Structured Analysis Using Decision Trees

Structured Analysis Using Decision Trees

James Scanlan; School of Engineering Sciences

1

Case Study Issues

James Scanlan; School of Engineering Sciences

2

Mechanical students Laboratory groups Aerospace Students Randomly generated teams. Electro-Mech Students 2 teams; let me know composition of these. Review meetings 1 each team after Easter to check on progress. Deadline Hand-in date for the case study report will be Monday 21st May. I will give further more specific guidelines on what I expect in the report after Easter but the case study document on my web-site gives the important details. No formal assessed presentations this year. Peer review; PLEASE NOTE! I will be using a peer review process to ensure all individual members of each team are graded according to how much effort they put in to the design case study! James Scanlan; School of Engineering Sciences

3

Roles and team organisation Team member 1 Team member 2 Team member 3 Team member 4 Team member 5 (Team member 6)

Design Activity leader for Customer requirements and cost estimating Concept design analysis and shortlisting QFD/CODA matrices Design embodiment/analysis alternative 1 Design embodiment/analysis alternative 2 Design embodiment/analysis alternative 3

Management role Chair and coordinate meetings Store and manage master data (sussed?) Meeting records Report compilation and coordination Timescale plan and barchart, background research Alternate meeting chair

James Scanlan; School of Engineering Sciences

4

Costing

James Scanlan; School of Engineering Sciences

5

Idealised VE Output A

Good

Upper constraint Comparative Assessment of Function

Alternative Design Concepts

C

E

D Bad

B Cost James Scanlan; School of Engineering Sciences

+ 6

DATUM (Design Analysis Tool for Unit-cost Modelling) £

James Scanlan; School of Engineering Sciences

7

Need to understand the difference between a cost estimate and a cost model

James Scanlan; School of Engineering Sciences

8


Changing price of materials 2005 - 2007

Factor 0.6

$ per tonne

$ per tonne

The problem of material price Factor 0.4

0il


The need: Cost modeling for material selection

Factor 3

$ per tonne

$ per tonne

40000

30000

Factor 4 20000

10000

James Scanlan; School of Engineering Sciences

9

Windturbine cost model Inputs: quantity of blades 8 blade chord 0.05 m Blade length 0.5 m qty of legs 4 Carbon composite cost 1500000 $/m^3 Aluminium cost 10 $/kg leg length 0.8 m

quantity of blades 8

blades 4500 $

individual blade cost 562.5 $

Blade length 0.5 m amount of material required 0.00038 m^3 Carbon composite cost 1500000 $/m^3

mean CSA 0.00075 m^2

blade chord 0.05 m mean thickness 0.015 m

Aluminium cost 10 $/kg legs 464.5152 $

Total cost of wind turbine 7258.0608 $ Structure 2008.0608 $

generator 500 $

base 1908.0608 $ upright 100 $

total leg cost 1858.0608 $ attachment 50 $

qty of legs 4

leg length 0.8 m Volume 0.02323 m^3

leg mass 46.45152 kg density 2000 kg/m^3

width 0.0762 m height 0.381 m

Bought out item 500 $

electronics 250 $

James Scanlan; School of Engineering Sciences

10

Generative…Parametric There is no clear boundary..

Performance-based Parametrics Whole-Product Parametrics System-level Parametrics

Cost Accuracy

High-level Part Parametrics Part-Feature Parametrics Process Parametrics Process Simulation

“Parametric” “Generative” Level of Detail

James Scanlan; School of Engineering Sciences

£ 11

Research Objectives • Hierarchical structure – Why? Even for complex design there has to be a single objective (Even Multi-objective simulation has to resolve down to a single higher level objective).

• Fuzzy (Uncertain) Data – Why? Often requirements/ design data is not precisely known

• Sensitivity Analysis – Why? To allocate effort proportionally

• Monte-Carlo capability – Why? To compute the net effect of uncertainty

James Scanlan; School of Engineering Sciences

12

Spreadsheets • Wrong representation (cannot model trees). • Complex, non-intuitive structure. • Difficult to maintain (Mix of structure and data) • Sensitivity, Monte-Carlo need Excel add-ins (such as @Risk) • Spreadsheets are computationally slow, memory hungry and inefficient. • Difficult to “roll-out” • Spreadsheet “Cottage Industry”.

