CH1EF
INTERNATIONAL SYO.SIUM ON THE DYNAMICS OF MARINE VEHICLES AND STRUCTURES IN WAVES
at
University College London, May l971
Paper number 30
DATA ON MIDSHIP BENDING STRESSES FROM FOUR SHIPS
by
G. Ward and M. Katory (with Appendix by P. Gedling)
Lab. y.
Schepsbouwkunde
Technische Hogeschool
Deth
1. INTRODUCTION & SUMMARY
Data loggers have been fitted by B.S.R.A. to four ships to
measure main hull and local strains together with other parameters used in
the general assessment of ship performance in service. This note presents
an analysis of some of the logger data on the main hull bending stresses
from three of these ships. A complementary project at B.S.R.A. is
the recording of main hull bending streeseS using statistical counting
gauges; these have been used on 38 ships over the last 19 years and data
from such gauges, fitted to the same ships as above, are also presented.
The main hull stresses, as measured by the loggers, have been analysed using a numerical filtering technique to separate the wave induced stresses from the springing stresses and the results have been presented in
cumulative probability forni. This form of presentation suggests the maximum
stress likely to be experienced in say a ship's life-time. A similar
presentation of the statistical strain gauge data has been given using a
simple cumulative analysis of the number of stress reversals counted;
also a method has been applied, to the same data, that fits a theoretical
double exponential distribution to the maximum stress over a number of watches
chosen at random to give projections of hull bending stresses which are likely to be experienced over specified time-spans and with a known degree
of statistical confidence.
Analysis of the combined signals of wave induced and springing
stresses has shown that an equation originally proposed by Rice (Ref. i)
and recently used by Miles (Ref. 2) defines the short term behaviour of
-2-Calculations have been carried out for the wave induced stresses, using programs developed at B.S.R.A., and these have been developed in the same cumulative probability form as the analysed ship data for
comparison purposes at a probability corresponding to a ship's lifetime.
Further work is required to complete the analysis of the measured
d.ata and the calculations and in particular to consider the combined wave induced and springing stresses.
3
1. AUTOMATIC DATA LOGGING SYSTEMS
The objective in using data loggers was to record a large sample of the required data for a ship in normal service with no interference to
the ship's routine and a minimum of extra tasks for the crew. The first
system used by B.S.R.A. was a 25 channel automatic data logger developed
from a successful aircraft flight recorder. Ships A and B were fitted with
these (see Table 1 for ship particulars).
This logger automatically records information in digital form
every 12 hours, for a period of 20 minutes, with a scanning rate of 25
channels per second, that is each channel is sampled once per second. To
provide for a sufficient sampling rate to allow springing stresses to be deduced,, five channels are used for one strain gauge measuring main hull
bending stresses, making the scanning period for this gauge 0.2 secs. The
only operation required by the crew is to change tapes about once every
15 days, and periodically post these back to B.S.R.A. In addition to this
task the crew are asked to fill ii log-sheets which provide general data on the ship's condition and also visual estimates of the sea conditions prevailing during each watch.
The initial requirement for data loggers was essentially from the speed and power performance point of view, but a later requirement to
measure local strains led to the choice of a second system which was
fitted to ships C and D (see Table 1 for ship particulars). This has
many more channels than the first system and is used in a mode which logs
information on 90 channels with a scanning rate of 30 channels per second.
Data is again logged automatically every 12 hours for a period of 20
transducers use more than one channel and in particular the main hull bending stress uses 15 channels to provide a scanning period for this
gauge of 0.2 secs.
3.
STATISTICAL STRAIN GAUGESThe investigation using statistical strain gauges has been
ii operation at B.S.R.A. for the last 19 years and in that time 38 ships
have supplied service stress data covering a total of over 50 years of
recording time. The equipment used for measuring the main hull bending
stresses (Ref. 3) will only be briefly described here. It is a strain
gauge, with a 100 in gauge length, arranged to record the number of
excedences, at predetermined stress levels, on electro-mechanical counters. The gauge is mounted on the deck structure near amidships, either near the centre line or near the deck edge, and the recorder is usually located
on the bridge, for the deck officers' convenience in logging the counts
each watch. The gauge does not record changes in the mean level of the
stress but only records the out-to-out longitudinal stress excursions experienced in the deck structure owing to wave bending action or to main
hull vibration.
In this note only the statistical strain gauge data is given
for those ships also fitted with data loggers. The position of the gauges
in these ships is given in Table 2. The main hull stresses given for the
data logger have also been measured by a 100 in strain gauge, which was
5
).. DATA PROCESSING OF LOGGER STRESSES
Fig. i shows the scheme for processing the data recorded by the
loggers. When received at B.S.R.A. the magnetic tapes which in both
systems are in a non-standard digital format are read into the computer and
operated on by pre-analysis programs. These check for parity, scanning
and formatting errors and make corrections before transferring the logged
data to magnetic tapes in standard computer format. These in turn are
further processed by a preliminary analysis routine which contains calibration information, to provide a summary of the data, such as mean value and standard deviation of each parameter, and also to provide a copy
of the data for further analysis such as that decribed below. The stress
data are analysed to provide spectra together with the area under the
spectra m and their second rn2 and fourth m moments. The latter are used
-
2'
\!
