diff --git a/CHANGELOG.md b/CHANGELOG.md index e9ad96e..1d5118d 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -163,6 +163,14 @@ All notable changes to CableDyn are recorded in this file. The format follows ### Fixed +- A finite-EI dynamic step with bending damping (`EI|Bb`) could fail to converge on a soft line: + the Newton residual stagnated just above the tolerance (1.06 times it on a 20 m riser on a + 20 kN/m tensioner, EA_eff 4e5 N, 80 elements, Bb 500 N m^2 s, dt 1e-3 s, in VIV), and the + step recovery's smaller steps only made it worse. The bending damping takes the curvature rate + from the Newmark velocity, so its forces round at about `Bb gamma/(beta dt) eps |r|/l0^3`, + which the tolerance's round-off floor (the axial forces' `(EA + BA gamma/(beta dt)) eps + |r|/l0`) did not cover. The floor now takes the bending forces, `(EI + Bb gamma/(beta dt)) + eps |r|/l0^3`, as well; it binds only where that term is the larger. - Finite-EI node tension, curvature and bend moment next to the touchdown point: the floor reaction of the first grounded node was spread over its grounded half-element alone, wherever the touchdown point lay in the touchdown element. The local tension error there fell only at diff --git a/doc/theory.rst b/doc/theory.rst index 19c6a68..97f6de3 100644 --- a/doc/theory.rst +++ b/doc/theory.rst @@ -2671,10 +2671,12 @@ residual decreases; its convergence norm divides each translational residual by frozen at the predictor and each tangent-DOF residual additionally by the nodal tributary length. The force scale is the largest inertial or residual entry, bounded below by 1 N, by the peak element weight, and by the round-off floor -:math:`16\,\epsilon\,\max\big((EA + BA\,\gamma/(\beta\,\Delta t))/L_0\big)\,x_\text{ref}/\text{tol}` -of the axial forces (:math:`x_\text{ref}` the largest coordinate magnitude or element length; -the axial damping takes the strain rate from the Newmark velocity, whose rounding is -:math:`\gamma/(\beta\,\Delta t)` times that of the coordinates). The +:math:`16\,\epsilon\,\max\big((EA + BA\,\gamma/(\beta\,\Delta t))/L_0,\ (EI + B_b\,\gamma/(\beta\,\Delta t))/L_0^3\big)\,x_\text{ref}/\text{tol}` +of the axial and bending forces (:math:`x_\text{ref}` the largest coordinate magnitude or element +length; the axial and bending damping take the strain and curvature rates from the Newmark +velocity, whose rounding is :math:`\gamma/(\beta\,\Delta t)` times that of the coordinates). The +bending term is the larger on a soft line (a tensioner in series, a fine mesh) with bending +damping at a small time step. The deck defaults are relative tolerance :math:`10^{-8}`, absolute tolerance :math:`10^{-14}`, 30 iterations, and 12 backtracks (option ``dynamic_solver``). Option ``modified_newton`` reuses the factorised tangent within a step while each iterate reduces the residual to at most a quarter diff --git a/src/CableDyn_HermiteCableDynamic.f90 b/src/CableDyn_HermiteCableDynamic.f90 index 68c153a..1bb731e 100644 --- a/src/CableDyn_HermiteCableDynamic.f90 +++ b/src/CableDyn_HermiteCableDynamic.f90 @@ -5746,7 +5746,7 @@ SUBROUTINE eval_res(qcur, use_trial, rn, esx, emx, stash) REAL(wp), POINTER :: Gout(:) INTEGER :: ii, nd INTEGER(INT64) :: ec0, ec1 - REAL(wp) :: xref, xd, xc, ea_r, ka, kt, tw_round + REAL(wp) :: xref, xd, xc, ea_r, ka, kb, kt, tw_round REAL(wp), PARAMETER :: RESID_ROUNDOFF = 16.0_wp REAL(wp), PARAMETER :: STATE_DIVERGED = 1.0e12_wp ec0 = 0_INT64 ! defined even with profiling off (see the step-entry note) @@ -5873,6 +5873,14 @@ SUBROUTINE eval_res(qcur, use_trial, rn, esx, emx, stash) ! held line damped near its first axial mode (BA = 2 zeta EA/omega_1, which at small dt ! puts the damping term well above EA) stagnated at a residual of a few times the ! tolerance and failed its step. + ! Bending takes the curvature from the same coordinate differences over l0^2, and its nodal + ! forces are the moment over l0, so an element's bending rounding is about eps*|r|/l0^3 + ! times EI, plus Bb gamma/(beta dt) for the bending damping (the curvature rate from the + ! Newmark velocity). Where EA/l0 is small (a soft tensioner in series with the line, a + ! fine mesh) this term is the