k=n

Dr. Shirsendu Mitra


Relation between Fibonacci Search Method and Golden
Section Search Method
• As we approach larger and larger numbers in the Fibonacci series, (F n-1 /Fn+1)
reduces asymptotically to the value 0.618.
• For achieving higher accuracy with Fibonacci search, large number of iterations are
to be performed which require large Fibonacci numbers to be calculated (not
desirable).
• If we want to enjoy the full advantages of Fibonacci search but do not want to
calculate and store the numbers and also do not want unequal interval reduction
across iterations, we can use Golden Section search.

Dr. Shirsendu Mitra


Multivariable Optimization

Systems for Process Optimization


1. Open-Ended Systems
• Any number and type of component are analysed for
optimization.
• This system makes no assumption of which components are
best for any process.

2. Close-Ended Systems
• Fixed number and type of component are analysed for
optimization.
• This is the simplest strategy but many different beneficial
components may not be considered.

Dr. Garima
Multivariable Optimization

• Optimization of functions involving more than one real variable.


• Eg:
1. Minimize f(x1,x2) = 3x1 - 5x2
2. Maximize f(x1,x2, x3 ,x4) = 20x1 + sin(x2) – 4/x3 + log(x4)

Popular Algorithms for multivariable optimization:


3. Borrowing
4. Component Replacing
5. One-factor-at-a-time
6. Factorial Design
7. Response Surface Methodology
8. Artificial Neural Network
9. Fuzzy Logic
10. Genetic Algorithm
11. Box-Behnken Design

Dr. Shirsendu Mitra


Borrowing

• This is an open ended system.


• The components and process conditions obtained from
literatures are analysed using this method.
• Advantage – Simple, easy and requires no mathematical skill.
• Disadvantage- Too many options for a single process.

Dr. Shirsendu Mitra


Component Replacing

• It is an open-ended system for process optimization.


• It is used to compare component of one type.
• In this method, one component is replaced by other at same
incorporation level.
• This method is useful for screening sources for improving
medium utilization.
• This method does not consider component interaction.

Dr. Shirsendu Mitra


One-factor-at-a-time

• It is a close-ended system.
• Only one factor or variable is varied at a time while keeping
other variables constant.
• This method has been the most popular method for designing
medium composition used in the initial stages.
• Based on the approach, it is sub-grouped into-

1. Removal Experiments- Here, all the medium components


are removed one-by-one, and after proper incubation period,
their effects on the production of secondary metabolite or
the product of interest is observed in terms of suitable
parameters.

Dr. Shirsendu Mitra


One-factor-at-a-time

2. Supplementation Experiments- These are generally performed


to evaluate the effects of various carbon and nitrogen
supplements on metabolite production.

3. Replacement experiments- For medium formulation,


carbon/nitrogen sources showing enhancement effect on the
desired metabolite production in supplementation
experiments are generally tried to be used as a whole
carbon/nitrogen source.

4. Physical parameters- In addition to chemical and biological


variables, several researchers used OFAT experiments to
standardize the physical parameters such as pH, temperature,
agitation and aeration requirements of the fermentation
process
Dr. Shirsendu Mitra
One-factor-at-a-time

Dr. Shirsendu Mitra


Factorial Design

• It is a close-ended system.
• Here, level of factors/parameters are independently varied, each
factor at two or more levels.
• The optimization procedure is facilitated by the construction of a
polynomial equation that describes the experimental results as a
function of the factor level and the co-efficient are related to the
effects and interactions of factors.
• Typical factors are microbial strain, medium components,
temperature, humidity, initial pH and inoculum volume.
• There are two types of factorial design-

1. Full Factorial Design – Every combination in factor level is tested.


The most common method is two factorial design which is
represented by 2n, where ‘n’ is the no. of factors.

Dr. Shirsendu Mitra


Factorial Design

Partial Factorial Design - This method provides a compromise when the


number of runs required in full factorial is impracticable. Two-level factorial
design is usually used. It is denoted by 2n-k, where ‘n’ is the no. of factors and
‘k’ is the fraction of complete factorial used.

