Strong enough?
In statics you found the forces on a part: the 50 kN pulling on a tie rod, the reaction at a hinge, the tension in a cable. Equilibrium told you how much force each part carries. It could not tell you whether the part survives. A 50 kN pull is nothing to a ship's anchor chain and fatal to a bicycle spoke.
What decides survival is how concentrated the force is inside the material, and how far the material stretches in response. Those two ideas, stress and strain, are the language of the rest of this level. With them you can answer the questions a mechanical engineer is asked every day: how thick must this bolt be? How much will this shaft twist? How far will this floor sag? When will this bracket crack?
This lesson builds the vocabulary. You will compute stresses and strains, link them with Hooke's law, read the stress–strain curve of a real metal, see why heating a constrained rail creates enormous forces, and discover that stress at a point is not one number but a small matrix, whose eigenvalues tell you where the material is most at risk.
Cutting a bar in your imagination
Take a steel rod pulled by a force at each end. Equilibrium of the whole rod is trivial: to the left, to the right. Now imagine slicing the rod anywhere along its length and looking at the left piece alone, as a free body. It is still in equilibrium, so the material on the other side of the cut must be pulling it with a total force . That pull is spread over the whole cut face, every square millimetre of steel holding on to its neighbour.
The intensity of that internal force, force per unit area, is the stress. Double the force and the stress doubles; double the area and it halves. A thick rope and a thin thread made of the same fibre fail at the same stress, which is why stress, not force, is what we compare with the material's strength.
The rod also gets longer. A 2 m rod stretches twice as much as a 1 m rod under the same stress, because every centimetre of it stretches by the same small amount. So the natural measure of deformation is stretch per unit length, the strain, which is dimensionless. And a stretched rod gets slightly thinner, like a rubber band, which is Poisson's effect.
Analogy, not a formal construction
Think of a crowd holding hands in long parallel lines, everyone leaning back to resist a pull. The total pull is the force. How hard each person must grip is the stress: add more lines of people (more area) and each grip gets easier. How much each arm stretches is the strain: a longer line stretches more in total, but each arm stretches the same.
Stress, strain and Hooke's law
Definition— normal stress
When a force acts perpendicular to a cut of area and is spread uniformly, the normal stress is . Tension is positive, compression negative. The SI unit is the pascal (1 Pa = 1 N/m²); engineers use megapascals, and , which makes hand calculations in newtons and millimetres painless.
Definition— shear stress
When a force acts along (parallel to) a cut of area , the average shear stress is . A pin in a clevis, a bolt in a lap joint and a punch cutting sheet metal are loaded mainly in shear.
Definition— normal and shear strain
A line of original length that stretches by has normal strain . A right angle in the material that closes by a small angle (in radians) has shear strain . Strains in metal parts are tiny, typically below , so they are often quoted in microstrain ().
Definition— Hooke's law, Young's modulus and Poisson's ratio
For most engineering materials at small strains, stress is proportional to strain. In uniaxial tension
where is Young's modulus (the stiffness of the material) and is Poisson's ratio (how much it narrows as it stretches). In shear, with shear modulus . For an isotropic material (the same in every direction) only two of these constants are independent:
Definition— thermal strain
A free piece of material warmed by expands with strain , where is the coefficient of thermal expansion. Thermal strain causes no stress by itself; stress appears only if something prevents it.
Some representative room-temperature values, rounded, to build intuition (always take design values from the supplier's data or a materials standard):
| Material | (GPa) | (/K) | |
|---|---|---|---|
| Structural steel | 200–210 | 0.30 | 12 |
| Aluminium alloys | 69–72 | 0.33 | 23 |
| Titanium alloys | 105–120 | 0.32–0.34 | 9 |
| Concrete | 25–40 | 0.15–0.20 | 10–12 |
| Glass (soda-lime) | 70 | 0.22 | 9 |
| Douglas fir, along the grain | 11–14 | — | 3–5 |
| Rubber | 0.001–0.01 | close to 0.5 | 100–200 |
Wood has no single Poisson's ratio because it is strongly anisotropic: stiff along the grain, soft across it. Rubber's ratio near 0.5 means it hardly changes volume as it deforms. Why atoms produce these numbers, and why some materials are stiffer than others, is the subject of elasticity, plasticity and strengthening in Lumenary Materials; here we take them as measured facts and design with them.
