Exam 2012: Solution SH problems Problem 1 a) * Principles regarding structural design: N-001 Actions and action effects: N-003 Code check and utilization: N-004 * For 80m depth we do expect the natural period of jacket to be reasonably small, say less than 2s, suggesting that dynamics can be skipped or at least estimated in a simplified way. The recommended approach would therefore be the design wave method. * Governing combination of environmental process is: 10-2 – annual probability wind, 10-2 annual probability wave, 10-1 – annual probability current and 10-2 – annual probability sea level (from N-003, Table 4. Page 16 in exam). Wind: Assuming wave induced response is governing, we could use the 1-minute mean speed corresponding to 10-2 annual exceedance probability (3s gust wind will also be accepted): 51m/s at 10m above sea level (Table 1 and 2 in Metocean Report, page 17 of exam). Waves: Design wave method requires the 10-2 – annual probability individual wave height and a 90% band for the associated wave period, i.e. Table 4, Metocean report, page 17: h0.01 = 29m and <90%> band for period: 14 – 18.2s Current: 10-1 annual surface current speed: 117cm/s (Table 5, metocean report, page 17/18 of exam) Surface level: Ttidal amplitude + 10-2 – annual probability storm surge: 1 + 0.9m = 1.9m. (Table 6, Metocean report, page 18 of exam). b) * We need the load history for the whole period, we need boundary conditions at the start of the simulation, we have to introduce some assumption regarding the change of acceleration from one time step to next. In order to make our assumptions accurate regarding acceleration, length of time step should be small as compared the variation of the response time history . * In order to obtain a 3 hour simulation of good quality simulation, simulation must in principle be somewhat longer. One should skip the first part of simulation due to resonant effects that may occur at start up. If a standard FFT with equidistant frequency components are used, the frequency resolution must be 1/T if T is total required length of simulation. 1 c) * The property of a probability paper for given distribution is that if this distribution is plotted in the paper, it will appear as a straight line. 𝑥 𝛾 * Given distribution: 𝐹𝑋 (𝑥) = 1 − 𝑒𝑥𝑝 {− } 𝑥 𝛾 Reorganizing: 1 − 𝐹𝑋 (𝑥) = 𝑒𝑥𝑝 {− } Taking the natural logarithm and change sign at both sides yields: − ln(1 − 𝐹𝑋 (𝑥)) = 1 𝛾 𝑥 Using the left hand side as the ordinate and x as the abscissa, the distribution will be straight line, i.e. the probability paper would be an exponential probability paper (since the actual distribution is the exponential distribution). Problem 2 a) The minimum airgap that can be used in order to avoid deck impact for the 10-4 – annual probability crest height is (see Table 6, Metocean report, page 18 of exam): 23.5m Thus we have for this case: d = 80+23.5+10 = 113.5m The natural period for this condition is given by: 𝑇113.5𝑚 = 2𝜋√ 𝑚 𝑘 = 2𝜋√ 𝑚 𝑐/𝑑 3 = 4.6𝑠 As deck is jacked to 50m above mean sea level, d is given by: d = 80+50+10 = 140m The natural period for case reads: T140m = 7s. It is seen that by increasing deck height above sea bed, the natural period increases considerably. (The important distance is from sea bed to center of gravity of topside.) a) Environmental contour method is used. Provided the worst sea state is correctly identified, one can estimate the long term 10-2 – annual probability deck displacement by calculating the 90% value of the extreme value distribution: exp( − exp( − 𝑥0.01 −1.14 0.14 )) = 0.9 x0.01 = 1.46m If we could neglect the short term variability we could use the most probable value or the median as a proper estimate. However, short term variability can usually not be neglected and in order to account for that we have to select a higher percentile (90%). The choice of 90% is an approximation. Since we assume that an exceedance probability of 0.1 correspond to a return period of 100-year, an exceedance probability of 0.01 should correspond to a return period of 1000 years and an exceedance probability of 0.001 should correspond to 10000-year return period. x0.001 = 2.11m 2 This approach is not tuned for estimating 10-4 – annual probability values from the 10-2 annual probability contour line. Since we are not accounting for the importance of the very severe sea states regarding the exceedance of the 10-4 – annual probability response, it is expected that this estimate is somewhat on the low side. b) One could use Monte Carlo simulation. The latest distribution function is based on merely 12 data points. By assuming this distribution to be correct, one can simulate many samples of size 12. Thus we get large number of valid samples from our fitted model. If – say - we generate 200 samples of size 12, we can establish a confidence band around our fitted model. If the old model (Eq. (2.2) falls inside this interval, one can conclude that the observed difference may well be a result of statistical uncertainties. A possible realization from the Gumbel distribution can be obtained by generating a variable for a uniform distribution between 0 and 1. Denoting this value by u and replacing F X(x) with this value, the corresponding realization is found from Eq. (2.1). This step is repeated 11 times in order to have one sample (of size 12) that could have been observed if our fitted Gumbel distribution was true. Thereafter this procedure is repeated many times – say 200. Then you can estimate the distribution function for each sample of size 12. Plotting all these in a Gumbel probability paper together with the two distrbutions, Eqs. (2.1 and 2.2) we can deviation may well be due limited sample size if Eq. (2.2) fall within “cloud” of simulated distribution functions. The best approach for an accurate estimate is to do a full long term analysis. By such an approach we don’t have to choose this approximate percentile. If contour method is to be applied, one should increase the number of simulated observations underlying the extreme value distribution to at least 20 preferably more (say 40). c) In order to use Eq. (2.3) for obtaining q-probability values for 𝑌̃, we have to determine the corresponding exceedance probability per storm. The mean annual number of storms is: 4.83storm/year. The target probability per storm thus reads: qstorm = q/4.83 The q-probability value for 𝑌̃ is then given by: 1 − 𝐹𝑌̃ (𝑦 ̃) 𝑞 = 𝑞𝑠𝑡𝑜𝑟𝑚 If this value shall be correct, the short term variability of storm maximum response must be negligible. In most cases that is not the case and the estimate is therefore on the low side . 3