Health & Safety Executive

HSE

Response spectra for explosion resistant design and assessment

Prepared by Atkins MSL Engineering Limited for the Health and Safety Executive 2006

RESEARCH REPORT 484

Health & Safety Executive

HSE

Response spectra for explosion resistant design and assessment

Steve Walker & Rashid Shahsavar Atkins MSL Engineering Limited Haleworth House Tite Hill Egham Surrey TW20 0LT Brian Corr & Vincent Tam BP Upstream Technology Group EPTG Building C Chertsey Road Sunbury on Thames Middlesex TW16 7LN

This Report describes a project to examine the explosion test results obtained at Spadeadam in the 1990's from the point of view of structural response. BP supplied the Flame Acceleration Simulator (FLACS) pressure traces corresponding to the experimental results obtained at the time. This enabled a comparison to be made between experimental and simulated traces taking into account the target structures' dynamic characteristics and plastic reserves of strength. The main objective of the project was to derive static design pressures for sizing main structures and barriers. The comparison Response Spectra has given a more robust measure of the conservatism or otherwise of the simulation traces than direct comparison of simulated and experimental pressure traces would yield. This report and the work it describes were funded by the Health and Safety Executive (HSE). Its contents, including any opinions and/or conclusions expressed, are those of the authors alone and do not necessarily reflect HSE policy.

HSE BOOKS

 Crown copyright 2006

First published 2006

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means (electronic, mechanical, photocopying, recording or otherwise) without the prior written permission of the copyright owner. Applications for reproduction should be made in writing to: Licensing Division, Her Majesty's Stationery Office, St Clements House, 2-16 Colegate, Norwich NR3 1BQ or by e-mail to hmsolicensing@cabinet-office.x.gsi.gov.uk

ii

EXECUTIVE SUMMARY
This report describes a project to examine the explosion test results obtained at Spadeadam in the 1990s from the point of view of structural response. BP has supplied the FLACS simulated pressure traces corresponding to the experimental results obtained at Spadeadam during the 1990s. The technical reviews by the HSE were performed by Roland Martland. Objective The main objective of the project was to derive static design pressures for sizing main structures and barriers. Motivation Structural engineers are presented with complex simulated or experimental pressure traces and are expected to design efficient structures to resist these loads. It is usually desirable to allow significant local plastic deformation in order to arrive at an efficient design. This often results in the need for a non-linear, dynamic design approach which may be expensive and can be prone to error. The Response Spectrum approach brings the explosion response checks within the framework of a conventional design or modification project. The information needed to apply the method is an estimate of the natural period of vibration of the target structure and its allowable ductility. The load is represented by the effective pressure Pe and the load duration td. Generation of response spectra for the experimental and simulated results has given a more robust measure of the conservatism or otherwise of the simulation traces than simple comparison of peak pressures would yield. The response spectrum approach has been in use for decades in the earthquake response context and was in fact pioneered in the Second World War to calculate ground motion effects and structural response from explosions [2]. The Use of Response Spectra The structural element to be assessed may be a panel, a deck, module or a whole topside if they can be idealised as a one degree of system oscillator. This process will be familiar to designers who use the Biggs response method [3] and is in routine use. The next stage is to assess the allowable ductility for the structural element. The allowable ductility of a structural element is a measure of the amount of deformation the element can sustain before rupture or when its performance standards cease to be satisfied. This is usually expressed as a multiple '' of the effective elastic yield displacement.

Next determine the load duration td and the effective pressure Pe from the available loading information. The explosion loading may be determined from local conditions by the use of nominal overpressures, previous experience, risk classification, simulations or experiment. The Figure below is a Response Spectrum obtained by averaging all Response Spectra for all sensors and all tests in the Phase 2 full scale explosion experiments carried out at the Spadeadam site. Each curve corresponds to a different allowable ductility value from 1 to 10.

Mean Response Spectrum for All Phase 2 Tests If the allowable ductility for the structural element was determined to be 2. The correct curve in the figure is the one corresponding to =2 (the purple curve). If the structures natural period turned out to be half the load duration (td/T=2) then the =2 curve would yield a dynamic load factor (DLF) of 1.15 which is the corresponding value on the vertical axis. In this case, the required static resistance R is 1.15 times the effective pressure Pe . This is also equal to the equivalent Static Design Load Pstat. For a trace with Pe = 1000 mbar (1 bar) the required static resistance for the target structure would be 1150 mbar. A static elastic analysis with a blanket load of 1150 mbar could then be performed to assess the structure in the normal way.

The Figure below is a Response Spectrum obtained by averaging all response spectra for the FLACS simulations performed to simulate the Phase 2 full scale explosion experiments.

The main difference between the experimental and simulated response spectra is that high frequency loading components which are present in the experimental traces and which show up as spikes on the experimental traces. The simulations are not able to represent these high frequency components. These high frequency components tend to stimulate response at small natural periods corresponding to large values of td/T. Conversely the test results do not exhibit the peak in the response spectra around resonance (between td/T = 1 and td/T=2). Use of simulation response spectra will give conservative results in this range. The response spectra for higher ductilitys have the resonance around td/T = 1 suppressed. Basis of the Derivation of Mean Response Spectra Each pressure trace was run through the response calculation for 23 natural periods chosen to give td/T ratios in the range 0.1 to 10. The traces were each run for each natural period above and for 100 structural resistances. This yielded a response surface with 2,300 points for each trace. The data examined included:Phase 2 tests 27 tests each with about 20 measurement positions giving about 500 traces. Phase 2 simulations 27 simulations. The first 20 traces for each simulation were run through the response calculations. Phase 3A tests 11 were selected from the 45 tests giving about 220 traces. Phase 3A simulations 9 simulations were selected from the 45 supplied to correspond to the chosen test results.

The construction of these response surfaces involved in excess of 3,000,000 sets of response calculations. Four workstations were kept fully occupied for a period of two months to provide these results. The responses were then averaged for each test and over all the tests. The corresponding simulated data was processed in the same way.

Contents
EXECUTIVE SUMMARY ....................................................................................................... 1 1. INTRODUCTION ............................................................................................................ 7 1.1 General ..................................................................................................................... 7 1.2 Objective .................................................................................................................. 7 1.3 Motivation................................................................................................................ 7 1.4 Report Contents........................................................................................................ 8 2. EXPERIMENTAL EXPLOSION PRESSURE TRACES................................................ 9 2.1 Spadeadam Tests...................................................................................................... 9 2.2 Experimental Pressure Trace Characteristics......................................................... 10 2.3 Phase 2 - Tests........................................................................................................ 13 2.4 Phase 2 - Comparison of Test and Simulated Pressure Traces .............................. 14 2.5 Phase 3 Simulations and Tests ............................................................................... 16 3. BIGGS RESPONSE CURVES THE PROTOTYPE PROBLEM ............................... 20 3.1 Derivation of the Biggs Curves for Peak Response ............................................... 20 3.2 Use of the Biggs Response Curves ........................................................................ 22 4. RESPONSE SURFACES ............................................................................................... 24 4.1 Examples Response Surface (Biggs) ..................................................................... 24 4.2 Response Spectra from Response Surfaces............................................................ 25 4.3 Response Versus Resistance .................................................................................. 26 4.4 Biggs Load and Response Scaling ......................................................................... 27 5. THE USE OF RESPONSE SPECTRA .......................................................................... 29 5.1 Overview of the Method ........................................................................................ 29 5.2 Structure Idealisation ............................................................................................. 30 5.3 Allowable Ductilities ............................................................................................. 31 5.4 Load/pressure Time History Idealisation ............................................................... 31 5.5 Required Resistance - Equivalent Static Loads...................................................... 31 5.6 Explosion Load Sources......................................................................................... 33 6. RESULTS ....................................................................................................................... 34 6.1 Discussion of the Features of Response Spectra.................................................... 34 6.2 Comparison between Simulated and Experimental Response Spectra .................. 35 6.3 Fourier Transforms and Response Spectra ............................................................ 36 6.4 Response Scaling ................................................................................................... 37 6.5 Generic Response Spectra...................................................................................... 37 6.6 Repeatability Test Results...................................................................................... 38 7. FUTURE WORK............................................................................................................ 39 7.1 The Earthquake Analogy ....................................................................................... 39 7.2 Extension to Multi-Degree of Freedom Systems ................................................... 40 7.3 Reliability and Probability of Failure..................................................................... 40 7.4 Membrane and Tension Effects ............................................................................. 40 7.5 Confirmation of the Response Spectrum Approach - Variability .......................... 40 7.6 Robust Software Simulation/Test Comparisons .................................................... 41 7.7 HSE JIP Case Study ............................................................................................... 41 8. CONCLUSIONS ............................................................................................................ 42 9. REFERENCES ............................................................................................................... 43

APPENDIX A CHARACTERISTICs OF THE PHASE 2 AND PHASE 3 A TRACES EXPERIMENTAL AND SIMULATIONS RESULTS .......................................................... 45 APPENDIX B SPADEADAM TESTS RELEVENT OTO REPORTS............................ 46 APPENDIX C AVERAGE RESPONE SPECTRA FOR THE PHASE 2 AND SELECTED PHASE 3A TESTS.................................................................................................................. 47 APPENDIX D TRACE PROCESSING ............................................................................ 48 APPENDIX E CLOSED FORM SOLUTIONS................................................................ 50 NOMENCLATURE AND DEFINITIONS T = Natural period the time required for a freely vibrating structure to complete one cycle of motion. (secs). (or MU) = Ductility The ratio of peak deflection to the deflection at first effective yield. td = Load duration (secs). R = Resistance at effective yield (mbar). Xe = Deflection at first effective yield. Xmax = Peak deflection. Pe = Effective pressure in a pressure trace (mbar). Ppeak = Peak pressure from pressure trace (mbar). Pstat = Equivalent Static Design Pressure. Sf = Shape factor for pressure trace (Ppeak/Pe). I+ = Positive impulse of pressure trace up to first intercept. DLF= Dynamic load factor (R/Pe).

