ce 257 discrete methods of structural analysis

68
CE 257 Discrete Methods of Structural Analysis 1 This lecture material is not for sharing/distribution outside of CE257 and CE297 class.

Upload: others

Post on 12-Nov-2021

0 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: CE 257 Discrete Methods of Structural Analysis

CE 257 Discrete Methods of Structural Analysis

1

This lecture material is not for sharing/distribution outside of CE257 and CE297 class.

Page 2: CE 257 Discrete Methods of Structural Analysis

Last meeting

• Eigenvalue Problems

2CE 257

Page 3: CE 257 Discrete Methods of Structural Analysis

For this meeting

• Time-dependent Problems

Reference for this lecture:

J. N. Reddy, An Introduction to the Finite Element Method (Chapter 6)

3CE 257

Page 4: CE 257 Discrete Methods of Structural Analysis

CE28 Analytical and Computational Methods in CEIII

4

Time-dependent Problems: Introduction

Page 5: CE 257 Discrete Methods of Structural Analysis

Introduction

• We develop the finite element models of one-dimensional time-dependent problems anddescribe time approximation schemes to convertODEs in time to algebraic equations.

5CE 257

Page 6: CE 257 Discrete Methods of Structural Analysis

Introduction

• F.E. models of time-dependent problems can bedeveloped in two alternative ways: (a) coupledformulation in which the time 𝑡 is treated as anadditional coordinate along with the spatialcoordinate 𝑥 and (b) decoupled formulation wheretime and spatial variations are assumed to beseparable. Thus, the approximation in the twoformulation takes the form

6CE 257

Page 7: CE 257 Discrete Methods of Structural Analysis

Introduction

• where 𝜓𝑗𝑒 𝑥, 𝑡 are time-space (two dimensional)

interpolation functions and ො𝑢𝑗 are the nodal values

that are independent of 𝑥 and 𝑡,

• where 𝜓𝑗𝑒 𝑥 are the usual one-dimensional

interpolation functions in spatial coordinate 𝑥 onlyand the nodal values 𝑢𝑗

𝑒 𝑡 are functions of time 𝑡

only.

7CE 257

Page 8: CE 257 Discrete Methods of Structural Analysis

Introduction

• In this section, we consider the space-timedecoupled formulation only.

8CE 257

Page 9: CE 257 Discrete Methods of Structural Analysis

Introduction

• The space-time decoupled finite elementformulation of time-dependent problems involvestwo steps:

• 1. Spatial approximation - where the solution 𝑢 ofthe equation under consideration is approximatedby the second expression and the spatial finiteelement model of the equation is developed usingthe procedures of static or steady-state problemswhile carrying all time-dependent terms in theformulation. This step results in a set of ODEs intime for the nodal variables 𝑢𝑗

𝑒 𝑡 of the element.

9CE 257

Page 10: CE 257 Discrete Methods of Structural Analysis

Introduction

• 2. Temporal approximation - where the system ofODEs are further approximated in time, oftenusing finite difference formulae for the timederivatives. This step allows conversion of thesystem of ODEs into a set of algebraic equationsamong 𝑢𝑗

𝑒 at time 𝑡𝑠+1 = 𝑠 + 1 Δ𝑡, where Δ𝑡 is

the time increment and s is a nonnegative integer.

10CE 257

Page 11: CE 257 Discrete Methods of Structural Analysis

Introduction

• All time approximation schemes seek to find 𝑢𝑗 at

time 𝑡𝑠+1 using the known values of 𝑢𝑗 from

previous times:

• Thus, at the end of the two-stage approximation,we have a continuous spatial solution at discreteintervals of time

11CE 257

Page 12: CE 257 Discrete Methods of Structural Analysis

Introduction

• Note that the approximate solution has the sameform as that in the separation-of-variablestechnique used to solve the boundary value andinitial value problems. By taking nodal values to befunctions of time, we see that the spatial points inan element take on different values for differenttimes

12CE 257

Page 13: CE 257 Discrete Methods of Structural Analysis

Introduction

• We study the details of the two steps byconsidering a model DE that contains both thesecond- and fourth-order spatial derivatives andfirst- and second-order time derivatives.

13CE 257

Page 14: CE 257 Discrete Methods of Structural Analysis

Introduction

• The above equation is subject to appropriateboundary and initial conditions. The boundaryconditions are of the form

• at 𝑥 = 0, 𝐿.