larger: without it a 20 m riser on a 20 kN/m tensioner + ! (EA_eff 4e5 N, 80 elements, Bb 500 N m^2 s, dt 1e-3 s) stagnated at 1.06 times the + ! tolerance and failed its step, and the step recovery's smaller steps only raised the + ! damping term further. xref = MAXVAL(model%l0) DO ii = 1, ndof IF (MOD(ii - 1, 6) < 3) xref = MAX(xref, ABS(qcur(ii))) @@ -5884,6 +5892,9 @@ SUBROUTINE eval_res(qcur, use_trial, rn, esx, emx, stash) ka = ka + model%torsion%couple(ii)**2*model%torsion%gj(ii) IF (model%has_axial_damping) ka = ka + ABS(model%BA(ii))*gm/(bt*dt) ea_r = MAX(ea_r, ka/model%l0(ii)) + kb = ABS(model%EI(ii)) + IF (model%has_bend_damping) kb = kb + ABS(model%Bb(ii))*gm/(bt*dt) + ea_r = MAX(ea_r, kb/model%l0(ii)**3) END DO fscale = MAX(fscale, RESID_ROUNDOFF*EPSILON(CD_ONE)*ea_r*xref/tol) ! Nodal twist: the torque scale of the angle rows, frozen with fscale and formed like it diff --git a/tests/test_hermite_bending_damping_dynamic.f90 b/tests/test_hermite_bending_damping_dynamic.f90 index 39ab091..bdbb14b 100644 --- a/tests/test_hermite_bending_damping_dynamic.f90 +++ b/tests/test_hermite_bending_damping_dynamic.f90 @@ -16,6 +16,12 @@ PROGRAM test_hermite_bending_damping_dynamic !! B5 a straight cable moving in its first bending mode with zero acceleration has no elastic !! curvature: the node recovery reports none although the bending damping acts (its viscous !! moment is not EI kappa), while the end force carries the damping. + !! B6 the residual's round-off floor: a soft line (EA 4e5 N, as a tensioner in series) with a + !! stiff bending damping (Bb 1e4 N m^2 s) at dt 1e-4 s. The damping takes the curvature + !! rate from the Newmark velocity, whose rounding is gamma/(beta dt) times the coordinates', + !! so its force rounds at about Bb gamma/(beta dt) eps |r|/l0^3, far above the axial floor + !! EA eps |r|/l0 the tolerance scale allowed: every step stagnated just above the + !! tolerance and failed. Every step converges and the motion stays bounded. USE CableDyn_Precision, ONLY: wp USE CableDyn_HermiteCableDynamic, ONLY: CD_HermiteCableDynType, CD_HermiteCable_Dyn_Init, & CD_HermiteCable_Dyn_Step, CD_HermiteCable_Dyn_Set_Bending_Damping, & @@ -32,6 +38,7 @@ PROGRAM test_hermite_bending_damping_dynamic CALL decay_case('B3 undamped', 1.0e4_wp, 0.0_wp) CALL check_refusals() CALL check_recovery() + CALL check_roundoff_floor() IF (nfail > 0) THEN WRITE (*, '(A,I0,A)') 'FAIL: ', nfail, ' assertion(s) failed' ERROR STOP 1 @@ -156,4 +163,34 @@ SUBROUTINE check_recovery() CALL require(NORM2(f_damped - f_plain) > 0.0_wp, 'B5: the end force carries the damping') END SUBROUTINE check_recovery + SUBROUTINE check_roundoff_floor() + TYPE(CD_HermiteCableDynType) :: model + REAL(wp), PARAMETER :: EA_SOFT = 4.0e5_wp, TENS = 6.0e3_wp, A0 = 1.0e-3_wp, BB = 1.0e4_wp, DT = 1.0e-4_wp + REAL(wp) :: l0(NE), ea_(NE), ei_(NE), rho(NE), w(NE), q(NDOF), eps, lt, x, res, zmax + INTEGER :: k, n, es, it + CHARACTER(512) :: em + eps = TENS/EA_SOFT; lt = (1.0_wp + eps)*LL + l0 = LL/REAL(NE, wp); ea_ = EA_SOFT; ei_ = EI; rho = RHOA; w = 0.0_wp + DO k = 1, NN + x = lt*REAL(k - 1, wp)/REAL(NE, wp) + q(6*k - 5:6*k) = [x, 0.0_wp, A0*SIN(PI*x/lt), (1.0_wp + eps), 0.0_wp, (1.0_wp + eps)*A0*PI/lt*COS(PI*x/lt)] + END DO + CALL CD_HermiteCable_Dyn_Init(model, l0, ea_, ei_, rho, w, q, [1, 2, 3, 6*NE + 1, 6*NE + 2, 6*NE + 3], & + -1.0e4_wp, 0.0_wp, 1.0_wp, es, em) + CALL require(es == CD_HCDYN_OK, 'B6 init: '//TRIM(em)) + CALL CD_HermiteCable_Dyn_Set_Bending_Damping(model, [(BB, k=1, NE)], es, em) + CALL require(es == CD_HCDYN_OK, 'B6 set damping: '//TRIM(em)) + zmax = 0.0_wp + DO n = 1, 200 + CALL CD_HermiteCable_Dyn_Step(model, DT, 50, 1.0e-8_wp, es, em, it, res) + IF (es /= CD_HCDYN_OK) THEN + CALL require(.FALSE., 'B6 round-off floor: step failed: '//TRIM(em)); RETURN + END IF + zmax = MAX(zmax, ABS(model%q(6*(MID - 1) + 3))) + END DO + WRITE (*, '(A,ES11.4,A,ES11.4)') 'B6 round-off floor: 200 steps converged, last residual ', res, & + ', largest midspan deflection ', zmax + CALL require(zmax <= A0*(1.0_wp + 1.0e-6_wp), 'B6 round-off floor: the damped motion stays bounded') + END SUBROUTINE check_roundoff_floor + END PROGRAM test_hermite_bending_damping_dynamic