Advanced structural models

for 3D stress analysis of

laminated composites
Ojo, Saheed Olalekan

University of Limerick

February 2019

The DRIVEN project has received funding from the

European Union’s Horizon 2020 research and
innovation programme under grant agreement No
811099
 Preliminaries

Objective
Structural models using advanced theories for stress analysis of laminated and variable
composites.
Highlight prospects for collaboration with the DRIVEN project.

Motivation
➢ High accuracy of advanced theories for analysis of complex structures.
➢ High order numerical approximation techniques
➢ Improved applications of variable stiffness and variable geometry composites.
➢ Excellent numerical properties of NURBS functions.

Methods
➢ Theory of unified formulation using hierarchical functions
➢ Strong formulation using Differential quadrature method (DQM)
➢ Weak formulation using Higher order finite element method (HOFEM)
➢ Exploring potentials of NURBS functions for structural composite applications
 Theory of unified formulation using Serendipity Lagrange
Functions

𝒖(𝑥, 𝑦, 𝑧) = 𝐹𝜏 (𝑥, 𝑧)𝒖𝜏 (y), 𝜏 = 0,1, … , 𝑁

𝒖𝜏 = 𝑢𝑥𝜏 𝑢𝑦𝜏 𝑢𝑧𝜏 𝑇

Serendipity Lagrange Functions

It combines the efficiency of Taylor’s function and the numerical stability of Lagrange functions.
𝑡
The SLE functions 𝐹𝜏 𝜉, 𝜂 are defined by three set of functions with specific requirements as follows:

• A set of four first-order 2D bilinear polynomials called type 𝐼 shape functions that take the value of 1 at the four nodes
and 0 everywhere else in Ξ2𝐷 :
𝐼 1
𝐹𝜏 𝜉, 𝜂 = 1 + 𝜉𝜉𝑠 1 − 𝜂𝜂𝑠 ,
4
where 𝜉𝑠 , 𝜂𝑠 are the coordinates of the four corner nodes in Ξ2𝐷 .

• A set of 𝑁 𝑡ℎ order polynomials named type 𝐼𝐼𝐴 and 𝐼𝐼𝐵 that vanish along the three sides of Ξ2𝐷 to satisfy the continuity
of displacements across cross-sectional subdomains:
𝑇
1−𝜂 𝛿1𝑠 0 0 0 𝑃𝑟 𝜉
𝐼𝐼𝐴,𝐼𝐼𝐵 1 1+𝜉 0 𝛿2𝑠 0 0 𝑃𝑟 𝜂
𝐹𝜏 =
2 1+𝜂 0 0 𝛿3𝑠 0 𝑃𝑟 −𝜉
1−𝜉 0 0 0 𝛿4𝑠 𝑃𝑟 −𝜂
 Theory of unified formulation using Serendipity Lagrange
Functions
Serendipity Lagrange Functions
𝐼𝐼𝐴,𝐼𝐼𝐵
where 𝛿𝑖𝑗 is the Kronecker delta and the argument of 𝑃𝑟 −𝜉 and 𝑃𝑟 −𝜂 is negative to ensure that all 𝐹𝜏
polynomials of odd order are identical and separated by a rotation of 90 degree; a property of shape functions to
ensure uniqueness and completeness.

• A set of 𝑁 𝑡ℎ order polynomials named type 𝐼𝐼𝐼 defined in the interior subset of Ξ2𝐷 that vanish along its four
sides.
𝐼𝐼𝐼
𝐹𝜏 = 𝑃𝑛 𝜉 𝑃𝑚 𝜂

with 𝑛, 𝑚 = 2,3, … , 𝑁, constrained by 𝑛 + 𝑚 = 𝑟 and 𝑛 + 𝑚 ≤ 𝑁. Therefore, an 𝑁 𝑡ℎ order SLE model is implicitly

