Nicoll Highway — an Under-Designed Deep Excavation and a 2004 Collapse That Killed 4

Summary

At about 15:30 on 20 April 2004, a 33-metre-deep braced excavation for Singapore's Circle Line MRT, dug through soft marine clay beside the Nicoll Highway near Merdeka Bridge, collapsed inward without warning, killing four workers and injuring three. Over roughly ten minutes the temporary retaining walls buckled, a 100-metre stretch of the six-lane highway dropped into a crater some 30 metres deep and more than 100 metres across, and gas mains, water lines and the partly built tunnel structure were destroyed. The collapse was not an act of nature. The Committee of Inquiry traced it to two compounding design errors made by the contractor's engineers — a geotechnical analysis that overestimated the clay's strength, and a steel waler connection detailed at a fraction of its required capacity.

The excavation was the Tunnel Service Area cofferdam, dug between two diaphragm walls braced apart by ten levels of steel struts. The governing error was in the soil model. The designers used an effective-stress analysis — known on the project as Method A — to represent the undrained behaviour of Singapore's soft marine clay. That method overestimated the clay's undrained shear strength by roughly half, which in turn underestimated the bending moments and deflections the diaphragm walls would actually experience by a comparable margin. The wall was, in plain terms, designed for loads far smaller than the ones the ground would impose.

The second error sat in the connections. The horizontal waler beams that gathered the strut loads were joined to the diaphragm wall through a splay assembly stiffened by steel C-channels. The designers misread the stiff-bearing length under BS 5950 and credited the channels with an effective length factor of 0.7 where the unrestrained end conditions demanded 1.2 — leaving the connection at roughly 70 percent of the load it was assumed to carry. When the ninth-level struts took up their share of the load, the under-strength waler connections yielded, the walls deformed, the struts above were overloaded in turn, and the bracing unzipped in a progressive collapse.

The Committee of Inquiry, convened by Singapore's Ministry of Manpower and reporting in 2005, concluded that the disaster was preventable and the product of human error and organisational failure, not unforeseeable ground. It became the canonical modern case of a deep braced excavation lost not to the soil's caprice but to a soil-model assumption — and it permanently changed how Singapore designs, checks, monitors and supervises deep excavations.

---

Timeline

2001–2003

Design of Contract C824

Nishimatsu–Lum Chang, as design-and-build contractor for the LTA's Circle Line, design the Nicoll Highway cut-and-cover works, including the Tunnel Service Area (TSA) excavation braced by ten strut levels between diaphragm walls.

2003

Method A adopted for the soil

The diaphragm wall is analysed with an effective-stress ('Method A') simulation of undrained marine clay, which overestimates undrained shear strength by about 50% and underestimates wall bending moments and deflections by a similar margin.

2003

Waler connection detailed

The strut-to-wall waler connections, stiffened by steel C-channels, are designed using an effective length factor of 0.7 and a misread BS 5950 stiff-bearing length, leaving the connection at roughly 70% of its assumed capacity.

Late 2003 – early 2004

Excavation proceeds

Digging advances downward, strut level by strut level. Inclinometers and instrumentation record wall movements that exceed predicted values as the cut deepens through the soft clay.

Early 2004

Warning movements normalised

Measured wall deflections approach and pass review levels. The monitoring data are reinterpreted — review levels are raised and a faulty inclinometer reading is discounted — rather than treated as a stop signal.

March 2004

Ninth strut level installed

Excavation reaches the depth requiring the ninth level of struts. The waler connections at this level are now carrying loads near their true, deficient capacity.

~18–20 April 2004

Distress accelerates

Struts and walers in the affected zone show visible distress; remedial measures, including additional plates, are attempted but cannot make up the connection's missing capacity.

20 April 2004, ~15:30

Collapse begins

A ninth-level waler connection yields. The wall deflects, overloading struts above; the bracing fails in sequence and both diaphragm walls fold inward.

20 April 2004, ~15:30–15:40

Highway drops into the crater

About 100 m of the six-lane Nicoll Highway subsides into a crater roughly 30 m deep and over 100 m across. Four workers are killed; three are injured.

21 April 2004

Rescue and stabilisation

Search-and-rescue and crater stabilisation begin amid ruptured gas and water mains; the Circle Line programme is suspended at the site.

22 April 2004

Committee of Inquiry convened

The Ministry of Manpower establishes a Committee of Inquiry to determine the cause and assign responsibility.