1.
1.1 GENERAL

INTRODUCTION

The work described in this Report was managed by MSL Engineering Limited (UK) with the assistance of BP Sunbury. The Technical Officer from the HSE was Roland Martland. The project was completed in early 2006. A specialist Seminar on the subject was held at the HSE offices in Rose Court in London to disseminate the results to the Offshore Industry. The spectral method is described in the OTC paper OTC 17242 presented at the OTC in Houston in May 2005 [1]. The method is also briefly discussed in the Commentary of the new API RP on Fire and Blast [15] which will be made publicly available during 2006. The results of this work are reported in OMAE2006-92512 [16] to be presented in Hamburg (June 4-6 2006) BP supplied the FLACS simulated pressure traces corresponding to the experimental results obtained at Spadeadam during the 1990s. This enabled a comparison to be made between experimental and simulated traces taking into account the target structures dynamic characteristics and plastic reserves of strength. The comparison Response Spectra has given a more robust measure of the conservatism or otherwise of the simulation traces than direct comparison of simulated and experimental pressure traces would yield. 1.2 OBJECTIVE

The main objective of the project was to derive static design pressures for sizing main structures and barriers taking into account the dynamic behaviour and the inherent reserves of strength present in these structures. 1.3 MOTIVATION

At an early project phase, there is a need for some quantitative guidance on the characteristic magnitudes of the explosion loads to be used in any given situation. Structural engineers are often presented with complex simulated or experimental pressure traces and are expected to design efficient structures to resist these loads. It is often desirable to allow significant local plastic deformation in order to arrive at an efficient design. This has resulted in the need for a non-linear, dynamic design approach which may be expensive and can be prone to error. The use of valid equivalent Static Design Pressures which are the output of the Response Spectrum approach avoid the need for such analyses. Plastic deformation is quantified in terms of the ductility, defined as the maximum deformation divided by the deformation at yield. The allowable ductility is limited by the function of the structural element considered and the nature of its connections to the surrounding structure.

The spectral approach enables explosion loads to be derived which can be used in a conventional static, elastic analysis with conventional code checks being applied. This brings the explosion response checks within the framework of a conventional design or modification project. A robust representation of the loading will aid re-assessment. 1.4 REPORT CONTENTS

Chapter 2 examines the available experimental data and the corresponding simulations, and describes a method of deriving the parameters which characterise the severity of the loading. Experimental and simulated pressure traces are compared and contrasted. Chapter 3 describes the Biggs approach to response determination The Biggs model assumes a triangular form for the pressure time histories. Chapter 4 introduces the concept of Response Surfaces, which are the basis of the deviation of Response Spectra. Response Spectra for the Biggs response surface are derived and the scaling of load duration and structural resistance is discussed. Chapter 5 explains the implementation of the Response Spectrum approach. Chapter 6 gives a review of the results and extracts the main conclusions. Generic Response spectra from FLACS simulations and tests have been compared and their properties explained. The effects of spurious frequency components introduced by the Biggs triangular load idealisation are identified and explained. Chapter 7 discusses future work and extensions to the Response Spectrum approach with reference to the Earthquake response analogy. Chapter 8 lists the main conclusions. Appendix A tabulates the characteristics of the pressure traces for experimental and simulated results. Appendix B lists the relevant OTO reports. Appendix C gives average response spectra by test and for all Phase 2 and Phase 3A tests. Appendix D describes the assembly of the response database from the data and the averaging process which has been employed in deriving averaged Response Spectra for single tests and for the whole test series. Appendix E contains a discussion of the features of Response Spectra. Analytic, closed form solutions have been derived for parts of the response surfaces for comparison with the results obtained by numerical simulation.

2.
2.1

EXPERIMENTAL EXPLOSION PRESSURE TRACES

SPADEADAM TESTS

A large number of full scale explosion tests were carried out at the Spadeadam test site in the UK between 1990 and 2002. Phase 2 of the tests was financed by a joint industry project managed by SCI, the Phase 3A tests were performed later and wholly financed by the HSE. These explosion events were also simulated using amongst others, the Computational Fluid Dynamics (CFD) predictive code FLACS. Direct comparison of the pressure time histories and in particular the peak pressure values has been inconclusive. The test results contained a large number of short duration spikes superimposed on a generally smooth curve, of a form similar to the typical pressure time history shown in Figure 2.1.
Pressure - Phase 2 Test 7 Trace 20 (averaged 1ms)
5000

4000

3000 pressure mbar

2000

1000

0 750 -1000 t 800 850 900 950 1000 1050

Figure 2.1

Typical Pressure Time History From Experiment

Whatever the cause of these spikes, investigations have shown that the influence of these short duration loads is insignificant for structural components with natural periods more than 0.02 seconds (natural frequency less than 50 Hz) - which includes most components. These spikes, which can take the observed peak overpressure Ppeak beyond 12 bar, make Ppeak an unreliable measure of explosion severity. Pressure traces are often smoothed before further processing by time averaging over a period of 1 millisecond. This effectively filters out many of the very high frequency components.

The effective pressure Pe is the measure of the severity of the pressure trace which has been used in this work. Pe is based on the positive impulse and is independent of the precise shape of the trace. The Duration td is also derived from the trace by considering the cumulative impulse of the trace. The derivation of Pe and td from experimental and simulated pressure traces is described in the next Section. 2.2 EXPERIMENTAL PRESSURE TRACE CHARACTERISTICS

An explosion will produce a large range of pressures and impulses in the affected area. Figure 2.2 gives an example of this variation for Phase 2 test 24. Different parts of the structure experienced different loads for the explosion scenario for this test. Any structure must be designed to withstand loads from a large range of possible scenarios.

Figure 2.2

Variation of Peak Pressure and Impulse Measured by Sensors for Spadeadam Phase 2 Test 24

In the assessment of the capacity of a structure to resist explosion loads, it is not sufficient to take only one scenario as defined by, release rate and location, gas cloud size, ignition position, ventilation conditions and ignition time. One method is to choose a number of scenarios with roughly equal probability of exceedance as representative of the explosion hazard. It is not acceptable to take benefit from low pressure regions for any particular scenario. In this situation the variability of the loading within the explosion region may be overcome by rational choice of Design pressure time histories at different locations. Representation of load severity Impulse, defined as the integral of the pressure trace, is an important measurement for explosion structural response and is an averaged measure as it includes all data in the pressure-time history. There is also a more consistent correlation between impulses for actual test results and CFD simulations. Measures of load duration td have been found useful in scaling the target structures natural period on the horizontal axis of Response Spectra and has been used as the time scaling in the Biggs response curves [3]. Figure 2.1 illustrates the difficulties in attaching a meaningful value to this measure for experimental traces. Two main options are available. The conventional approach is illustrated in Figure 2.3 from reference 4.

10

Figure 2.3

Conventional Idealisation of Pressure Trace [4]

The duration td is denoted as in the figure. The MEGGE protocol for idealising pressure traces follows a similar approach. The weakness in this approach is the assumption that Ppeak is easily identifiable and is a valid measure of the severity of the explosion trace. A more rigorous approach, tailored towards actual experimental and simulated traces is shown in figure 2.4. This method was originally presented in reference 5.

11

Impulse - Integral of Pressure to time t Phase 2 Test 7 Trace 20

1.20E+09

1.00E+09 I+ positive impulse 8.00E+08 Impulse (msec*mbar)

6.00E+08

4.00E+08

2.00E+08 5% of I+ 0.00E+00 600 -2.00E+08 Time t (msec) 700 800 tstart tend 900 1000 1100 1200 1300

Figure 2.4

Method for Calculation of Duration and Positive Impulse for an Irregular Pressure Trace

The graph shows the accumulated impulse to time t for a realistic pressure trace. The spikes in the original curve are smoothed by the process of integration and do not appear in the graph. The start time of the pressure pulse tstart is assumed to be the time at which the positive impulse reaches about 5% of the peak value. This is because pressure traces often have an extended lead phase where the pressure is negligible whilst the explosion is developing. The time at which the pressure trace effectively crosses the axis is given by the time of maximum impulse tend on the graph. These times may be extracted automatically and the difference is taken to be the load duration td. Once an effective duration has been established, Response Spectra may be plotted in terms of td/T, as is the case for the Biggs approach. The effective peak pressure Pe is then calculated directly from the positive impulse I + using:I+ = td x Pe or Pe = 2I+/td

There is an implied triangular idealisation in the relations above. The effective pressure, Pe has however been found to be very useful as a load severity scaling parameter. It is also easily related back to the prototype Biggs formulation discussed in Chapter 3.