assumed to include all the shape functions of orders 1 to 𝑁

Stress-strain relation
𝜺 = 𝑫𝐹𝜏 (𝑥, 𝑧)𝒖𝜏 (y), ෩𝜺 ,
𝝈=𝑪

𝜕 𝜕 𝜕
0 0 0
𝜕𝑥 𝜕𝑧 𝜕𝑦
𝜕 𝜕 𝜕
𝑫𝑇 = 0 𝜕𝑦
0
𝜕𝑧
0
𝜕𝑥
𝜕 𝜕 𝜕
0 0 0
𝜕𝑧 𝜕𝑦 𝜕𝑥

SLE functions and SLE discretization method

 Strong formulation of the strain energy
Derivation of energy equations
From the principle of virtual work: 𝛿𝐿𝑖𝑛𝑡 − 𝛿𝐿𝑒𝑥𝑡 = 0,

𝛿𝐿𝑖𝑛𝑡 = ‫ 𝝈 𝑇𝜺 𝛿 𝑉׬‬dV = ‫ 𝝈 𝜏𝐹 𝑇𝑫 𝜏𝑇𝒖𝛿 𝑉׬‬dV

Applying partial integration by parts:

‫ 𝝈 𝜏𝐹 𝑇𝑫 𝜏𝑇𝒖𝛿 𝑉׬‬dV = − ‫ 𝝈 𝜏𝐹 𝑇𝑫 𝜏𝑇𝒖𝛿 𝑉׬‬dV + ‫ 𝝈 𝜏𝐹 𝑛𝑇𝑰 𝜏𝑇𝒖𝛿 𝑆׬‬dS
𝑦=𝐿
𝛿𝐿𝑖𝑛𝑡 = ‫ 𝑠𝒖 𝑠𝜏𝑲 𝜏𝑇𝒖𝛿 𝐿׬‬dy + 𝛿𝒖𝑇𝜏 𝜫𝜏𝑠 𝒖𝑠 ȁ𝑦=0 , where 𝑲𝜏𝑠 = 𝑫𝑇 𝐹𝜏 𝑪𝑫𝐹𝑠 , 𝜫𝜏𝑠 = 𝑰𝑇 𝐹𝜏 𝑪𝑫𝐹𝑠
Explicitly, the equilibrium equations for 𝑠, 𝜏 = 1, 2, … , 𝑀

𝜏𝑠
𝐾𝑥𝑥 𝜏𝑠
𝐾𝑥𝑦 𝜏𝑠
𝐾𝑥𝑧 𝜏𝑠
Π𝑥𝑥 𝜏𝑠
Π𝑥𝑦 𝜏𝑠
Π𝑥𝑧
𝜏𝑠 𝜏𝑠 𝜏𝑠 𝜏𝑠 𝜏𝑠 𝜏𝑠
𝑲𝜏𝑠 = 𝐾𝑦𝑥 𝐾𝑦𝑦 𝐾𝑦𝑧 , 𝚷𝜏𝑠 = Π𝑦𝑥 Π𝑦𝑦 Π𝑦𝑧 , 𝛿𝒖𝑇𝜏 : 𝑲𝜏𝑠 𝒖𝑠 (𝑦) = 𝒑𝑘𝑠 , 𝛿𝒖𝑇𝜏 : 𝜫𝜏𝑠 𝒖𝑠 (𝑦) = 0,
𝜏𝑠 𝜏𝑠 𝜏𝑠 𝜏𝑠 𝜏𝑠 𝜏𝑠
𝐾𝑧𝑥 𝐾𝑧𝑦 𝐾𝑧𝑧 Π𝑧𝑥 Π𝑧𝑦 Π𝑧𝑧

𝜏𝑠 𝜕 𝜕2
𝐾𝑥𝑥 = 𝐸𝜏55
,𝑧 𝑠,𝑧
+ 𝐸𝜏15
,𝑧 𝑠,𝑥
+ 𝐸𝜏15
,𝑥 𝑠,𝑧
+ 𝐸𝜏11
,𝑥 𝑠,𝑥
+ 𝐸𝜏56
,𝑧 𝑠
− 56
𝐸𝜏𝑠,𝑧
+ 𝐸𝜏16
,𝑥 𝑠
− 16
𝐸𝜏𝑠,𝑥 𝜕𝑦
− 66
𝐸𝜏𝑠 2.
𝜕𝑦