14CE 257

Page 15: CE 257 Discrete Methods of Structural Analysis

Introduction

• The initial conditions involve specifying

• where ሶ𝑢 ≡𝜕𝑢

𝜕𝑡.

15CE 257

Page 16: CE 257 Discrete Methods of Structural Analysis

Introduction

The above D.E. describes the following physicalproblems:

16CE 257

Page 17: CE 257 Discrete Methods of Structural Analysis

CE28 Analytical and Computational Methods in CEIII

17

Semidiscrete F.E. models

Page 18: CE 257 Discrete Methods of Structural Analysis

Semidiscrete F.E. models

• The semidiscrete formulation involvesapproximation of the spatial variation of thedependent variable. The formulation followsessentially the same steps as described in theprevious lessons.

• The first step involves the construction of theweak form of the equation over a typical element.

• In the second step, we develop the F.E. model byseeking approximation of the form in thedecoupled formulation.

18CE 257

Page 19: CE 257 Discrete Methods of Structural Analysis

Semidiscrete F.E. models

19CE 257

Page 20: CE 257 Discrete Methods of Structural Analysis

Introduction

• Recall: The boundary conditions are of the form

• at 𝑥 = 0, 𝐿.

20CE 257

Page 21: CE 257 Discrete Methods of Structural Analysis

Semidiscrete F.E. models

21CE 257

Page 22: CE 257 Discrete Methods of Structural Analysis

Semidiscrete F.E. models

• Next, we assume that 𝑢 is interpolated by anexpression of the decoupled formulation whichimplies that, at any arbitrary fixed time 𝑡 > 0, thefunction 𝑢 can be approximated by a linearcombination of 𝜓𝑗

𝑒 and 𝑢𝑗𝑒 𝑡 , with 𝑢𝑗

𝑒 𝑡 being

the value of 𝑢 at the 𝑗th node of the element Ω𝑒.

• We obtain the F.E. solution in the form

22CE 257

Page 23: CE 257 Discrete Methods of Structural Analysis

Semidiscrete F.E. models

• where (𝑢𝑗𝑠)𝑒 is the value of 𝑢(𝑥, 𝑡) at time 𝑡 =

𝑡𝑠 and node 𝑗 of the element Ω𝑒.

23CE 257

Page 24: CE 257 Discrete Methods of Structural Analysis

Semidiscrete F.E. models

24CE 257

Page 25: CE 257 Discrete Methods of Structural Analysis

Semidiscrete F.E. models

25CE 257

Page 26: CE 257 Discrete Methods of Structural Analysis

Semidiscrete F.E. models

• The above equation is a hyperbolic equation, andit contains the parabolic equation as a special case(set [𝑀2] = [0]).

26CE 257

Page 27: CE 257 Discrete Methods of Structural Analysis

CE28 Analytical and Computational Methods in CEIII

27

Semidiscrete F.E. models:Parabolic Equation

Page 28: CE 257 Discrete Methods of Structural Analysis

Parabolic Equations

• Time Approximation

• Consider the following D.E.:

• In the finite difference of the D.E., we replace thederivatives with their finite differenceapproximation.

28CE 257

Page 29: CE 257 Discrete Methods of Structural Analysis

Parabolic Equations

• The equation can be expressed as

• When 𝛼 = 0:

29CE 257

Forward difference approximation

Page 30: CE 257 Discrete Methods of Structural Analysis

Parabolic Equations

• The equation can be expressed as

• When 𝛼 = 1:

30CE 257

Backward difference approximation

Page 31: CE 257 Discrete Methods of Structural Analysis

Parabolic Equations

31CE 257

Page 32: CE 257 Discrete Methods of Structural Analysis

Parabolic Equations

32CE 257

• Approximations to the D.E.:

Page 33: CE 257 Discrete Methods of Structural Analysis

Parabolic Equations

33CE 257

• Time marching equation:

Page 34: CE 257 Discrete Methods of Structural Analysis

Parabolic Equations

34CE 257

Time approximation scheme using finite element method:

• Approximation of solution:

Page 35: CE 257 Discrete Methods of Structural Analysis

Parabolic Equations

35CE 257

• Time approximation scheme using finite element method:

• Approximation of solution:

• Approximation of 𝑓(𝑡):

Page 36: CE 257 Discrete Methods of Structural Analysis

Parabolic Equations

36CE 257

• Time approximation scheme using finite element method:

• If 𝑢𝑠 is known:

Page 37: CE 257 Discrete Methods of Structural Analysis

Parabolic Equations

37CE 257

• Stable and Conditionally Stable Schemes

• When the error remains bounded for any time step, the scheme is stable.

