i i

“master” — 2020/5/14 — 10:02 — page 1 — #1

i i

0.0.1 Inertia relief

Inertia relief1 refers to an analysis procedure that allows unconstrained
systems – or systems otherwise susceptible to stress-free motions – to
be subjected to a quasi-static analysis by taking rigid body inertia
forces into account.
Conventional static analysis cannot be performed for such systems
since, in the absence of constraints, the stiffness matrix is singular. The
structure response is measured relative to a steady state accelerating
frame, whose motion is induced by the (usually nonzero) external load
resultants.
The inertia relief solution procedure provides for three steps, namely
i) the rigid body mode evaluation, ii) the assessment of the inertia re-
lief loads, and iii) the solution of a supported, self-equilibrated static
loadcase within the moving frame.
A set of nodal degree of freedom (dof)s is supplied, one each ex-
pected rigid body motion, whose imposed displacements values uniquely
define the structure positioning in space; also, they may be employed in
supporting the structure to untangle the stiffness matrix rank-deficiency.
The t l rigid body modes are evaluated by sequentially setting each
of these support dof to unity, while retaining the others to zero, and
solving for the system of nodal equilibrium equations

K d = F, (1)

where K is the structure stiffness matrix, in the absence of further

external loads. Since the tied/retained condition of the structure dofs
does not vary throughout the sequence of aforementioned loadcases,
comprised of the final step introduced in the following, a single L L >
Cholesky system matrix decomposition is required by the procedure,
whose computational burden is thus not significantly increased with
respect to the usual static solution.
A rigid body, steady state acceleration field is defined as the linear
1
XXX some cut and paste from the MSC.Marc vol A manual, please rewrite as
required to avoid copyright infringement.

i i

i i
 i i

“master” — 2020/5/14 — 10:02 — page 2 — #2

i i

combination of the so defined t l rigid body modes

..
 
  . 
d̈ = · · · t l · · ·  αl , (2)
} . 
..
| {z
T
| {z }
α

whose αl coefficients define the modal acceleration vector α . Those

acceleration terms are then evaluated according to the inertial equilib-
rium of the structure under the applied F external loads, condition,
this, that may be stated as

T>M T α = T>F (3)

The projection of the equilibrium equations onto the subspace defined

by the linear span of the t l rigid body mode vectors – i.e. the left
multiplication of both the equation sides by the T > matrix, is solved
in place of the overdetermined linear system

M T α = F [+ R l ]

since the R l reaction forces associated to the rigid body constraints

balance the equilibrium residual components that are orthogonal2 to
such allowed configuration subspace.
The inertia relief forces may then be quantified as M T α , and su-
perposed to the initial external loads, thus leading to a self equilibrated
loading condition in the context of the steady state accelerating frame;
by employing the support dofs to establish a positioning constraint
set, the elastic problem may finally be solved in the form

K d = F − M T α, (4)

The d displacement components are expressed with respect to a ref-

erence frame that clings to the possibly accelerating structure through
the support dofs; due to the self-equilibrated nature of the applied
loads in the moving frame, reaction forces at supports are zero.
As a closing comment, the MSC.Marc solver employs a lumped
definition for the system mass matrix for evaluating inertia relief forces.
2
We note that T > R l = 0 is a smooth constraint condition, i.e. the R l reactions
are work-orthogonal to the allowed motions.

i i

i i
 i i

“master” — 2020/5/14 — 10:02 — page 3 — #3

i i

0.0.2 Harmonic response analysis

The equilibrium equations of a multiple dof system subject to elastic,
inertial and viscous actions may be stated in the general form

M d̈ + C ḋ + K d = f (t), d = d (t) (5)

where:

• M is the mass matrix, which is symmetric and positive definite;

• C is the viscous damping matrix, which is symmetric and posi-

tive semidefinite;

• K is the elastic stiffness matrix, which is symmetric and posi-

tive semidefinite: complex terms may appear within the stiffness
matrix to represent structural damping contributions;

• f (t) is the vector of the external (generalized) forces;

• d (t) collects the system dofs, which vary in time.