James Scanlan; School of Engineering Sciences

13

DecisionPro

James Scanlan; School of Engineering Sciences

14

Decision Tree Analysis For choosing the best course of action when future

Increase promotion 1 -49,000

10% Competitor is superior -25,000

Abandon product 1 -25,000

outcomes are uncertain. 70% Competitor enters 8,900

Maintain course 1 -30,000

Increase promotion 2 8,000 60% Competitor is equal 8,000

Maintain course 2 6,000 Abandon product 2 -25,000 Increase promotion 3 10,000

Choose high price 17,330

30% Competitor is inferior 22,000 10% Sales high 1 95,000 30% No competition 37,000

Product pricing strategy 17,330

10% Sales high 2 42,000 Choose low price 13,000

Maintain course 3 22,000 Abandon product 3 -25,000

80% Sales typical 1 35,000 10% Sales low 1 -5,000

80% Sales typical 2 12,000 10% Sales low 2 -8,000

Don_t launch product 0

James Scanlan; School of Engineering Sciences

15

General Modeling and Problem Solving • For tackling complex problems and communicating ideas clearly. Free Cash Flow FCF := EBIT * ( 1 - t ) + Dep - WC - CapEx 21.29

Profit After Tax PAT:=13.25 Earnings Before Interest and Taxes EBIT := PAT + Interest + Tax 32.26

Interest paid Interest:=8.16 Taxes paid Tax:=10.85

Tax rate t:=34% Depreciation Dep:=15.14 Working capital change WC:=0 Capital expenditures CapEx:=15.14

Company value FCF Value := WACC 283.24

Weighted Average Cost of Capital WACC 7.5%

Debt after leverage Debt:=102 Equity after leverage Equity:=291

Equity Debt WACC1 := * Ke + * Kd Debt + Equity Debt + Equity 7.5%

Cost of equity Ke 13.9%

Risk-free return Rf:=5.1%

Ke1 := Rf + Beta * ( Rm - Rf ) 13.9%

Systematic risk Beta:=1.6 Average market return Rm:=10.6%

Cost of debt Kd := Id * ( 1 - t ) 5.3%

Debt interest rate Id:=8%

James Scanlan; School of Engineering Sciences

Tax rate 34.0%

16

Sensitivity Analysis • For determining which assumptions drive decisions most. Input Sensitivity 1,600

1,400

Payment

1,200

1,000

800

600

400 -60

-40

-20

0

20

40

60

% Change in Input Interest

Term

Price

James Scanlan; School of Engineering Sciences

Down Payment

17

Monte Carlo Simulation Frequency Distribution 250

Occurrences

• For modeling uncertainty to help manage business risk and simulate complex systems.

200 150 100 50 0

Cumulative Distribution

20 40 60 80 100 120 140 160 180 200 220 240 100%

Profit

Probability

80% 60% 40% 20% 0% $0

$50

$100

$150

$200

$250

Profit

James Scanlan; School of Engineering Sciences

18

Advanced Analytics Underlying security price P:=29.25 Cumulative Normal distribution

Normal distribution function

z

N( z ) :=

∫

Normal( x ) dx

Normal( x ) :=

-∞

1 2π

2

* exp

x -2

P

Black-Scholes option pricing model. Value := P * N( d1 ) - E * exp( -r * T ) * N( d2 ) $0.92

log d1 :=

P E

Exercise price E:=33

2

+

Volatility r+ 2

Volatility *

*T

T

-0.51644

Volatility 36.0%

E

Years to expiration T:=90/365

r T d2 := d1 - Volatility * -0.69521

Risk-free interest rate r:=5%

d1 T Volatility T

James Scanlan; School of Engineering Sciences

19

Web Delivery • Easy conversion of Model • Hosting of Models on server • DP use of JavaScript

James Scanlan; School of Engineering Sciences

20

DecisionPro Demo • • • • • • •

Construction of tree Insertion of stochastic input Monte Carlo Sensitivity analysis Alternative inputs Max branch List processing

James Scanlan; School of Engineering Sciences

21