2(J-cz)
exj
where (z,n) = the probability density function of z
z = the height of the peak stress divided by the root-mean-square of the record
2
m2 m
o
erf error function
exp exponential
(1)
in the Miles equation (Ref.
vTi
2). 2 --z + [l + erf{z = . e 2(1-n) 2The stress data are also filtered to separate the wave induced from the springing stresses before passing through the spectral analysis
routine. The requirement is for a band pass digital filter that will only
pass the wave encounter frequencies, using the 20 minute records with a
sampling period of 1 second. Also a band pass filter is required that will
only pass the data in the vicinity of the two node mode of vibration
slid, for this, any 5 minutes of a 20 minute record are used with a
sampling period 0.2 second. The selected 5 minute records are also
filtered by a band pass filter to obtain the wave induced stress in order to determine the corresponding amplitudes of combined, wave induced and
springing stresses. Filtering techniques have been described by Martin
(Ref. ) and the output of such a filter is defined by
N2
6
N1
where y1(t) is the input, T is the sampling time interval and the numerical filter is the (N2-N1l) weights Bk. The filtering of a set of data is
achieved by moving equation (2) along the data points (N2-N1)/2 times. This
gives a weighted value to each point in turn with a contribution from the N points before and after where 2NN2-N1. Ref. 4 provides information
the weights Bk to be used for various types of filter. Fig. 2 illustrates
diagramatically the combined and filtered signals and also defines the stress
peaks and reversals.
Numerical procedures are used to define the stress reversals from which histograms for the various records are computed and also their
root-mean-square (R.M.S.). Long-term probability trends are calculated
following the methods used by Lewis and Hoffman (Ref. 5). These assume
that the peak-to-trough distribution of stresses within a record are
Rayleigh distributed and the R.M.S. stress values are normally distributed
within any one weather group. The weather groupings used, for both the
logger data and the statistical gauge data, are given in Table 3.
Long-term probability trends are calculated based on the R.M.S. of thr stress
7
reversals in each record, the measured maximum stress reversal in each record and the predicted distribution of maxima of stress reversals assuming a double exponentional form for this distribution.
DATA PROCESSING OF STTISTICAL GAUGE STRESSES
The stress counts per watch together with other relevant information, such as Beaufort Number, wind direction, ship's heading,
sea area, ship's loading condition, are first checked for consistency before
the data is put into punched card form and then transferred onto magnetic
tape. A computer sorting routine is then used to group the data into
various categories such as Load or Ballast condition and Weather Groups. Detailed sortings are being carried out on the total data but for the
purposes of this note only Weather Groups and Load and Ballast conditions
are presented separately.
Figs. 3 to 6 show the data, for each of the four ships, presented
in histogram form for each Weather Group to allow the effect of weather to be seen, but for the probability distributions based on these data and
shown in Fig. 7 the combined data were used. The cumulative probability
values Q were derived from the formula
Q=l
nXCn+l
E n+l
where C = cumulative count up to the n th stress level and En = the total
n
Register
-8
Table 1 shows the results of a method. of analysis which
is described in detail in the Appendix. In this method the analysis is
carried out on the maximum stress logged in each watch. From these a
frequency distribution is formed using the gauge interval of i ton/in2
for the groups. That is, a maximum stress of less than 1 ton/in2 would
be in the O-i group and a stress of 3 ton/in2 would be in the 3-1f group.
The maximum stress data is assumed to follow a double exponential form for which the parameters are derived in the manner described in the
Appendix. Goodness-of-fit tests on the derived equations did not generally
show up well when the method was applied to the total data sample per ship or to the data divided into Weather Groups but a considerable improvement was achieved by applying the method to the maximum stress logged in ten
watches picked at random from the total data.
A particularly large sample of data, which is used in Table )4,
was obtained by grouping together the statistical strain gauge data for six
sister ships*, one of which is ship 'B' in Table 1. The data, which
is divided into two groups to correspond to the periods of operation
before and after major structural modifications, have been further anälysed
by treating the maximum stress in twenty watches and. the maximum in one
hundred watches all chosen at random. For this large sample, satisfactory
and consistent fits were obtained for each of these higher groupings.
6. THEORETICAL CALCULATIONS
To support the work on ship measurements, calculation
procedures have been programmed at B.S.R.A. to determine the wave induced
motions and loads. This work has provided a suite of programs that will
-9
calculate, for seas from various directions, the pitch, heave, roll,
sway and yaw motions (with a close fit method for calculating the
hydrodynamic coefficients), the vertical and transverse bending moments
and shear forces and the twisting moments. Short and long term
distributions can also be calculated using the Pierson-Moskowitz formulation
of sea states. In addition to these the springing bending moments can be
calculated using a program based on the theory by Goodman (Ref.
6).
The objective of this work is to provide designers with the capability of calculating the various loads on the hull structure for rational
structural design procedures.
In the various full scale measurements taken by B.S.R.A. on
ships in service the sea state is not directly recorded. There is no
suitable instrument for this and hence the weather is identified by the measurement of wind speed together with the deck officer's estimate of
the sea state and the Beaufort number which are used to corroborate the
measured data. However, this means that measured stresses cannot be used
directly to verify calculated stresses in the short term but it is anticipated that a satisfactory comparison of long term predictions, by methods based on measured and calculated stresses, should establish overall
confidence in a completely theoretical approach.