𝜏𝑠 = 𝐸 56 + 𝐸16 + 𝐸 66 𝜕
Π𝑥𝑥 𝜏𝑠,𝑧 𝜏𝑠,𝑥 𝜏𝑠 .
𝜕𝑦

𝑖𝑗 𝑖𝑗
𝐸𝜏,𝜗 𝑠,𝜇 = ‫׬‬Ω 𝐶ሚ𝑖𝑗 𝐹𝜏,𝜗 𝐹𝑠,𝜇 dΩ 𝐸𝜏𝑠,𝜇 = ‫׬‬Ω 𝐶ሚ𝑖𝑗 𝐹𝜏 𝐹𝑠,𝜇 dΩ.
𝑖𝑗 𝑖𝑗
𝐸𝜏,𝜗 𝑠 = ‫׬‬Ω 𝐶ሚ𝑖𝑗 𝐹𝜏,𝜗 𝐹𝑠 dΩ, 𝐸𝜏𝑠 = ‫׬‬Ω 𝐶ሚ𝑖𝑗 𝐹𝜏 𝐹𝑠 dΩ. 𝑖, 𝑗 = 1, 2, … , 6 and 𝜗, 𝜇 = 𝑥, 𝑧
 Strong formulation of the strain energy
Discretization along the beam by DQM
𝑦=𝐿
𝑲𝑠𝜏(𝑘) σ𝑚 𝑖𝜏
𝑖=1 𝐿𝑖 𝑦 𝒖𝑖𝜏 = 𝐩𝑘 , 𝚷 𝑠𝜏(𝑘) σ𝑚
𝑖=1 𝐿𝑖 𝑦 𝒖𝑖𝜏 ȁ𝑦=0 = 0

𝐿𝑖 𝑦 is the high order Lagrange polynomial which is 𝑚 − 1 th order accurate everywhere in the domain.
𝑦 =𝐿
𝑲𝑠𝜏𝑖𝑗(𝑘) 𝒖𝑖𝜏 = 𝐩𝑖𝜏
𝑘, 𝚷 𝑠𝜏𝑖𝑗(𝑘) 𝒖𝑖𝜏 ȁ𝑦𝑖𝑖=0 = 0

𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖

𝐾𝑥𝑥 𝐾𝑥𝑦 𝐾𝑥𝑧 Π𝑥𝑥 Π𝑥𝑦 Π𝑥𝑧
𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖 𝑦 =𝐿
𝑲𝑠𝜏𝑗𝑖 = 𝐾𝑦𝑥 𝐾𝑦𝑦 𝐾𝑦𝑧 , 𝚷 𝑠𝜏𝑗𝑖 = Π𝑦𝑥 Π𝑦𝑦 Π𝑦𝑧 , 𝛿𝒖𝑇𝜏 : 𝑲𝑠𝜏𝑖𝑗(𝑘) 𝒖𝑖𝜏 = 𝐩𝑖𝜏
𝑘, 𝛿𝒖𝑇𝜏 : 𝚷 𝑠𝜏𝑖𝑗(𝑘) 𝒖𝑖𝜏 ȁ𝑦𝑖𝑖=0 = 0
𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖
𝐾𝑧𝑥 𝐾𝑧𝑦 𝐾𝑧𝑧 Π𝑧𝑥 Π𝑧𝑦 Π𝑧𝑧
𝜏𝑠 = 𝐸 55 + 𝐸15 + 𝐸15 + 𝐸11 56 56 16 16 66 1 2
𝐾𝑥𝑥 𝜏,𝑧 𝑠,𝑧 𝜏,𝑧 𝑠,𝑥 𝜏,𝑥 𝑠,𝑧 𝜏,𝑥 𝑠,𝑥 + 𝐸𝜏,𝑧 𝑠 − 𝐸𝜏𝑠,𝑧 + 𝐸𝜏,𝑥 𝑠 − 𝐸𝜏𝑠,𝑥 𝑎𝑖𝑗 − 𝐸𝜏𝑠 𝑎𝑖𝑗 .