Worked examples: a tie rod and a hot rail
Example— A steel tie rod
A round steel tie rod of diameter 20 mm and length 2.0 m carries 50 kN. Take GPa and .
The area is , so MPa. The strain is , about 800 microstrain. Over 2.0 m the rod stretches mm, and its diameter shrinks by mm, about a twentieth of a human hair.
Is 159 MPa safe? That depends on the steel. A common structural steel yields at around 235–355 MPa, so the rod is working at roughly half to two-thirds of its yield strength: reasonable for a static load. The lesson on failure and fatigue makes "safe" precise.
Example— A rail that cannot expand
The same rod is now fixed between two rigid walls at a comfortable temperature and then warmed by 40 K on a summer afternoon. Free, it would expand by . The walls keep its total strain at zero, so the mechanical strain must be , and
That is a compressive force of about 30 kN on each wall, from a change in temperature alone, and notice that the length of the rod does not appear. Continuously welded railway track lives with exactly this: the rails are anchored firmly to the sleepers and laid at a chosen "stress-free" temperature so that summer compression and winter tension both stay within limits, and engineers watch for the sideways buckling that compressed rails can suffer on very hot days. Bridges avoid the problem the other way, with expansion joints that let the deck move.
The stress–strain curve
Hooke's law holds only up to a point. To find out what happens beyond it, laboratories pull a carefully machined specimen in a tensile-testing machine, measuring force and extension until it breaks, and plot engineering stress against engineering strain . The method is standardised (ISO 6892-1 and ASTM E8/E8M) so that results from Pune, Pittsburgh and Pohang can be compared.
Reading the curve from left to right:
- Elastic region. Straight, with slope . Unload anywhere here and the specimen returns to its original length.
- Yield. The curve bends: the material starts to deform plastically, permanently, because dislocations in the crystals begin to glide. Some steels show a sharp upper yield point; most metals bend over gradually, so the yield strength is defined by convention as the stress at which the permanent strain is 0.2%. Draw a line parallel to the elastic part starting at ; where it meets the curve is , the 0.2% proof stress.
- Strain hardening. Plastic deformation multiplies and tangles dislocations, so the material resists further deformation and the stress keeps rising.
- Ultimate tensile strength , the maximum engineering stress. Beyond it the specimen necks: deformation concentrates in one place, the local cross-section shrinks faster than the material hardens, and the force falls until fracture.
- Ductility, the strain at fracture (elongation after fracture, as a percentage of the gauge length) or the reduction of area at the neck. A ductile part bends and stretches visibly before it breaks, which is a warning; a brittle part does not.
The fall after the ultimate strength is partly an illusion of bookkeeping: engineering stress divides by the original area , but the neck's real area is much smaller. The true stress keeps rising until fracture. Designers of ordinary parts use engineering values (a part that has necked has already failed); people who simulate forming, crashes and machining need true-stress curves.
Computing it: a tensile test in NumPy
Real test data comes as columns of strain and stress. Here we generate a realistic curve for a mild steel from a simple hardening law, true stress with MPa, and hardening exponent , convert it to engineering values as a testing machine would report them, and then do what a test laboratory does: find the 0.2% offset yield and the tensile strength.