7. DISCUSSION OF MAIN HULL STRESSES MEASURED BY DATA LOGGER
Figs. 8 to 15 present examples for ships 'A' and 'D' of
the analysed information for peak stresses for both 20 min and 5 min
records. The peak stresses are non-dimensionalised with respect to the
variance of the record and are given in histogram form; also peak stresses
are given in spectral form for the filtered and unfiltered data, for both 20 min and 5 min records. The theoretical distributions superimposed on the histograms are those derived from the use of equ. (1) using the spectral
'I' 10 'I'
-properties as quoted on each spectral diagram. The data given in these
figures, and others like them, verify the filtering techniques used.
The theoretical distributions generally compare well with the histograms for both the filtered and unfiltered data as was found by Miles
ÇRef. 2). The spectral width parameter a was found to be quite close to
unity, that is the peaks are Rayleigh distributed, for all the filtered springing records but was generally between
0.6
and 0.7 for the filteredwave induced stress records. This is a result of negative peaks in the
latter as illustrated in Fig. 2. The corresponding histograms show the
negative peaks in the filtered wave induced stresses but not in the
filtered springing stresses. By using stress reversals and not peak
stresses in the long term probability trends of wave induced stresses, the assumption of a narrow band Rayleigh distribution will be more nearly
satisfied.
The other assumption made in the long term probability trends is that the R.M.S. of the reversals is normally distributed within each Weather
Group. An example of the R.M.S. data is shown plotted on a normal base in
Fig. 16 and the straight line distribution of the data justifies the
assumption.
An examination of the combined wave induced and springing stresses and the filtered stresses supports the conclusion given by Lewis in Ref. I that the combined stress is less than the sum of the corresponding wave
induced and springing stress amplitudes.
Equ. (i), in its cumulative probability form, has been used to predict the maximum peak stress in a record and Table 5 shows the results
of these calculations against the analysed maximum in the corresponding
records. The records were chosen at random and it seems that the spectral
data in Miles' equation predicts the maximum well enough for further applications. A suitable method for combining individually calculated wave induced and springing stresses still needs to be tried and proved for the long term probability trends.
The spectral analysis for ships 'A' and 'C' showed that there
was some springing in nearly every record. For ship 'D' there were
many records which showed the presence of 2 node mode vibration stress
in moderate weathers, in the same way as was observed for ships 'A'
and 'C'; the level of this stress, in terms of R.M.S. of the filtered
record for the higher frequency stress, was similar over various 5 minute portions of the records, in those cases where this was checked.
Also for ship 'D' there were several records where the R.M.S. of the filtered high frequency stress was considerably larger than that of
the rest of the records and it is taken that these were caused by impulsive loading.
Figs.
r(, 18
and 19 show the long term trends of wave inducedstress reversals. The base is in terms of cumulative probability, that
is a probability of exceeding a particular stress once, but this can alternatively be considered in terms of number of cylces which on average
will experience that particular stress once. It is generally assumed
8
that a ship s lifetime of 20 years is covered by about 10 cycles, which
is equivalent to a probability of 10
The trend curves shown in Figs. 11, 18 and 19 have been calculated
12
-considered in Ref. 5. These methods are based
on:-the R.M.S. of on:-the stress reversals in each record
the measured maximum stress reversal in each record and the predicted distribution of maxima of stress reversals assuming a double exponential form for this distribution
The data have also been analysed by Weather Groups (W.G.'s) but only for
method (a) above is a full set of these shown. For methods (b) and (c)
only the combined W.G. Curve is shown. It should be noted that the trend
lines have been presented on a conunon base for methods (a) and (b)
determined by the number of cyôles and this accounts for the separation of log (300) at the origin as the analysis in method (b) is based on
records with an assumed 300 cycles per record. Method (c) is also based
this
on number of records and in / case the presentation shows a separation tending to log (300) at the low probability levels. Method (b) does not seem to be consistent with the other two and this is presumably due to the assumption of a normal distribution of the maximum stress within each
Weather Group.
Also shown in these Figs. is the maximum stress reversal
recorded and this has been plotted at the point corresponding to the number
of cycles over which this maximum stress applies. This point should lie
on the trend line based on methods (a) and (b) for combined W.G.ts and in fact in the three cases given it lies just below the trend line for
method (a) but above that for method (b). This is further evidence that
13
-Also shown in Figs. 1, 18 and 19 is the statistical gauge projections in a ship's lifetime and in Figs. 18 and 19 the corresponding
theoretical values. Figs. li and 19 show reasonable consistency at
the lifetime point but the logger data in Fig. 18 is much different
from the other two. This is considered to be due to lack of data for
ship 'A' in the higher W.G.'s; in fact most of the data recorded was
for W.G.2.
8. DISCUSSION OF STATISTICAL STRAIN GAUGE DATA
Figs. 3 to 6 show histograms of stress reversals collected
over about two years of operation by each ship. The data have been
presented in this form for possible use in building up loading informatifor
fatique calculations; that is the loading on similar ships in various
combinations of the Weather Groups can be constructed to represent the
ship's life. The sanie data have also been presented in Fig. 7 in a
cumulative proportion form, that is in terms of probability. For the
lower probability values, ships 'A' and 'C', show the lowest stresses, with ship 'B' somewhat higher, and ship 'D' well above the rest. The
figure shows data for ship 'A' in the Load and the Ballast conditions.
