| Abstract Scope |
Cold rolling pass schedules for aluminum coils are traditionally set from experience and validated templates, yet the mechanical properties obtained depend on the reductions actually achieved on the mill. This study uses the production data of a multi-alloy cold rolling line — 8,984 coils in 8,675 lots across 28 individual alloys in all series — linked pass by pass to cast chemistry and certified tensile strength, yield strength and elongation. Models from multiple linear regression to gradient-boosted trees are fitted per alloy–temper group; on AA5754-H22, for example, tree ensembles cut the prediction error against the group median by up to 44 %, and feature importance follows the metallurgical temper rather than the alloy. An inverse roll-gap analysis of 58,462 logged passes identifies plant-specific flow curves and an effective friction coefficient at no additional instrumentation cost. A set-versus-achieved gauge audit shows tight and highly repeatable control: first-pass targets are adapted to the incoming material and every later pass is re-planned onto the gauge actually realized — the 95th-percentile deviation collapses from 154 to 24 µm, with no detectable effect on the certified mechanical properties. Coupled to a physics-informed pass schedule generator with force screening and a property gate, the chain recommends feasible pass schedules for a target temper and gauge, extending schedule optimization to industrial, multi-alloy aluminum practice. |