Design, failure and safety: how engineering knowledge grows: Four failures and what each one taught
All courses / Understanding Engineering / Module 12: Design, failure and safety: how engineering knowledge grows Part 2 of 5Four failures and what each one taught
Four public cases follow, told from official inquiries and reliable reports. Each shows a different link in the chain: what was not known, what was changed without checking, what the data said and was not heard, and what was not tested. The numbers come from the investigation records and the arithmetic is ours. Questions of legal blame belong to the courts; the aim here is the engineering lesson.
2.1 The Comet airliners: nobody knew how fast a pressurised skin would tire
In the early 1950s the de Havilland Comet was the first jet airliner in passenger service. To fly high and fast it needed a pressurised cabin, so every flight swelled the fuselage like a balloon and every landing let it down. In 1954 two Comets broke up in the air: one near Elba on 10 January, killing 35, and one near Naples on 8 April, killing 21. A court of inquiry under Lord Cohen reported in 1955. The Royal Aircraft Establishment had put an entire fuselage in a water tank and cycled the pressure to mimic flights. It failed after 1,830 further cycles, which with its earlier flights made 3,060, and the crack began at a corner of the forward escape hatch. The Elba aircraft had made 1,290 pressurised flights and the Naples aircraft 900. The makers' aim had been 10,000.
What was missing was knowledge, not carelessness. Metal fatigue was understood in principle, but nobody had a method to predict the life of a whole pressurised fuselage, and the fatigue requirements had not been changed for aircraft with pressurised cabins. An earlier fatigue test had also misled: its fuselage section had first been proof-loaded to twice the working pressure, which left the metal at its corners in a stronger state, and it went on to survive about 16,000 cycles. It was a stand-in that left out something that mattered.
Two lessons followed. Fatigue life must be shown by testing a complete structure under the repeated loads of service. And a test result must be cut by a scatter factor before it becomes a licensed life, because identical parts do not fail at identical cycle counts. Estimate 2 shows what that would have meant. The module Structures and materials explains fatigue and stress concentration; the point here is how one case became a rule.
| Quantity | Working | Result |
|---|---|---|
| Tank-test fuselage, total pressure cycles to crack | 1,230 earlier flights + 1,830 tank cycles | 3,060 (calculated) |
| Designers' target life | stated aim | 10,000 cycles |
| Shortfall against the aim | 10,000 / 3,060 | about 3.3 times (calculated) |
| Scatter factor, chosen for illustration | one test article, wide scatter in fatigue lives | 4 |
| Life that could be licensed from the tank result | 3,060 / 4 | about 765 flights (calculated) |
| Flights made by the Naples aircraft | from the inquiry record | 900, above 765 |
| Flights made by the Elba aircraft | from the inquiry record | 1,290, well above 765 |
Check: simple division throughout. Real certification uses factors that depend on the number of test articles and the material; 4 is a round figure. Even so, an honest scatter factor on such a test, had it been run before service, would have put both aircraft beyond their safe lives.
Source: Aerossurance, 'Common Comet Misconceptions and Collaborative Contribution to Safety' (read November 2026), summarising Court of Inquiry report CAP 127 (1955), dates, fatalities, 1,230 + 1,830 = 3,060-cycle tank test with failure at the forward escape hatch, 1,290 and 900 flights, 10,000-cycle aim, earlier test section proof-loaded and run to 16,000 cycles, unchanged fatigue requirements; https://aerossurance.com/?p=10066- Petroski, To Engineer Is Human (Macmillan, 1985) · ch. 10, 'When Cracks Become Breakthroughs'; ch. 14, 'Forensic Engineering and Engineering Fiction'
- Aerossurance, 'Common Comet Misconceptions and Collaborative Contribution to Safety' (read November 2026), summarising Court of Inquiry report CAP 127 (1955) · web
2.2 The Hyatt Regency walkways: a small change in a connection doubled the load on its weakest detail
On the evening of 17 July 1981 the lobby of the Hyatt Regency hotel in Kansas City, Missouri, was crowded for a dance. Two suspended walkways crossed the atrium, one at the second floor and one at the fourth, hung from the roof by steel rods. Both fell, and 114 people died. The US National Bureau of Standards carried out the technical investigation and published it in 1982.