Dr. Shirsendu Mitra


Response Surface Methodology

• It uses factorial designs to optimize the production processes of


the desired metabolites.
• It is a sturdy, robust and efficient mathematical approach which
includes statistical experimental designs and multiple regression
analysis, for seeking the best formulation under a set of
constrained equations.
• It employs several phases of optimization and it can be
performed in three basic steps, i.e., experiments designed for the
screening of the factors followed by the path of steepest
ascent/descent and finally quadratic regression model is fitted
and optimized using canonical regression analysis method.
• One of the important inputs of RSM is representation of the yield,
as a surface plot. It can provide multiple responses at the same
time by considering the interactions between the variables, which
is utmost necessary for designing and process optimization.
Dr. Shirsendu Mitra
Response Surface Methodology

• + +
Where Y = predicted response, a0 = intercept coefficient, aiXi = linear
terms, aijXiXj = interaction terms and aiiX2 = square terms.
• It is simple, efficient, less time consuming and capable of predicting
the optimization of various processes of metabolite production.
• It is used to determine the factor levels which can simultaneously
satisfy a set of desired specifications.
• This method helps us to determine, how a specific response is
affected by changes in the level of the factors over the specified
levels of interest and to achieve a quantitative understanding of the
system behaviour over the region tested.
• Using this method, we can predict the product properties
throughout the region, even at factor combinations not actually run
and to find conditions for the process stability.

Dr. Shirsendu Mitra


Response Surface Methodology

Advantages
1. It needs fewer experiments to study the effects of all the factors
and the optimum combination of all the variables can be revealed.
2. The interaction (the behaviour of one factor may be dependent on
the level of another factor) between factors can be determined.
3. It also requires less time and effort.

Limitations
4. The prediction of responses based on second-order polynomial
equation is often limited to low levels and results in poor
estimation of optimal formulations.
5. It is quite complicated to study the interactions of more than five
variables and large variations in the factors can give misleading
results possibly due to error, bias, or no reproducibility.

Dr. Shirsendu Mitra


Response Surface Methodology

Dependent Determining
Problem
Variables Independent
Identification
Determination Extraction Variables

Factor level Proper Design Running the


Decision Selection experiment

Model Evaluation Model Optimization Model Validation

Dr. Shirsendu Mitra


Response Surface Methodology

Typical RSM model

Dr. Shirsendu Mitra


Box-Behnken Design

• It is a class of rotatable or nearly rotatable second-order


response surface design based on three-level incomplete
factorial design.
• The sample size is kept to a value, which is sufficient for the
estimation of coefficients in a second degree least squares
approximating polynomial.
• A block of samples corresponding to two-level factorial design is
repeated over different sets of parameters. The parameters
which are not included in the factorial design remain at their
mean level throughout the block.
• The type (full or fractional), the size of factorial and number of
blocks which are evaluated depend on the number of
parameters, and it is chosen so that the design meets the
criterion of rotatability. An experimental design is said to be
rotatable if the variance of the predicted response at any point is
a function of the distance from
Dr. Shirsendu Mitra the central point alone.
Box-Behnken Design (continues)

• In BBD, the graphical representation can be seen in two forms for three
factors, viz. a cube that consists of the central point and the middle points
of the edge and a figure of three interlocking 2 2 factorial designs and a
central point.
• The number of experiments (N) required for the development of BBD is
defined as N =2k(k - 1)+C0, where k is number of factors and C 0 is the
number of central points.

Advantages
1. It is slightly more efficient than CCD but much more efficient than the
three-level full factorial designs, where the efficiency of one experimental
design is defined as the number of coefficients in the estimated model
divided by the number of experiments.
2. It does not have axial points, thus all design points fall within the safe
operating zone.
3. BBD ensures that it does not contain combinations for which all factors
are simultaneously at their highest or lowest levels.
Dr. Shirsendu Mitra
Box-Behnken Design (continues)

Typical Box-Behnken Design

Dr. Shirsendu Mitra


Genetic Algorithm

Dr. Shirsendu Mitra


Artificial Neural Network

Dr. Shirsendu Mitra


Dr. Shirsendu Mitra 37

You might also like