# Design, failure and safety: how engineering knowledge grows: Why failure teaches more than success

Source page: https://www.materialinsights.org/learn/engineering/design-failure-safety/p1/

[All courses](https://www.materialinsights.org/learn/) / [Understanding Engineering](https://www.materialinsights.org/learn/engineering/) / [Module 12: Design, failure and safety: how engineering knowledge grows](https://www.materialinsights.org/learn/engineering/design-failure-safety/) Part 1 of 5

## Why failure teaches more than success

Engineers do not learn the limits of a design from the designs that stand. They learn them from what falls, cracks or leaks, and from tests that push things until they do. This part shows why a long record of success says little about the margin left, where failures come from, and how the swing between daring and caution makes knowledge grow.

### 1.1 A standing structure is a prediction not yet contradicted, and a failure marks the limit exactly

Think of the hook in a ceiling that carries a fan. It has held for twenty years. That tells you the hook can carry at least one fan. It does not tell you whether it could carry two, or whether it is a whisker from letting go. Every design is a prediction that a thing will carry what is put on it, and a thing that stands is a prediction not yet contradicted. A failure is different. It marks the one load at which capacity and demand were equal, and shows how the part gave up. Henry Petroski's account of design rests on this asymmetry: what an engineer most needs is a clear picture of the ways a design could go wrong.

The loads that matter most are rare, which is why survival is weak evidence. Engineers describe a rare load by its return period. A wind that is exceeded once in 50 years on average has a 1 in 50 chance in any single year. Wind codes, including the Indian code IS 875 Part 3, set the design wind on a return period of this kind. Over a building's life the small yearly chances add up: the chance of at least one such wind in L years is 1 minus (1 minus 1/T) to the power L. Estimate 1 puts numbers on it. A building designed for the 50-year wind has about a two in three chance of meeting it during a 50-year life. After 30 quiet years, the chance that it has simply never been tested is still 55 per cent.

So a quiet record shows only that the worst day has not come yet. That is why engineers keep margins, as the module Structures and materials showed, and test parts on purpose. The module What engineers do described design as a bet on incomplete knowledge; this module asks how the bet is checked.

Data Worked estimate 1: the chance a structure has met its design event

| Return period T of the event | Chance of at least one in 10 years | In 25 years | In 50 years | In 100 years |
|---|---|---|---|---|
| 10 years | 65 per cent (calculated) | 93 per cent (calculated) | 99.5 per cent (calculated) | about 100 per cent (calculated) |
| 50 years | 18 per cent (calculated) | 40 per cent (calculated) | 64 per cent (calculated) | 87 per cent (calculated) |
| 200 years | 5 per cent (calculated) | 12 per cent (calculated) | 22 per cent (calculated) | 39 per cent (calculated) |
| 500 years | 2 per cent (calculated) | 5 per cent (calculated) | 9.5 per cent (calculated) | 18 per cent (calculated) |
| 1,000 years | 1 per cent (calculated) | 2.5 per cent (calculated) | 4.9 per cent (calculated) | 9.5 per cent (calculated) |

Working: chance = 1 - (1 - 1/T)^L, treating each year as independent. Check: for T = 50 and L = 30, (1 - 0.02)^30 = 0.545 (calculated), so a 30-year-old building has a 55 per cent chance of never having met its 50-year wind. All values rounded.
 Source: Bureau of Indian Standards, IS 875 (Part 3):1987, Wind loads, cl. 5.2 and 5.3.1, Table 1: basic wind speed is a 3-second gust at 10 m with a 50-year return period, and k1 = 1.0 for general buildings with a 50-year design life; https://law.resource.org/pub/in/bis/S03/is.875.3.1987.pdf. The probability arithmetic is the author's  Common misreading

#### It has stood for 40 years, so it must be safe

Old structures earn trust by lasting, and the trust is only partly deserved. Their record covers only the loads that have arrived, while the structure changes under it: steel rusts, joints loosen, fatigue cracks grow. The Morbi bridge in Part 2 had stood for 135 years when it fell, after repairs that altered it. A long record is a reason to inspect the structure as it is today.

- Petroski, To Engineer Is Human (Macmillan, 1985) · ch. 4, 'Engineering as Hypothesis'; ch. 5, 'Success Is Foreseeing Failure'

### 1.2 Failures trace to a few families of cause, and a missing theory is only one of them

When investigators take a failure apart, they usually find a chain of causes, not one. Writers sort causes in different ways, and no list is complete. One useful sort has four families, shown in the tree. First, the load was larger than assumed: a crowd bigger than the deck was meant for, a flood higher than the record, a room put to a new use. Second, the capacity was smaller than assumed: a weld never made well, a bar that rusted, a connection nobody checked. Third, a mechanism was missing from the model: nobody had a theory or a test for the way a pressurised skin tires or a bridge deck flutters. Fourth, the system around the design failed: a change was made without checking, a warning did not travel upward, an inspection was skipped, or schedule pressure won an argument.

The fourth family matters most to a reader who judges projects rather than drawing them. Missing mechanisms get discovered, written into codes and stop being surprises. The other families are old ways of going wrong, and they recur because they are about people, money and time more than about physics. Anyone assessing a plant, a bridge or a vehicle line can ask across the families. Who checks changes to the design? Who can stop work? What test was done on the thing as built, rather than as drawn?

Forensic engineers, who investigate failures for courts and regulators, read a failure backwards. They begin with the broken surfaces and the records of loading, then ask what the design assumed, who decided it and what the organisation knew. One cause is rarely enough to explain a disaster, and the cause easiest to blame, such as one worker's slip, is rarely the deepest.

