Timing Decomposition and Strategy Trade-Offs in Contrast Detection Autofocus Under Platform Capability Constraints

Contrast detection autofocus (CDAF) performance in industrial machine vision is shaped by platform capability as well as by the focus measure and search strategy. CDAF is analyzed through a platform capability framework and a unified frame-level transaction chain across three platforms: a capability upper-bound platform (P1), a bridging platform (P2), and an industrial black-box platform (P3). In experiments covering six scene categories, four initial conditions, five fixed-rule strategies, and 30 repetitions per condition, the dominant observable tail on P3 is localized after control submission, in the command-to-actuation segment. On P2, controlled one-factor perturbations using a physically calibrated sample position mismatch intensity (σ_align) and an actuation chain variability coordinate (λ_act) reproduce the main P3 degradation directions, providing a mechanism-level account in terms of sample position mismatch and command-to-actuation variability. Platform capability sets the reachable performance boundary, within which strategies trade speed, final quality, and failure risk. On P3, S1-S4 form the main engineering trade-off band, whereas S5 shows condition-dependent upper-quantile quality gains without a stable frontier advantage. The resulting deployment logic combines capability tiering, segment-wise bottleneck localization, and strategy band selection and treats CDAF as a capability-conditioned speed-quality-risk trade-off rather than a platform-independent strategy ranking.