The number of high stress counts was not considered large enough to justify other than a mean line through the data but there seems to be a tendency for the Load condition to be the more highly stressed, and
this is supported by the projections given in Table 1.
To use Fig. to predict the maximum stress in say a ship's
life one needs to know the number of stress counts over some unit of
time. For ship 'D' an average stress reversal period was found to be
12 secs which gives for a ship's life of 20 years about 1 x 101 cycles
or, in probability terms, 0.25 x 10 This figure could also apply
l4
-'A' and 'C', however, the average stress reversal period was about twice
that for ship 'D' and this gives for a life of 15 years about lO7 cycles or a
probability of 10 A comparison of the projected lifetime stress maxima
at these probability levels with those calculated using the method described
in the Appendix and listed in Table 4 shows that the projection lines
of Fig. 7 give results about 10% lower than those in Table 1. The results
given in Table 1 are considered the more accurate for reasons referred to
below and also for the fact that the projected lifetimes used in Table 1 are precise in terms of number of watches rather than of estimates of the
variable time period for stress reversals.
In addition to the projected maximum in a lifetime. Table 1
shows a 95% control interval for this value as well as the maximum value to
be expected in one hundred ship-lives. The 95% control interval is a range
within which the lifetime maximum will occur with a probability of 95%. An
increase from one life to 100 ship-lives increases the projected maximum
stress by about 35%.
It should be noticed that in Table 1 the data for ship 'B'
has been increased by including similar data from five sister ships. This
has allowed the method. the results of which are summarised in Table 14, to
be thoroughly tested. Generally the use of the maximum stress from every
watch was not satisfactory but by treating the watches, chosen at random, in groups of 10, 20, 100 and 200 the samples,from whìch the maximum stress was taken,became more homogeneous and the stress projections became
acceptable. It would seem that groups of at least 10 are probably necessary
and such groupings are probably sufficient, as even groups of 200, which were possible with the large sample under ship 'B', did not change the
-
15-The data for ship 'B' and its sisters have been entered in two groups, one for the ships as originally built and one for the ships after
a substantial modification which increased the midship sectional modulus.
The ratio of the original section modulus, at the Upper Deck, to the
modified modulus is
0.773.
This relates to a mean projected stress for alifetime of 255 NN/m2 before rnodification and 195 /m2 after modification,
which gives a corresponding ratio of 0.765.
Table does not contain maximum stress projections for ship
'D because the data for this ship has obviously suffered from truncation at the high stress end as can be seen in Fig. 6b and such truncation would
seem to distort the double exponential fit to the data.
CON CLUS IONS
Although all the data in hand have hot been analysed and not all the theoretical predictions of the stress likely to be experienced in a ship's lifetime have been completed, the comparisons which have been made support the use of completely theoretical methods for the calculation of
wave induced stresses for new ship designs.
The use of statistical strain gauges together with the use of a double exponentional distribution applied to the maximum stress in 10 watch-periods provides a relatively cheap and efficient means of deriving
data on maximum likely design loads. The gauge also provides histograms
i6
-The Miles eq.u. (i) has been shown to fit a conibine signal of
wave induced end springing stresses quite well. Further work is required to assess the use of this equation in long term predictions.
REFERENCES
Rice, S.O., Mathematical Analysis of Random Noise. Selected
Papers on Noise and Stochastic Processes, Dover Publications Inc., 1951t.
2. Miles, M.D., On the Short-Teim Distribution of the Peaks of
Combined Low Frequency and Springing Stresses. Soc. Nay. Arch.
& Mar. Engrs., Symposium in Ottawa, 1971.
Johnson, A.J. and Taylor, K.V., Wave Induced Stresses in Ships
in Service. Proceedings of Conference on Stresses in Service,
Inst. of Civil Engrs., 1966.
Martin, M.A., Freqiency Domain Applications in Data Processing. Technical Information series No. 5ÎSD 31t0, General Electric Co., Missile & Ordnance Systems Dept. Phil, U.S.A. 1957.
5 Hoffman, D., van Hooff, R. and Lewis, E.V., Evaluation of Methods
for Extrapolation of Ship Bending Stress Data. Ship structure
Committee, SSC-231t, 1972.
Goodman, R.A., Wave Excited Main Hull Vibration in Large Tankers
and Bulk Carriers, R.I.N.A. 1970.
Lewis, E.V. and Wheaton, J.W., Study of Combined Wave and
Springing Stress of the 'tEdward L. Ryerson". Soc. Nay. Arch.
8. Gumble, E.J., Statistics of Ectremes, Columbia University
THOD OF ANALYSIS OF MAXIMUM STRESSES PER SHIP-WATCH PERIOD USING STATISTICAL GAUGE DATA
Reduced Variate
The Double Exponential distribution is
-c(x-u)
P(x) = e e , for all real x.
The distribution has two parameters, a and u, and by setting a = i and u = O, we obtain a standard form with no parameters, termed
the reduced form:
P.(x)
ee,
for all realSuppose we have the general form and particular values of
a and u have been chosen. Then for any value of x with cumulative
probability P(x), a value can be found for the reduced form variate y,
say, which has the same cumulative probability. Thus,
P(x) = P(y),
ee
y =
and y is termed the reduced variate and can be obtained from x by a
change of origin nd scale.