𝜏𝑠 = 𝐸 56 + 𝐸16 + 𝐸 66 𝑎 . 1
Π𝑥𝑥 𝜏𝑠,𝑧 𝜏𝑠,𝑥 𝜏𝑠 𝑖𝑗

1
1 𝑆 1 𝑦𝑖 2 1 1 𝑎𝑖𝑗
𝑎𝑖𝑗 = 𝑎𝑖𝑗 = 2 𝑎𝑖𝑗 𝑎𝑖𝑖 − 𝑆 1 = ς𝑁
𝑗=1 𝑗≠𝑖 𝑦𝑖 − 𝑦𝑗
𝑆 1 𝑦𝑗 𝑦𝑖 −𝑦𝑗 𝑦𝑖 −𝑦𝑗
 Weak formulation of the strain energy

Derivation of energy equations

𝒖 = 𝐹𝜏 (𝑥, 𝑧) σ𝑚
𝑖=1 𝐿𝑖 𝑦 𝒖𝑖𝜏

෩ 𝑫𝑖𝜏 𝒖𝑖𝜏
𝜺 = 𝑫𝑖𝜏 𝒖𝑖𝜏 , 𝝈 = 𝑪

From the principle of virtual work: 𝛿𝐿𝑖𝑛𝑡 − 𝛿𝐿𝑒𝑥𝑡 = 0,

𝑇 𝑖𝑗𝜏𝑠
𝛿𝐿𝑖𝑛𝑡 = ‫ 𝝈 𝑇𝜺 𝛿 𝑉׬‬dV = ‫ 𝝈 𝜏𝐹 𝑇𝑫 𝜏𝑇𝒖𝛿 𝑉׬‬dV = ‫𝜏𝑗𝒖 𝛿 𝑙׬‬ 𝑲 𝒖𝑖𝑠 dy

Explicitly, the equilibrium equations for 𝑠, 𝜏 = 1, 2, … , 𝑀

𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖

𝐾𝑥𝑥 𝐾𝑥𝑦 𝐾𝑥𝑧
𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖
𝑲𝑠𝜏𝑗𝑖 = 𝐾𝑦𝑥 𝐾𝑦𝑦 𝐾𝑦𝑧 , 𝛿𝒖𝑇𝜏 : 𝑲𝑠𝜏𝑖𝑗(𝑘) 𝒖𝑖𝜏 = 𝐩𝑖𝜏
𝑘,
𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖 𝑠𝜏𝑗𝑖
𝐾𝑧𝑥 𝐾𝑧𝑦 𝐾𝑧𝑧

𝑖𝑗𝜏𝑠
𝐾𝑥𝑥 = 𝐸𝜏55 𝑠
,𝑧 ,𝑧
+ 𝐸𝜏
15
𝑠
,𝑧 ,𝑥
+ 𝐸𝜏
15
𝑠
,𝑥 ,𝑧
+ 𝐸𝜏
11
𝑠
,𝑥 ,𝑥
𝐻𝑖𝑗 + 𝐸𝜏
56
,𝑧 𝑠 + 𝐸𝜏
16
,𝑥 𝑠 𝐻𝑖𝑗,𝑦
+ 𝐸 56
𝜏𝑠,𝑧
+ 𝐸 16
𝜏𝑠 ,𝑥
𝐻𝑖,𝑦 𝑗 + 𝐸 66
𝜏𝑠 𝐻𝑖,𝑦 𝑗,𝑦

