A fresh shape-memory alloy shows the one-way effect only: cool it and nothing visibly happens (twinned martensite forms as a random mix of self-accommodating variants that cancel out macroscopically), and it only reveals a shape change if you deform it while cold and then reheat it.
Repeatedly cycling the alloy between austenite and martensite while it is mechanically biased (by a load, or by internal constraint) leaves behind an oriented population of dislocations and residual stress fields. That oriented defect structure biases which martensite variants form on the next cooling — no external load required. This is training, and the result is the two-way shape-memory effect (TWSME): the part spontaneously curls on cooling and spontaneously straightens on heating, cycle after cycle, on its own.
Training strain: ε_TW(N) = ε_max · (1 − e^(−N/N₀))
Curvature: κ(T) = κ_max · ε_TW(N) · (1 − T̄)
N = completed training cycles T̄ ∈ [0,1] = smoothed temperature
N₀ = training rate constant (1 = fully austenite, 0 = fully martensite)
- Heat / Cool — manually swing the temperature at any time to test how much two-way memory the strip currently has, without adding more training.
- Run Training Cycle — takes the strip through one full heat→cool round trip; each completed round trip nudges the oriented defect population further, following a saturating exponential (real SMAs typically saturate after a few dozen cycles).
- Erase Training (Anneal) — models a high-temperature anneal that removes the retained dislocation structure, resetting the alloy back to plain one-way behaviour.
- The small tilted plates along the strip represent oriented martensite variants — they only appear once training and cold temperature coincide, visualising the very texture responsible for TWSME.
Real-world relevance: TWSME is what lets NiTi and Cu-based SMA parts act as reusable, no-bias-spring self-actuating hinges, couplers and grippers that flex and recover on their own as they cycle with ambient or resistive heating.