• If the error remains bounded only when the time step remains below certain value, the scheme is said to be conditionally stable.

Page 38: CE 257 Discrete Methods of Structural Analysis

Parabolic Equations

38CE 257

• Stable and Conditionally Stable Schemes

• For different values of 𝛼 , the following are the well-known time approximation schemes:

Page 39: CE 257 Discrete Methods of Structural Analysis

Fully discretized finite element equations

39CE 257

• Matrix equation:

• Subject to initial condition:

• We obtain:

Page 40: CE 257 Discrete Methods of Structural Analysis

Consistency, Accuracy, Stability

40CE 257

• The numerical scheme is said to be consistent if the round-off and truncation errors go to zero as Δ𝑡 → 0 .

• Accuracy is a measure of closeness between the approximate solution and exact solution.

• Stability of a solution is a measure of the boundedness of the approximate solution with time

• A time-approximation scheme is said to be convergent if, for fixed 𝑡𝑠, the numerical value 𝑢 𝑠 converges to its true value {𝑢(𝑡𝑠)} as Δ𝑡 → 0 .

Page 41: CE 257 Discrete Methods of Structural Analysis

CE28 Analytical and Computational Methods in CEIII

41

Semidiscrete F.E. models:Hyperbolic Equation

Page 42: CE 257 Discrete Methods of Structural Analysis

Hyperbolic Equations

• Time Approximation

• Consider matrix equations of the form

42CE 257

Page 43: CE 257 Discrete Methods of Structural Analysis

Hyperbolic Equations

• Such equations arise in structural dynamics, where[𝑀] denotes mass matrix, [𝐶] the damping matrix,and [𝐾] the stiffness matrix. The damping matrix[𝐶] is often taken to be a linear combination of themass and stiffness matrices, [𝐶] = 𝛽1[𝑀] +𝛽2[𝐾], where 𝛽1 and 𝛽2 are determined fromphysical experiments.

43CE 257

Page 44: CE 257 Discrete Methods of Structural Analysis

Hyperbolic Equations

Related Problems:

• Axial Motion of Bars

• Transverse Motion of Euler-Bernoulli Beams

• Transverse Motion of Timoshenko Beams

44CE 257

Page 45: CE 257 Discrete Methods of Structural Analysis

Hyperbolic Equations

• There are several numerical methods available toapproximate the second-order time derivativesand convert differential equations to algebraicequations.

• Newmark family

• Wilson method

• Houbolt method

45CE 257

Page 46: CE 257 Discrete Methods of Structural Analysis

Hyperbolic Equations

• Newmark's Scheme

• In the Newmark Method, the function and its firsttime derivative are approximated according to

where

46CE 257

Page 47: CE 257 Discrete Methods of Structural Analysis

Hyperbolic Equations

• The 𝛼 and 𝛾(= 2𝛽) are parameters thatdetermine the stability and accuracy of thescheme. Equations can be viewed as Taylor's seriesexpansion of 𝑢 and ሶ𝑢.

47CE 257

Page 48: CE 257 Discrete Methods of Structural Analysis

Hyperbolic Equations

• Newmark's Scheme

• The following are special cases:

48CE 257

Page 49: CE 257 Discrete Methods of Structural Analysis

Hyperbolic Equations

• Newmark's Scheme

• For all schemes in which 𝛾 < 𝛼 and 𝛼 ≥1

2, the

stability requirement is

• where 𝜔max is the maximum natural frequency ofthe system from (if without [𝐶])

49CE 257

Page 50: CE 257 Discrete Methods of Structural Analysis

Fully Discretized Finite Element Equations

50CE 257

Page 51: CE 257 Discrete Methods of Structural Analysis

Fully Discretized Finite Element Equations

51CE 257

Page 52: CE 257 Discrete Methods of Structural Analysis

Fully Discretized Finite Element Equations

• Initial conditions: 𝑢 0 , ሶ𝑢 0, ሷ𝑢 0

• Calculate ሷ𝑢 0 from:

