applications of numerical methods in engineering cns 3320
TRANSCRIPT
Applications of Numerical Methods in EngineeringCNS 3320
James T. Allison
University of Michigan
Department of Mechanical Engineering
January 10, 2005
University of Michigan Department of Mechanical Engineering
Applications of Numerical Methods in Engineering
Objectives:
B Motivate the study of numerical methods through discussion ofengineering applications.
B Illustrate the use of Matlab using simple numerical examples.
University of Michigan Department of Mechanical Engineering January 10, 2005
Lecture Overview
• Quantitative Engineering Activities: Analysis and Design
• Selected Categories of Numerical Methods and Applications
– Linearization– Finding Roots of Functions– Solving Systems of Equations– Optimization– Numerical Integration and Differentiation
• Selected Additional Applications
• Matlab Example: Fixed Point Iteration
• Matlab Example: Numerical Integration
University of Michigan Department of Mechanical Engineering January 10, 2005
Quantitative Engineering Activities: Analysis and Design
Engineering: Solving practical technical problems using scientific andmathematical tools when available, and using experience and intuitionotherwise.
B Mathematical models provide a priori estimates of performance— verydesirable when prototypes or experiments are costly.
B Engineering problems frequently arise in which exact analytical solutionsare not available.
B Approximate solutions are normally sufficient for engineering applications,allowing the use of approximate numerical methods.
University of Michigan Department of Mechanical Engineering January 10, 2005
Quantitative Engineering Activities: Analysis and Design
BAnalysis Predicting the response of a system given a fixed system design and operating
conditions.
• 0–60 mph acceleration time of a vehicle (Mechanical Engineering)
• Power output of an electric motor (Electrical/Mechanical Engineering)
• Gain of an electromagnetic antenna (Electrical Engineering)
• Maximum load a bridge can support (Civil Engineering)
• Reaction time of a chemical process (Chemical Engineering)
• Drag force of an airplane (Aerospace Engineering)
• Expected return of a product portfolio (Industrial and Operations Engineering)
BDesign Determining an ideal system design such that a desired response is achieved.
• Maximizing a vehicle’s fuel economy while maintaining adequate performance levels
by varying vehicle design parameters.
• Minimizing the weight of a mountain bike while ensuring it will not fail structurally
by varying frame shape and thickness.
University of Michigan Department of Mechanical Engineering January 10, 2005
Categories of Numerical Methods and Applications
• Linearization
• Finding Roots of Functions
• Solving Systems of Equations
• Optimization
• Numerical Integration and Differentiation
University of Michigan Department of Mechanical Engineering January 10, 2005
Linearization
• Nonlinear equations can be much more difficult to solve than linear equations.
• Taylor’s series expansion provides a convenient way to approximate a nonlinear equation
or function with a linear equation.
• Accurate only near the expansion point a.
f(x) = f(a) + f′(a)(x − a) +
f ′′(a)
2!(x − a)
2+ . . .
• Linear approximation uses first two terms of the expansion.
University of Michigan Department of Mechanical Engineering January 10, 2005
Linearization Example: Swinging PendulumSum forces in tangential direction:∑
Ft = Wt = mgsinθ = mat
= md2θ
dt2L = mθL
⇒ θ −g
Lsinθ = 0
Linearize sinθ:
sinθ ≈ sin(0)+ cos(0)(θ− 0) = θ
Equation of motion valid for small
angles:
θ −g
Lθ = 0
m
T
W=mg
m
Wt
Wr
University of Michigan Department of Mechanical Engineering January 10, 2005
Finding Roots of Functions
Find the value of x such that f(x) = 0
• Frequently cannot be solved analytically in engineering applications.
– Transcendental equations
– Black-box functions
• May have multiple or infinite solutions
Example: static equilibrium problems must satisfy∑F = 0∑M = 0
University of Michigan Department of Mechanical Engineering January 10, 2005
Root Finding Example- Statically Indeterminate Structural Analysis
beam 1
beam 2
beam 3
beam nb
.