Value chain Four families of cause, with a case of each from Part 2

- **A structural or system failure**

  - **Load larger than assumed** Morbi: about 300 people on a deck rebuilt without a load test
  - **Capacity smaller than assumed** Hyatt Regency: connections able to carry about 27 per cent of the code requirement
  - **Mechanism missing from the model** Comet: fatigue of a pressurised skin. Tacoma Narrows: wind-driven twisting of a slender deck
  - **The system around the design failed**

    - **A change made and not checked** Hyatt Regency
    - **A warning discounted** Challenger
    - **An inspection or approval skipped** Morbi

 Source: Petroski, To Engineer Is Human (Macmillan, 1985), ch. 14, 'Forensic Engineering and Engineering Fiction'; cases from the public inquiries cited in Part 2

- Petroski, To Engineer Is Human (Macmillan, 1985) · ch. 14, 'Forensic Engineering and Engineering Fiction'; ch. 16, 'Connoisseurs of Chaos'

### 1.3 Success breeds daring, daring meets its limit, and knowledge grows in the swing between them

Petroski noticed a rhythm in the history of bridges. A new type succeeds. Designers, trusting it, build the next ones lighter, longer and cheaper, because material saved is money saved. Each success removes a little margin. In time one design goes beyond what the knowledge supports, and it fails. The inquiry adds a lesson, designers turn cautious, and the cycle starts again with a better-informed boundary. Economy pushes toward thinner margins and failure pushes back. Nobody knows in advance exactly where the line is, so it is found by crossing it, sometimes in a laboratory and sometimes in public. The pattern is old. Medieval masons had no theory of stress; they worked from rules of proportion passed from master to apprentice, and a rule changed when a wall cracked or a vault came down.

Walter Vincenti, who studied how aircraft designers came to know what they know, described the same process in general terms. Designers make variations: a different wing section, another joint, a new layout. Something then picks the survivors. At first the only judge is use itself: a new plough shape is tried in the field, and failures are paid for in full. As knowledge grows, engineers build stand-ins for use: a calculation, a model in a tank or tunnel, a rig that loads one part. Most bad ideas then die cheaply. This is the reason testing and simulation exist.

Stand-ins can be wrong, and their failure teaches too. On 23 August 1991 the concrete base of the Sleipner A gas platform sank in a Norwegian fjord during a controlled ballasting test, a loss of about 700 million dollars. The investigators found that the finite element model used in design had understated the shear stress in one wall junction by 47 per cent, so the walls there were too thin. The model had been too coarse exactly where it mattered. Every stand-in is a bet that it leaves out nothing important. Part 4 returns to this.
 Case

#### Tacoma Narrows, 1940: a gap in the model

The Tacoma Narrows Bridge, with its slender deck, fell in a moderate wind four months after opening. The module Vibration, waves and resonance gives the physics. The knowledge lesson is that designers had a good theory for the deck's weight and none for how wind would twist it. A board of inquiry in 1941 pointed to the need for wind-tunnel tests of models, and these became standard for long suspension spans. The failure created a stand-in where none had existed.

- Petroski, To Engineer Is Human (Macmillan, 1985) · ch. 9, 'Safety in Numbers'; ch. 13, 'The Ups and Downs of Bridges'
- Vincenti, What Engineers Know and How They Know It (Johns Hopkins University Press, 1990) · ch. 8, 'A Variation-Selection Model for the Growth of Engineering Knowledge'
- Hammack, The Things We Make (Sourcebooks, 2023) · ch. 1, 'The Invisible Method: How to Build a Cathedral'
- D. N. Arnold, University of Minnesota, 'The sinking of the Sleipner A offshore platform', excerpting the SINTEF investigation report · 23 August 1991, shear underestimated by 47 per cent, loss about 700 million dollars; https://www-users.cse.umn.edu/~arnold/disasters/sleipner.html

### See it in the table
[St Steel](https://www.materialinsights.org/commodities/steel/)

### Check your understanding

A building was designed for the 50-year wind and has stood for 30 years. What does that record show?
 With a 1 in 50 chance each year, the chance of no such wind in 30 years is (0.98)^30, about 55 per cent. Surviving 30 years is weak evidence about the margin.

Which of these is a failure of the system around the design, rather than of the physics?
 The Hyatt Regency change was an organisational failure: the arithmetic existed, but no one redid it after the change.

What does 'selection by stand-in' mean in Vincenti's account of design?
 Engineers test ideas on paper, in tunnels and on rigs so that most bad variations die before they reach use. The stand-in must itself be checked, as the Sleipner platform showed.

[← Module contents](https://www.materialinsights.org/learn/engineering/design-failure-safety/)[Mark read · Part 2: Four failures and what each one taught →](https://www.materialinsights.org/learn/engineering/design-failure-safety/p2/)

## Other sections

- [Module overview](https://www.materialinsights.org/learn/engineering/design-failure-safety/index.md)
- [Four failures and what each one taught](https://www.materialinsights.org/learn/engineering/design-failure-safety/p2/index.md)
- [What a factor of safety really buys](https://www.materialinsights.org/learn/engineering/design-failure-safety/p3/index.md)
- [The design cycle, testing and how knowledge is stored](https://www.materialinsights.org/learn/engineering/design-failure-safety/p4/index.md)
- [Risk, redundancy and reliability numbers](https://www.materialinsights.org/learn/engineering/design-failure-safety/p5/index.md)
- [Key terms, quiz and sources](https://www.materialinsights.org/learn/engineering/design-failure-safety/review/index.md)
