MBDyn Bugs & Unresolved Issues

Disclaimer

• differences between mathematical models and reality
• numerical approximation of the mathematical models
• errors in the implementation of the numerical approximations

Known Problems

• The multithread assembly broke with g++ >= 4.4; it is not yet clear whether the bug is in MBDyn or in the compiler (or in libatomic_ops, or whatever).
• The deformable hinge joint used an incremental strain formulation which could drift under large periodical loads. In 1.1.2 it has been rewritten using an absolute strain measure; now the drawbacks are a little more computational overhead and the limitation to 'small' strains (say less than +/-180 degrees). Viscous and viscoelastic versions have been ported to the new formulation.
• I realize most of the comments inside the code are in Italian (many are also out of date; this might be a feature for many non-Italian hackers :)
• There's a bug in the C81 unsteady aerodynamic forces in 1.1.1, which has been FIXED in 1.1.2. Users interested in aeroelasticity are encouraged to upgrade.
• There's a library cross-dependency that's not appropriately handled by recent versions of libtool (e.g. 1.4.1). The file mbdyn/Makefile can be hand-crafted by linking twice the base/.libs/libbase.a library in mbdyn target, before and after the libaero/libstruct/libelec/libhydr block.
• (FIXED in 1.1.0) There's a problem in the memory handling of y12/harwell solver that makes solution inefficient when large sparse matrices are used in versions up to 1.0.1; users are encouraged to explicitly set the solver to "solver: y12, workspace size, <n>;" where <n> is more than three times the estimated number of nonzero terms of the matrix.
  The fix makes use of the non-zero count from the previous factorization to adaptively change the optimal size; in case of large variations of non-zeroes it may fail. The problem can be fixed by forcing the use of a given workspace size (see above). Unfortunately there's no simple automatic fix because when the solver realizes there's not enough room, the coefficients of the matrix have already been altered. Another solution is to use the Umfpack3 solver instead of y12, because it dynamically allocates the workspace.
• There's a number of features included in the distribution that are known to work only partially or have not been thoroughly checked. Among these are:
  • modal aerodynamics
  • unsteady aerodynamic correction (built-in; the C81 version has been FIXED in 1.1.2)
  • some of the hydraulic elements
  • the kinematic joint
• There's a number of non-reentrant parts in MBDyn. As a consequence there's no chance to easily turn it into a multi-thread software (this may be considered a feature considering the issues of porting thread-based code among platforms :).
• The math interpreter symbol table is local. This means that run-time changes of a symbol are not shared among processors during parallel simulations. This will eventually be fixed when MBDyn is interfaced to a symbolic math toolbox.

Miscellaneous

• Go to our Issue tracking system.
• Browse the change log:
  • CHANGELOG-1.3.
  • CHANGELOG-1.2.