𝐻𝑖,𝑦 𝑗,𝑦 = ‫׬‬L 𝐿𝑖,𝑦 𝐿𝑗,𝑦 dy, 𝐻𝑖𝑗,𝑦 = ‫׬‬L 𝐿𝑖 𝐿𝑗,𝑦 dy

𝐻𝑖,𝑦 𝑗 = ‫׬‬L 𝐿𝑖,𝑦 𝐿𝑗 dy, 𝐻𝑖𝑗 = ‫׬‬L 𝐿𝑖 𝐿𝑗 dy. 𝑖, 𝑗 = 1, 2, … , 𝑛

 Validation of the strong and weak formulations

A simply supported composite beam of different stacking

sequences under sinusoidal loads at the top and bottom surfaces.

𝑚𝑎𝑥
Laminate Model 𝜎ത𝑦𝑦 ҧ 𝑚𝑎𝑥
𝜏𝑦𝑦

A Pagano 0.7913 3.3167

FEM 0.7913 (0.0000) 3.3167 (0.0000)
Material Ex Ey Ez Gyz Gxz Gxy 𝜈yz 𝜈xz 𝜈xy
DQM 0.7913 (0.0000) 3.3167 (0.0000)
GPa HOFEM 0.7913 (0.0000) 3.3167 (0.0000)
B Pagano 0.8672 3.2228
p 1.0 25.0 1.0 0.5 0.2 0.5 0.25 0.25 0.01
FEM 0.8672 (0.0000) 3.2228 (0.0000)
DQM 0.8672 (0.0000) 3.2228 (0.0000)
Laminate Layer thickness ratio Materials Stacking sequence HOFEM 0.8672 (0.0000) 3.2228 (0.0000)
C Pagano 1.6307 5.3340
S 0.23 p3 03 FEM 1.6307 (0.0000) 5.3340 (0.0000)

Symmetric DQM 1.6307 (0.0000) 5.3340 (0.0000)

HOFEM 1.6307 (0.0000) 5.3340 (0.0000)
A 1Τ3 3 p3 0/90/0
I Pagano 2.0870 4.8799
B 0.25 p5 0/90/0/90/0
FEM 2.0870 (0.0000) 4.8789 (0.001)
C 0.25 p5 90/0/90/0/90 DQM 2.0870 (0.0000) 4.8771 (0.0028)

Anti-Symmetric HOFEM 2.0870 (0.0000) 4.8771 (0.0028)

J Pagano 1.2175 4.3539
FEM 1.2175 (0.0000) 4.3538 (0.0001)
I 0.3/0.7 p2 0/90
DQM 1.2175 (0.0000) 4.3539 (0.0000)
J 0.254 p4 0/90/0/90 HOFEM 1.2175 (0.0000) 4.3539 (0.0000)
 Computational efficiency

Direct solvers Iterative DOFs Laminate Model Bandwidth Conditioning* PD DOFs

Laminates Model solvers Space A FEM 679 10−8 Yes 15,561
Time~𝑂 𝑛𝑏 2 Time*~𝑂 𝑛2 ~𝑂 𝑛𝑏 𝑛 DQM 1078 10−6 No 1,197
A FEM 109 108 107 15,561 HOFEM 1135 10−5 Yes 1,197
DQM 109 106 106 1,197 B FEM 1083 10−8 Yes 24,843
HOFEM 109 106 106 1,197 DQM 1720 10−6 No 1,911
B FEM 1010 108 107 24,843 HOFEM 1811 10−5 Yes 1,911
DQM 109 106 106 1,911 G FEM 1487 10−10 Yes 34,125
HOFEM 109 106 106 1,911 DQM 2362 10−9 No 2,625
G FEM 1010 109 107 34,125 HOFEM 2483 10−7 Yes 2,625
DQM 1010 106 106 2,625 K FEM 881 10−9 Yes 20,202
HOFEM 1010 106 106 2,625 DQM 1399 10−7 No 1,554
K FEM 1010 108 107 20,202 HOFEM 1473 10−6 Yes 1,554
DQM 109 106 106 1,554
HOFEM 109 106 106 1,554
Comparison between strong and weak formulations

Strong form Weak form

Good convergence but slower Excellent convergence with
than weak form for small degrees small degrees of freedom.
convergence of freedom.

