Kaplan, Sargent and Goodman 
tionally'' efficient method for system identification purposes, 
Various techniques for reducing the time (i.e. real time 
of the observed data) for convergence of the parameters, which would 
then also reduce total computer time, were also found. Thus an im- 
proved convergence time is found when the constant matrix elements 
on the right hand side of the gain equations (i.e. the P-equations, as 
shown in Equation (30) ) are made larger, but that improvement does 
not show significant gains for larger equation systems. The most 
useful approach is to vary the magnitudes of the elements in the 
weighting Q-matrix that appears in Equations (29) and (30) in 
order to reflect the importance of measured state variable data that 
are well known. Thus larger values of the matrix elements are used 
for particular variables that are known to be measured accurately 
and directly, rather than data that is not directly observable or has 
a lower degree of accuracy due to instrumentation difficulties. Thus 
these particular strategies are useful means of achieving more rapid 
convergence for this type of system identification technique. 
CONCLUSIONS 
The present paper has demonstrated the feasibility of 
using different system identification techniques to determine the 
values of major parameters and coefficients in a mathematical model 
representing the motion of different naval craft. This was demons- 
trated by application to a number of diverse vehicles, suchasa 
surface ship, a hydrofoil craft, and an SES craft, using data that 
was generated on a computer (with known coefficients and a known 
mathematical model) as well as from full scale tests. Different 
techniques are used, in accordance to the extent of the influence of 
noise on the system (and its measured responses), and their limi- 
tations as well as capabilities are described in the paper. Certain 
virtues of the two different methods used are quite important, such 
as having a means of determining a level of confidence for different 
converged parameters while carrying out the identification, as well 
as an indication that a particular mathematical model is not fully 
appropriate for representing certain features of the craft motions. 
The two techniques demonstrated here are generally applicable toa 
number of different stability and control problems of naval craft, for 
both full scale and model scale data and analyses. Depending upon 
the degree of accuracy and the procedures used for data acquisition, 
these methods can be applied to determine stability derivatives, 
nonlinear coefficients, etc. in a structured mathematical model 
representation, from data taken in model tanks. Such data would 
only involve motion trajectory measurements of free models and could 
1668 
