# Design, failure and safety: how engineering knowledge grows: What a factor of safety really buys

Source page: https://www.materialinsights.org/learn/engineering/design-failure-safety/p3/

[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 3 of 5

## What a factor of safety really buys

The module Structures and materials defined the factor of safety and listed typical values. This part asks what that margin buys in terms of risk, what it cannot buy, and how aircraft designers live with a small factor by letting parts crack safely and finding the cracks in time.

### 3.1 A factor of safety buys a small chance of failure, and the spread in loads matters as much as the factor

A factor of safety is a ratio of two averages, but real strengths and real loads are not single numbers. Two bars from the same factory differ a little, and the crowd on a floor changes from hour to hour. Picture each as a bell-shaped spread about its average. Failure happens when a weak part meets a heavy load, so the chance depends on how much the two spreads overlap. A bigger factor pushes the strength spread away from the load spread. Narrower spreads do the same job without adding any material.

A spread is measured by the coefficient of variation, the standard deviation as a fraction of the average. For the margin (strength minus load), the reliability index is the average margin divided by its standard deviation. It counts how many standard deviations separate the average case from failure, and the bell curve turns it into a probability. Estimate 5 runs three factors against two levels of knowledge about the load, with strength known to 10 per cent.

The result is striking. With the load known to 25 per cent, going from a factor of 2 to a factor of 3 cuts the chance of failure from about 9 in 10,000 to about 1.5 in 10 million, nearly 6,000 times. But if the load is known only to 50 per cent, even a factor of 3 gives about 3 in 10,000, no better than a factor of 2 with a well-known load. Knowing the load is worth as much as a great deal of extra steel. That is the economic case for weighing trucks, counting visitors and testing materials: knowledge is often cheaper than margin. One caution: the bell curve is a model, and the extreme tails that decide failure are the part we know least about.

Data Worked estimate 5: chance of failure against factor of safety and knowledge of the load

| Factor of safety (mean strength / mean load) | Load known to | Reliability index | Chance of failure |
|---|---|---|---|
| 1.5 | 25 per cent | 1.71 (calculated) | 4.3 in 100 (calculated) |
| 1.5 | 50 per cent | 0.96 (calculated) | 17 in 100 (calculated) |
| 2 | 25 per cent | 3.12 (calculated) | 9 in 10,000 (calculated) |
| 2 | 50 per cent | 1.86 (calculated) | 3.2 in 100 (calculated) |
| 3 | 25 per cent | 5.12 (calculated) | 1.5 in 10 million (calculated) |
| 3 | 50 per cent | 3.43 (calculated) | 3 in 10,000 (calculated) |

Working for the first row: mean load 1, mean strength 1.5; spreads 0.15 (strength) and 0.25 (load); spread of the margin = sqrt(0.15^2 + 0.25^2) = 0.29; index = (1.5 - 1) / 0.29 = 1.71 (calculated). Chances read from the normal curve and rounded; they apply to the largest load in the structure's life. Illustrative: real loads are not exactly bell-shaped. Compare the Eurocode target index of 3.8 (Part 4).
 Source: Petroski, To Engineer Is Human (Macmillan, 1985), ch. 9, 'Safety in Numbers'; calculation by the author using the standard normal distribution  Common misreading

#### A factor of 2 is not simply twice as safe as a factor of 1

The table shows the chance of failure falling by orders of magnitude as the factor rises, so safety does not grow in proportion to the factor. It also works in reverse: a small loss of margin near the middle of the table raises the risk steeply. With the load known to 25 per cent, taking the factor from 2 down to 1.5 multiplies the chance of failure by nearly 50.

- Petroski, To Engineer Is Human (Macmillan, 1985) · ch. 9, 'Safety in Numbers'
- Hammack, The Things We Make (Sourcebooks, 2023) · ch. 3, 'Uncertainty'

### 3.2 A factor cannot cover a changed structure, a missing check or a margin eaten away

A factor is calculated for the structure as drawn and as maintained. The moment the real structure drifts from that picture, the factor on paper stops describing it. Hyatt is the extreme case: the detail as built had 27 per cent of the capacity the code required, so a margin of about 1.7 had turned into a margin below 1. Morbi is the slow version. Corrosion thinned the cable, a new deck changed the weight and stiffness, and a crowd arrived that nobody had verified the bridge for.

Margins are used up in steps, each modest. Estimate 6 follows a margin of 2 through four changes of the kind reported at Morbi: a heavier deck, a rusted section, a poor repair and a bigger crowd. None is dramatic, and by the end the margin is below 1. This is why inspection is not a formality but a way of recalculating the factor for the structure that exists today.

Margins also cost money and weight, which is the other half of the trade-off. The module Structures and materials compared the large factor on a lift rope with the small one on an airliner, where every kilogram is paid for in fuel. The point here is that a small factor is safe only if the checking around it is strong. An airliner lives with 1.5 because its loads are measured and its skin is inspected on a schedule; a renovated footbridge with no survey and no load test has no such support. Hammack describes engineers as working around uncertainty rather than waiting for it to go away, and a margin kept honest by inspection is one way they do it. The module Cost, trade-offs and resources takes up the economics.