Good numerical stability of the A slightly better numerical

Numerical stability stiffness matrix but not positive stability of the stiffness matrix
definite. than strong form and positive
definite.
Computationally efficient with Computationally efficient with a
Efficiency low pre-processing times. higher pre-processing times
compared to strong form.

Not suitable to treat finite Suitable for finite discontinuities

discontinuity. of various degrees.
Treatment of finite
discontinuities
 Multidomain DQM for treatment of discontinuity
Variation of strain energy

‫𝝈 𝜏𝐹 𝑇𝑫 𝜏𝑇𝒖𝛿 𝑉׬‬ 𝑚
dV = − ‫𝝈 𝜏𝐹 𝑇𝑫 𝜏𝑇𝒖𝛿 𝑉׬‬ 𝑚
dV + ‫׬‬Γ 𝛿𝒖𝑇𝜏 𝑰𝑇𝑛 𝐹𝜏 𝝈 𝑚
dΓ 𝑏 𝑖𝑛𝑡
𝑏 𝑖𝑛𝑡

Γ 𝑏 𝑖𝑛𝑡 represents either the strain energy at the boundary or at the interface.

ഥ
𝝈 𝑚 ഥ
=𝑪 𝑚 𝜺𝑚

න 𝛿𝒖𝑇𝜏 𝑫𝑇 𝐹𝜏 𝝈 dV
𝑉

= − න 𝛿𝒖𝑇𝜏 𝑫𝑇 𝐹𝜏 𝝈 𝑚 dV + න 𝛿𝒖𝑇𝜏 𝑰𝑇𝑛 𝐹𝜏 𝝈 𝑚 dΓ 𝑏 + න 𝛿𝒖𝑇𝜏 𝑰𝑇𝑛 𝐹𝜏 𝝈 𝑚 Γ 𝑖𝑛𝑡

𝑉 Γ𝑏 Γ 𝑖𝑛𝑡
𝛿𝐿𝑖𝑛𝑡
Interfacial compatibility for heterogeneous beams
𝑚 𝑚+1 (𝑚) (𝑚+1)
𝒖 ห𝑦=𝑦 =𝒖 ห𝑦=𝑦 , 𝜎𝑦𝑦 ቚ = 𝜎𝑦𝑦 ቚ
𝑖𝑛𝑡 𝑖𝑛𝑡 𝑦=𝑦𝑖𝑛𝑡 𝑦=𝑦𝑖𝑛𝑡
(𝑚) (𝑚+1) (𝑚) (𝑚+1)
𝜏𝑦𝑧 ቚ = 𝜏𝑦𝑧 ቚ , 𝜏𝑥𝑧 ቚ = 𝜏𝑥𝑧 ቚ
𝑦=𝑦𝑖𝑛𝑡 𝑦=𝑦𝑖𝑛𝑡 𝑦=𝑦𝑖𝑛𝑡 𝑦=𝑦𝑖𝑛𝑡

(𝑚)𝑇 ഥ 𝑚 (𝑚) (𝑚)𝑇 𝜏𝑠(𝑚) (𝑚)

𝛿𝐿𝑖𝑛𝑡 = ‫׬‬Γ 𝛿𝒖𝜏 𝐹𝜏 𝑪 𝑫𝐹𝑠 𝒖𝑠 Γ 𝑖𝑛𝑡 , 𝛿𝐿𝑖𝑛𝑡 = 𝛿𝒖𝜏 𝑲𝑖𝑛𝑡 𝒖𝑠
𝑖𝑛𝑡

𝜏𝑠(𝑚) (𝑚) 𝜏𝑠(𝑚+1) (𝑚+1)

𝑲𝑖𝑛𝑡 𝒖𝑠 ቚ = 𝑲𝑖𝑛𝑡 𝒖𝑠 ቚ
𝑦=𝑦𝑖𝑛𝑡 𝑦=𝑦𝑖𝑛𝑡
 Application of MDQM to modelling of patched laminates
Features of pathed laminates
• Efficient way to reduce fibre consumption in load path optimized design.
• limitations associated with process induced defects in continuous fibre technology is eliminated.
• Mechanical properties are strongly dependent on the distribution of discontinuities in the patched laminates.
• Numerical models for stress analysis of patched laminates is an emerging research interest.

