Module 2 · Lesson 2.1
Design under uncertainty
Why safety factors exist at all, and what each one is actually paying for.
Why this matters
Ask an engineer why the load factor is 1.35 and you will usually get the answer 'because the code says so'. That is not an engineering answer, and it is dangerous, because an engineer who does not know what a factor is for cannot tell when it is insufficient. The factors are not arbitrary and they are not a single lump of conservatism. Each one is buying something specific, and knowing which is which is what lets you judge an unusual situation the code did not anticipate.
By the end of this lesson you should be able to
- State the three irreducible sources of uncertainty in a design
- Define a characteristic value as a fractile
- Explain why loads use an upper fractile and strengths a lower one
- Say what γF and γM each cover
- Explain why the two are kept separate rather than combined
What you should already know
- How reinforced concrete works (Module 1)
- The idea of a design calculation as a comparison of effect against resistance
Every structural design is a bet. You are asserting that a structure which does not yet exist, made of materials not yet mixed, loaded by people not yet born, will not fail during a working life of fifty years or more. Nothing about that is certain, and pretending otherwise is the first mistake.
The uncertainty has three irreducible sources, and they are genuinely different in kind.
1. The loads are not known. You do not know how many people will stand in the room, what they will bring, how the owner will change the use in twenty years, or how hard the wind will blow on the worst day of the structure's life. You can characterise these statistically, but you cannot know them.
2. The materials are not what the drawing says. Concrete is mixed on a lorry, poured in weather, compacted by someone at the end of a long shift, and cured or not cured. A specified C30/37 will produce cubes ranging over several newtons per square millimetre. Reinforcement is more consistent, but the bars may be placed 15 mm from where you drew them, and in a 400 mm deep slab that is a 4% error in lever arm.
3. The model is not the structure. You analysed a line diagram. The real thing has finite joints, partial fixity, a floor slab that stiffens the beams, cladding that was never meant to be structural but is, and a construction sequence that loaded members in an order your analysis never considered.
Design does not eliminate these uncertainties. It allocates them — deciding how much of each the calculation will carry, and how much is accepted as residual risk.
Characteristic values: a fractile, not an average
If loads and strengths are variable, a design must pick a number from a distribution. The choice made throughout the Eurocodes is a fractile — a value with a defined probability of being exceeded.
For actions, the characteristic value is an upper fractile: typically the value with a 5% chance of being exceeded in a reference period, or for climatic actions, a value with a 50-year return period. You want a load that is unlikely to be beaten.
For material strengths, the characteristic value is a lower fractile: the 5% fractile, so 95% of test results exceed it. You want a strength that is very likely to be achieved.
This is why the concrete class C30/37 does not mean the concrete is 30 N/mm² strong. It means the mix is designed so that only 5% of cylinder tests fall below 30. The mean strength of that concrete is around 38 N/mm² — a difference of roughly eight newtons that the designer never uses and never sees.
Two factors, doing two different jobs
The partial factors are deliberately split, and the split is not cosmetic.
γF, on actions. Covers the possibility that the load exceeds its characteristic value, and the possibility that your analysis under-estimates the effect that load produces. It is applied to the load, but part of what it pays for is your model being wrong.
γM, on material strengths. Covers the possibility that the material is weaker than its characteristic value in the actual structure — which is not the same as in a test cube. It absorbs the difference between a laboratory specimen and concrete cast in a real beam, the effect of dimensional deviations, and the crudeness of the resistance model.
Keeping them separate matters because they are not interchangeable. A structure that is unusually sensitive to load uncertainty is a different problem from one whose material behaviour is unusually uncertain.
What it calculates: The single statement that every design check is an instance of.
- Ed
- Design value of the effect of actions — the demand
- Rd
- Design value of the resistance — the supply
- Fk
- Characteristic action
- fk
- Characteristic material strength
- Partial factor on actions (increases the demand)
- Partial factor on materials (decreases the supply)
This assumes
- The effect E can be computed from the actions — which requires an analysis model
- The resistance R can be computed from the strengths — which requires a resistance model
- Both models are adequate, which the factors partly, but only partly, cover
In plain terms: The demand is pushed up and the supply is pushed down, and the design is adequate if they still do not meet. Notice that the factors act at different points: γF before the analysis, γM inside the resistance. That ordering matters whenever the analysis is non-linear.
The two limit states
Ultimate limit state (ULS) — states associated with collapse or with other forms of structural failure: loss of equilibrium, rupture, instability, fatigue. These concern the safety of people. They are checked with factored loads and reduced strengths, because the consequences of being wrong are severe and irreversible.
Serviceability limit state (SLS) — states beyond which the specified service requirements are no longer met: deflection that alarms occupants or cracks a partition, crack widths that look wrong or admit water, vibration that makes a floor unpleasant. These concern function and appearance.
The SLS uses unfactored loads, and this is the point most often missed. Applying a 1.5 factor to a deflection check would be meaningless: you are not asking 'could it deflect this much in a catastrophe', you are asking 'how much will it actually deflect'. A safety factor on a question about reality is a category error.
Predict first
Two concrete mixes are both specified as C30/37. Mix A is produced by a plant with tight quality control; mix B by one with poor control. Which mix has the higher MEAN cube strength?
Worked example
What a characteristic strength means for a real mix
Given
- Concrete specified as C30/37, so fck = 30 N/mm²
- Cube test results from the plant have a standard deviation of 5 N/mm²
- The characteristic value is defined as the 5% lower fractile
Find
The mean strength the plant must target, and the design strength you may use.
Assumptions
- Strengths are normally distributed — a reasonable approximation for a controlled plant
- The 5% fractile of a normal distribution lies 1.64 standard deviations below the mean
Practice
A concrete is specified as C40/50. Using the code relationship, what is its mean compressive strength fcm, in N/mm²?
Practice
For C40/50 concrete with γC = 1.5 and αcc = 0.85, what is the design compressive strength fcd, in N/mm²?
Check yourself
Which of these is γM NOT intended to cover?
Summary
- Uncertainty has three sources: loads, materials, and the model itself
- A characteristic value is a fractile — upper for loads, lower for strengths
- C30/37 concrete is on average about 38 N/mm² strong; the design uses 17
- γF covers load and analysis uncertainty; γM covers material and resistance uncertainty
- Factor the loads BEFORE a non-linear analysis, never after
- ULS asks whether it survives; SLS asks whether it is fit to use, with unfactored loads
This is educational material. It uses simplified examples to teach principles, and must not be relied on for real design or safety-critical decisions. Module overview and checkpoint