The Bauschinger Effect: Why Yielding in One Direction Softens the Reverse

What actually happens on the stress–strain curve

Take a virgin ductile metal and load it monotonically in tension. It yields at the initial yield strength σ_y0 — say 300 MPa for a low-carbon steel — and work-hardens as you strain it further, reaching a flow stress σ_f of 400 MPa at 2% plastic strain. Now unload elastically (the line comes down with slope E, Young's modulus, ≈ 200 GPa for steel) and reload in compression.

Isotropic hardening — the simplest textbook model — says the material now yields in both directions at ±400 MPa: the yield surface has grown symmetrically. The Bauschinger effect says otherwise. The metal begins to yield in reverse at a stress σ_r of magnitude well below 400 MPa, typically 150–250 MPa. The curve rounds over early and rejoins the hardening trajectory with a smooth, elongated elbow rather than a sharp knee.

The bookkeeping is clean. Define the back stress X as the center of the elastic range on the stress axis. After forward flow, X ≈ ½(σ_f − σ_r) with σ_r the magnitude of the reverse yield; the practical relation is:

• σ_r = σ_f − 2X — reverse yield equals forward flow stress minus twice the back stress.
• The Bauschinger factor β = |σ_r| / σ_f, so β = 1 means no effect (pure isotropic) and β → 0 means the reverse yields the instant load reverses.
• Typical measured β: 0.7–0.9 for annealed pure metals, 0.4–0.6 for dual-phase and cold-worked alloys, brass, and precipitation-hardened aluminum.

Equivalently, the total drop in reverse yield equals 2X, split as the missing hardening (didn't reach 400) plus the extra softening (yields before 300). For σ_f = 400 and σ_r = 180 MPa, X = (400 − 180)/2 = 110 MPa.

The mechanism: directional back stress in the dislocation field

Plastic flow in metals is dislocation glide on slip planes. When you strain forward, mobile dislocations sweep across grains and get stopped at obstacles: grain boundaries, hard second-phase particles, forest dislocations, and each other. They stack into pile-ups and leave Orowan loops wrapped around precipitates. Every one of these arrangements stores a long-range internal (back) stress that points against the direction you were pushing — it is the elastic reaction of the stacked, tangled field.

Two directional sources dominate:

• Dislocation pile-ups (Orowan / Nabarro). A pile-up of n dislocations at a boundary exerts a back stress on the source that scales as the applied stress times n. On reversal, that stored stress now helps the dislocations run backward, so they move at a lower applied stress — reverse yielding starts early.
• Composite / Orowan-loop back stress. In alloys with hard particles (Al–Cu θ′, carbides, martensite islands in dual-phase steel), the soft matrix plastically flows while the hard phase stays elastic. The hard phase builds an elastic misfit stress — a kinematic back stress X that literally shifts the yield surface. This is why two-phase and precipitation-hardened alloys show the strongest Bauschinger effect.

On reversal, some of these dislocations also annihilate or unpin easily (they were pushed against obstacles, not through them), and geometrically-necessary structure relaxes. The net result is a lower reverse yield and a rounded, gradual re-yielding transition — the elastic-to-plastic knee is smeared over tens of MPa. The effect is essentially athermal at room temperature but fades as recovery and recrystallization erase the stored structure, which is why a stress-relief anneal (e.g., 550–650 °C for steel) largely resets it.

Modeling it: kinematic and combined (Chaboche) hardening

Finite-element plasticity needs a rule for how the yield surface evolves. Von Mises yield with pure isotropic hardening writes the surface as f = √(3/2 · s:s) − σ_y(ε̄ᵖ) = 0, where s is deviatoric stress and the radius σ_y grows with accumulated plastic strain ε̄ᵖ. That surface only expands, so it cannot capture reverse softening.

Kinematic hardening fixes this by translating the surface: f = √(3/2 · (s − α):(s − α)) − σ_y0 = 0, where α is the back-stress tensor (deviatoric X). The surface keeps its size 2σ_y0 but its center moves in stress space, so forward yielding drags the compression side down with it — exactly the Bauschinger effect.

• Prager / Ziegler linear rule: dα = C·dεᵖ. Simple, one constant C (units of stress), but predicts constant β and a ratcheting/plateau that overshoots real data.
• Armstrong–Frederick nonlinear rule: dα = (2/3)C·dεᵖ − γ·α·dp, where p is accumulated plastic strain and γ sets saturation. The back stress saturates at X_sat = C/γ, giving a realistic rounded reversal.
• Chaboche combined model: superpose several Armstrong–Frederick terms, α = Σαᵢ, plus a small isotropic term. Three back-stress components with (C₁, γ₁)…(C₃, γ₃) — e.g., C₁≈60 GPa/γ₁≈2000, C₂≈8 GPa/γ₂≈200, C₃≈1 GPa/γ₃≈10 for a steel — capture the sharp initial transient, the mid-range curvature, and the long linear tail simultaneously.

The trade-off is calibration cost: you need reversed/cyclic test data (tension–compression or bauschinger-loop specimens per ASTM E606 low-cycle fatigue protocol) to fit C and γ. Fitting only monotonic data and switching on kinematic hardening 'for free' gives you the wrong X and worse predictions than honest isotropic.

Where it bites: springback, forming, and residual stress

Any process that bends metal past yield and then relaxes or reverses is governed by the Bauschinger effect. The outer fiber of a bend is in tension, the inner in compression; on unloading, both fibers try to recover elastically, and the reverse-loading softness changes how much moment is stored — i.e., the springback angle.

• Sheet-metal bending / stamping. Advanced high-strength steels (DP600, DP980) and 6xxx/7xxx aluminum have strong Bauschinger effects. An isotropic model can mispredict springback by 3°–8° on a 90° flange — enough to blow a ±0.5 mm assembly tolerance on an automotive B-pillar. Production FE (LS-DYNA, AutoForm, PAM-STAMP) uses Yoshida–Uemori two-surface hardening precisely to nail the reversal on the draw-then-flange path.
• UOE and roll-formed pipe. A large-diameter line pipe (e.g., API 5L X70, 1067 mm / 42 in OD) is U-formed, O-formed, welded, then mechanically expanded. The reverse plastic passes lower the compressive yield strength of the pipe by 5–15% versus the parent plate — critical because collapse of deepwater pipe is a compressive-buckling limit state (DNV-OS-F101 explicitly accounts for the Bauschinger-degraded compressive yield).
• Autofrettage of gun barrels and high-pressure vessels. The bore is overpressurized to yield the inner wall in tension, planting beneficial compressive residual stress. On release, the reverse-loaded inner wall can yield in compression at a lower stress — so ignoring Bauschinger over-predicts the residual compression by 20–40% and the fatigue benefit with it.

Sizing the numbers: how much softening, and how to measure it

To quantify the effect on a real material you run a Bauschinger test: load a specimen to a chosen pre-strain εᵖ in tension, unload, then reload in compression, recording the reverse yield. Because thin sheet buckles in compression, this is done in plane bending, cyclic shear, or tension–compression on stubby specimens with anti-buckling supports.

• Step 1 — forward flow. Strain to εᵖ (say 2–5%); record σ_f. Larger pre-strain generally deepens the effect (more stored structure) up to a saturation.
• Step 2 — define reverse yield. Because the reversal is rounded, pick a consistent offset (0.2% proof, or a fixed 0.05% offset for finer resolution). Record σ_r.
• Step 3 — extract parameters. Back stress X = (σ_f − σ_r)/2; Bauschinger factor β = σ_r/σ_f; a Bauschinger energy parameter compares the area under the reversed curve to the isotropic prediction.

Order-of-magnitude anchors for a cold-worked steel: σ_y0 ≈ 300 MPa, forward flow at 3% strain σ_f ≈ 450 MPa, reverse yield σ_r ≈ 200 MPa → X ≈ 125 MPa, β ≈ 0.44. The permanent softening (the offset of the reversed curve from the monotonic one at large reverse strain) is a separate, smaller quantity — often 20–50 MPa — that the Chaboche third back-stress term captures. Note the modulus story too: on reversal the apparent chord (unloading) modulus is measurably lower than E — 10–20% down for AHSS — an additional 'modulus degradation' effect that Yoshida–Uemori bundles with the Bauschinger transient for springback accuracy.

Failure modes, limits, and best practice

The Bauschinger effect is neither good nor bad — it is a fact you must account for on the correct side of the ledger. It hurts when reverse-loading strength matters and helps when reverse-load ductility or residual stress matters.

• Low-cycle fatigue. Under fully reversed strain control (R = −1), every cycle traverses a Bauschinger reversal. The stabilized hysteresis loop is fatter than an isotropic model predicts, meaning more plastic strain energy per cycle and shorter life. A Coffin–Manson life estimate built on the wrong loop shape can be non-conservative by 2×. Fit the cyclic curve, not the monotonic one (ASTM E606 / E2714 for creep-fatigue).
• Deepwater pipe collapse. The reduced compressive yield after cold expansion lowers the collapse pressure. Best practice: measure post-forming compressive yield, or apply the code knock-down; never use the parent-plate tensile yield for a compression limit state.
• Over-crediting autofrettage. Assuming the bore stays elastic on depressurization over-states beneficial compressive residual stress and the fatigue win. Use an elastic–plastic (kinematic) unload.
• Erasing it unintentionally. A stress-relief or recrystallization anneal (steel 550–700 °C, aluminum 300–400 °C) recovers the stored dislocation structure and largely resets β toward 1. Good if you want isotropic behavior; bad if you were counting on planted residual stress.

Bottom line for the analyst: if a part sees any strain reversal — cyclic load, a forming pass followed by service tension the other way, or an unload from plasticity — default to a combined isotropic–kinematic (Chaboche) model calibrated on reversed test data. Pure isotropic hardening is only defensible for monotonic, proportional loading, and even then it lies about the springback.

Isotropic vs kinematic hardening: how each treats a reverse-loading yield surface after forward plastic strain

Feature	Isotropic hardening	Kinematic hardening	Reality (Bauschinger)
Yield surface	Expands uniformly	Translates (shifts center)	Translates + mild distortion
Reverse yield σ_r	= forward flow σ_f (raised)	σ_f − 2X (lowered)	Lowered, β·σ_f, β≈0.4–0.9
Elastic reverse range	2σ_f	2σ_y0 (unchanged span)	Reduced, often < 2σ_y0
Back stress X	None (X = 0)	Linear (Prager) or nonlinear	Nonlinear, saturates
Best for	Monotonic proportional loads	Cyclic / reverse loading	Combined Chaboche model