import numpy as np
# A model tensile test of a mild steel: Swift hardening, true stress vs true plastic strain
E = 200e3 # MPa
K, eps0, n = 600.0, 0.005, 0.20
ep = np.linspace(0.0, 0.30, 3001) # true plastic strain
sig_true = K * (eps0 + ep) ** n # true stress, MPa
eps_true = sig_true / E + ep # true total strain
e_pl = np.exp(eps_true) - 1.0 # engineering strain
s_pl = sig_true / (1.0 + e_pl) # engineering stress (volume conserved)
# prepend the elastic line from the origin to first yield
e_el = np.linspace(0.0, e_pl[0], 50, endpoint=False)
e = np.concatenate([e_el, e_pl])
s = np.concatenate([E * e_el, s_pl])
# 0.2 % offset yield: where the curve meets the line s = E (e - 0.002)
gap = s - E * (e - 0.002)
i = np.argmax(gap < 0) # first point below the offset line
t = gap[i - 1] / (gap[i - 1] - gap[i]) # interpolate between the two points
Rp02 = s[i - 1] + t * (s[i] - s[i - 1])
k = np.argmax(s_pl) # ultimate tensile strength
print(f"end of the linear part = {s_pl[0]:4.0f} MPa")
print(f"0.2% offset yield Rp0.2 = {Rp02:4.0f} MPa")
print(f"tensile strength Rm = {s_pl[k]:4.0f} MPa at engineering strain {e_pl[k]:.3f}")
print(f"maximum at true plastic strain {ep[k]:.3f}; Considere predicts n - eps0 = {n - eps0:.3f}")
print(f"true stress there = {sig_true[k]:4.0f} MPa")
print("\n strain eng. stress (MPa)")
for target in [0.0005, 0.001, 0.002, 0.005, 0.02, 0.05, 0.10, 0.15, 0.20, 0.25]:
j = np.argmin(abs(e - target))
print(f" {e[j]:6.4f} {s[j]:5.0f} " + "#" * int(s[j] / 10))The program reports a yield strength MPa and a tensile strength MPa at about 22% strain, figures typical of a low-strength structural steel. The maximum falls exactly where Considère's criterion predicts, at true plastic strain , and the true stress there, 435 MPa, is about a fifth higher than the engineering value. The little bar chart shows the two personalities of the curve: in the first 0.1% of strain the stress shoots up to 200 MPa, and over the next 20% it creeps up by only another 150 MPa.
Stress in every direction: transformation and Mohr's circle
So far the cut was perpendicular to the rod. Cut it at an angle instead, and the force on the cut face has components both normal and parallel to it: a tension rod carries shear stress on inclined planes. This is not academic. Ductile metals yield by slip on planes of high shear stress, which in a tensile test lie at 45° to the axis, and bands of slip at roughly 45° to the axis can sometimes be seen on the polished surface of a yielding steel specimen.
In general, at any point of a loaded part, the state of stress needs more than one number. In two dimensions (plane stress, the usual case at a free surface or in a thin plate) it is three numbers: normal stresses , on faces perpendicular to and , and the shear stress on those faces. Arrange them as a symmetric matrix, the stress tensor,
On a face whose outward unit normal is , the force per unit area (the traction) is . Its normal component is and its shear component along is . Multiplying out and using double-angle identities gives the transformation equations
Theorem— Principal stresses
At every point there are two perpendicular directions (three in 3-D) on which the shear stress is zero. The normal stresses on them, the principal stresses , are the largest and smallest normal stresses on any plane through the point, and they are the eigenvalues of :
The maximum in-plane shear stress is , on planes at 45° to the principal directions.
Proof
The transformation equations say that the point is the centre on the -axis plus a vector of fixed length that turns through angle as the face turns through : its components are a rotation of by in the plane. So as varies, the stress on the face traces a circle of radius about . The normal stress is extreme, , exactly where the circle crosses the -axis, where ; the shear stress is extreme, , at the top and bottom of the circle, a quarter-turn () away. Because is symmetric, the spectral theorem gives the same conclusion: an orthonormal basis of eigenvectors in which is diagonal, with the eigenvalues as the extremes of over unit vectors.
∎
The circle in the proof is Mohr's circle, published by the German engineer Otto Mohr in 1882 and drawn by hand by generations of engineers to read off principal stresses with a compass and ruler. Each face of the element is a point on the circle, and turning the face by moves the point by .