Data Worked estimate 6: how a margin of 2 is used up

| Step | Change | Margin after the step |
|---|---|---|
| Start | strength = 2.0 x the load in service | 2.00 |
| Heavier deck | load up 20 per cent: 2.00 / 1.20 | 1.67 (calculated) |
| Corrosion | 25 per cent of the steel section lost: 1.67 x 0.75 | 1.25 (calculated) |
| Poor repair detail | capacity of a connection cut by a further 20 per cent: 1.25 x 0.80 | 1.00 (calculated) |
| Bigger crowd | load in service up by 50 per cent: 1.00 / 1.50 | 0.67 (calculated) |

Illustrative round numbers, not Morbi data; loads are treated as one total for simplicity. Check: 2 x (1/1.2) x 0.75 x 0.8 x (1/1.5) = 0.67. A margin below 1 means failure is expected at that load, though scatter means it is not certain.
 Source: Petroski, To Engineer Is Human (Macmillan, 1985), ch. 9, 'Safety in Numbers'; arithmetic by the author

- Petroski, To Engineer Is Human (Macmillan, 1985) · ch. 9, 'Safety in Numbers'
- Hammack, The Things We Make (Sourcebooks, 2023) · ch. 3, 'Uncertainty: The Art of the Unknown and How Engineers Use It to Create Solutions'

### 3.3 Fail-safe design assumes parts will crack and arranges that a crack is found before it matters

The older way to handle fatigue was safe life: test the part, divide the test life by a scatter factor and retire it at the result. It works when the scatter is well known, but Comet showed how one hidden crack could end an aircraft before its safe life was reached. Aircraft designers moved to a second idea, called fail-safe or damage-tolerant design. It assumes cracks will start and asks three questions. Can the structure carry its limit load with a crack of a stated size in it? Is there a second load path, or a crack stopper, so that a crack in one member does not take the whole? How often must the structure be inspected so that a crack is found while it is small?

Fail-safe thinking reaches well beyond aircraft. A 'Railway air brake' holds its shoes off the wheels with air pressure, so if a hose bursts or a coach breaks away, the pressure falls and the brakes apply. A 'Safety elevator' holds its car on rope tension; if the rope breaks, spring-loaded clamps grip the guide rails. In both, the failure of the working part triggers the safe state.

Cracks under repeated load grow faster as they lengthen, in a simple case doubling in length after a fixed number of cycles. The consequence is that most of the life of a cracked part is spent while the crack is tiny, and the last stage is quick. Inspection must catch the crack in the middle of its life, so the interval depends on how large a crack an inspector can reliably find. Estimate 7 sets an interval from this reasoning, and shows why better inspection methods, such as ultrasonic testing, can be worth more than extra metal. The module Structures and materials explains why cracks grow.

Data Worked estimate 7: how often to inspect for a skin crack

| Quantity | Working | Result |
|---|---|---|
| Growth rate assumed | crack doubles in length every 1,000 cycles | k = ln 2 / 1,000 = 0.00069 per cycle (calculated) |
| Length at which the skin would tear | assumed | 60 mm |
| Case A: inspectors reliably find 5 mm | cycles = ln(60 / 5) / k = 2.485 / 0.00069 | about 3,600 cycles (calculated) |
| Interval allowing three chances to find it | 3,600 / 3 | about 1,200 cycles (calculated) |
| At 4 flights a day | 1,200 / 4 | 300 days, about 10 months (calculated) |
| Case B: inspectors reliably find only 10 mm | ln(60 / 10) / k = 1.79 / 0.00069 | about 2,600 cycles (calculated) |
| Interval and calendar time | 2,600 / 3 = 860 cycles; 860 / 4 | about 215 days, 7 months (calculated) |

Illustrative growth law and numbers; real programmes use measured growth data and detection probabilities. Check: 5 mm doubled 3.6 times is 5 x 12 = 60 mm. The coarser the inspection, the shorter the interval must be.
 Source: Petroski, To Engineer Is Human (Macmillan, 1985), ch. 10, 'When Cracks Become Breakthroughs'; arithmetic by the author  Worked example

#### The weak link in a pressure cooker

A pressure cooker has several fail-safe features. The weighted valve lifts at its working pressure and lets steam escape. If the vent clogs, a second safety valve in the lid opens at a higher pressure, and in many designs the rubber gasket is the last line: it is made to bulge out through a slot in the lid and vent the steam, aimed away from the cook, long before the pot itself is stressed. Designers choose which part will give way first and make that part harmless. This is the same idea as a crack stopper in a fuselage: failure is allowed, but steered.

- Petroski, To Engineer Is Human (Macmillan, 1985) · ch. 10, 'When Cracks Become Breakthroughs'; ch. 11, 'Of Bus Frames and Knife Blades'
- 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'

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

### Check your understanding

Strength is known to 10 per cent. Which pair of cases gives about the same chance of failure?
 Both are in the region of a few in 10,000 (3 in 10,000 and 9 in 10,000). Knowing the load better buys as much as a 50 per cent larger factor.

A margin of 2 meets a deck load up 20 per cent, 25 per cent of the steel lost to corrosion, and a crowd 1.5 times the intended one. What margin is left?
 2 / 1.2 x 0.75 / 1.5 = 0.83. Each change is modest, yet together they take the margin below 1.

Why is a crack inspection interval set at a fraction of the time the crack takes to grow from detectable to critical?
 Detection is never certain and cracks grow faster as they lengthen, so the interval allows two or three looks before the critical size is reached.

[← Part 2: Four failures and what each one taught](https://www.materialinsights.org/learn/engineering/design-failure-safety/p2/)[Mark read · Part 4: The design cycle, testing and how knowledge is stored →](https://www.materialinsights.org/learn/engineering/design-failure-safety/p4/)

## Other sections

- [Module overview](https://www.materialinsights.org/learn/engineering/design-failure-safety/index.md)
- [Why failure teaches more than success](https://www.materialinsights.org/learn/engineering/design-failure-safety/p1/index.md)
- [Four failures and what each one taught](https://www.materialinsights.org/learn/engineering/design-failure-safety/p2/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)