Plotting a sample
The general form is -a (x-u) -e
P(x) = e
and taking natural logarithms twice gives - log {- log P(x)} = a(x-u).
If we have a sample of observations x,...xn, the cuinulat5,..
probabilities P(x) can be estimated from the ranks of the sample in
ascending order. For example, P(x5) is estimated by 5/(n+l), P(xl
by lO/(n+l) and so on (it can be shown that these are better estimat,,3 than 5/n, 10/n, etc.).
A plot of - log
I-
log {m/(n+i)}7 against xm should therefore be linear except for sampling fluctuations if the underlyinìdistribution is, in fact, Double Exponential. An example of such a
is shown in Fig. Al.
Since
-loa
i-log P(x)} = a(x-u) = y (the reduced variate)the expression -log
L
-log {rn/(n+l)} / can be regarded as an estimateof the reduced variate corresponding to x.
Estimating ci and u
For the linear regression equation x = a + by, it can be shown that the least squares estimates of a and b can be expressed as: b = ro
xy
¡G andax-by
where and and 2r correlation coefficient of x and y = standard deviation of x
= standard deviation of y x = mean of x
y = mean of y
For r close to unity, the estimate of b becomes
approximately
We have seen that theplotted data should be close to the
line y = cx(x-u), where y = -log
L
-log {m/(n+l)}I,
and we can w-rite thisline as x = u+ y/a..
Applying the approximate least square formulae we obtain estimates of a and u as follows:
1/ci = o ¡o
xy
or a = o ¡oyx
u = =
Grouped Data
A complication arises when the sample members are not
individually but only in groups. In this case, to derive the plott
positions, the procedure is to plot each group as one point, the
appropriate rank being the geometric mean of the highest and lowest
in the group. For example, if a group consisted of observation; ".
ranks 3, 4, 5 ... 12, the_plotting position of the group on log (i"
linear paper would be /x 12/(n+1) against a group average, or c'i;
and
3
In the estimation of a and u, grouping does not affect the
computation of y and o since these quantities only depend on the sample
size. y
Thus,
= E
[_ioS
C -log(
J I
.th
x. = central value of i group
i
th
f. = frequency in i group
i
Estimation of Group Central Values
Since the best central values can only be found after the theoretical distribution is known, an iterative method is used which can
be shown as
follows:-Set initial estimates of central values
4.
Obtain theoretical distribution
++
Find new estimates of central values
4.
Leave loop after three iterations
In all cases considered so far, the method has converged with
sufficient accuracy in three iterations.
The revised estimate of xj. the central value of the th group
L''-- is taken to beexpected value of the variate within the group. This
can be written:-2 L
ml
[_loS
[ -log ( a..)]_2
in
2However, for the evaluation of x and a ith. grouped data, central values must be assigned to the groups to gve the following:
1k
=nL Lx.
ii
i=l1k
2-
2 -anL
f.x.-x
X.
1=1ii
where andx. = E{xi-i<x<l}
=fP(x)dx fP(x)dx
where p(x) is the Double Exponential probability density function. Goodness -o f- fit
The components of the Chi-squared goodness-of-fit criterion are evaluated, one for each group. These are summed to give a single
value which can be compared with a suitable x2 percentage point from tables, with degrees of freedom three less than the number of groups involved (number of groups less number of parameters estimated less one). Groups with small expected frequencies (less than about five) are not used but are pooled to
give fewer, larger groups.
Projected Maximum Stress
In a ship's life of N watches the stress x which is expected to occur once is given by
l-F(x) = 1/N A(l)
where F(x) is the cumulative distribution of x, this result being
independent of the form of F(x). If F(x) is the double exponential
distribution it can be shown that the x in equation A(i) is also the modal maximum, i.e. the most likely maximum in N watches. In this
particular case A(l) becomes:
log N
xu+
A(2)a
To obtain an upper limit for the lifetime maximum, the method described
in reference (8), p.2)4, is used, in which N in equation A(2) is divided
by a small risk of exceedance,
u,
giving= log(N/a) A(3)
For example, for = 0.05 equation (3) becomes
x u log (20N) A( 14)
and this gives the stress which has a i in 20 chance of being exceeded
in N watches. By comparing equations A(2) and A(14) it is obvious that
this stress can also be interpreted as the stress expected to occur
exactly once in twenty ships' lives.
To find a lower limit for the lifetime maximum, the following
x
u+Jo
N i log L..1w
A( 5)where w takes values near unity. This equation is also valid for upper limits, i.e. for small values of w, but in these cases it reduces to equation A(3), which is easier to compute.