The original drawing showed long rods running from the roof through the fourth-floor walkway and on down to the second-floor one, so each walkway hung from its own nuts on the same rod. Getting a nut to the middle of a very long threaded rod was awkward, and the fabricator proposed another arrangement: one rod from the roof to the fourth-floor walkway, and a second rod hanging from that walkway down to the second. On a drawing the change looks trivial. It made the nut and washer under the fourth-floor beam carry the weight of both walkways, which doubled the load on the weakest detail. Each walkway beam was a pair of steel channels welded into a box, with the rod through the joint between them. The investigators found that neither the materials nor the workmanship started the collapse. The detail was simply overloaded: the box beam deformed, and the rod and its nut pulled through it.
The causes were not exotic. Nobody recalculated the connection after the change, and responsibility for checking such drawings was unclear. The structure also lacked a second load path: when one connection let go, its load passed to neighbours that were no stronger. Estimate 3 uses the investigators' figures to show how little margin the built detail had.
| Quantity | Pounds (as reported) | Kilonewtons | Ratio |
|---|---|---|---|
| Design load on one connection (walkway weight plus design live load) | 40,700 | 181 (calculated) | 1.00 |
| Ultimate capacity the code expects of such a connection | 68,000 | 302 (calculated) | 68,000 / 40,700 = 1.67 (calculated) |
| Average ultimate capacity of the as-built connection | 18,600 | 83 (calculated) | 18,600 / 68,000 = 27 per cent of the code requirement |
| Capacity of the as-built connection against the design load | 18,600 | 83 (calculated) | 18,600 / 40,700 = 0.46 (calculated) |
| Largest load on any connection at the collapse | 21,400 | 95 (calculated) | 21,400 / 18,600 = 1.15 (calculated); and 21,400 / 40,700 = 0.53 of the design load (calculated) |
Conversion: 1 pound-force = 4.448 newtons. Check: 21,400 / 68,000 = 31 per cent, matching the investigators' statement that the load at collapse was 31 per cent of the ultimate capacity expected under the code. The built detail could not carry even half of its design load.
Source: National Bureau of Standards, Investigation of the Kansas City Hyatt Regency Walkways Collapse, NBSIR 82-2465 (1982), figures 40,700 lb, 68,000 lb, 18,600 lb, 21,400 lb, the 27 and 31 per cent ratios, and the finding that materials and workmanship did not start the collapse; https://nvlpubs.nist.gov/nistpubs/Legacy/IR/nbsir82-2465A.pdf- Petroski, To Engineer Is Human (Macmillan, 1985) · ch. 8, 'Accidents Waiting to Happen'
- National Bureau of Standards, Investigation of the Kansas City Hyatt Regency Walkways Collapse, NBSIR 82-2465 (1982) · web
- Linda Hall Library, 'Hyatt Regency walkways collapse' · date and deaths (114; the NBS report, written earlier, gives 113); https://www.lindahall.org/about/news/scientist-of-the-day/hyatt-regency-walkways-collapse/
2.3 Challenger: the data had been seen, and the decision was reached with the burden of proof the wrong way round
On 28 January 1986 the space shuttle Challenger broke apart 73 seconds after lift-off, and all seven crew died. The presidential commission chaired by William Rogers traced the cause to a joint between sections of a solid rocket booster. Rubber O-rings sealed each joint. When the motor fired, the steel casing swelled slightly and the gap opened, and the ring had to spring across it within a fraction of a second. On a cold morning the rubber was stiff and slow, hot gas leaked past it, and the leak grew into a jet of flame that destroyed the structure.
The weakness was not unknown. A NASA memo of 17 July 1985 counted twelve cases of primary O-ring erosion on earlier flights. All those flights had returned safely, and over time the damage came to be read as a sign that the joint worked well enough. The sociologist Diane Vaughan named this drift normalisation of deviance: a departure from the design, once survived, becomes the new normal. On the evening before the launch, engineers at the booster maker, Morton Thiokol, advised against flying in the forecast cold. After an off-line discussion the company's managers reversed the advice. The burden of proof had swung round: the engineers were left needing to show the launch was unsafe, when the managers should have needed to show it was safe.
Estimate 4 shows a second error, extrapolation. Every earlier flight had been warmer. The coldest earlier joint was at about 53 degrees Fahrenheit, and Thiokol's calculation after the accident put the failed joint at about 16. There was no experience at all in that range, so no trend line through the old data could be trusted there.
| Quantity | Fahrenheit | Celsius |
|---|---|---|
| Coldest earlier launch, O-ring temperature | 53 | (53 - 32) x 5/9 = 11.7 (calculated) |
| Failed joint at launch, Thiokol's calculation after the accident | 16 | (16 - 32) x 5/9 = -8.9 (calculated) |
| Gap between the two | 37 degrees (calculated) | 37 x 5/9 = 20.6 degrees (calculated) |
Check: -8.9 C is below the freezing point of water, while all earlier flights lay above 11 C. A fitted line through data in that range says nothing reliable about a point more than 20 degrees outside it. The lesson is the habit of asking how far a conclusion lies beyond the evidence.