12

This effective pressure Pe has been found to be the parameter which enables generic response spectra to be derived. 2.3 PHASE 2 - TESTS

Table 2.1 summarises the experiments conducted during Phase 2 of the Blast and Fire Engineering for Topside Structures JIP over the period from January 1994 to January 1998. The data in this table was taken from Appendix C of Reference 6. The test rig boundary conditions used during these tests are shown on page III-6 of this Reference. The ignition point locations are given in metres from the co-ordinate origin located at the south west corner of the test rig as defined in Figure C1 of the Reference. The column headed MER gives the mean equivalence ratio of the gas concentration in each test. The Water deluge column gives the type of nozzle used in those tests that included water deluge. The column headed Cut? Describes whether the polythene sheets applied to the exterior faces of the test rig to retain the gas mixture before ignition were cut before ignition took place. The last three columns of information in Table 2.1 summarise the instrumentation used during each test.
Test No. Modellers test case(1) Equipment ConfiDensity nement A A A A A A A A A A A A E E A A F F G G G G E D1 D2 C2 C2 C Open Open Closed Closed Ignition Position x y Z 12.76 0.52 12.76 0.52 0.52 0.52 0.52 0.52 0.52 12.76 12.76 12.76 12.76 0.52 12.76 12.76 12.76 0.52 12.76 0.52 12.76 0.52 12.76 12.76 14 14 11.5 4 4.29 4 4.29 4.29 4.29 4.29 4.29 4.29 4 4 4 4 4.29 4 4 4 4.29 4 4.29 4 4.29 4 4 9 8 10 4.23 4.26 4.23 4.26 4.26 4.26 4.26 4.26 4.26 4.23 4.23 4.23 4.23 4.26 4.23 4.23 4.23 4.26 4.23 4.26 4.23 4.26 4.23 4.23 0.13 4.23 4.23 F Open 4<x<16 Open Closed Open Internal External MER Water Cut? Pressure pressure deluge transducers transduce rs 1.07 no 25 10 1.1 yes 5 11 1.07 no 5 11 1.14 yes 6 3 1.14 yes 7 4 1.11 no 27 4 1.07 yes 30 11 1.07 LDN yes 29 11 1.07 MV57 yes 30 11 1.06 LDN no 28 10 1.07 MV57 no 29 10 1.08 no 30 10 1.05 no 27 11 1.05 yes 28 11 0.73 no 22 10 1.22 no 23 10 1.07 no 20 10 1.05 no 21 10 1.08 MV57 no 21 11 1.06 MV57 yes 19 11 1.03 yes 20 11 1.04 yes 19 7 1.04 no 17 0 1.06 no 28 11 1.06 no 39 11 1.06 no 39 11 1.06 no 39 11 G Open 4<x<20 Open Closed Open Ionisation probes 40 30 40 40 39 0 40 37 37 40 35 39 37 38 40 38 46 46 46 48 46 46 20 50 90 90 90

1 1 Low 2 Low 3 Low 4 Low 5 Low 6 2 Low 7 4 High 8 High 9 High 10 High 11 High 12 3 High 13 High 14 High 15 High 16 High 17 High 18 High 19 High 20 High 21 High 22 High 23 High 24 10 High + 25 11 High + 26 12 High + 27 13 High + Confinement Configuration A North Closed East South West Open Closed Open

E Open 12<x<16 Open Closed Open

Table 2.1

Phase 2 Tests

13

2.4 PHASE 2 - COMPARISON OF TEST AND SIMULATED PRESSURE TRACES BP have supplied the FLACS simulated pressure traces corresponding to the experimental situations for the Phase 2 and Phase 3A Spadeadam tests. This has enabled a comparison to be made between experimental and simulated traces taking into account the target structures dynamic characteristics and plastic reserves of strength. This then gives a more robust measure of the conservatism or otherwise of the simulation traces than simple comparison of peak pressures. EExisting CFD explosion simulation codes do not generate the spikes present in the experimental pressure traces, because they do not represent any of the processes thought to cause them.
Pressure - Phase 2 Test 7 Trace 20 (averaged 1ms)
5000

4000

3000 pressure mbar

2000

1000

0 750 -1000 t 800 850 900 950 1000 1050

, however t

Figure 2.5 Typical Pressure Trace (Phase 2 Test 7 Trace 20)

14

Pressure - Phase 2 Simulation (test 7) Trace 20

9.00E+03 8.00E+03 7.00E+03 6.00E+03

pressure mbar

5.00E+03 4.00E+03 3.00E+03 2.00E+03 1.00E+03 0.00E+00 1100.00 -1.00E+03

1150.00

1200.00

1250.00

1300.00

1350.00

1400.00

Figure 2.6 Simulated Pressure Trace (Phase 2 Test 7 Trace 20) Further work [7,8,9], has been performed to examine the experimental (Spadeadam) and CFD derived pressure traces, partly to determine common characteristics which could be used to derive some envelope curves to cover a variety of situations. One conclusion of this work is that the impulse (integrated pressure over time) is a characteristic which is relatively invariant for any particular scenario. Work has also been undertaken by BP to investigate the spectral characteristics of pressure traces, both experimental and simulated. The two frequency (Fourier) traces, experimental and simulated, were found to be of a similar shape, with the experimental trace having a larger contribution at higher frequencies (see Figure 2.7). This shape is sustained for a large range of scenarios and for traces obtained at different times within any particular scenario. This similarity of shape has led the authors to consider the use of the response spectrum approach.

15

Figure 2.7 Fourier transform of typical test pressure trace 2.5 PHASE 3 SIMULATIONS AND TESTS

The experiments conducted during Phase 3A of the Spadeadam explosion test series are summarised in Table 2.2. The information in this Table was taken from OTO reports published by HSE. A list of these reports is given in Appendix C. The notation used in Table 2.2 is similar to Table 2.1. The Peak iop and Peak eop columns give the peak internal and external overpressure respectively recorded in each experiment. These peaks are taken from the un-smoothed experimental data. Additional information is included for the water deluge cases in Table 2.3 to describe the number of nozzles used in the mezzanine and cellar deck areas, and the locations of the water curtains used in nine of the experiments.

16

Test

Report

1 prelim results** 2 prelim results 3 prelim results 4 prelim results 5 prelim results 6 prelim results 7 prelim results 8 prelim results 9 prelim results 10 prelim results 11 prelim results 12 prelim results 13 prelim results 14 prelim results 15 prelim results 16 prelim results 17 prelim results 18 prelim results 19 prelim results 20 OTO-98-116 21 prelim results 22 prelim results 23 prelim results 24 OTO-98-134 25 OTO-98-135 26 OTO-98-136 27 OTO-98-123 28 OTO-98-124 29 OTO-98-137 30 OTO-98-125 31 OTO-98-126 32 OTO-98-138 33* OTO-98-127 34 OTO-98-128 35 OTO-98-129 36 OTO-98-130 37 OTO-98-131 38 OTO-98-132 39 OTO-98-139 40 OTO-98-140 41 OTO-98-141 42 OTO-98-142 43 OTO-98-143 44 OTO-98-144 45 OTO-98-133 Confinement Configurations North East South West Roof Floor C1 Open Open Open Open Confined Confined

Test Series A A A A B B B C C C C D C D D E E E E F F F F R1 R2 R3 G H R4 H H R5 I I J J J J R1 R2 R3 R4 R5 R6 K C2 Open Open Confined Open Confined Confined

Confinement C1 C1 C1 C1 C1 C1 C1 C2 C2 C2 C2 C2 C2 C2 C2 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3 C3

Obstacle Layout O1 O1 O1 O1 O1 O1 O1 O1 O1 O1 O1 O1 O1 O1 O1 O1 O1 O1 O1 O1 O1 O1 O1 O2 O2 O2 O2 O2 O2 O2 O2 O2 O2 O2 O2 (+S) O2 (+S) O2 (+S) O2 (+S) O3 O3 O3 O3 O3 O3 O3

Ignition x y 13.5 5 13.5 5 9 11.6 0.4 6 13.5 5 0.4 6 13.5 5 13.5 5 9 11.6 13.5 5 13.5 5 13.5 5 13.5 5 13.5 5 0.4 6 13.5 5 9 11.6 28 6 0.4 6 13.5 5 0.4 6 0.4 6 0.4 6 13.5 5 13.5 5 13.5 5 13.5 5 13.5 5 13.5 5 13.5 5 0.4 6 13.5 5 0.4 6 0.4 6 13.5 5 0.4 6 13.5 5 13.5 5 0.4 6 0.4 6 0.4 6 0.4 6 0.4 6 0.4 6 13.5 5

z 0.4 4.25 0.4 2 4.25 2 4.25 4.25 0.4 4.25 4.25 4.25 4.25 4.25 2 4.25 0.4 8 2 4.25 2 2 2 4.25 4.25 4.25 4.25 4.25 4.25 4.25 2 4.25 2 2 4.25 2 4.25 4.25 2 2 2 2 2 2 4.25

m.e.r. Peak iop 1.12 2.953 1.09 1.604 1.07 3.388 1.08 17.096 1.09 0.472 1.09 0.288 1.06 0.357 1.08 3.243 1.07 3.87 1.06 0.599 1.08 0.6912 1.07 2.535 1.07 0.8839 1.06 2.044 1.08 2 1.10 0.943 1.10 3.204 1.09 2.868 1.08 5.7 1.09 0.39 1.09 0.202 1.09 4.704 1.09 0.452 1.09 5.304 1.12 12.717 1.08 13.05 1.09 0.411 1.08 1.921 1.09 8.018 1.06 0.301 1.08 0.221 0.98 9.361 0.98 0.451 1.11 13.138 1.03 0.339 1.09 0.253 1.10 18.867 1.04 25.358 1.10 21.313 1.10 19.975 1.10 22.759 1.10 9.972 1.10 8.183 1.10 27.611 1.00 0.488