Control intervals for the lifetime maximum can be
constructed using equations A(3) and A(5). For example a 95% control
interval is found by setting w =
0.915
in A(5) for the lower limit andw
0.025
in A(3) for the upper lirait. These control limits have beenTABLE i SHIP PABTICULARS L B D T C C I I/ B W 'Y pp m m m Tonnes 2 2 m
cm m
Upper Deck 2 cm m Ship "B" Loaded 213.363oJ8
i6.1t6 9.1k 36,500 .600 -1.69x106 193,000 (Containership) *1. 38x106*l19 000 H.T. Steel in D'k Mild Steel in B'tm
Ship "C" Loaded 281.98 26 23.39
i8.18
199,300 .8L6 .909 7.069x106 75,130 (0.B.0.)H.T. Steel in D'k Mild Steel in B'tm
Ship "G" Ballast 281.98 114.26 23.39 12.8 135,000 .826 .869 1.069x106 75,130
H.T. Steel in D'k Mild Steel in B'tm
Ship "D" Loaded 218.0 30.L8 18.6 9.558
39,00
.727 1.2x106 117,830 (Containership)H.T. Steel in D'k Mild Steel in B'tm
*
As originally built After structural modifications
Ship "A" Loaded
330.71 51.82 25.603 19.577 284,700 .828 .906 10.09xl06 177 200 (Tanker)
H.T. Steel in Deck and Bottom
Ship "A" Ballast
330.71 51.81 25.603 6.02 76,100 .765 .817 lO.09xl06 717 ,200
TABLE 2
POSITIONS OF 100 INS. LONG BASE STRAIN GAUGES
(The statistical and logger gauges on each ship are adjacent to each other).
B Midships Port & Stbd 0.8 in Inboard
On longitudinal on Upper Dk. (Logger gauge on Port side only)
TABLE 3 WEATHER GROUPINGS
Weather Group Beaufort No. Wind Velocity, Knots
I
Oto3
itolO
II
4to5
llto2l
III
6to7
22to33
IV
8to9
31to17
V
lOtol2
48toT1
A i1 m aft of Midships 1.01 in to port off
On Upper Dk. Plating above a long'l
C
3.65
in aft of Midships Portside, 1.0 in InboardOn Upper Dk. Plating above a long'l
D Midships Portside, 1.2 in Inboard
On Longitudinal on Upper Dk.
Data Processing Scheme for Data Logger
Ilagnetic tape from Ship's Data Logger
Check tapes and re-write in computer compatible form
Probability distribution for all Weather Groups using ship weather distributions
FiG.
i
Apply calibrations, perform preliminary analysis and write to tape for further analysis
Filter data:
(a) 20 min. records for wave frequencies (b) Any 5 min. of a 20 min. record for
vibration frequencies
Plot filtered data on time base,
Spectral Analysis
Calculation of moments for 20 min.
or 5 min. records.
Histograms of peaks and their R.M.S. Probability density distribution of peaks
using Miles1 equationlbn each record
t
YHistograms of stress reversals and their R.M.S.
1
Probability distribution for 5 min. and 20 min. filtered data assuming Rayleigh distribution of reversals and normal distribution of R.M.S.;
for each Weather
Group-V
Probability distribution accoiìting for all Weather Groups using ship weather distributions
Probability distribution for 5 min. and 20 min. filtered data based on
measured maxima & ptedlcted extreme stress reversal for each RJ-LS.
assuming a double exponential distribution of extremes; for each Weather Group,
(
peaks
potie)
Pci±cie
&!k
iVe.. ¡ve pek
R
('o5ìVPid
(y
peaks)
flG.2
ILLUSTRATION or COMßIND & flLTERD STRESS 3IGNALS
COMB/I\1E20 S14t'iÌL
Ò1c 57Es
FLyERED /fßRMYIoN
5r,zEs3
FI-TEe?ED w,qv,
5Tfl?55
G No DRAWN BY
Tos/ ti4
TONS/FIG.3
3 25NiP 'P
LLOD
-t-A1..LAST)
HLSTOGRAM5 OF 5TRSS CYCLES
Y WEATH
GQUP5
SrATLSTtCAL GAUGE
W.G. i N HZS72Z2
RESEARCH ITEM Iw.c. 3
NC H'Q3; 1043
j
4 o LØ OF NUVt6EOF CYC.LS
(..5w.c. 2
Ñ k'S
2804
W.Ca. 5ERSr 6
2 3 - 50 so too-BRITISH SHIP RESEARCH ASSOCIATION
REPORT No
DRO. No DRAWN BY,.
s1.ss
7o S/lu'
Src s
* ON S/INFIG.4
3 -1 o I 2. 3W.C. 3
RESEARCH ITEM REPORT No
W
34i-(s9Ñ° HS Io1(.
I'1-4
0 LO NL)Mf3E.f b2
BRITISH SHIP RESEARCH ASSOCIATIO
WALLSEND RESEARCH STATION
NtHS
4.4
1 7 s 4 3 2 3w.. 2
Ñ HS 3160
I
w,c, S
?"JH'S 4
- ici
MN'
loo
MW/ loo aSHIP
'(ST6D C*AL))
HiSTOC,R.AM5 OF 5TR.SS CL
6Y JEA1kE.Z GÇWL)PS
5TATISflCAL
AUDRGNo DRAWN 8V £TR.ta ss TONS/IN2 7 S-rR..sc -ThÑ5/IN& 1 3
FIG.
5-i
ISNIP
(LQAD +.57)
ISTOcR.AM
OF STRESS C..YCLES ¿Y WE.ATHE
C,POU15 (w.cm.); FROM
STATISTICAL
GIALJGE W.CA.i
N t-S
357
w.. 3
N' k'RS
647
H
7 4 3 2 1 o i z 3 4LOC OF NUMP.