Source: US House of Representatives, Committee on Science and Technology, Investigation of the Challenger Accident, House Report 99-1016 (1986), 53 F seal temperature of the coldest earlier flight (51-C), 16 F aft field joint temperature calculated by Thiokol after the accident, 17 July 1985 memo on twelve primary O-ring erosion cases; https://www.govinfo.gov/content/pkg/GPO-CRPT-99hrpt1016/pdf/GPO-CRPT-99hrpt1016-2.pdf- US House of Representatives, Committee on Science and Technology, Investigation of the Challenger Accident, House Report 99-1016 (1986) · web
- Aerossurance, 'NASA Challenger Launch Decision' (read November 2026) · Rogers Commission finding, Thiokol recommendation and its reversal, Vaughan's normalisation of deviance; https://aerossurance.com/safety-management/challenger-launch-decision-30/
2.4 Morbi, 2022: the knowledge to check the bridge was old, and the gate that would have required it was missing
The Morbi bridge in Gujarat was a pedestrian suspension bridge over the Machchhu river, opened in 1887. It was closed in March 2022 for renovation and reopened on 26 October. On the evening of 30 October it fell with about 300 people on it, and 135 died. A five-member Special Investigation Team (SIT) submitted a preliminary report in December 2022, and its findings became public in February 2023.
The module Structures and materials sets out what the SIT found in the steel and works out what the corroded cable could still carry. In short, a main cable was badly corroded, the hangers and deck had been changed, and nobody had measured what the altered bridge could bear. The question here is why no step in the process caught it.
Compare the shape of this case with the others. Hyatt had a design error and Comet had a gap in knowledge. At Morbi, the knowledge to survey a cable, review a repair and load-test a bridge before opening it has existed for generations. What was missing was a gate: someone competent, with the authority to say no, whose approval was required before a changed structure carried the public again. A factor of safety is only as good as the last time anyone checked that it was still there. The table lists the checks that would normally have stood in the way.
| Check | What it can find | What the SIT reported |
|---|---|---|
| Condition survey of the main cables before work | lost wire section, corrosion, broken wires | 22 of 49 wires in one main cable corroded, probably already broken |
| Engineer's review of the repair design | changes in weight, stiffness and load path | aluminium panels replaced timber; old and new hangers welded together |
| Load test before reopening | actual behaviour under a controlled crowd load | none carried out |
| Approval and inspection by the owner or authority | independent confirmation that the work is complete and safe | reopened on 26 October without prior approval or inspection |
| Crowd limit and control on the day | keeps the load within what was verified | about 300 people on the deck, far more than its load-bearing capacity in the SIT's view |
The right-hand column follows the SIT findings as reported by the Press Trust of India in February 2023. The middle column is general engineering practice, not a statement of what any individual should have done.
Source: Press Trust of India, 'Corroded wires, welded suspenders behind Morbi bridge tragedy: SIT report', ThePrint (20 February 2023), SIT findings; https://theprint.in/india/corroded-wires-welded-suspenders-behind-morbi-bridge-tragedy-sit-report/1384720- Press Trust of India, 'Corroded wires, welded suspenders behind Morbi bridge tragedy: SIT report', ThePrint (20 February 2023) · SIT findings, preliminary report of December 2022; https://theprint.in/india/corroded-wires-welded-suspenders-behind-morbi-bridge-tragedy-sit-report/1384720
- Petroski, To Engineer Is Human (Macmillan, 1985) · ch. 14, 'Forensic Engineering and Engineering Fiction'
See it in the table
H₂O WaterSt SteelCheck your understanding
The Comet tank test failed after 3,060 cycles. With a scatter factor of 4, what life could be licensed from that result?
3,060 / 4 is about 765. Both lost aircraft had already made more flights than that, which shows why the factor matters.By the investigators' figures, what share of the code's required ultimate capacity did the as-built Hyatt Regency connection have?
The average as-built capacity was 18,600 lb against 68,000 lb expected, about 27 per cent. The original arrangement would have given about 60 per cent.What was the main point of difference in the Morbi case compared with Comet?
At Comet the knowledge was missing. At Morbi the reported failures were of surveying, reviewing and testing, and of approval before reopening.