13 May 2005

COI reports

The Committee concludes a string of critical design errors — chiefly the Method A under-design of the wall and the under-designed waler connection — caused a preventable collapse. Charges and fines follow.

The Build: A Braced Box in Marine Clay, Held Apart by Ten Levels of Steel

The Nicoll Highway works belonged to Contract C824 of the Circle Line, a deep underground rail line driven through some of the softest ground in Singapore. Where the line passed beneath the Nicoll Highway near Merdeka Bridge, the stations and tunnels were built by cut-and-cover: a long, deep trench excavated between two reinforced-concrete diaphragm walls, with the walls braced apart by horizontal steel struts installed in stages as the dig went down. The Tunnel Service Area cofferdam was the deepest part — roughly 33 metres of excavation requiring ten levels of bracing, all of it temporary works the ground would press against until the permanent tunnel was complete.

The ground was the problem the design existed to solve. Beneath the site lay thick deposits of soft marine clay, a weak, water-saturated soil that flows toward any unsupported face and loads a retaining wall heavily. The diaphragm wall and its struts had to hold that clay back across the full depth. The decisive engineering question was therefore not the steel or the concrete but the soil: how strong is the clay, and how hard will it push?

To answer it, the designers ran a soil-structure analysis using effective-stress strength parameters — the project's Method A — to model the clay's short-term, undrained response during excavation. It was the wrong tool for the condition. Applying drained strength parameters to an undrained problem overestimated the clay's undrained shear strength by about 50 percent. Because a stronger clay appears to push less, the calculated bending moments and deflections in the diaphragm wall came out roughly half of what the real, weaker clay would impose. The most heavily loaded element on the site — the wall holding back the marine clay — was sized against a fiction of the ground.

The Failure Sequence: A Yielding Connection and a Bracing That Unzipped

The second defect waited at the joints. Each level of struts delivered its compressive load to the diaphragm wall through a horizontal waler beam, and the waler met the wall at a splayed connection stiffened by steel C-channels. Designing those channels, the engineers misapplied BS 5950: they took the stiff-bearing length as 400 mm rather than the correct 65 mm, and they assigned an effective length factor of 0.7 — appropriate to a restrained member — when the actual end conditions were unrestrained and demanded 1.2. The result was a connection with an axial capacity near 70 percent of the load it was presumed to carry. The waler joints were the weakest link in a system designed as though they were not.

For months the excavation deepened and the instruments told the truth the calculations had missed. Inclinometers in the diaphragm walls recorded movements that ran ahead of predictions, climbing toward and past the review thresholds set to trigger intervention. But the readings were explained away rather than acted on: review levels were revised upward, and one inclinometer showing alarming movement was dismissed as faulty. The monitoring system that should have caught the under-design was instead used to rationalise it.

By April 2004 the cut had reached the ninth strut level, and the connections there were loaded near their real, deficient capacity. On 20 April, at about 15:30, a ninth-level waler connection yielded. The wall it braced deflected inward; that deflection transferred load to the struts above, which were now overstressed in turn; their connections failed; and the bracing came apart in sequence, top following bottom, in a textbook progressive collapse. Both diaphragm walls folded into the trench. A 100-metre length of the Nicoll Highway dropped into a crater some 30 metres deep and more than 100 metres wide, taking gas mains, water lines and the half-built tunnel with it. Four workers in the excavation were killed; three were injured. From first yield to full collapse was a matter of minutes.

The Reckoning: A Preventable Failure, and a Discipline Put on Notice

The Committee of Inquiry, convened by the Ministry of Manpower on 22 April 2004 and reporting in 2005, heard months of expert testimony and reached an unambiguous verdict. The collapse was preventable, and it was caused by human error and organisational failure — not by unforeseeable ground conditions. Two design errors were primary. First, the under-design of the diaphragm wall flowing from Method A, which had overestimated the marine clay's undrained shear strength by about half and so underestimated the wall's bending moments and deflections by a comparable margin. Second, the under-design of the waler-to-wall connection, left at roughly 70 percent of the load it carried by the misread BS 5950 stiff-bearing length and the wrong effective-length factor.