.
.
F1
rod 1
rod 2
rod nb-1
Ø dnb
Ø d3
Ø d2
Ø d1
Ø dr(nb-1)
Ø dr2
Ø dr1
L
lr1
lr2
lr(nb-1)
.
.
.
University of Michigan Department of Mechanical Engineering January 10, 2005
Root Finding Example- Statically Indeterminate Structural Analysis
• Force applied to lower beam known
• All other forces and displacements unknown
• Solution process:
1. Make a guess for the force on the top beam
2. Calculate the required applied force to generate this top beam force
3. Compare to actual applied force, iterate until they match
Solve: F1(F3)− F1 = 0
F33
r2a d2
f
F2b
r1e
1 F1g c
ˆ
University of Michigan Department of Mechanical Engineering January 10, 2005
Solving Systems of Equations
Solve for the value of the vector x that satisfies the given system of equations.
Linear Systems:
a11x1 + a12x2 = b1
a21x1 + a22x2 = b2⇒
[a11 a12
a21 a22
] x1
x2
=
b1
b2
⇒ Ax = b
Non-Linear Systems:
f1(x) = b1
f2(x) = b2⇒ f(x) = b
University of Michigan Department of Mechanical Engineering January 10, 2005
Linear Systems Example: Circuit Analysis
Kirchhoff’s Laws:
1. The sum of all voltage changes around any closed loop is zero:
ne∑i=1
∆Vi = 0
2. The sum of all currents at any node is zero.
nb∑i=1
∆Ii = 0
Application of these two laws to an electrical circuit facilitates the formulation of a system
of n linear equations when n unknown quantities exist.
University of Michigan Department of Mechanical Engineering January 10, 2005
Linear Systems Example: Circuit Analysis
Given that R1 = 2Ω, R2 = 4Ω, R3 = 1Ω, E1 = 6V , E2 = 9V , and using
equations from Loop 1, Loop 2, and Node A we find:
6− 2I1 − I3 = 0
9 + 4I2 − I3 = 0
−I1 + I2 + I3 = 0
⇒
−2 0 −1
0 4 −1
−1 1 1
I1
I2
I3
=
−6
−9
0
⇒ Ax = b
Loop 1 Loop 2
Node A
Node B
R1 R3 R2
I1 I3 I2
1 2
University of Michigan Department of Mechanical Engineering January 10, 2005
Linear Systems Example: Circuit Analysis
Matlab Implementation
University of Michigan Department of Mechanical Engineering January 10, 2005
Nonlinear Systems Example: Turbine Blade Analysis
• Turbine blades are components of
gas-turbine engines (used for aircraft
and electricity generation)
• Subject to high temperatures, high
inertial forces, and high drag forces.
• Commonly constructed of
monocrystalline alloys such as
Inconel.
• Structural and thermal analyses
must be performed simultaneously
(coupled non-linear equations).
University of Michigan Department of Mechanical Engineering January 10, 2005
Nonlinear Systems Example: Turbine Blade Analysis
Methods apply to arbitrary non-linear
equations (black-box functions)
T (x) = f1(L)
L = f2(T (x))
w
t
L0
x
vg, Tg
fac
ThermalAnalysis
StructuralAnalysis
T(x) (temperatureprofile)
L (dilated length)
University of Michigan Department of Mechanical Engineering January 10, 2005
Optimization
Find the values of the input variables to a function such that the function is minimized
(or maximized), possibly subject to constraints.
Negative Null Form: minx
f(x)
subject to g(x) ≤ 0
h(x) = 0
Applications:
• Engineering Design
• Regression
• Equilibrium in Nature
University of Michigan Department of Mechanical Engineering January 10, 2005
Optimization- Engineering Design
Maximize performance criteria subject to failure constraints:
• Minimize bicycle frame weight subject to structural failure constraints by varying frame
shape and thickness.
Minimize cost subject to performance and failure constraints.