F CYCL
2N HS
I'5O
I 2 14
.5-- LOO'Q
RESEARCH ITEMBRITISH SHIP RESEARCH ASSOCIATION
REPORT No
WALLSEND RESEARCH STATION
M
DRG No DRAWN BY
To
N 4 3 2 'oFIG. 6
3 2 oRESEARCH ITEM REPORT No
,1 I
w.c. 3
NJHS
t?(0 oj
LOG 0F NUMßE.f. OF
CYcL-BRITISH SHIP RESEARCH ASSOCIATIO
WALLSEND RESEARCH STATION
SHIP L
tSTORM5 OF TE55 C1CLS ß
VATHR GOUP5
o
STATISTICAL
GAL1E
Wc,
j
wc,. 2
ÑH'IH5
N l-RS1800
WG,. 4
H'S
77'
ls-o2
4-s-ísr
- loo
-6 4 4rtc. 6b
¡0 g ¿SHt'
S N I I I 1 2 3 ÇLO
OF NU1EvZ OF CYCLES
I*STOCP..AM OF STRS5 CYCLES 6Y WaPTL-4EYZ G ROUP
(w G); FÇ2OM
5TAT1STIeAL GAUGE
!50
-50
BRITISH SHIP RESEARCH ASSOCIATION
WALLSEND RESEARCH STATION
UK(j NO ,SC)CMrÇLfl I
I JI
DRAWN BYjREPORT No
CUMULATIVE PROPORTION DISTRI8UTON FOR STATISTICAL GAUGE STRESS CYCLES
o
,1FIG. 7
NtP St-i P 'b TIo-',
I- -3 IoCL)MULATIVE
PRO8AL3ILITIES
ThP
Lot' CCNTNTION
S HIP ,A', 6ALLAST
4' SI-412 'L3PCRTGAUGE
SHIP '6'SThb
o
SHIP 'C
+ St-liP
8'
STE.SS
go_4 I - So4-
2loo -
lLT)
Oc
°'854
,Zì45
722
OF gECOIZ)'0
20
2o
t'T,Ñ5
LCO/D
flG.8
SPECTRA or STRESSES
UÑ/.LYRQ
0609
3t3
OF RECOOIo
2o
30
'8
/6'4-
2Io
WPVE ÌNbUCD
t-Fo
EQu()
-2
o2
3
4
O 2. 32otmvs. ,CCO(?O
5/-//P A
FK. 9
bISTRL&TIONS
OF PEAK
ST?ESSES
Peak 57±res5
/-'eak 5L1ve5s
57re.s
Var ¡c
3tYe5s Vr/ance
u,vrlL-,-E ÇI D
ç7L-T-I
(ED
100
¡00
'Q
L tSJ1Jfv5 icoRO
Ç/fVfLTI/
D
7(Q.Q227
M5 3-36
CF ?ECOß bp. ¡q.
FO
SPEC1RP O
SRESSES
II
I t t IF /LTiiU
0.93
&.JG. O RECO!Zb o50
lOo sovi
32
FgOM QOE(I)3re5s Vr,dM
o
¡ Peak. .5tre.5$ .S PR NG t N G5tes
V'ariafrlce¿J NF/L7&ic'&L
FL-TRED
5î'mv RLECOD
51-I/p A
FiGli
1$TRIEUTIONS O
PEAk STI?ESSS
o
2
3
Peak 5ve.5
IIo
/0000
/000
loo
=
f
ILYO
loo
FK. 2
5lP D
cO.69
íM5= 2187
OF ECoRbIo
o r»7/'.
4cO
2o
J
o
c_ p. î'.'?.Io
SPECTRA OF STESSE5
UNF(I-TZER2D
oÇO 432
N15224-8
OF2o
/4
/2
2oî'i/Ìv'S. /LCQRD
'D'
flG.
3
DSTRUTIONS OF PEAK STRESSES
2
3
Qz
3
4
Peek SLV-e55Pedk 5tve5s
5Lvej.ç/ay/.r7ce
5t'ess V-,ice
F,LTE,'ED
¿J,VF/ LTERED
10000 Iø O
'co
'ô
q-F1k.To
51'4/N5. ICoiO
I
C. P. t-i.FICi. 14
SPECTRA 0F STRESSES
2o
-1ç3 16 ¡2-
(o-SPING ING
C) 23
4
2
o
PeQk ..5L,-e,sPeÁ j&eç
iireSj
5yes Veri,ìce
FILTERED
UNF/L-TLF?ED
5 ¡'-'1/NS. RECOR D
51-1/P D
RIG. IS
F1G. 16
_I I-5H/P iO
(J.ave £ncDLJfl L:
r-$? ve$55}
,Ci'IO,t-lÑh.,/Ty
0F
¿3î15 57,E55 y-1fT/u/N
7/i'1.R
cRoUP iii
I I I iUNUL,PF(VE PRo5F1/L/7ES
J I 98 9990
95
2o
3o4o 50
6 70 20.5
5 so 7050
ft'15
40
30
2o
l0
0-J 0 0fFIG. 17
139
MP4/2
roLL W.G BASED ON MEASuRE
MAX. STiESS PER RECORDWEATHER
J
CROUP I AL.L WC.sW63
's'F'OR ALL W.(i e}AE:D ou CLCuLATEb
-MAX, STRESS DITRiß.UTIOÑ
lo_%
I0'
I0'
k,-'
IOto.'