Around those technical faults the inquiry found a ring of organisational failures. The contractor designed, built and monitored its own temporary works, so the same party that made the modelling error also judged whether the instruments justified stopping. When the inclinometers flagged excessive movement, the response was to raise review levels and discount data rather than halt and reassess. The back-analysis — a corrected Method B effective-stress check — that should have caught the strength overestimate was not given controlling weight in time. The Land Transport Authority, as owner, had not secured an independent check on the temporary works that would have caught either error.

The legal aftermath fixed responsibility on individuals and firms. Three Nishimatsu engineers were charged under the Factories Act and an LTA project officer under the Building Control Act; the defendants were convicted and fined. The deeper indictment, though, was of a practice that let one consultant's soil-model assumption go unchecked from design through monitoring to collapse. Nicoll Highway became the modern teaching case for deep excavation in soft ground: proof that the controlling failure mode of a braced cut is not the soil surprising the engineer, but the engineer mis-modelling the soil — and then explaining away the instruments that said so.

---

Contributing Factors

01

Wrong soil model for an undrained problem

The diaphragm wall was analysed with an effective-stress method (Method A) using drained strength parameters to represent the undrained behaviour of soft marine clay. This overestimated the clay's undrained shear strength by roughly 50 percent, which underestimated the wall's bending moments and deflections by a similar margin. The single most load-governing assumption on the site was systematically unconservative.

02

An under-designed waler connection

The strut-to-wall waler connections, stiffened by steel C-channels, were detailed with a misread BS 5950 stiff-bearing length (400 mm instead of 65 mm) and an effective length factor of 0.7 where unrestrained end conditions required 1.2. The connection's axial capacity was about 70 percent of the load assumed to pass through it — the weakest link in the bracing and the place the collapse began.

03

Instrument warnings normalised away

Inclinometers recorded wall movements that exceeded predictions and passed review thresholds for weeks before failure. Rather than triggering a stop, the data were reinterpreted: review levels were raised and an alarming reading was dismissed as a faulty instrument. A monitoring system intended to catch the under-design was instead used to justify continuing.

04

No independent check of the temporary works

The contractor designed, constructed and monitored its own temporary excavation support, so the party that made the modelling error also judged whether the readings warranted stopping. No genuinely independent review of the temporary works existed to catch either the soil-strength overestimate or the connection deficiency before they were built and loaded.

05

Progressive collapse with no redundancy

The braced excavation had no reserve load path. When the under-strength ninth-level waler connection yielded, its share of the thrust redistributed to struts above that were already at capacity, overloading them in sequence. The system was a chain in which one failed connection condemned the rest; nothing in the design arrested the cascade once it began. ---

Aftermath

The collapse killed four workers, injured three, destroyed a 100-metre length of the Nicoll Highway and delayed Singapore's Circle Line by years; the rebuilt Nicoll Highway station was realigned and its retaining walls driven to about 60 metres, roughly twice the previous depth, with walls some 1.5 metres thick. The lasting legacy was regulatory and professional. Singapore's authorities moved temporary works out of the contractor's sole control: the Land Transport Authority ceased allowing contractors to design and supervise their own temporary works and to engage their own monitoring, requiring independent consultants for design and independent firms for instrumentation. The use of effective-stress 'Method A' analysis to represent undrained behaviour of soft clay was discredited; the case entered geotechnical practice and the BS 5950 / Eurocode teaching canon as the standard warning against applying drained strength to an undrained excavation. The Building and Construction Authority tightened oversight of deep excavations, mandating accountable qualified persons and independent checks on temporary earth-retaining structures. Nicoll Highway is now the byword, taught worldwide, for a braced excavation lost to a soil-model assumption rather than to the ground itself.

---

Lessons

Match the soil model to the drainage condition, not to convenience: never use drained effective-stress strength to represent the undrained, short-term behaviour of soft clay in a deep excavation — it will overestimate strength and underestimate the wall loads that govern.
Design and check the connections to the same rigour as the members they join; a waler or splay detail at 70 percent of its load is the place a braced system unzips, so verify stiff-bearing lengths and effective-length factors against the real end conditions.
Treat instrument readings as the structure speaking: when monitoring exceeds review levels, stop and reanalyse — do not raise the thresholds or discount the data, because the instruments are usually right and the calculations are usually the thing that was wrong.
Put an independent party between the designer and the decision to proceed; the team that modelled the soil should not also be the team that judges whether the readings justify continuing to dig.
Build redundancy into temporary works, not just permanent ones, and assume a single connection can fail — a bracing system with no reserve load path turns one yielded joint into a progressive collapse in minutes. ---