Peak eop 1.32 2.172 2.544 3.891 0.32 0.147 0.229 1.616 2.8 0.4075 0.6389 1.047 0.5676 1.644 4.038 0.863 1.5562 0.697 3.764 0.331 0.144 2.417 0.415 1.503 1.851 1.797 0.322 2.102 1.852 0.244 0.39 1.723 0.332 2.184 0.238 0.151 2.708 3.067 16.352 5.895 1.804 3.28 2.82 1.972 0.234

C3 Open Open Open Open 1/3 Open Confined

(+S) = with scaffolding *Test 33 was a Propane explosion ** These tests were reported by 'Preliminary Results Reports' published for each test by the HSE

Table 2.2

Phase 3A Tests - Major Characteristics

17

Test Number 1 2 3 4 5 6 7 8* 9* 10* 11 12* 13* 14 15*

Water Sprays None None None None mv25 mv25 mv25 None None mv25 mv25 mv57 mv25 fan nozzles fan nozzles

Nozzles Mezz. Cellar

Position

Cut??

full area full area full area

75 75 75

17 17 17

full area full area curtain 1 curtain 2 full area curtain 1 curtain 2 curtain 1 curtain 2 curtain 3

75 75 8 7 75 8 7 8 8 7

17 17 3 3 17 3 3 3 3 3

x=7m x=21m x=7m x=21m x=7m x=14m x=21m

no yes no no yes no no no no no yes

16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33* 34 35 36 37 38 39 40 41 42 43 44 45

None None None None mv25 mv25 None fan nozzles None None None perimeter Vessel specific None

full area full area curtain 1 curtain 2

75 75 8 7

17 17 3 3 x=7m x=21m

no yes yes yes no yes yes yes

no no no 44 27 no 20 12 no no full area 84+20 54+12 no full area 84+20 54+12 yes None no full area 84+20 54+12 yes Vessel specific 20 12 yes full area 84+20 54+12 no full area 84+20 54+12 yes None no None no None West end only None West end only none West end only none West end only none West end only none West end only full area 84+20 54+12 no Table 2.3 Phase 3A Tests - Water Sprays and Confinement at Ignition 18

There was not time to utilise the full set of test results from the Phase 3A series. A number of characteristic tests were examined and response spectra were obtained for these tests and the corresponding simulations. The selected Tests are listed in Table 2.4 below. Test Number 1 2 3 4 5 8 11 17 41 43 44 Reason for selection Typical test no deluge Typical test no deluge Typical test no deluge High peak pressure no deluge Low peak pressure - with deluge Typical test no deluge Low peak pressure - with deluge Typical test no deluge Repeatability test 22.759 bar peak overpressure Repeatability test 8.183 bar peak overpressure Repeatability test 27.6 bar peak overpressure Table 2.4 Phase 3A Tests selected for investigation Simulation 1 2 3 4 5 8 11 41 -

The Repeatability tests were performed in order to test whether the peak pressure could be reproduced under the same nominal scenario conditions. These tests were chosen for examination, because of the unexplained differences between the peak overpressures. The corresponding response spectra have been examined to see if they represent the basic same basic characteristics for the repeatability tests. The Response Spectra being consistent between the repeatability tests indicates that they are a robust representation of the characteristics of the explosion event.

19

3.

BIGGS RESPONSE CURVES THE PROTOTYPE PROBLEM

3.1

DERIVATION OF THE BIGGS CURVES FOR PEAK RESPONSE

Components may be analysed in isolation using the Biggs method, as long as the interaction with the surrounding structure through fixity and the applied loads is negligible or is represented in the component model. The Biggs method, illustrated in Figure 3.1, requires two basic inputs, the resistancedisplacement curve and a loading time history. Design charts are available for the calculation of the peak response, Xmax or , given the load duration to natural period ratio td/T and the ratio of peak overpressure load Ppeak to component resistance R. The method is based on a triangular idealisation of the pressure time history. The response spectra generated as part of this work have avoided the limitations of this assumption by using experimental, or simulated time histories generated using the FLACS CFD code. The Biggs charts are based on a simplified, bi-linear resistance behaviour, which may be inaccurate for fully or partially fixed members as well as for members where tension effects are significant. Under increasing uniform loading, the member deflects elastically up to its yield displacement Xe (part 1 of the curve in figure 3.1). Further loading results in no increase in the resistance of the member it is assumed that the member is deforming purely plastically (part 2 of the curve). If the displacement reduces after plastic deformation then it is assumed that the resistance of the member returns to the pre-plastic (or elastic) form on a line parallel to the line representing the initial elastic deflection (part 3 of the curve). Limitations of the Biggs method Current practice also often adopts an idealization of the pressure time history that is inappropriate involving the peak pressure Ppeak. The prediction of rupture using this method is unreliable as there is no detailed representation of the strain distribution within the structure. Although some of the limitations seem severe, the technique has been successfully used in design situations, particularly where the resistance function has been accurately determined. If the resistance function is determined via a static non linear finite element analysis (NLFEA) the deflection time histories can compare well with an explicit dynamic NLFEA. This model may be unable to give accurate values at high deflections due to the spread of plasticity which results in a change of the shape function used in formulating the spring mass idealization. A full description of the method is given in references 3 and 11.

20

Figure 3.1

Biggs' single degree of freedom response method

21

3.2

USE OF THE BIGGS RESPONSE CURVES

The Biggs model idealisation was executed for a range of triangular load durations, natural periods, peak pressures (Ppeak) and structural resistances, R. The resulting peak response curves are shown in figure 3.2. For a truly triangular pressure time history, Ppeak is equal to the effective pressure Pe.

Biggs response
td/T

0.1

10 100
Mu=

10

R/Pe=0.2 R/Pe=0.3 R/Pe=0.4 R/Pe=0.5 R/Pe=0.6 R/Pe=0.7

Mu

R/Pe=0.8 R/Pe=0.9

R/Pe=1 R/Pe=1.2 R/Pe=1.6 R/Pe=2

0.1

Figure 3.2

Biggs' design chart - Overpressure Rise Time Equals Half Load Duration

This chart enables the peak response of the member to be calculated. The vertical axis is in terms of the ductility MU or , which is the number of multiples of the yield displacement the member will reach under the loading. To use this chart: Determine the peak resistance R, effective stiffness and yield displacement Xe for the member. 22

Either: Determine the peak pressure Ppeak using the conventional triangular load idealisation method shown in figure 2.3 and described in Section 2.2. Or:Use the refinement introduced in the general Response Spectrum method using the more rigorous approach described in Section 2.2. This involves the use of the actual pressure time history, the impulse, the generalised load duration td and the effective or design pressure Pe. If a pressure trace is idealised following the rigorous approach, the Biggs curves may still be used and will give more realistic results (but see Section 6.3). We shall see that the Biggs curves themselves may be converted to Response Spectra. Calculate the equivalent mass of the member Me Determine the natural period of the member T from the effective mass and stiffness.

T = 2

Me. .(3.1) Ke

Choose the nearest curve corresponding to the ratio of R/Pe (or R/Ppeak.) Read off the ductility for the peak response corresponding to the td/T for the member from the vertical axis. (Include an example of use) The reaction loads at the ends of the member may also be estimated using this method although the accuracy may be doubtful as detailed strain information is not represented in the model. These formulae are not accurate for two way spanning panels but are reasonable for the assessment of structures which may be represented as one way spanning panels or beams. Similar charts have been constructed for idealised asymmetric triangular loading where the rise time (the time to maximum pressure) is not equal to half the load duration. Similar charts may be constructed for partially fixed members by applying a rotational spring stiffness, Ks to the ends of the member [11]. Ks tends to infinity for a fully fixed member and to zero for a simply supported member. The degree of end fixity for a member is determined from the flexural stiffness of the connecting structure.

23

4.
4.1

RESPONSE SURFACES

EXAMPLES RESPONSE SURFACE (BIGGS)

Response surfaces have been calculated for a large number of experimental and simulated pressure traces. A typical response surface, for a single idealised triangular load time history is shown in Figure 4.1. The horizontal axes are the target structures natural period T and the resistance R. The surface is constructed by calculating the peak response (ductility) for all values of T and R within chosen ranges of these variables. The ranges are chosen to give ratios of td/T from 0.1 to 10 and ranges of R/Pe (or R/Ppeak) from 0.2 to 2. The vertical axis is hence the peak deflection expressed in terms of ductility for the particular values of T and R for the point considered.

Response surface for a Triangular pulse load.