The system response is assumed linear – a strong assumption, this,

that hardly holds in complex structures as the automotive chassis under
scrutiny. The lack of nonlinear analysis tools whose modeling and
computational effort is comparable with respect to the one presented in
the present section, pushes for some laxity in the linearity prerequisite
check, and for the acceptance of a certain extent of error.
The applied force is assumed periodic in time, and so is the long
term solution, if linearity holds. Moreover, Fourier decomposition may
be applied, and there is no lack in generality in further assuming an
harmonic forcing term, and hence an harmonic solution. We have

f̄ ejωt + f̄ ∗ e−jωt
f (t) = = Re( f̄ ejωt ) (6)
2
where the asterisk superscript denotes the complex conjugate variant
of the base vector. We recall that the compact notation

f (t) = f̄ ejωt (7)

i i

i i
 i i

“master” — 2020/5/14 — 10:02 — page 4 — #4

i i

extensively employed below defines a complex form for the driving

force, whose real part is the portion which is physically applied to
the nodes over time, i.e.

Re( f̄ ejωt ) = Re( f̄ ) cos ωt − Im( f̄ ) sin ωt (8)

This compact formalism is not rigorous but still it is effective, and

hence commonly employed. Any phase difference amongst the applied
nodal excitations may be described by resorting to the complex nature
of the f̄ vector terms.
In the neglection of the transient response, the harmonic tentative
solution
d (t) = d̄ ejωt (9)
is substituted within Eq. 5, thus obtaining

−ω 2 M + jω C + K d̄ = f̄


(10)

where the ejωt time varying, generally nonzero factors are simplified
away.
Expression 11 defines a system of linear complex equations, one
each dof, in the complex unknown vector d̄ ; equivalently, each com-
plex equation and each unknown term may be split into the associated
real and imaginary parts, thus leading to a system of linear, real equa-
tions whose order is twice the number of the discretized structure dofs.
The system matrix varies with the ω parameter, and in particular
its stiffness contribute K is dominant for low ω values, whereas the
C , M terms acquire relevance with growing ω.
In distributed inertia systems, however, it is a misleading claim
that the stiffness matrix contribution becomes negligible with high ω
values, since – with the notable exception of external loads that are
directly applied to concentrated masses or rigid bodies – the pulsation
is unphysically high above which such behaviour arises.
Since Eqns. 11 are independently solved for each ω value, it con-
stitutes no added complexity to let M , C , K and f̄ vary according to
the same parameter.
Finally, in the absence of the damping-related imaginary terms
within the system matrix, the Eq. 11 problem algebraic order is led
back to the bare number of system dofs; in fact, two independent real

i i

i i
 i i

“master” — 2020/5/14 — 10:02 — page 5 — #5

i i

system of equations – that share a common L L > matrix decompo-

sition – may be cast for the real and the imaginary parts of d̄ and
f̄ .

0.0.3 Modal analysis

The present paragraph briefly deals with the structure’s natural modes,
i.e. those periodic3 motions that are allowed according to Eq. 5, in the
further absence of externally applied loads.
A necessary condition for a motion to endure in the absence of
a driving load is the absence of dissipative phenomena; it is hence
necessary to have a zero C damping matrix, whereas the K stiffness
matrix must be free of imaginary terms. This hypothesis holding, Eq.
11 is reduced to the following real-term algebraic form

−ω 2 M + K d̄ = 0


(11)

whose nontrivial solutions constitute a set of (ωi2 , d̂ i ) generalized eigen-

value/eigenvector pairs, one each system dof, if eigenvalue multiplicity
is taken into account.
In the context of each (ωi2 , d̂ i ) pair, ωi is the natural pulsation
(ωi = 2πfi , where fi is the natural frequency), whereas the d̂ i vector
of generalized displacemts is named natural mode.
The extraction of the Eq. 11 nontrivial solutions reduces to a stan-
dard eigenvalue problem is the algebraic form is left-multiplied by the
mass matrix inverse, i.e.