• Minimize vehicle cost subject to acceleration, top speed, handling, comfort, and safety
constraints by varying vehicle design variables.
University of Michigan Department of Mechanical Engineering January 10, 2005
Optimization- Regression
Regression: technique for approximating an unknown response surface (function).
• Sample several points experimentally
• Fit an approximating function to the data points, minimizing the error between the
approximating function and the actual data points.
Criteria for best fit:
SSE =
n∑i=1
(fi − fi(p))2
⇒ minp
SSE(p)
University of Michigan Department of Mechanical Engineering January 10, 2005
Optimization- Equilibrium in Nature
• Gravitational Potential Energy
– Objects seek position of minimum gravitational potential energy: V = mgh
• Bubbles
– Energy associated with surface area.
– Bubbles seek to minimize surface area ⇒ spherical shape.
– Many small bubble coalesce to form fewer large bubbles.
• Atomic Spacing
– Atoms seek positions that minimize ‘elastic’ potential energy.
– At large separation distances attractive forces pull atoms together (depends on
bonding type).
– At small separation distances repulsive forces due to positively charged nuclei push
atoms apart.
– The net force results in an energy well. The steepness of this well determines
material properties, such as thermal expansion.
University of Michigan Department of Mechanical Engineering January 10, 2005
Numerical Integration and Differentiation
Solve: ∫ b
a
f(x)dx
df(x)
dx
where f(x) is an arbitrary continuous function.
Numerical approaches may be required when:
• f(x) is an analytical function that yields the integration unsolvable
• f(x) is known only through discretely sampled data points
University of Michigan Department of Mechanical Engineering January 10, 2005
Numerical Integration Example: Falling Climber
Falling rock climber possesses kinetic energy
T (energy of motion) that must be absorbed
by the belay system.
T =
∫F·dr
T at time of impact can be determined
analytically. Since F · v = mdvdt ·v, and
using the product rule ddt (v · v) = 2dv
dt ·v ⇒dv · v = 1
2(v · v).
F·dr = mdv · v =1
2md(v · v) =
1
2md
(v
2)
T =
∫ r2
r1
F · dr =
∫ v22
v21
1
2md
(v
2)
=1
2m
(v
22 − v
21
)
University of Michigan Department of Mechanical Engineering January 10, 2005
Numerical Integration Example: Falling Climber
T can be determined analytically, how
the rope deflects requires numerical
methods.
T = V =
∫ δf
0
F·dr
The rope behaves as a nonlinear spring,
and the force the rope exerts F is an
unknown function of its deflection δ.
• F(δ) determined experimentally with
discrete samples.
• Approximation of F(δ) necessitates
numerical integration.
• Solving for δf requires a root finding
technique.
V=T
F
f
University of Michigan Department of Mechanical Engineering January 10, 2005
Numerical Integration Example: Position Calculation
Accelerometer: measures second time derivative of position.
Application: determining position from discrete set of acceleration values (robotics).
a = x =d2x
dt2=
dx
dt
x = x0 +
∫ t
0
xdt
x = x0 +
∫ t
0
xdt
University of Michigan Department of Mechanical Engineering January 10, 2005
Numerical Differentiation Example: Solid Mechanics
Objective: Determine stress within a loaded
object to predict failure.
Constitutive Law:
σ = Eε = Edu
dx
Photoelasticity Example:
Displacement u determined experimentally at
discrete points, facilitating the calculation ofdudx and σ.
University of Michigan Department of Mechanical Engineering January 10, 2005
Selected Additional Applications
• Numerical solutions to differential
equations
– Finite Difference Method
∗ Computational Fluid Dynamics
(Navier–Stokes Equations)
∗ Dynamics (Newton-Euler &
Lagrange’s equations)
– Finite Element Method
∗ Solid Mechanics (Elasticity
equations)
∗ Heat Transfer (Heat equation)
• Kinematics Simulation
• Complex System Optimization
University of Michigan Department of Mechanical Engineering January 10, 2005