1000 100 10 Response 1 0.0098 0.0151 0.0328 0.0787 0.1405 Natural period T 0.2810 0.6556 565 444 505 323 384 Resistance R 262 202 141 81 20 0

Figure 4.1

Response Surface for a Triangular Pressure Time History (Biggs Idealisation)

24

The contours on the surface correspond to ductilities of 1, 10 and 100. This response surface is an alternative representation of the Biggs curves. Vertical sections parallel to the natural period axis are the conventional Biggs curves for fixed R/Pe values as shown in figure 3.2 of Chapter 3. The contours on the surface represent the values of natural period and resistance which give rise to the required ductility levels. 4.2 RESPONSE SPECTRA FROM RESPONSE SURFACES

Unlike the Biggs response curves, the response spectra are obtained from the Biggs response surface by taking horizontal sections at allowable ductility values as shown in Figure 4.2. Here Mu is the allowable ductility, R/Pe is the required static resistance to effective pressure ratio, td/T is the load duration natural period ratio.

Response Spectrum from Biggs curves (Triangular pressure time history)

td/T

MU=

0.1

10 1.8
MU=

1.6 1.4
R/Pe or R/Ppeak
MU=1 MU=2

1.2 1 0.8 0.6 0.4 0.2 0

MU=3 MU=4 MU=5 MU=6 MU=7 MU=8 MU=9 MU=10

Figure 4.2

Response Spectrum from Biggs Curves

The required static resistance R for a structure with a given natural period and predetermined allowable ductility may be read off from the vertical axis. The value of R/Pe may be read from the vertical axis using the appropriate allowed ductility curve and load duration/natural period ratio. If Pe is known then R may be calculated. A subsequent static elastic analysis may then be performed with the static pressure of magnitude R being applied. These curves are an extension of the dynamic load factor (DLF) curves specific to a triangular pressure trace of height Pe and duration td. 25

The dynamic load factor DLF is defined by:DLF = R/Pe..(4.1) This represents the dynamic amplification or the resistance required to resist the triangular load with height Pe. These curves also include the alleviation of the required load due to the allowed plastic deformation corresponding to the chosen curve for ductilities 2 to 10. The vertical axis may hence be labelled DLF and the equivalent static load Pstat will be defined in the usual way by:Pstat = R = Pe x DLF....(4.2)

It can be seen that the curves can be bounded for a large range of natural periods and so a required resistance may often be obtained without precise knowledge of the structures natural period or load duration. It can also be seen that the required resistance decreases as the allowable ductility increases. Note the vertical axis is now not logarithmic indicating the insensitivity of the required resistance to variations in allowed ductility and natural period. Example If the structure has natural period of 50 msec and the load has a duration of 100 msec then td/T is 2. For a structure with an allowable ductility of 2 the curve corresponding to = 2 should be used. The DLF is then 0.87 For a triangular load time history with peak pressure equal to 1000 mbar the required static resistance R is then Pe x DLF = 870 mbar. This structural element may then be tested against an applied static load Pstat of 870 mbar to see if it fails the code checks at this load level. Note the oscillations on these curves are not meaningful, they are just a result of the assumption of the triangular shape of the loading (Section 6.3). In this work Response Spectra have been generated from actual simulated and experimental pressure traces with a realiostic spread of component frequencies. 4.3 RESPONSE VERSUS RESISTANCE

For completeness figure 4.3 shows the curves obtained by taking vertical sections of the response surface parallel to the resistance R axis. This figure shows the variation of response with static resistance R for a given td/T ratio. Note the logarithmic scale on the response axis. Variations in resistance have a very marked effect on peak response particularly at large td/T ratios where pseudo-static response occurs.

26

Response Vs. Resistance Resistance/Pe

0.5

1.5

2.5 100

MU=

td/T=0.1 td/T=0.15 td/T=0.2 td/T=0.25

10

td/T=0.3 td/T=0.4 td/T=0.5 td/T=0.6

Ductility MU

td/T=0.7 td/T=0.8 td/T=0.9 td/T=1 td/T=1.5 td/T=2 td/T=2.5

td/T=3 td/T=4 td/T=5 td/T=6 td/T=7 td/T=8 td/T=9 td/T=10

0.1

Figure 4.3

Variation of Peak Response with Resistance R for Fixed td/T Ratios

These curves are not very useful as the ductility varies rapidly with variations of resistance/Pe ratios between 0 and 1. The sensitive td/T range is td/T between 1 and 10. The structure should ideally be outside these ranges. 4.4 BIGGS LOAD AND RESPONSE SCALING

The characteristics used for the definition of the loading intensity in the Biggs approach are td, the load duration and Pe the effective pressure of the triangular pressure time history. The dimensionless parameter td/T is used in many areas of dynamic response relating to dynamic amplification and earthquake response, for example. This enables the range of response to be extended for traces of varying duration td for ranges of natural period T. The equations of motion which form the basis of the response calculations may be nondimensionalised by using this form of scaling. In the Biggs formulation the pressure trace has a triangular time history as shown in figure 3.1. For the triangular load time history the peak pressure, Ppeak and the effective pressure 27

Pe (as defined in section 2.2) are identical. This is not the case for a general experimental pressure trace. In the majority of cases the pressure trace has a Pe less than Ppeak as the area under the trace (the impulse) is less than that of a triangular trace with the same peak and duration. This has given rise to a measure of the departure of the trace from the triangular shape called the shape factor Sf which is defined by:Sf = Ppeak/Pe ..(4.3) This factor is generally greater than unity. Appendix A lists the calculated shape factors for the pressure traces derived in Phase 2 tests and their corresponding simulations. In the Biggs formulation, the structural resistance R is scaled by dividing it by Pe. Responses for double the resistance and double the effective peak pressure are identical so long as the td/T ratio is also the same. This non-dimensional scaling of the resistance has also been found to be suitable for experimental and simulated pressure traces. The ductility is a non-dimensionalised form of the response being defined by: = Xmax/Xe..(4.4) Where Xmax is the peak deflection Xe is the effective yield deflection It is also possible to scale the response spectra in such a way that the response curves for ductilities of two or more may be represented as a single curve. The elastic case corresponding to an allowable ductility of one has some special features associated with the lack of structural damping which occurs during plastic deformation. Because of the difference between the elastic curve (=1) and the curves for higher allowable ductility, this scaling has not been implemented.

28

5.
5.1

THE USE OF RESPONSE SPECTRA

OVERVIEW OF THE METHOD

The response spectrum approach takes into account the variations in response of structural elements resulting from their differing natural periods and enables the reserves of strength released when elements are allowed to deform plastically to be taken into account. The response spectrum approach has been in use for decades in the earthquake response context and was in fact pioneered in the Second World War to calculate ground motion effects and structural response from explosions. The purpose of the method is to enable equivalent static blanket loads to be derived for use in a conventional design check for blast loads of given severity and duration. Figure 5.1 describes the use of blast response spectra in determining an equivalent static design pressure and shows how the use of Response Spectra would fit in with a design or assessment process. Figure 5.2 illustrates this same process in a more graphic way. The severity of the blast loading may be determined from local conditions by the use of nominal overpressures, previous experience, risk classification, simulations or experiment. The accuracy of the response calculation will naturally depend on the accuracy of the loads used as input. The structural element to be assessed may be a panel, a deck, module or a whole topsides idealised as a one degree of system oscillator. This process will be familiar to designers who use the Biggs response method and is in routine use. The structural element is represented by its natural period and resistance at effective yield R. A further important parameter is the allowable ductility of the element which is a measure of the amount of plasticity which the element can sustain before rupture and depends on connection details and the function of the element.

29

The Use of Response Spectra

Performance Standards Structural Function Displacement limits Xmax Allowable Ductility

Load Characteristics Positive Impulse I+ Duration td Effective Pressure Pe

Structural Idealisation Effective Stiffness Ke Effective Mass Me Natural Period T=2 Me/Ke

td Pe

Response Spectra Choose allowable ductility curve Extract dynamic load factor - DLF from curve Required static resistance Rstat = DLF x Pe = Equivalent static load Pstat = DLF x Pe Pstat Static Elastic Structural Analysis Effective yield displacement Xe Structural Resistance Re Response under co-existant loads/restraints Refined effective stiffness Ke

No re-calculate Pstat

Input parameters confirmed?

Yes

Satisfies code checks? Re>Rstat

yes

NO

Re-size structure or mitgate loads

Satisfies Performance Standards?

NO

Re-size structure or mitgate loads

Yes

Next Structure

Figure 5.1 5.2

The Use of Response Spectra

STRUCTURE IDEALISATION

The structure is idealised in the same way as is familiar to any user of the Biggs response method [3] and is described in Section 3.2.

30

5.3

ALLOWABLE DUCTILITIES

The allowable ductility of a structural element is a measure of the amount of deformation the element can sustain before rupture or when its performance standards cease to be satisfied. This is usually expressed as a multiple '' of the effective yield displacement and referred to as the ductility. In explosion response, an allowable ductility may be associated with a particular structural type or configuration. A high allowable ductility will have the effect of reducing the effect of dynamic amplification through energy absorption and will result in reduced equivalent static load and a reduced required static resistance. The allowable ductility is assessed for the structure by consideration of the connection detailing and framing. The allowable ductility may also be derived from a known reserve strength ratio in combination with the performance standards appropriate for the situation. For example, small allowable displacements ( ~1) may be appropriate for the Temporary Refuge and supports of critical systems, and large allowable deformation ( ~5 to 10) for equipment and piping supports, cladding and blast walls. 5.4 LOAD/PRESSURE TIME HISTORY IDEALISATION

The recommended method for obtaining the load duration td and effective peak pressure Pe from a pressure trace is described in Section 2.3. The pressure trace may alternatively be used as input to a single degree of freedom non-linear oscillator model to obtain response surfaces and Response Spectra directly as described in Sections 4.1 and 4.2. Alternatively, the Generic Response Spectra published in this report may be used directly. 5.5 REQUIRED RESISTANCE - EQUIVALENT STATIC LOADS

The response spectra generated in the work described in this Report have used the actual pressure time histories. The curves are constructed by considering the allowable ductility (from 1 to 10) and then plotting the resistance R required to limit the peak displacement to this value of ductility. The vertical axis is scaled by the effective peak pressure of the pressure trace Pe. R/Pe is plotted on the vertical axis labelled DLF or dynamic load factor. The element natural period and load duration determines where we are on the horizontal axis of the Response Spectrum (Figure 5.2) and the dynamic load factor (DLF) may be read off the relevant curve representing the allowable ductility.

31

Phase 2 tests Generic Response Spectrum

1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 0.1 1

td/T

MU=

MU=1 MU=2 MU=3 MU=4

DLF

MU=5 MU=6 MU=7 MU=8 MU=9 MU=10

10

Figure 5.2

Mean Response Spectrum for all Phase 2 Tests

The design pressure or required static resistance R may be calculated from the DLF using:R = DLF x Pe(5.1) Structures with natural periods less than one third of td, are able to respond to the load time history point by point and are referred to as responding in a pseudo-static way. This corresponds to the right hand side of the figure. Structures with low natural frequency or with natural periods more than about 3 times the load duration are said to respond impulsively to the load. This response is virtually independent of the shape of the load time history and is referred to as impulsive response. These structures are represented by points on the left side of the figure. Generally impulsive response is much smaller than that indicated by the peak load as the structure does not have time to respond. This is indicated by the much reduced DLF at this end of the response spectrum.

32

5.6

EXPLOSION LOAD SOURCES

The severity of the blast loading is determined from local conditions by the use of nominal overpressures, previous experience, risk classification, simulations or experiment. Specialized CFD gas explosion codes (e.g. FLACS) are often used to derive explosion loads, the other two methods in common use are empirical and phenomenological models. The CFD codes give pressure-time histories for the loading on different parts of a structure on a platform. These codes require accurate geometric representation of equipment, pipework, structure and deck arrangements and can predict overpressure loadings for a wide range of scenarios. In any case Pe and td are required. The accuracy of the estimate of R will depend on the quality of the load input information.

Generic response spectrum

Phase 2 Test 7 Mean Response Spectrum
MU=
MU=1

1.80 1.60 1.40

MU=2 MU=3 MU=4 MU=5 MU=6 MU=7

Local conditions Risk level Ignition point Mass release Confinement Congestion

1.20

Impulse Duration td Effective pressure Pe

R/Pe

1.00 0.80 0.60 0.40 0.20 0.00 0.1 1

Required static resistance R Design pressure R

MU=8 MU=9 MU=10

td/T

10

Idealize structure Mass Me Stiffness Ke Yield deflection Xe

Natural period T= 2 sqrt (Me/Ke) Allowable ductility

Figure 5.3

The use of Response Spectra in Explosion Resistant Design

33

6.
6.1

RESULTS

DISCUSSION OF THE FEATURES OF RESPONSE SPECTRA

Typical Generic response spectra are shown in figures 6.1 and 6.2 to aid discussion.
Phase 2 tests Generic Response Spectrum

1.6 1.4 1.2 1

DLF

MU=
MU=1 MU=2 MU=3 MU=4 MU=5 MU=6 MU=7 MU=8 MU=9

0.8 0.6 0.4 0.2 0 0.1 1

td/T

MU=10

10

Figure 6.1

Mean Response Spectra for all Phase 2 Tests

Phase 2 Simulations Generic Response Spectrum

1.8 1.6 1.4 1.2

MU=
MU=1 MU=2 MU=3 MU=4

D LF

1 0.8 0.6 0.4 0.2 0 0.1 1

td/T

MU=5 MU=6 MU=7 MU=8 MU=9 MU=10

10

Figure 6.2

Mean Response Spectra for all Phase 2 FLACS Simulations

34

These are the Response Spectra obtained by averaging over all traces and all tests for the Phase 2 experiments at the Spadeadam site and the corresponding simulations. Each curve corresponds to a different allowable ductility value from 1 to 10. The horizontal axis represents the variation of the load duration td divided by the target structures natural period T. The required static resistance in any case can be read off from the vertical axis as a DLF which is to be multiplied by Pe to get the required resistance R. In this case, the effective peak pressure is defined as the mean Pe for all the traces obtained throughout the compartment for all the tests. This average Pe has little physical meaning for a whole compartment but will be a useful technique when examining the response of a blast wall where a number of traces are available at different parts of the wall. The left hand side of each response spectrum below td/T = 0.3 corresponds to the impulsive regime where the shape of the pressure trace does not contribute to the peak response, this is related only to the impulse (or a combination of Pe and duration td). The middle section, between td/T = 0.3 and td/T = 3 is the fully dynamic part of the response spectrum. The right hand side where td/T > 3 corresponds to the pseudo static part of the response spectrum where the response is determined by the work done by the applied load and the potential energy absorbed at the moment of maximum deflection. The curves for simulations and experimental results will still differ in the middle region where fully dynamic response occurs. This region has also been represented analytically for = 1 and for a trace with a dominant frequency component around td. The analytical results are in good agreement with the simulations (Appendix E). 6.2 COMPARISON BETWEEN SIMULATED AND EXPERIMENTAL RESPONSE SPECTRA The main difference between the experimental and simulated response spectra is due to the inclusion in the experimental loading of high frequency loading components which are the spikes in the pressure traces discussed in section 2.1. These high frequency components tend to stimulate response at small natural periods corresponding to large values of td/T. This suggests that a valid correction to a simulated response spectrum would be to elevate the values at the right hand side of the spectrum to those near the td/T=1 values. A simulated pressure trace, being smooth, will have a predominant frequency content around the period corresponding to the range td to 2 x td. This gives rise to a peak in the response spectrum in the middle region, particularly for the elastic case ( =1). The higher allowable ductility curves have this peak suppressed because of the damping introduced by plastic deformation.

35

6.3

FOURIER TRANSFORMS AND RESPONSE SPECTRA

The ability to replace 1/T by the more general td/T is a result of the fact that the frequency components of the loading (Fourier components) amplify, by resonance, the response of those systems whose natural periods match the frequency component considered. Consider the triangular pulse f(t) assumed for the Biggs response calculations with the following characteristics:-

f (t ) = 2 Pe t t d f (t ) = 2 Pe

; ;

0<t<

(1 t )
td

td (6.1) 2

td < t < t d ... (6.2) 2

The response spectrum for this loading is shown in figure 7.3.

MU=

Figure 6.3

Response Spectrum for Triangular Load Time History (Biggs Idealisation)

The spectra show elastic dynamic peaks at resonance td/T =1 and at td/T = 3, 5, 7..etc.

36

The triangular form of the load may be treated as a periodic function by being repeated to the left and right. This can be decomposed into its Fourier (frequency components) as below:-

f (t ) =

pe 16 p e 1 2 1 6 2 { 2 cos + 2 cos + .......} (6.3) (t d t ) 6 (t d t ) 2 2

The triangular pulse has frequency components at td/t = 1, 3, 5.etc. These discrete frequency components give rise to minor peaks in the response spectrum at these points. Illustrating the principle that the frequency components of the load will give rise to peaks of response at the corresponding points on the response spectrum. These peaks are a direct result of the triangular idealisation used in the construction of the Biggs curves and have no real physical significance. 6.4 RESPONSE SCALING

It has been found that the response spectra for differing ductilities may be scaled using energy and impulse arguments to lie close to each other. As described in Appendix E, the extreme ends of the spectra represent regimes where approximations may be made to derive the response. The elastic curve at = 1 shows some departure from the others at natural periods around td/T=1 (resonance) as it follows the loading more closely. This is because the response occurs without plastic deformation which itself is an energy absorbing mechanism. It is possible to scale the response spectra in such a way that the response curves for ductilities of two or more may be represented as a single curve. Because of the difference between the elastic curve (=1) and the curves for higher allowable ductility, this scaling has not been performed. A brief discussion of response scaling is given in reference 1. 6.5 GENERIC RESPONSE SPECTRA

Mean response spectra for each test and simulation set have been derived. Standard deviations and coefficients of variation have been derived for each test and for the test series 2 and 3A. See Appendix C. The results indicate that the Biggs response surface is a good benchmark against which the response surfaces for realistic load time histories can be gauged. The essential step is to derive a representative maximum pressure Pe from the pressure time history by calculating the positive phase impulses. This impulse is then related back to a representative Pe through the positive phase load duration td. This Pe will rarely be as large as the peak pressure for the time history. Mean Response Spectra for each test are given in Appendix C with curves for the test based Standard Deviations and Coefficients of variation (COV). There is some variability between Response Spectra for the traces in a given test reflecting the variability of loading during the explosion development and throughout the test compartment. 37

Typical COVs for the elastic curve (=1) in the important region td/T between 0.5 and 5 are about 20% and standard deviations for the Response Spectra could be as much as 0.4. COVs and standard deviations for the higher allowable ductility curves are consistently smaller. It is proposed to examine the Response Spectra in more detail in order to identify the anomalous curves which are contributing to these variations and identify their special features. It is likely that the variation of Response Spectra for traces of comparable effective peak pressure (in the range 100mbar to 5 bar) will be smaller than the variability for the whole population of traces. 6.6 REPEATABILITY TEST RESULTS

The Repeatability tests were performed in order to test whether the peak pressure could be reproduced under the same nominal scenario conditions. These tests were chosen for examination, because of the unexplained differences between the peak overpressures. The corresponding response spectra have been examined to see if they represent the basic same basic characteristics for the repeatability tests. The overall peak internal overpressures for the selected repeatability tests were as follows: Test 41 Test 43 Test 44 22.759 bar 8.183 bar 27.6 bar.