Computing it: rotating the element and finding principal stresses
The transformation is just , so a few lines of NumPy can turn the element through every angle, and numpy.linalg.eigh, the eigenvalue solver for symmetric matrices, finds the principal stresses directly. The last part does the same for a full three-dimensional stress state, where hand formulas become painful and an eigenvalue solver is simply the right tool.
import numpy as np
# Plane stress at a point (MPa): sigma_x, sigma_y, tau_xy
sx, sy, txy = 80.0, -40.0, 30.0
S = np.array([[sx, txy],
[txy, sy]])
print(" theta sigma_x' tau_x'y'")
for deg in range(0, 181, 15):
th = np.radians(deg)
n = np.array([np.cos(th), np.sin(th)]) # normal of the rotated face
m = np.array([-np.sin(th), np.cos(th)]) # direction along that face
print(f" {deg:4d} {n @ S @ n:8.1f} {m @ S @ n:8.1f}")
# Mohr's circle
c = (sx + sy) / 2
R = np.hypot((sx - sy) / 2, txy)
print(f"\nMohr's circle: centre {c:.1f} MPa, radius {R:.1f} MPa")
print(f"principal stresses {c + R:.1f} and {c - R:.1f} MPa; max in-plane shear {R:.1f} MPa")
print(f"principal angle {np.degrees(0.5 * np.arctan2(2 * txy, sx - sy)):.1f} degrees")
# The same answer from an eigenvalue solver
vals, vecs = np.linalg.eigh(S) # ascending order for symmetric matrices
print("eigenvalues:", np.round(vals[::-1], 1))
v = vecs[:, 1]
print(f"direction of sigma_1: {np.degrees(np.arctan2(v[1], v[0])) % 180:.1f} degrees")
# A full 3-D stress state at a point inside a part (MPa)
S3 = np.array([[ 50.0, 20.0, 0.0],
[ 20.0, -10.0, 15.0],
[ 0.0, 15.0, 30.0]])
p = np.linalg.eigvalsh(S3)[::-1]
print("\n3-D principal stresses:", np.round(p, 1), "MPa")
print(f"absolute maximum shear (s1 - s3)/2 = {(p[0] - p[2]) / 2:.1f} MPa")
print(f"invariant check: trace {np.trace(S3):.1f} = sum of principals {p.sum():.1f}")The table shows the normal stress climbing from 80 MPa to a peak of 87 MPa near 15° and the shear stress passing through zero there; near 58° the shear reaches its extreme of MPa. Mohr's formulas and the eigenvalue solver agree: principal stresses 87.1 and MPa, with acting on the plane turned 13.3° from . In three dimensions the principal stresses are 56.8, 33.3 and MPa, and their sum equals the trace of the matrix, 70 MPa, as it must: the trace is an invariant, the same in every set of axes.
How engineers measure and use stress
Nobody can see stress. Engineers infer it from strain, which they can measure, and from models, which they can compute.
- Tensile testing gives the material's , , and elongation. Test laboratories follow ISO 6892-1 or ASTM E8/E8M, which fix specimen shapes, strain rates and how to compute each property, and they report results on test certificates that travel with each batch of steel or aluminium. Design codes and handbooks (for example the ASM Handbooks, or the material tables in national structural codes) give minimum guaranteed values, not averages.
- Strain gauges are thin foil grids glued to a surface whose electrical resistance changes in proportion to strain. A rosette of three gauges at 0°, 45° and 90° measures the three in-plane strain components at a point; Hooke's law and the transformation equations then give the principal stresses. Strain gauges are used everywhere from bridge monitoring to the load cells in kitchen scales.
- Digital image correlation tracks a speckle pattern painted on a part with cameras and computes a whole field of surface strains, which is now routine in research and aerospace testing.
- Finite element analysis, the subject of the last lesson of this level, computes the full stress tensor at millions of points in a complicated part. The software then reports principal stresses and a combined "equivalent stress", which you will meet in failure and fatigue.
When you see a stress contour plot, every colour on it came from the ideas in this lesson: a tensor at each point, its eigenvalues, and a comparison with a strength measured in a tensile test.
A short history
Robert Hooke stated his law in 1678 as a Latin anagram resolved to ut tensio, sic vis, "as the extension, so the force", from experiments with springs, wires and watch springs. Thomas Young described the modulus that bears his name in 1807, though not yet as stress over strain. Claude-Louis Navier and Augustin-Louis Cauchy, in Paris in the 1820s, turned elasticity into a field theory: Cauchy introduced stress as a quantity acting on every plane through a point and showed that it transforms as what we now call a tensor. Siméon Denis Poisson predicted the lateral contraction ratio from a molecular model (his model gave for every material, which experiments later showed was too simple). Otto Mohr's graphical circle followed in 1882, and William Rankine, Barré de Saint-Venant and many others turned the theory into engineering practice for the railways, bridges and boilers of the nineteenth century.