L0'
IOCUMI)LATIVE POAIL%TY
_____L..o4(300)
UP
BALLAST STRESS REVERSAL TREÑDS BA E
ON LOGGER DATA
140 120loo
STRESSM
40 2Q 's' 5.' s''o
.5'STAT C.AuE
'orEcTl Or.«fZoM
TALE 4)i2Oi$
pzorE.c.TIoN rROM
.MS. ALL w.G.
s..' s... s..' .5' 5% '.5.. .5' MAX. MEA5ORED STWSS-
5% . 5' 5%. 5. .5-BAUD ON R.M.S Or STRS REVERSALSFOR ALL W.G. BASEb ON
J
MEASURED MAX. STRESS
PER RECORD.
WEATHL GRoup I
7
s-' w.c ALL W.Cs w.4.2 s' -S-.5». '5 S--MPX. MEASURED STRE.$S 5' -s--SsN
BA5Eb QN Ç.M.S. or STRESS REVERSALS'
I0 IO tO so 10CUMULATIVE
PRO6AiLITY
L
LO(3oo
FIG. IS
SHIP
BALLAST.
STRESS REVERSAL TRENDS ,ASED ON LOGGER DATA
STRESS M
40
30
20'o
TgES -i2
MÑ/70
G0VOR ALL WATH
P1.SED ON
CALCUL.ATE. MAX STRESS DTRIB.UTiO4
.50
STAr. GAU
PROTECT I(.*.1 (FoM
TAtL
4) =I3
MN/
¶-4EogETicRL CL(.
FOR VETJCAL 6E'JbIÑ(.
MiI/
"S
P«OCTION FROM RMS,ALL W.Gs.
b
(O
STAI Au6E
pOTEC TIûN(FìOMF
WEATHE4
32o MÑ/i1 N
pPOrEcTIoÑ rZoM
M.SJ ALL '.'f(5
FOc ALL WC YASED ON
MEAUcCb MAX. TßESS PER
W.61. 3RECORD
W,G, £
W.s
Foc ALL WGs ,P5E..tb ON
4 ALL W. CALCULATED MPX. SISSDSTRI6LifloN
-'f
N MAX 'S..-5. NMAWED
's '5S 5.-STESS
N 'S.+
.5... 's\
Io_' Io_eCUMULATIVE pgoeABILIyY
N N'
\
N\
"S.'\
N\
\
TgENDS BASED ON I.M.S OF
'TRESS REvEALS
k
LOG.C300) 32o 280240
200 S TR E P...'i 2 160 120 pOSHP,
STRE:SS REVERSA TREÑDSBASEDONLOGGER
DATA
flc
Al
015o
ss
ÎÖNS/1Ñ2
100MÑ,/2
4-- SO 2SHIP
tONO. OrwATcS 550
t I I I 04 0.6 0.8 0.9 0.95 C.UMULATIVEPcor3LrTy
EXAMPLE ooUaLE
.XPONENTtPL. RIT TO STRESS iv1AXIMA RROrvi S1/,Ti5TIC?1L. GAUGE
+ I I I
O9
0.999
0.9999
TABLE 4
LoNC TERM POZTECTEDST.ESSVALU1ES BYF(TTNG DOL)&LE EXP0NNTLDST SLJTJOÇ'S To
SY?E.SS MAXIM/A MEASoRE
s'y' THE STATsTKAL STZA(N CUGES
NOrE: T X2VLUE SHOULD ÇE.P'JEALLy &E LESS TI-IAN TI-tE SX StC.ÑIFICANT Poir-..'T VALuE Fòi
A P4EAt'JINcrUL POTE-CTI0N OF STRESS T0LO%W Ö6AßILrÍY VALUES.
t
NOT SIG NIiCANT AT T'-tE j0/LEVEL,YHEREFORE A STRESS POECflON HAS ße-.N MADESHiPS LiFEflME Th<EN AS 30000 WATCHES(15 000 IN GALLASTI0o0 ¡N LoDCcTIors)
ro.sNIpsA
*
APPEARS Toe>E AGOob FT Of'd PL0TTING(ACruIL FIT SHN N FKJ¼).AS -r-tIS HI(,I-LY I(*NWCANT I)UE To A L.AQE CONT6UTI0N FROM ONLY ONE STrESS
A MAXIMtJM 5TES Po.TIoN I4'.5 ßEJ'4 tvlADE.
Se
bA-rA Tyç'5piPtC
Ssz 2 bÇ.IZ
FJ,oM
5%
Sui.po,j-p'OCÍEr) MAXIIt.IL)M STsS P.EVERSALS
'osit-uzwi' V#.c-' IN Lo,./EZ 2-SZ COMTOL tx'c'- 2-6 CONTROL MOST LIKELY VAL'. IN
S,jic' Li Poit'.rr POINT tooS.eL'
Ç WPÇTCI-t
1N-48-tt5
109
i-' 2(.o
I 3
122
186
198
Lobb
3HS
I 46
I ¡ 3 7-8I 39
I 2 2 1 8 31 99
'A48s
us
18
2 te-0136
120
179
190
MLLAST36is
¡SS72
378
i 35
(20
179
190
C ALLW#TCI1E 92S ,4-.54
9-5 LoADED J ipj (0 923'9
I38
131I (8
¡69
(79
'C ALL4JATCHES S0 (G-2 20
i2o
106 1S9 1686,LLAST I IN (0 55 NOT EÑOU(,I-i PTP
è; ALLWATCHIS S5l