Patched composite configuration with resin pockets

Test layup Sequence Load type BC

1 Symmetric 90 0 90 0 90 0 𝑠 Axial compression C-F
 Validation of MDQM and HOFEM for patched laminate analysis

Axial stresses at different sections of the patched composite for ABAQUS, MDQM and HOFEM models
 Potential areas of collaboration with DRIVEN project

Finite discontinuity
• What prospects lie in the handling of discontinuity of various degrees, regarding
geometry and material of the structure?
• How effective and accurate can we treat different boundary conditions?
Numerical stability and accuracy
• Exploiting the merits of the monotonic and variation diminishing properties of
NURBS for improved numerical stability and accuracy.

Function approximation with Lagrange and NURBS Structural discontinuities

 Potential areas of collaboration with DRIVEN project

Refinement process
• Methods of DQM and HOFEM uses 𝑝-refinement along the beam
to increase the accuracy of the system solution. Can we efficiently
explore the computational gain in 𝑘 -refinement technique as
opposed to 𝑝-refinement?
• SLE functions are accurate but numerical stability and
computational effort can still rise significantly subject to increased
order.
Irregular geometry
• Modelling of non-prismatic and non-straight beams.
 Model reduction of structural models

Original high order model Motivation for MOR

Linear first order system • Low fidelity model with reduced computational effort
𝐸 𝝌ሶ + 𝑨𝝌 = 𝑭𝑢(𝑡) ∈ ℝ𝑁
• Possibility to adapt to parameter dependent systems
Linear second order system • Structure preserving capacity of reduced models
𝑴𝝌ሷ + 𝑫𝝌ሶ + 𝑲𝝌 = 𝑭𝑢(𝑡) ∈ ℝ𝑁

Nonlinear second order system

𝑴𝒙ሷ = 𝒇𝑒𝑥𝑡 + 𝒇𝑖𝑛𝑡 (𝒙) ∈ ℝ𝑁

Reduced low order model via a projector 𝑽

Linear first order system

𝑬𝑟 𝝌ሶ 𝑟 + 𝑨𝑟 𝝌𝑟 = 𝑭𝑟 𝑢(𝑡) ∈ ℝ𝑛

Linear second order system

𝑴𝑟 𝝌ሷ 𝑟 + 𝑫𝑟 𝝌ሶ 𝑟 + 𝑲𝑟 𝝌𝑟 = 𝑭𝑟 𝑢(𝑡) ∈ ℝ𝑛

Nonlinear second order system

𝑴𝑟 𝒛ሷ 𝑟 = 𝒇𝑒𝑥𝑡 𝑖𝑛𝑡
𝑟 + 𝒇𝑟 (𝒙) ∈ ℝ𝑛 s.t. 𝑛 ≪ 𝑁
 Potential areas of collaboration with DRIVEN project

Model order reduction applied to variable composite structures

• Previous experience involves Proper Orthogonal Decomposition (POD), method of Krylov subspace, Discrete

Empirical Interpolation Method (DEIM) for nonlinear systems, structure-preserving techniques.

• Exploring parametric model reduction methods for variable composite structures.

 Acknowledgment

The DRIVEN project has received funding from the European Union’s
Horizon 2020 research and innovation programme under grant
agreement No 811099.

The Temporally VARIable COMPosite (VARICOMP) Grant No.

(15/RP/2773) project acknowledge funding from the Science
Foundation Ireland (SFI) under its Research Professor programme.