M −1 K − ω 2 I d̂ = 0 ;


(12)

the availability of solvers that specifically approach the generalized

problem avoid such computationally uneconomical preliminary.
It is worth to recall that in the case of eigenvalues with non-unit
multiplicity – concept, this, that is to be contextualized within the lim-
ited precision floating point arithmetics4 – the associated eigenvectors
must be considered only through their linear combination; the specific
selection of the base elements for representing such a subspace (i.e.,
3
harmonic in the context of linearly behaving systems
4
XXX

i i

i i
 i i

“master” — 2020/5/14 — 10:02 — page 6 — #6

i i

each single eigenvector) derives in fact from the unpredictable inter-

action between the truncation error and the inner mechanics of the
numerical procedure.
Also, the eigenvectors that are associated to eigevalues of unit mul-
tiplicity are returned by the numerical solver in the misleading form of
a definite vector, whereas an arbitrary (both in sign and magnitude)
scaling factor has to be prepended.
In particular, any speculation which is not robust with respect to
such arbitrary scaling (or combination) is of no engineering relevance,
and must be avoided.
Finally, in continuous elasticity, no upper bound exists for natural
frequencies; in Finite Element (fe) discretized structure, an apparent
upper bound exists, which depends on local element size5 .
A common normalizing rule for the natural modes is the one that
produces a unit modal mass mi , i.e.

mi = d̂ >
i M d̂ i = 1 (13)

this rule is e.g. adopted by the MSC.Marc solver in its default config-
uration.
The resonant behaviour of the system in correspondence with a
natural frequency may be investigated by substituting the following
tentative solution
x (t) = a d̂ i sin(ωi t) (14)
within the dynamic equilibrium equations 5, with

f (t) = f̂ cos(ωi t), (15)

and thus obtaining

−ωi2 M + K d̂ i ai sin(ωi t) + ωi ai C d̂ i cos(ωi t) = f̄ cos(ωi t). (16)

| {z }
=0

By simplifying away the generally nonzero time modulating factors,

and by left-multiplying both equation sides by d̂ > – i.e. by projecting
5
In particular, the natural oscillation period for the highest dynamic mode is
estimated with order of magnitude precision as the minimum time it takes a pressure
wave to travel between two different nodes in the discretized structure.

i i

i i
 i i

“master” — 2020/5/14 — 10:02 — page 7 — #7

i i

the equation residual along the subspace defined by the eigenvector

itself, we obtain an amplitude term in the form

d̂ > f̄
ai = (17)
ωi d̂ > C d̂ i

whose singularity is prevented only a) in the presence of a damping

matrix that associates nonzero and non-orthogonal viscous reactions
to the motion described by the natural mode under scrutiny, or b) if
the driving load is orthogonal to such natural mode, i.e. it unable to
perform periodic work on such a motion. The nature of the expression
17 numerator will be further discussed in the following paragraph.

0.0.4 Harmonic response through mode superposition

In the case the eigenvalues associated with the dynamic modes are all
distinct6 , the following orthogonality conditions hold
2
d̂ >
j M d̂ i = mi δij d̂ >
j K d̂ i = mi ωi δij (18)

where δij is the Kroneker delta function, and mi = 1 is the i-

th modal mass, which is unitary due to the the d̂ i unit modal mass
normalization.
It is further assumed that it is possible to describe the elastic body
motion through a linear combination of a (typically narrow) subset of
the dynamic natural modes. Such assumption may be rationalized in
two equivalent ways: on one hand, the contribution of the neglected
modes is assumed negligible, and hence ignored; on the other hand, it
is imagined that a set of kinematic constraints is imposed, that rigidly
impede any additional system motion with respect to the chosen set.
According to this latter explanation, reaction forces will be raised that
absorb any equilibrium residual term which is orthogonal with respect
to the allowed displacements.
The subset defined by the first m eigenvectors (1 ≤ m  n) are
commonly employed, whereas different assortments are possible; a con-
trol calculation perfomed with a wider base may be employed for error
estimation.
6
condition, this, that is assumed to hold; a slightly perturbed FE discretiza-
tion may be effective in separating the instances of a theoretically multiple natural
frequency.

