Report no. 352
r
March 1972
LABORATORIUM VOOR
SCHEEPSBOUWKUNDE
TECHNISCHE HOGESCHOOL DELFT
DATA ANALYSIS OF FULL SCALE MEASUREMEWPS
WITH A HYBRID COMPUTER
by
Ir. F.J. Pasveer
Ir. C.C. Glansdorp
M. Buiterihek
L
I
Scientific Officer Computation Center, Deift University of Technology
Members of the Staff of the Shipbuilding Laboratory
Summary
A short description of data reduction and analysis of full scale
w,
Introduction.
Planned full scale motion measurements with a containership lead to a careful
re-examination of the method of data reduction in view of the new hybrid
computer facilities available at the Computation Center of the Deift University of Technology.
The method of data reduction used so far was very time consuming due to
separate real time tape punching of all time histories of the measured
signals. Furthermore punching errors could not be entirely avoided because
of the large amount of numbers to be punched.
In 1969 a hybrid computer ADl/IBM18OO was installed and now a vast software
package is available or in development for the users of this facility.
It was found that existing progranis for the calculation of power spectra
together with data supply by a taperecorder could be applied with a number
of modifications as to meet the requirements of full scale motion measurements.
After some trialruns with old measurements it was decided to adopt this new method for data reduction and analysis of the measurements.
It appeared that when a suitable division in analog and digital calculation
work was made a substantial decrease in total analyzing time could be
realised. Furthermore, comparing to the old method which made use of punch tapes the hybrid method causes no unallowable losses due to transmission
errors.
2-The old method.
During motion measurements various signals are measured.
The measured signals and a reference signal were simultaneously recorded
on magnetic tape of an instrumentation tape recorder. The internal modulator
transforms an- analog signal in a frequency modulated signal. Generally
speaking, when reproducing the signal is internally demodulated and an analog
signal is produced.
Because of the sometimes very severe motions during the measurements the
recording speed of the tape recorder is not constant. This fact leads to
decreasing accuracy when demodulating in the usual way with the internal
frequency demodulator. The reason is that the carrier frequency on tape is no more sufficiently constant.
To overcome this difficulty a digitizer was made controlled by the reference
signal put on the tape during the measurements.
Generahly speaking the digitizing of each signal was successively done
on a reãl time base; at the same time the speach channel gave the particulars
of the specific run. In fig. i a sketch of the set up is given.
The punch tapes were fed into a digital computer in which autocovariance
functions and power spectra were calculated according to the formula's
given in appendix 1.
The new method
The same recording principle will be used in this case.
Deviations of the uniform recording speed are expressed in frequency changes
of both frequency modulated measured signal and reference signal. These speed
deviations can be compensated when the demodulator is controlled by the
reference signal of the tape.
In fig. 2 the principle pf' this demodulator is given, while fig. 3 presents the time chart of relevant signals and commands.
The measured signal can be written as fcn ± Af in which is the
carrier-frequency and is the modulating frequency.
In the timing- and controlcircuit both the sampling time and the counting
time are determined from the reference signal
r
During the counting time the leading edges of (r ±
f) and
are fedinto two pulsers with constant pulse width T. These pulsers control the
electronic switches Sn and Sr
During the pulsertime T the voltages and _Ur are fed into integrator I,
which is in the "operate mode" during the counting time.
Becaus of the opposite signs of and Ur the output voltage (Vn - Vref)
of I is proportional to the modulating frequency if the ratio fr/fc
equals the coefficient setting 1cc (see appendix II).
The sc.ling of integrator I is controlled by coefficient setting Kr. The
value of Kr is chosen in such a way that the integrator output is at a maximun
when tfn is maxima]. (see appendix II). The values V - Vref of all integrators
of all measured signals are read into the digital computer on a command
at the end of the counting time. During the time between two conunandpulses
the digital partner calculates the contributions of the autocovariance function for each signal.
As the ratio between digital calculation time and sampling time is rery small, it is possible to increase the reproducing speed.
With the present analog hardware it is expected that at least an increase in tape spetd by a factor 8 can be reached when 6 signals are simultaneously processed. The total time consumption of data reduction and analysis is
Prelimenary results.
From old motion measurements a pitch spectrum has been calculated according
to the new method. This spectrum is compared with the spectrum calculated
according to the old method. The differences between the two spectra are
marginal and they are mainly caused bythe fact that the sampling time
was not equal in both cases. Fig. 4 gives an impression of the spectra.
In fig. 5 and fig. 6 a photograph of the autocovariance function and the
power spectrum are given, taken from the oscilloscope, immediately after
calculation in order to have a visual check. 5
Appendix I
Formula's for the digital computation of the autocövariance function
and the power spectrum.
The autocovariance function of a signal x(t) is given by
T/2
I f
= hm
Jx(t).x(t+T)dt
(i)T- -TI2
The power spectrum is calculated according to:
Sxx(e)
=(Rxx(t)
cos eT
(2)o
These formal expressions (i) and (2) are approximated by
n=N-p x(tn).x(tn+p)
(3)
n= i L = {K.ca?
E } n 'IT P m P p=o (4)with K is the integration constant of the trapezoidrule,
=
if p0 arid pm
K
= i in any other caseIt is to be noted that
= and t
t -t
e p P-1
The raw spectrum is now smoothed by the window of Henning giving the
following modified coordinates of the power spectrum:
G0 O..5L0 + 0.5L1
(5)
Gb = O.25Lh...1 + O.5Lh + O.2SLh.,..
During the processing of the tapes the digital computer calculates the
values according to (3).
At the end of a run the digital computer calculates the power spectra according
to
(4)
and (5) and the results are presented on a scope.=
N-p
Appendix II
IÍ the counting time is x sea then the integrator I gives an output
voltage V Vref as follows:
- = x(f ± Af)T Kc Kr K -
T X
r Kr K1 (i)where: f ± Af is the mod1ating signal
T 15 the pulsertime
K1., K are coefficient settings
K is the integration speed of all integrators I
If one requires that
V- - Vref = O when n O
fr
then (i) gives K
= - with K1. + O and
I( + 0;C fc
Eq. (i) can be written as:
V
- Vref = xf Kc
T Kr Ki (2)When (Af) is at the maximum (Af) the output voltage reaches the
umax maximum Vmax.
The value of the coefficientsetting Kr is now given by:
Vmax
Kr (Af)max Kc X T
A digital service program calculates and controiB the coefficient settings
s
Ô
.
.
Timebase Frequency (Reference) Frequency to be counted (FM-signal) iming Unit Gatefigure 1 Block Diagran Old Method.
Counter
Memory
Punchdrivers
p
Punch Tape Digital
- signal
f c± ofc
Channel 2
Reference
Signal fr
Gate
Pulser
timing - contro
Coef.
Switch
Kc
Coef.J Kr
+1Integra t or
Vref
o Q) Q) bO C) o1
m c.J u) bD .r-4 CI-1figure 3 Time Chart.
from tape reference
FM-signal fc±f
and PuLser t
Integrator VoLtage
without
fr
Reference SignaL fr
Integrator Voltage
'wìthout fc
±&f
Integrator VoLtage
Vref
Command PuLse to
DigitaL Computer
___p
1
sampUnq time
counting
time
4
rA4
_____I-Avcn
/
y/
/
/
/
ri
ri
Vref
VcnVref
LV
\
\
A
/
o
w In 3t
o
o
w
i
PITCH
)
J
I'
I'
old method
new method
'I : I I\
Iy'
t1
\
O 0,5 _, 1,0We
sec