Validation and Testing of Performance
A HydroComp Technical Report
Product link: NavCad,
This report comments
on the various methods applied by HydroComp to validate performance
prediction algorithms. It also lists algorithms that have been
corrected or revised.
public data (parametric models) for its numerical software algorithms.
Reliability of these algorithms takes two forms - one, the faithful
duplication of the intended data, and two, successful application
to real vessels. Two different testing and validation efforts
are undertaken to evaluate each algorithm or method.
Many of the methods
are derived from technical literature. Testing is performed to
seek out publication errors and to insure that the codes conform
with the intended methodology. Such testing follows the process
- Evaluate the statistical development
of the parametric model. Review regression process and look
for statistical "instability".
- Perform hand calculations of
methods to establish test variables and results.
- Determine boundary conditions
of algorithms (e.g., points of instability or possible calculation
error, locations of divide by zero, etc.).
- Hand check codes against method.
- Use an algorithm test driver
to confirm results of codes.
- Test installed algorithms within
the larger program.
Coding errors, of
course, are corrected. Suspected errors within the methods will
either cause the method to be rejected, or will initiate a dialog
with the original author of the method. HydroComp has established
in-depth relationships with many of the international research
institutions and experts, and this network is exploited to uncover
possible errors and solutions. HydroComp has identified and corrected
publication errors in the following algorithms:
and propulsive coefficients
Howe barge train
USNA YP series
ducted propeller series
Comparison of the
algorithms with model testing or full scale results provides feedback
on the applicability of the various methods to contemporary vessels.
HydroComp maintains a library of hundreds of model tests and trials
for such testing. When trends are discovered, a description of
the behavior is noted (typically in the help system and User's
In some cases, poor behavior can
be attributed to numerical problems which are caused when the
vessel's parameters are outside the range of the method's dataset.
It is often possible to improve this type of poor behavior "at
the edges" without compromising the integrity of the method through
careful additions to the algorithms. Application behavior has
been improved in the following methods: