Chapter VIII: Slew Maneuver Dynamics
8.3 Determination of Kelvin-Voigt Damping Coefficients
is due to either the available torque or momentum, not the structure, regardless of the requirement on the residual angular velocity. Similarly, with the larger reaction wheel and the coarser pointing requirement, Fig. 8.5b shows that the available torque again drives the minimum slew time. The only exception is the case with the larger reaction wheels and the finer pointing requirement; see Fig. 8.5d. In this case, the structure-based performance limit constrains the minimum slew time for spacecraft at length scales below approximately 40 m. Above 40 m, the slew times are again torque-constrained. Even so, below the crossover point, the structure-based performance limit results in minimum slew times on the same order of magnitude as those from the torque limit.
Decreasing either the fraction of the momentum and torque available for slews or the maximum momentum and torque shifts the corresponding curves in Fig. 8.5 up. Similarly, decreasing the requirement on the amplitude of the residual angular velocity also shifts the corresponding curves in Fig. 8.5 up. Increasing the maximum momentum and torque is likely to require replacing reaction wheels with control moment gyroscopes.
Based on Fig. 8.5, the capabilities of each spacecraftβs attitude control system are often significantly more limiting than the dynamics of the structure. When this is the case, the results suggest that a lighter-weight, less-stiff, and lower-cost structure can be used to shift the structure-based performance limit closer to that of the attitude control system, at least as far as slewing is concerned. The figure likewise emphasizes that SSPP-like flexible spacecraft can likely achieve slew times on the order of 10 min or less for 90 deg, single-axis maneuvers at length scales as large as 50 m. If this is indeed the case, these slew times are realistically an order of magnitude or more faster than the existing state-of-practice. To gain confidence in these results, Sec. 8.4 compares the predictions from the reduced-order model for the 24 mΓ24 m spacecraft with those from geometrically nonlinear simulations of the corresponding full finite element model. In the interim, Sec. 8.3 develops the optimization approach for determining the viscoelastic damping coefficients that are inputs to the full finite element model.
simulations in Sec. 8.4.
Unlike linear finite element formulations, where it is straightforward to determine a global damping matrix, geometrically exact finite element formulations require modifications at the element level to guarantee that the damping formulation is in- variant to superposed rigid body motions. Invariant damping formulations only di- rectly dissipate energy associated with the elastic motion. A recent numerical study demonstrates that these formulations are important for correctly modeling damping effects in large-deformation simulations [259]. However, these damping models are often troublesome due to the difficulties associated with determining appropriate damping coefficients [260]. To that end, this section proposes an optimization-based approach for determining these coefficients. Even though this section focuses on geometrically exact beam finite elements, the approach readily generalizes to other types of geometrically exact finite elements (e.g., plates or shells) by appropriately modifying the tangent damping matrix.
The simplest damping model for geometrically exact finite elements is referred to as Kelvin-Voigt damping. For geometrically exact beams, Kelvin-Voigt damping augments the constitutive relation [Eq. (4.42)] for the force resultantNand moment resultantM with terms proportional to the material strain rate πͺΒ€ and the material curvature rateKΒ€ [148]. In other words,
S=CE+DEΒ€ (8.1)
whereSπ = Nπ,Mπ
, Eπ = πͺπ,Kπ
, Cβ R6Γ6is the sectional stiffness matrix, and D β R6Γ6 is the matrix of to-be-determined sectional damping coefficients.
To the authorβs knowledge, there are two systematic approaches in the literature for determiningD, although most studies instead tend to use βreasonable guessesβ
[213] or sensitivity studies. The first approach [261] derives closed-form expres- sions for the damping coefficients for geometrically exact beams with homogeneous, isotropic material properties. However, these expressions assume the availability of viscoelastic material properties, specifically the viscoelastic bulk and shear vis- cosities. These material properties are unavailable for the equivalent beam models of the strips, and hence, this approach is not applicable here. The second approach [213] applies modal analysis to the linearized partial differential equations govern- ing the dynamics of geometrically exact beams with simple boundary conditions to derive expressions for the unknown damping coefficients. The optimization- based approach developed here generalizes this approach to finite element models of arbitrary complexity.
The formulation of the optimization problem starts from the dynamic equilibrium equations for a nonlinear finite element model:
Finer(g,gΒ€,gΒ₯) +Fint(g,gΒ€) =Fext (8.2) wheregβRπis the vector of generalized coordinates,Finer(g,gΒ€,g) βΒ₯ Rπis the vector of generalized inertia forces,Fint(g,g) βΒ€ Rπis the vector of generalized viscoelastic forces, and Fext β Rπ is the vector of generalized external forces. The tangent damping matrixCπ βRπΓπthen follows as
Cπ(g) = πFint(g,gΒ€)
πgΒ€ . (8.3)
For simplicity (and without any loss of generality), these developments only consider generalized coordinates in a vector space and neglect external constraints, e.g., due to the joints in flexible multibody systems. The treatment of generalized coordinates in a Lie group entails straightforward modifications to Eq. (8.3) and what follows.