Common misconceptions
Misconception
"Stress is force." Stress is force per unit area, and it is what the material feels. A 10 kN load is harmless on a 100 mm diameter column and destructive on a 3 mm wire. Compare stresses with strengths, never forces with strengths.
Misconception
"Stiff means strong." Stiffness () says how much a part deflects; strength (yield or tensile strength) says how much stress it survives. Every steel, from soft wire to a hardened bearing race, has nearly the same GPa, yet their yield strengths differ by a factor of ten. Glass is about as stiff as aluminium and far less tough.
Misconception
"The engineering stress–strain curve falls after the peak, so the metal gets weaker as it is stretched." The true stress keeps rising; the engineering stress falls because the force falls as the neck thins, while we keep dividing by the original area.
Misconception
"Poisson's ratio can be anything." For a stable isotropic material, . Values approaching 0.5 mean the material is nearly incompressible (rubber); cork, close to 0, hardly bulges when squeezed, which is why it seals a bottle neck so well. Engineered auxetic structures can even have a negative ratio and widen when stretched.
Misconception
"If there is no shear stress on my x and y faces, there is no shear stress anywhere." A rod in pure tension carries a shear stress of half the axial stress on 45° planes, and that shear is what makes ductile metals yield.
Exercises
1. Stress and strain
Warm-upHumanA flat aluminium hanger bar with a cross-section of 6 mm × 25 mm carries a tensile load of 12 kN. What is the normal stress in the bar, in MPa?
2. Stress and strain
Warm-upHumanTwo rods have identical dimensions and carry the same load, well within their elastic range. Rod A is a high-strength steel (yield strength 900 MPa); rod B is a soft, low-carbon steel (yield strength 250 MPa). Which statement is right?
3. Stress and strain
CoreHumanAn aluminium bar ( GPa, K⁻¹) is fitted snugly between two rigid supports and then warmed by 30 K. Assuming it stays straight and elastic, what is the magnitude of the compressive stress in the bar, to the nearest MPa?
4. Stress and strain
CoreHumanAt a point on the surface of a bracket, MPa, MPa and MPa. What is the largest principal stress , in MPa?
5. Stress and strain
CoreHumanWhich of these statements about Poisson's ratio of an isotropic, linear-elastic material are true? Select all that apply.
6. Stress and strain
CoreComputationalWrite principal_stresses(S), which takes a symmetric 3 × 3 stress tensor (a NumPy array or nested list, in MPa) and returns a pair:
- a tuple of the three principal stresses
(s1, s2, s3)in descending order, and - the absolute maximum shear stress
(s1 - s3) / 2.
Use an eigenvalue solver. The tests include uniaxial tension, hydrostatic pressure, a case where the largest shear acts out of the plane, and a rotated tensor (the principal stresses must not depend on the axes).
7. Stress and strain
StretchComputationalA test laboratory gives you a tensile curve as two arrays, strain (dimensionless, increasing) and stress (MPa), and Young's modulus E (MPa). Write offset_yield(strain, stress, E, offset=0.002) that returns the offset yield strength: the stress where the curve first crosses the line stress = E * (strain - offset), found by linear interpolation between the two data points either side of the crossing. If the curve never crosses the line, return None.
Further reading and resources
- ISO, the International Organization for StandardizationPublishes ISO 6892-1, the international method for tensile testing of metals at room temperature.
- ASTM InternationalPublishes ASTM E8/E8M (tension testing of metallic materials) and many other test methods.
- ASM InternationalThe materials society behind the ASM Handbooks, a standard source of measured material properties.
- MIT OpenCourseWare 2.001, Mechanics and Materials I · Massachusetts Institute of TechnologyLecture notes and problem sets for a first course in the mechanics of materials.
- A History of Strength of Materials · Stephen P. Timoshenko · 1953How Hooke, Young, Navier, Cauchy, Poisson and Mohr built the subject; reprinted by Dover.