Tag Archives: stress

Thermal stress failure

Take a glass, preferably a cheap glass, and set it in a bowl of ice-cold water so that the water goes only half-way up the glass. Now pour boiling hot water into the glass. In a few seconds the glass will crack from thermal stress, the force caused by heat going from the inside of the glass outside to the bowl of cold water. This sort of failure is not mentioned in any of the engineering material books that I had in college, or had available for teaching engineering materials. To the extent that it is mentioned mentioned on the internet, e.g. here at wikipedia, the metric presented is not derived and (I think) wrong. Given this, I’d like to present a Buxbaum- derived metric for thermal stress-resistance and thermal stress failure. A key aspect: using a thinner glass does not help.

Before gong on to the general case of thermal stress failure, lets consider the glass, and try to compute the magnitude of the thermal stress. The glass is being torn apart and that suggests that quite a lot of stress is being generated by a ∆T of 100°C temeprarture gradient.

To calcule the thermal stress, consider the thermal expansivity of the material, α. Glass — normal cheap glass — has a thermal expansivity α = 8.5 x10-6 meters/meter °C (or 8.5 x10-6 foot/foot °C). For every degree Centigrade a meter of glass is heated, it will expand 8.5×10-6 meters, and for every degree it is cooled, it will shrink 8.5 x10-6 meters. If you consider the circumference of the glass to be L (measured in meters), then
∆L/L = α ∆T.

where ∆L is the change in length due to heating, and ∆L/L is sometimes called the “strain.”. Now, lets call the amount of stress caused by this expansion σ, sigma, measured in psi or GPa. It is proportional to the strain, ∆L/L, and to the elasticity constant, E (also called Young’s elastic constant).

σ = E ∆L/L.

For glass, Young’s elasticity constant, E = 75 GPa. Since strain was equal to α ∆T, we find that

σ =Eα ∆T 

Thus, for glass and a ∆T of 100 °C, σ =100°C x 75 GPa x 8.5 x10-6 /°C  = 0.064  GPa = 64MPa. This is about 640 atm, or 9500 psi.

As it happens, the ultimate tensile strength of ordinary glass is only about 40 MPa =  σu. This, the maximum force per area you can put on glass before it breaks, is less than the thermal stress. You can expect a break here, and wherever σu < Eα∆T. I thus create a characteristic temperature difference for thermal stress failure:

The Buxbaum failure temperature, ß = σu/Eα

If ∆T of more than ß is applied to any material, you can expect a thermal stress failure.

The Wikipedia article referenced above provides a ratio for thermal resistance. The usits are perhaps heat load per unit area and time. How you would use this ratio I don’t quite know, it includes k, the thermal conductivity and ν, the Poisson ratio. Including the thermal conductivity here only makes sense, to me, if you think you’ll have a defined thermal load, a defined amount of heat transfer per unit area and time. I don’t think this is a normal way to look at things.  As for including the Poisson ratio, this too seems misunderstanding. The assumption is that a high Poisson ratio decreases the effect of thermal stress. The thought behind this, as I understand it, is that heating one side of a curved (the inside for example) will decrease the thickness of that side, reducing the effective stress. This is a mistake, I think; heating never decreases the thickness of any part being heated, but only increases the thickness. The heated part will expand in all directions. Thus, I think my ratio is the correct one. Please find following a list of failure temperatures for various common materials. 

Stress strain properties of engineering materials including thermal expansion, ultimate stress, MPa, and Youngs elastic modulus, GPa.

You will notice that most materials are a lot more resistant to thermal stress than glass is and some are quite a lot less resistant. Based on the above, we can expect that ice will fracture at a temperature difference as small as 1°C. Similarly, cast iron will crack with relatively little effort, while steel is a lot more durable (I hope that so-called cast iron skillets are really steel skillets). Pyrex is a form of glass that is more resistant to thermal breakage; that’s mainly because for pyrex, α is a lot smaller than for ordinary, cheap glass. I find it interesting that diamond is the material most resistant to thermal failure, followed by invar, a low -expansion steel, and ordinary rubber.

Robert E. Buxbaum, July 3, 2019. I should note that, for several of these materials, those with very high thermal conductivities, you’d want to use a very thick sample of materials to produce a temperature difference of 100*C.

Statistics of death and taxes — death on tax day

Strange as it seems, Americans tend to die in road accidents on tax-day. This deadly day is April 15 most years, but on some years April 15th falls out on a weekend and the fatal tax day shifts to April 16 or 17. Whatever weekday it is, about 8% more people die on the road on tax day than on the same weekday a week earlier or a week later; data courtesy of the US highway safety bureau and two statisticians, Redelmeier and Yarnell, 2014.

Forest plot of individuals in fatal road crashes over 30 years. X-axis shows relative increase in risk on tax days compared to control days expressed as odds ratio. Y-axis denotes subgroup (results for full cohort in final row). Column data are counts of individuals in crashes. Analytic results expressed with 95% confidence intervals setting control days as referent. Results show increased risk on tax day for full cohort, similar increase for 25 of 27 subgroups, and all confidence intervals overlapping main analysis. Recall that odds ratios are reliable estimates of relative risk when event rates are low from an individual driver’s perspective.

Forest plot of individuals in fatal road crashes for the 30 years to 2008  on US highways (Redelmeier and Yarnell, 2014). X-axis shows relative increase in risk on tax days compared to control days expressed as odds ratio. Y-axis denotes subgroup (results for full cohort in final row). Column data are counts of individuals in crashes (there are twice as many control days as tax days). Analytic results are 95% confidence intervals based on control days as referent. Dividing the experimental subjects into groups is a key trick of experimental design.

To confirm that the relation isn’t a fluke, the result of well-timed ice storms or football games, the traffic death data was down into subgroups by time, age, region etc– see figure. Each groups showed more deaths than on the average of the day a week before and after.

The cause appears unrelated to paying the tax bill, as such. The increase is near equal for men and women; with alcohol and without, and for those over 18 and under (presumably those under 18 don’t pay taxes). The death increase isn’t concentrated at midnight either, as might be expected if the cause were people rushing to the post office. The consistency through all groups suggests this is not a quirk of non-normal data, nor a fluke but a direct result of  tax-day itself.Redelmeier and Yarnell suggest that stress — the stress of thinking of taxes — is the cause.

Though stress seems a plausible explanation, I’d like to see if other stress-related deaths are more common on tax day — heart attack or stroke. I have not done this, I’m sorry to say, and neither have they. General US death data is not tabulated day by day. I’ve done a quick study of Canadian tax-day deaths though (unpublished) and I’ve found that, for Canadians, Canadian tax day is even more deadly than US tax day is for Americans. Perhaps heart attack and stroke data is available day by day in Canada (?).

Robert Buxbaum, December 12, 2014. I write about all sorts of stuff. Here’s my suggested, low stress income tax structure, and a way to reduce/ eliminate income taxes: tariffs– they worked till the Civil war. Here’s my thought on why old people have more fatal car accidents per mile driven.