The Response Spectra being consistent between the repeatability tests indicates a robust representation of the characteristics of the explosion event. Examination of the mean Response Spectra for these tests indicates that the Response Spectra for test 43 and 44 are very similar. This is very encouraging. Test 41 has some anomalous behaviour which may be due to lack of information as a result of the high local pressures.

38

7.
7.1

FUTURE WORK

THE EARTHQUAKE ANALOGY

The structure to be analysed is idealised as a one degree of freedom system or as a series of one degree of freedom systems corresponding to each mode of response. The natural period and yield resistance for each of these one degree of freedom systems is calculated from the effective mass, stiffness and static yield deflection. The spectral acceleration SA for the earthquake to be used in design is dependent on the location of the structure. API [17] divides the USA into a number of zones within which the ratio G of the effective horizontal ground acceleration to the acceleration due to gravity is prescribed as a value between 0.05 and 0.4. The spectral acceleration SA, at short periods is then equal to G in units of the acceleration due to gravity. The variation of the response spectrum then represents the dynamic amplification variation for structures with varying natural periods. An assumed damping in the region of 5% of critical is implicit in this approach. The local soil (support) conditions will determine the magnitude of the response for structures with longer natural periods. Three spectra are shown for 3 differing soil types A, B and C varying from rock to sands, silts and clays. Soils of type C tend to amplify the response more than the stiffer local support conditions. Once the structure natural period and resistance has been determined the corresponding design acceleration may be read from the response spectrum curve. Equal horizontal accelerations are considered and combined from two perpendicular directions, together with a vertical acceleration of half the value. These responses are combined using a SRSS (square root of the sum of the squares) method. The modal responses are combined using the CQC (complete quadratic combination) method. Two levels of earthquake are usually considered in earthquake resistant design. The strength level earthquake demands elastic non-yielding response of the structure. This earthquake is likely to occur within the lifetime of the structure. The more onerous rare intense earthquake is taken to have an intensity (as defined in terms of the averaged spectral velocity near the natural period of the structure) of less than twice the strength level earthquake. Local plastic deformation of the structure is allowed for the rare intense earthquake. As in explosion response, an allowed ductility may be associated with a particular structural type. A high allowed ductility will have the effect of reducing the effect of dynamic amplification and will result in reduced design response spectra accelerations. Some authors have incorporated this effect into their design response spectra. A number of scaling schemes have been used to generalise earthquake response spectra to represent ductile response beyond yield whilst only requiring elastic response analyses to be performed.

39

7.2

EXTENSION TO MULTI-DEGREE OF FREEDOM SYSTEMS

The Response Spectra method may also be extended to multi-degree of freedom structures by analogy with the related earthquake response spectrum methods. The modal periods of a multi-degree of freedom structure and the corresponding load components (which give rise to the deflected shape corresponding to each mode). Responses may be superposed as in Earthquake analysis. 7.3 RELIABILITY AND PROBABILITY OF FAILURE

If the probability of exceedance associated with a given set of pressure time histories are known, then the reliability (or probability of failure) associated with a given static resistance may be calculated. The Response Spectra curves for a given ductility may be considered to be failure surfaces. Points below the curves represent resistances which give higher ductilities. Conventionally, the (space averaged) peak overpressure is plotted in the form of an exceedance diagram for a representative number of explosion scenarios to determine the SLB and DLB load levels. Characterising the explosion intensity by impulse and energy is likely to give more representative and consistent exceedance curves. 7.4 MEMBRANE AND TENSION EFFECTS

The current simulations have assumed elastic / perfectly plastic behaviour. For large deflections, tension and membrane effects may be important. Membrane effects and geometric hardening effects may be incorporated into the method by the use of a modified resistance displacement formulation. One approach is to follow the NORSOK procedure [18] of representing the plastic phase of response in the resistance/displacement function as a line with non-zero slope. It would be useful to see how this would affect the response spectra. 7.5 CONFIRMATION OF THE RESPONSE SPECTRUM APPROACH VARIABILITY

Some variability between the Response Spectra between traces for a given test has been observed associated with:Incompleteness of information Frequency components in the trace (e.g. Biggs wobbles) These variations should be distinguished from legitimate variability between tests. Some selective averaging may be instructive. (low intensity traces not considered). The variability should be related to the physical test conditions. A parametric representation of the effective pressure Pe related to the test physical conditions would extend the applicability of the method at an early project phase.

40

There is still a large amount of data (experimental and simulated) available for further analysis arising from consideration of partial gas clouds and deluge, this could be processed and examined. 7.6 ROBUST SOFTWARE SIMULATION/TEST COMPARISONS

The relative invariance of response spectra for similar physical situations (repeatability tests) indicates that comparison of these spectra would be a more robust method of comparison in the context of dynamic non-linear response. 7.7 HSE JIP CASE STUDY

The HSE has suggested the usefulness of the application of the method to actual existing designs to assess the conservatism of otherwise of the conventional methods of load simplification/combination. Multiple loadings (on an extended blast wall) may be superposed by considering the averaged response spectrum and the load factor associated with the assumed deflected shape. (c.f. Biggs). This is much simpler than processing the traces directly.

41

8.

CONCLUSIONS

Experimental and simulated explosion pressure traces have been processed and compared and their underlying structure examined. A method has been identified which enables the duration and effective pressure to be calculated for irregular experimental and simulated traces. The shape of the Response Spectra has been found to be almost invariant between traces and between tests with the correct scaling, which is compatible with the implied scaling in the Biggs approach. Mean Response Spectra for complete tests and series of tests have been constructed giving a suggested form for the Generic Response Spectrum. The common spectral form for experimental and simulated results has been identified and the main differences identified. A correction has been proposed to bring the simulation Response Spectra into line with those to be expected from experimental results. The construction of response spectra from response surfaces has been demonstrated. The common spectral form for experimental and simulated results has been identified. The response spectrum approach has been described, which differs from the nominal overpressure approach in that the dynamic characteristics and allowable deformations of the target structure are represented. The design response spectra give the required static resistance and the Design load level to be used for simplified explosion resistant design and assessment. Careful evaluation of the allowable ductility for a particular structural element avoids expensive over-design based on current approaches. This enables the critical details to be developed with the required plastic deformation requirements.

42

9.
1.

REFERENCES

Walker S, Corr B, Tam V and Shearer M, OTC 17242, Response Spectra for Explosion Loading and Response, Houston May 2005. J. M. Nau, W. J. Hall, 1982, An Evaluation of Scaling Methods for Earthquake Response Spectra, Department Civil Engineering of University of Illinois, US. Biggs J.M., Introduction to Structural Dynamics, McGraw Hill, 1964. Bowerman H, Owens G W, Rumley J H, Tolloczko, Interim Guidance Notes for the Design and Protection of Topside Structures Against Explosion and Fire, Steel Construction Institute Document SCI-P-112/487, January 1992. OTO 1999 047, Explosion loading on topsides equipment Part 2 Determination of explosion loading on offshore equipment using FLACS, March 2000. 'Structural Response Model Evaluation Specification', BG Doct. RT616, Version 01, February 1997. Shearer M J, Tam V Y H, Corr B, Analysis of results from large scale hydrocarbon gas explosions, J. Loss Prev. Process Ind. 13, 167-173, 2000. Shearer, M.J., Tam, V.H.Y., Corr, B., 1998, Analysis of Results From Large Scale Hydrocarbon Explosion, 7th Conference on Offshore: Fire and Explosion Engineering, Shearer, M.J., Analysis of Large Scale Gas Explosion Tests and Assessment of Potential Impact on Structures, PhD Thesis, University of Surrey, 2003. UKOOA/HSE, Preparation of Updated Guidance for Fire and Explosion Hazards Part 1 Guidance on Design and Operational Considerations for the Avoidance and Mitigation of Explosions, Commentary, December 2002. (available from www.fireandblast.com) Walker S and Haworth M.W., Sensitivity of response of topside structures to fires and explosions, HSE OTO 97 043, 1996. Selby C A, Burgan B A, Blast and Fire Engineering for Topside Structures Phase 2, Final Summary Report, Steel Construction Institute, ISBN 1 85942 078 8, 1998 Design of Offshore Facilities to Resist Gas Explosion Hazard, Engineering Handbook, Ed. Czujko J., Corrocean 2001. Clough, R.W, Penzien, J. Dynamics of Structures, McGraw-Hill, 1975. API RP 2FB, 1st Edition Design of Offshore Facilities Against Fire and Blast Loading.

