Weibull Strength Size Scaling

A bigger part made of the same material is weaker. Not weaker in total load — weaker in stress, in megapascals, measured the same way, before any design decision has been made. It is the most counterintuitive result in brittle materials and it is a direct consequence of the weakest-link argument: ten times the volume holds ten times as many chances of containing the one flaw that decides the strength, and the worst of ten draws is worse than the worst of one.

Put a number on it, because the number is what makes it real. Take a typical engineering ceramic at $m = 10$, qualified on a bend bar, and build a component of ten times the stressed volume. The strength ratio is $10^{-1/10} = 0.794$. The component is 20.6% weaker than the bar that qualified it — 400 MPa on the bar becomes 318 MPa in the part — and nothing was done wrong, no defect was introduced, no process drifted. A safety factor of 1.25 applied to the test result would have been entirely consumed by the change in size alone, before the first real uncertainty was accounted for. Scale up by a hundred and the loss is 37%.

The modulus sits in the exponent, and that is where the design lesson lives. Repeat the tenfold scale-up at $m = 5$ and the loss is 36.9% instead of 20.6%. At $m = 20$ it is 10.9%, and at $m = 40$ just 5.6%. A material with poor consistency is punished twice by size — once by having a wide distribution to begin with, and again by losing more of it on scale-up. For anything that has to be built full size, a high Weibull modulus is worth more than a high mean strength, and that is the real argument for spending money on process control rather than on a headline number.

This is also, finally, why ceramic design does not use safety factors. A safety factor is a statement about a single number, and there is no single number here to be safe against. The honest design statement is a probability of survival, chosen against the consequence of failure — one in a hundred may be perfectly reasonable for a kiln shelf and unacceptable for a turbine rotor — and then the stress and the size that deliver it. That reframing is the practical content of the whole category.

Two cautions that decide whether the answer is right.

Both volumes must be effective volumes. Not the volume of the object — the volume genuinely carrying tension, weighted by the $m$-th power of the stress. For a bend bar those differ by more than two orders of magnitude, and using the bar's physical volume against a component's stressed volume produces an error in the unsafe direction. For a real component the effective volume comes out of a finite-element stress field integrated against the Weibull weighting, which is what commercial ceramic reliability codes do.

And the scaling must be on the right quantity. Volume flaws scale with volume, surface flaws with area. If grinding damage sets the strength, this equation should be run on the ratio of stressed areas instead, and for geometrically similar parts that is a materially different answer. Fractography settles it. Nothing else does.

Read backwards — solving for $m$ from two sizes — the same relation becomes a way to test the weakest-link assumption rather than assume it. An $m$ from the size effect and an $m$ from the slope of a single-size Weibull plot ought to agree. When they do not, the usual reason is that the flaw population differs between the two sizes: bigger specimens are machined differently, fired differently, or cooled more slowly. That disagreement is a genuine result about the process, not a defect in the arithmetic.