• Update vectors:

52CE 257

Page 53: CE 257 Discrete Methods of Structural Analysis

CE28 Analytical and Computational Methods in CEIII

53

Mass Lumping

Page 54: CE 257 Discrete Methods of Structural Analysis

Mass Lumping

• Recall from the time approximation of theparabolic equations that use of the forwarddifference scheme (i.e., 𝛼 = 0 ) results in thefollowing time marching scheme

54CE 257

Page 55: CE 257 Discrete Methods of Structural Analysis

Fully discretized finite element equations

55CE 257

• Matrix equation:

• Subject to initial condition:

• We obtain:

Page 56: CE 257 Discrete Methods of Structural Analysis

Mass Lumping

• The matrix [𝑀𝑒] derived from the weighted-integral formulations of the governing equation iscalled the consistent mass matrix, and it issymmetric positive-definite and non-diagonal.Solution of the global equations requires inversionof the assembled mass matrix.

• If the mass matrix is diagonal, then the assembledequations can be solved directly (i.e. withoutinverting a matrix)

56CE 257

Page 57: CE 257 Discrete Methods of Structural Analysis

Mass Lumping• There are several ways of constructing diagonal

mass matrices, also known as lumped massmatrices.

• The use a lumped mass matrix in transientanalyses can save computational time in two ways:

• First, for the forward difference schemes, lumpedmass matrices result in explicit algebraic equationsnot requiring matrix inversions.

• Second, the critical time step required forconditionally stable schemes is larger, and henceless computational time is required when lumpedmass matrices are use.

57CE 257

Page 58: CE 257 Discrete Methods of Structural Analysis

Mass Lumping

Row-Sum Lumping

• The sum of the elements of each row of theconsistent mass matrix is used as the diagonalelement

• where the property σ𝑗=1𝑛 𝜓𝑗

𝑒 = 1 of the

interpolation functions is used.

58CE 257

Page 59: CE 257 Discrete Methods of Structural Analysis

Mass Lumping

Row-Sum Lumping

• When 𝜌 is constant:

59CE 257

Page 60: CE 257 Discrete Methods of Structural Analysis

Mass Lumping

Row-Sum Lumping

• Consistent mass matrices

60CE 257

Page 61: CE 257 Discrete Methods of Structural Analysis

Mass Lumping

Proportional Lumping

• Here the diagonal elements of the lumped massmatrix are computed to be proportional to thediagonal elements of the consistent mass matrixwhile conserving the total mass of the element:

61CE 257

Page 62: CE 257 Discrete Methods of Structural Analysis

Mass Lumping

Proportional Lumping

• For constant 𝜌, proportional lumping gives thesame lumped mass matrices as those obtained inthe row-sum technique for the Lagrange linearand quadratic elements.

62CE 257

Page 63: CE 257 Discrete Methods of Structural Analysis

CE28 Analytical and Computational Methods in CEIII

63

Page 64: CE 257 Discrete Methods of Structural Analysis

Homework

• Derive the fully discretized finite element equation

using Newmark beta (𝛾 =1

2, 𝛽 =

1

4):

64CE 257

1

2

1

2

44242 ++ ++

++

+

=

+

+ nnnnn

tttttfuMuMCuMCuMCK

Page 65: CE 257 Discrete Methods of Structural Analysis

CE28 Analytical and Computational Methods in CEIII

65

Page 66: CE 257 Discrete Methods of Structural Analysis

Finite Element Method

References:

1. J. Fish, T. Belytschko, A First Course in Finite Elements

2. J. N. Reddy, An Introduction to the Finite Element Method

3. O.C. Zienkiewicz, R.L. Taylor, J.Z. Zhu, The Finite Element Method, Its Basis & Fundamentals

4. K-J. Bathe, Finite Element Procedures

5. K. Leet, C-M., Uang, A.M. Gilbert, Fundamentals of Structural Analysis

66CE 257

Page 67: CE 257 Discrete Methods of Structural Analysis

This lecture

• Time-dependent Problems

Reference for this lecture:

J. N. Reddy, An Introduction to the Finite Element Method (Chapter 6)

67CE 257

This lecture material is not for sharing/distribution outside of CE257 and CE297 class.

Page 68: CE 257 Discrete Methods of Structural Analysis

CE28 Analytical and Computational Methods in CEIII

68