2.

3. 4.

5.

6.

7.

8.

9.

10.

11.

12.

13.

14. 15.

43

16.

Steve Walker, Brian Corr, Vincent Tam, Roland Martland, Rashid Shahsavar, Response spectra for explosion resistant design and assessment, OMAE200692512, OMAE Conference, Hamburg, Germany, June 2006. Recommended Practice for the planning, Designing and Construction Fixed Offshore Platforms Working Stress Design, API RP 2A-WSD, 21st Edition. API (American Petroleum Institute). NORSOK n00401 Design of Steel Structures, Annex A, Design against accidental loads. 2000.

17.

18.

44

APPENDIX A CHARACTERISTICS OF THE PHASE 2 AND PHASE 3 A TRACES EXPERIMENTAL AND SIMULATIONS RESULTS
(May be found in Volume 2)

45

APPENDIX B SPADEADAM TESTS RELEVENT OTO REPORTS

Title OTO OTO OTO OTO OTO OTO OTO OTO OTO OTO OTO OTO OTO OTO OTO OTO OTO OTO OTO OTO OTO OTO OTO OTO

Test 98 98 98 98 98 98 98 98 98 98 98 98 98 98 98 98 98 98 98 98 98 98 98 98 48 116 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 20 27 28 31 33 34 35 36 37 38 45 R1 R2 R3 R4 R5 R1 R2 R3 R4 R5 R6

39 40 41 42 43 44

Title Phase 2 HSE Phase 3a HSE Phase 3a HSE Phase 3a HSE Phase 3a HSE Phase 3a HSE Phase 3a HSE Phase 3a HSE Phase 3a HSE Phase 3a HSE Phase 3a HSE Phase 3a HSE Phase 3a HSE Phase 3a HSE Phase 3a HSE Phase 3a HSE Phase 3a HSE Phase 3a HSE Phase 3a HSE Phase 3a HSE Phase 3a HSE Phase 3a HSE Phase 3a HSE Phase 3a

Test 24 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

46

APPENDIX C AVERAGE RESPONE SPECTRA FOR THE PHASE 2 AND SELECTED PHASE 3A TESTS
(May be found in Volume 2)

47

APPENDIX D
D.1 Trace Pre-processing

TRACE PROCESSING

The pressure traces supplied for the Phase 2 and 3A experimental and simulated results were processed in the following ways to enable the response calculations to be performed for each of the traces supplied. Phase 2 tests 27 tests each with about 20 measurement positions giving 540 traces each with about 64,000 points. The traces were processed by calculating moving averages on a 20 point basis. The filtered traces were compared with the experimental traces. The response characteristics were unchanged for selected filtered traces. Phase 2 simulations Each trace was supplied in a form which could be used for response calculations without modification. Phase 3A tests 45 tests of which 11 were selected. calculating moving averages on a 20 point basis. The traces were processed by

Phase 3A simulations These simulations were supplied in a form which included variable time steps between sample points. Linear interpolation was used where the time steps were large (5msec) to produce numerically stable inputs to the response calculations.

D.2

Assembly of Response Database

The duration td, and effective pressure Pe was derived for each pressure trace. Each pressure trace was run through the response calculation for 23 natural periods chosen to give td/T ratios in the range 0.1 to 10. The traces were each run for each natural period above and for 100 resistance values chosen to give an R/Pe ratio in the general range 0.2 to 2. This yielded a response surface with 2,300 points for each trace. Phase 2 tests 27 tests each with about 20 measurement positions giving about 500 traces. Phase 2 simulations 27 simulations. The first 20 traces were run through the response calculations. Phase 3A tests 11 were selected from the 45 tests as described in section 3.5 giving about 220 traces. Phase 3A simulations 9 simulations were selected from the 45 supplied to correspond with the chosen test results. The construction of these response surfaces involved in excess of 3,000,000 calculations. Four workstations were kept fully occupied for a period of two months to provide these results. Response surfaces and response spectra were generated for each trace.

48

The responses R/Pe were then averaged for each test over all the response spectra for each trace and at each point corresponding to a fixed td/T value as follows:Mean (DLFi )

1 n DLFi n i =1

The Pe to be used to determine the required static resistance for the averaged response spectrum is defined to be the mean Pe over all traces. The standard deviations of the R/Pe values or DFLs were similarly calculated for each test from Standard deviation (DLFi )

n (DLF mean (DLF ))

1
i i i =1

The coefficient of variation (COV) defined by:COV = Standard deviation/Mean Was then calculated for each test from the standard deviations and means for that test. Appendix D contains the averaged response spectra for each of the Phase 2 and Phase 3A tests. The plots of the standard deviations and coefficients of variation are also plotted in this Appendix. The mean response spectra for each test were then averaged across tests in the same way to obtain a test series mean response spectrum.

D.3

Necessary Modifications to the Biggs Algorithm

The special feature of the response algorithm was that a large range of resistances and natural periods needed to be run through the model. This required a very stable algorithm at a large range of time step/natural period values. The irregular nature of the experimental pressure traces meant that for large td/T values (long duration, pseudo-static response) the response was very complex. The full range of resistance displacement curves was required as the load could momentarily be in the negative direction with a magnitude greater than the static resistance of the target structure. For elastic response at displacements near to or below yield ( = 1) a small amount of damping was introduced at a level of 0.01% of critical. This is acceptable for welded steel structures in air. This in fact made a negligible difference to the peak response but maintained the stability of the response model.

49

APPENDIX E

CLOSED FORM SOLUTIONS

Analytic studies of response have been performed to check the results in various regions of the response surfaces. Good agreement was found with the numerically calculated results. The shape of the response spectra has been theoretically verified.

E.1

The Pseudo-static Asymptote

For long duration loadings with respect to the target structure natural period in the range td/T > 3, the pseudo-static approximation applies. This is referred to as the pseudo-static asymptote. At the moment of maximum deflection the structure is stationary and all its energy is in the form of potential or strain energy. Equating the work done in a triangular pulse to the stored energy gives:-

R 2 = DLF = (E.1) 2 1 Pe
This is an approximate result, Implementation of this simple formula gives the response spectra shown in figure E.1. This indicates the expected behaviour of the response spectra for td/T values greater than about 3.

Figure E.1

Expected Behaviour of the Response Spectra for td/T > 3 (Pseudo-static or Quasi-static Asymptote)

50

This approach is described in many references including reference 13.

E.2

The Impulsive Asymptote

The impulsive asymptote corresponds to situations where the load duration is much shorter than the natural period, td/T < 0.3. In this regime the shape of the load does not affect the response which is entirely determined by the load impulse. For these short period loads the structure does not have time to respond before the load has subsided. All the energy of the explosion is in the form of kinetic energy at the time at which the structure starts to move. This energy is converted to potential/strain energy at the point of maximum deflection. Impulsive response is much less than would be indicated by the peak value of the load. Equating the kinetic energy at the start with the final potential energy gives:-

1 t d R = 2 1 T (E.2) Pmax
Implementation of this formula gives the response spectra shown in figure E.2. This is the expected behaviour of the response spectra for td/T < 0.3.

Figure E.2

Expected Behavior of the Response Spectra for td/T < 0.3 The Impulsive Asymptote

This approach is described in many references including reference 13. 51

E.3

Elastic dynamic Response

The intermediate regime where td and T are compatible is the elastic dynamic regime. The most appropriate reference is reference 14 where the case of the response of a single degree of freedom oscillator to a half sine wave is considered for the elastic case. The equations of motion including both the mass (inertia) and stiffness terms is solved directly to obtain the peak response Xmax/Xe = . tmax is defined as the time at which the maximum deflection occurs. If the peak response occurs before the end of the loading (tmax < td) then the peak response is given by:-

P max 1 2 2 {sin 2 sin + } ..(E.3) R 1 1 /

Where =

td

the loading angular frequency

=2 =

The natural angular frequency of the structure

the angular loading and natural frequency ratio

2 the time of maximum deflection ( + )

t max =

If the peak response occurs during the time of the loading (td > tmax) then the peak response is given by:-

Pmax 2 cos 2 2 ..(E.4) R 1

Implementation of these formula gives the response spectra shown in figure E.3. This is the expected behaviour of the response spectra for 0.3 < td/T < 3. Resonance occurs around td/T = 1 as expected. The formulae above have been interpreted as applying to nonelastic response for values of greater than unity.

52

Figure E.3 Expected Behavior of the Response Spectra For 0.3 < Td/T < 3 Elastic Dynamic Response These curves represent a generalization of conventional shock spectra [14] for elastic systems which are restricted to ductility 1. The shape function for a half sine wave Ppeak/Pe is / 4 (0.785). Hence there is a need to multiply the response by this factor to get the equivalent response for the same Pe for a triangular idealisation.

53

Printed and published by the Health and Safety Executive C30 1/98 Published by the Health and Safety Executive 08/06

RR 484