Additive manufacturing knowledge

What is this?

• A page for RPI, Sandia and Simmetrix colloborators
• to debug additive manufacturing runs

Problem

The example problem we are primarily interested in is located on SCOREC machines in

/lore/granzb/for_ottmar

The problem is a squid-like structure sitting atop a hollow cylinder attached to a solid base-plate. Currently we are building up the cylinder layer by layer when failure occurs.

Current knowledge

• SI units lead to poorly conditioned systems with thermo-mechanical coupling
• Diagonal/Abs row sum scaling preconditioning greatly improves our ability to solve these coupled thermo-mechanical systems:

• GMRES iterations is likely strongly correlated with # of mesh elements

as well as the heat source that is being put into the problem.

run type	# of gmres iterations (1st solve)	failure mode
no scaling	328	? (Ottmar please fill in)
diagonal scaling	61	continuation step 136, linear solve (1000 GMRES iters)
abs row sum scaling	61	unknown

Diagonal scaling case

Specifically, it appears all runs "fail" due to a lack of convergence in the iterative GMRES solver. In this case of the diagonal scaling example, this is directly after (<-- suspicious!) a layer has been added. Important to note that, up to this point, there have been 26 successful layer additions.

For completeness, the convergence history for the failed linear solve is included.

  Iter    0, [ 1] :    1.000000e+00
  Iter   20, [ 1] :    9.988737e-01
  Iter   40, [ 1] :    9.871794e-01
  Iter   60, [ 1] :    9.725014e-01
  Iter   80, [ 1] :    8.828681e-01
  Iter  100, [ 1] :    5.384302e-01
  Iter  120, [ 1] :    3.478822e-01
  Iter  140, [ 1] :    2.171355e-01
  Iter  160, [ 1] :    1.542421e-01
  Iter  180, [ 1] :    9.596347e-02
  Iter  200, [ 1] :    7.881647e-02
  Iter  220, [ 1] :    6.109802e-02
  Iter  240, [ 1] :    5.166380e-02
  Iter  260, [ 1] :    4.025756e-02
  Iter  280, [ 1] :    2.405680e-02
  Iter  300, [ 1] :    1.516800e-02
  Iter  320, [ 1] :    1.335365e-02
  Iter  340, [ 1] :    1.240837e-02
  Iter  360, [ 1] :    7.786410e-03
  Iter  380, [ 1] :    4.458504e-03
  Iter  400, [ 1] :    2.976416e-03
  Iter  420, [ 1] :    2.142764e-03
  Iter  440, [ 1] :    1.659814e-03
  Iter  460, [ 1] :    1.422948e-03
  Iter  480, [ 1] :    1.357753e-03
  Iter  500, [ 1] :    1.153012e-03
  Iter  520, [ 1] :    1.153012e-03
  Iter  540, [ 1] :    1.153004e-03
  Iter  560, [ 1] :    1.152988e-03
  Iter  580, [ 1] :    1.152945e-03
  Iter  600, [ 1] :    1.152875e-03
  Iter  620, [ 1] :    1.152559e-03
  Iter  640, [ 1] :    1.152152e-03
  Iter  660, [ 1] :    1.151905e-03
  Iter  680, [ 1] :    1.151352e-03
  Iter  700, [ 1] :    1.150487e-03
  Iter  720, [ 1] :    1.149541e-03
  Iter  740, [ 1] :    1.132207e-03
  Iter  760, [ 1] :    1.101848e-03
  Iter  780, [ 1] :    1.096479e-03
  Iter  800, [ 1] :    1.092905e-03
  Iter  820, [ 1] :    1.090804e-03
  Iter  840, [ 1] :    1.080092e-03
  Iter  860, [ 1] :    1.003904e-03
  Iter  880, [ 1] :    9.613925e-04
  Iter  900, [ 1] :    9.095563e-04
  Iter  920, [ 1] :    8.454028e-04
  Iter  940, [ 1] :    7.253190e-04
  Iter  960, [ 1] :    6.348435e-04
  Iter  980, [ 1] :    4.667210e-04
  Iter 1000, [ 1] :    3.930348e-04 

 The Belos solver of type ""Belos::BlockGmresSolMgr": {Flexible: true, Num Blocks: 500, Maximum Iterations: 1000, Maximum Restarts: 20, Convergence Tolerance: 1e-06}" returned a solve status of "SOLVE_STATUS_UNCONVERGED" in 1000 iterations with total CPU time of 184.466 sec
WARNING: LOCA::BorderedSolver::LowerTriangularBlockElimination::solve() -
Return Type = NotConverged
WARNING: LOCA::MultiContinuation::ConstrainedGroup::applyJacobianInverseMultiVector() -
Return Type = NotConverged
WARNING: LOCA::MultiContinuation::ConstrainedGroup::computeNewton() -
Return Type = NotConverged
NOX::Solver::SingleStep - Unable to solve Newton system

************************************************************************
-- The "Nonlinear" Solver Step --
 (Failed!)
************************************************************************

Previous progress

• uncovered bug in how scaling is applied: #77
• fix scaling bug ONLY for Nonlinear Solver = Single Step : in Albany, in Albany, and in Trilinos

Should be investigated

• non-dimensionalizing the input material parameters
• one-way coupling with CTM approach
• stress anomalies in the geometry - are these physical?
• Does the mesh drastically change right before failure modes?