Page 42 - Failure Analysis Case Studies II

28
together they can carry 1.59MN. The total load that corresponds to the 19.5kN/m2 maximum
pressure applied simultaneously across the whole firewall is 5.77 MN. The clamps are at midheight,
and can be expected to carry at least half the total load. It follows that the clamps are not strong
enough to carry the total load on the wall, and that the clamps would break if the the wall had not
already broken up by failure of the frame bolts.
The analysis is based on plate theory, which is approximate because the deflection is not necessarily
small by comparison with the effective thickness of the firewall. The effective thickness of the wall
was estimated by finding the thickness which gives the same ratio between the fully-plastic membrane
stress resultant at collapse in pure tension and the fully-plastic membrane stress resultant at collapse
in pure bending [3], both calculated for the governing mode of frame-bolt failure in tension. The
efective thickness turns out to be 80mm for one direction of bending and 120mm for the other.
Moreover, the sides of the wall segments are not rigidly fixed at the top and bottom. It is known [4]
that small inward movements at the edges of transversely-loaded plates much reduce the stiffening
effect of membrane action, and an approximate calculation showed that in this instance an inward
edge movement of the order of 1 mm would be enough to eliminate a significant increase in strength
because of membrane effects. It was concluded that these effects could be neglected.

10. RESPONSE OF C/D AND A/B FIREWALLS

The wall between modules C and D was much stiffer and stronger than the wall between modules
B and C. The estimated collapse pressure of one of its triangular panels under quasi-static slow
loading is about 12 kN/m* (0.12 bars), compared to the peak pressure of 19.5 kN/m2 at PI in module
C. The lowest natural frequency of one of its triangular segments is about 410 rad/s, corresponding
to a period of 15 ms, and its response is not far from quasi-static.
The control room was in D module to the north of the C/D firewall, and had an additional wall
of steel plate. Two survivors were in the control room at the time of the explosion. They were blown
across the room, and saw that equipment near the wall had been damaged and that smoke was
apparently entering at the top part of the wall. Accordingly, since the C/D wall is stronger than the
B/C wall, it can be concluded independently that the B/C wall would have been more severely
damaged by an explosion in C module than the C/D wall was.
The A/B wall was similar to the B/C wall in construction and arrangement. There is evidence
from survivors that the A/B wall was not damaged. This supports the conclusion that the initial
explosion was in C module. If the initial explosion had been in B module, it cannot be explained
how the explosion leaves A/B intact but breaks down the stronger C/D wall. This is a particularly
robust conclusion, and is of course independent of the calculations.

I I. CONCLUSIONS
The analysis of the B/C firewall is consistent with the conclusion of the public inquiry, that an
initial explosion in C module was followed by breakup of the firewall and projection of panel
fragments into B module.

AcknowledgementsThe author thanks Elf Aquitaine and Paul1 and Williamsons for permission to publish this paper, and
records his gratitude to David Allwright, Derek Batchelor. Roger Fenner, Lesley Gray, Colin MacAulay, Alan Mitchison
and Rod Sylvester-Evans for helpful discussions.

REFERENCES
I. The Honourable Lord Cullen, The Public Inquiry into the Piper Alpha Disaster, HMSO, 1990, Command 1310.
2. Bakke, J. R., Gas Explosion Simulation in Piper Alpha Module C Using FLACS. Christian Michelsen Institute, 1989,
Report CM1 no. 25230-1.
3. Jones, N., Structural Dynamics, Cambridge University Press, 1989.
4. Jones, N., International Journal of Mechanical Sciences, 1973, 15, 547-561.
5. Gradshteyn, I. S. and Ryzhik, I. M., Table of Integrals, Series, and Products. Academic Press, 1979.
6. Mansfield, E. H., Proceedings ofthe Ro-Val Society A, 1957,241,311-338.