The formulation is independent of any external constraints on the system.
The tangent damping matrix Cπ for a finite element model with ππ elements is defined by a standard finite assembly step. Thus,
Cπ(g) =
ππ
Γ
π=1
LππCππ(Lπg)Lπ (8.4) whereπβ {1, 2, . . . , ππ};LπβRπ
π πΓπ
is the Boolean matrix that indexes the element nodal coordinatesgπ βRπ
π
π from the generalized coordinatesg, i.e., gπ = Lπg; and Cππ β Rπ
π
πΓπππ is the elemental tangent damping matrix. For a geometrically exact beam element of lengthβπ,Cππ is given by
Cππ(gπ) =
β« βπ
0
Bπ(gπ, π )DπB(gπ, π )dπ (8.5) whereBβR6Γ12is the discrete strain gradient matrix [Eq. (5.48)],Dπ βR6Γ6is the unknown matrix of viscoelastic damping coefficients for elementπ, andπ is the arc length coordinate along the elementβs reference axis. The integral in Eq. (8.5) is normally approximated by a quadrature rule withππ weightsπ€π at arc lengthsπ π. It follows that
Cπ(g) =
ππ
Γ
π=1
Lππ
βπ 2
ππ
Γ
π=1
π€πBπ(Lπg, π π)DπB(Lπg, π π)
!
Lπ (8.6)
which is a linear function ofDπ.
From here, an optimization problem determines the ππ unknown matrices of ele- mental damping coefficientsDπ that minimize the error between the global tangent damping matrix Cπ and a reference damping matrix Cref. In doing so, the finite element model inherits the damping characteristics ofCref in the specified config- uration (here, the initial configuration). A particularly simple choice for Cref is stiffness-proportional damping; i.e.,
Cref= π½Kπ (8.7)
whereπ½=2π1/π1,π1is the first-mode natural frequency,π1is the desired fraction of critical damping in the first mode, andKπ =πFint/πgis the tangent stiffness matrix.
With stiffness-proportional damping, successive modes are more heavily damped.
Specifically, the fraction of critical damping in theπth mode isππ =π1(ππ/π1)where ππ is theπth modeβs natural frequency;π =1, . . . , π; andπis the number of modes.
In the undeformed configuration, only the material tangent stiffness [Eq. (5.55)]
contributes to Kπ, from which it follows that Cπ and Kπ share the same matrix structure (sparsity pattern). This implies that there is an optimal set of elemental damping coefficients that results in negligible errors betweenCπ andCref. For this reason, stiffness-proportional damping is used to determine the elemental damping coefficients for the dynamic simulations in Sec. 8.4.
Following Sec. 7.3.1 and [243], the optimization problem enforces constraints on the structure of Dπ. In general, Dπ is a symmetric positive definite matrix, i.e., Dπ > 0 [148, 261, 262]. For simplicity, however, Dπ is taken to be diagonal, i.e., Dπ =diag{dπ}, where the coefficients ofdπare strictly positive. With diagonalDπ, the optimization problem takes the form
minimize
dπβπβ {1,2, ..., ππ}
kCπ(g0,dπ) βCrefkπΉ (8.8) subject todπ > 06Γ1. Here,
Cπ(g0,dπ) =
ππ
Γ
π=1
Lππ
βπ 2
ππ
Γ
π=1
π€πBπ(Lπg0, π π)diag{dπ}B(Lπg0, π π)
!
Lπ, (8.9) the subscript 0 denotes the initial (undeformed) configuration, and πΉ denotes the Frobenius norm.
To solve Eq. (8.8), the constraintdπ > 06Γ1is relaxed from strictly positive to simply nonnegative, i.e., dπβ π16Γ1 > 06Γ1 where16Γ1 is a vector of ones and π 1 is a small positive scalar. This results in a convex program with a unique, globally
optimal solution. SinceCπ(g0,dπ)is linear in dπ, the convex program can then be manipulated into the more standard form of a constrained least squares problem, i.e., an optimization problem of the form
minimize
d
kAdβbkπΉ (8.10)
subject todβπ16ππΓ1 > 06ππΓ1 wheredπ = dπ1, . . . , dππ
βR6ππ. Equation (8.10) is solved numerically using convex programming, in this case implemented using CVX [244, 245] with the SDPT3 solver [246, 247].