i i

i i
 i i

“master” — 2020/5/14 — 10:02 — page 8 — #8

i i

By stacking those first m normalized column eigenvectors into the

Ξ matrix below,
 
Ξ = d̂ 1 · · · d̂ l · · · d̂ m , (19)

any d̄ configuration belonging to the linear span of the selected modes

may be expressed through a vector of m modal coordinates ξ¯l , as in

d̄ = Ξ ξ¯ (20)

.
Due to the natural modes orthogonality conditions 18, the Ξ tran-
formation matrix diagonalizes both the mass and the stiffness matrices,
since

Ξ>M Ξ = I Ξ > K Ξ = Ω = diag(ωl2 ); (21)

by appling such transformation to the damping matrix, however, a

dense matrix is generally obtained.
The Rayleigh or proportional damping matrix definition assumes
that the latter may be passably represented as a linear combination of
the mass matrix and of the stiffness matrix: in particular

C = αM + βK (22)

where α and β are commonly named mass and stiffness matrix multi-
pliers, respectively; according to such assumption, the damping matrix
is also diagonalized by the Ξ tranformation matrix.
Equation 11 algebraic problem may be cast in terms of the m ξl
modal unknowns, thus obtaining

Ξ > −ω 2 M + jω C + K Ξ ξ¯ = Ξ > f̄


(23)

which reduces to the diagonal form

−ω 2 I + jω α I + β Ω + Ω ξ¯ = Ξ > f̄ ,



(24)

or, equivalently, to the set of m independent complex equations

−ω 2 + jω α + βωl2 + ωl2 ξl = ql , j = 1 . . . m



(25)

i i

i i
 i i

“master” — 2020/5/14 — 10:02 — page 9 — #9

i i

D E
where ql = d̂ l , f̄ is the coupling factor between the external load
and the l-th natural mode.
The algebraic equation above may be interpreted as the charac-
teristic equation of an harmonically driven single dof oscillator that
exhibits the following properties:

• its mass is unity;

• its natural frequency equals that of the l-th natural mode ωl ;

• its damping ratio ζl is a combination of the two Rayleigh damping

coefficients, i.e.  
1 α
ζl = + βωl ;
2 ωl

• the external load real(imaginary) term is defined as the cyclic

work that the external load performs upon a system motion de-
scribed as the sinusoidal (cosinusoidal) modulation in time of the
l-th modal shape, divided by π.7

The uncoupled equations 25 may be solved resorting to complex

division arithmetics, thus leading to the definition of the ξ¯l modal am-
plitude and phase terms; in particular we have that the l-th modal
shape is modulated in time according to the function

ξl (t) = Re(ξ¯l ) cos ωt − Im(ξ¯l ) sin ωt

= ξ¯l cos (ωt + ψl − φl )

whose terms are detailed in the following.

The auxiliary parameters
ω
al = 1 − rl2 bl = 2ζl rl rl =
ωl
are first defined; we then have the oscillation amplitude and phase
7
In the case of a concentrated load that act on a single dof, qj equates the prod-
uct of the load magnitude with the associated component in d̂ l , i.e. the generalized
displacement at the specific node, as shown by the FE postprocessor.

i i

i i
 i i

“master” — 2020/5/14 — 10:02 — page 10 — #10

i i

terms
|q̄l | 1
ξ¯l = 2 q
ωl a2l + b2l
ψl = arg(q̄l )
φl = arg(al + jbl )

or, equivalently, the real and imaginary parts

1 al Re(q̄l ) + bl Im(q̄l )
Re(ξ¯l ) =
ωl2 a2l + b2l
1 al Im(q̄l ) − bl Re(q̄l )
Im(ξ¯l ) = 2 .
ωl a2l + b2l

10

i i

i i
 i i

“master” — 2020/5/14 — 10:02 — page 11 — #11

i i

11

i i

i i
 i i

“master” — 2020/5/14 — 10:02 — page 12 — #12

i i

Bibliography

[1] T. J. Hughes and T. Tezduyar, “Finite elements based upon mindlin

plate theory with particular reference to the four-node bilinear
isoparametric element,” Journal of applied mechanics, vol. 48, no. 3,
pp. 587–596, 1981.

12

i i

i i